Selecting a comparison group for 5-year oral and pharyngeal cancer survivors: Two methods

MISSING IMAGE

Material Information

Title:
Selecting a comparison group for 5-year oral and pharyngeal cancer survivors: Two methods
Physical Description:
Mixed Material
Language:
English
Creator:
Logan, Henrietta L.
Tomar, Scott L.
Chang, Myron
Turner, Glenn E.
Mendenhall, William M.
Riggs, Charles E. Jr.
Publisher:
BioMed Central
Publication Date:

Notes

General Note:
doi:10.1186/1471-2288-12-63 Cite this article as: Logan et al.: Selecting a comparison group for 5-year oral and pharyngeal cancer survivors: Two methods. BMC Medical Research Methodology 2012 12:63.
General Note:
Logan et al. BMC Medical Research Methodology 2012, 12:63; Pgs.1-8 http://www.biomedcentral.com/1471-2288/12/63
General Note:
Publication of this article was funded in part by the University of Florida Open-Access publishing Fund. In addition, requestors receiving funding through the UFOAP project are expected to submit a post-review, final draft of the article to UF's institutional repository, IR@UF, (www.uflib.ufl.edu/UFir) at the time of funding. The Institutional Repository at the University of Florida community, with research, news, outreach, and educational materials.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
All rights reserved by the source institution.
System ID:
AA00013223:00001


This item is only available as the following downloads:


Full Text


Logan et al. BMC Medical Research Methodology 2012, 12:63
http://www.biomedcentral.com/1471-2288/12/63


BM
Medical Research Methodology


Selecting a comparison group for 5-year oral and

pharyngeal cancer survivors: Two methods

Henrietta L Logan*, Scott L Tomarl, Myron Chang2, Glenn E Turner3, William M Mendenhall4 and
Charles E Riggs Jrs


Abstract
Background: To assess potential long-term consequences of cancer treatment, studies that include comparison
groups are needed. These comparison groups should be selected in a way that allows the subtle long-range effects
of cancer therapy to be detected and distinguishes them from the effects of aging and other risk factors. The
purpose of this investigation was to test two methods of recruiting a comparison group for 5-year oral and
pharyngeal cancer survivors (peer-nominated and listed sample) with emphasis on feasibility and the quality of the
match.
Methods: Participants were drawn from a pool of 5-year survivors treated at a large Southeastern hospital. A peer
nominated sample was solicited from the survivors. A listed sample matched on sex, age, and zip code was
purchased. Telephone interviews were conducted by a professional call center.
Results: The following represent our key findings: The quality of matching between survivors and listed sample was
better than that between survivors and peer-nominated group in age and sex. The quality of matching between
the two methods on other key variables did not differ except for education, with the peer method providing a
better match for the survivors than the listed sample. The yield for the listed sample method was greater than for
the peer-nominated method. The cost per completed interview was greater for the peer-nominated method than
the listed sample.
Conclusion: This study not only documents the methodological challenges in selecting a comparison group for
studies examining the late effects of cancer treatment among older individuals but also documents challenges in
matching groups that potentially have disproportionate levels of comorbidities and at-risk health behaviors.


Background
Individuals are surviving cancer for longer time periods
[1,2]. However, survivors face treatment protocols that
may produce late effects that negatively affect quality of
life [3-5]. Some effects appear transient and end with treat-
ment [6], but many may persist for months and even years
[3,7-9]; others develop later [10,11]. The greatest collection
of work on long-term treatment effects is related to child-
hood cancers but evidence on adult-onset malignancies is
accumulating (e.g., [3,4,12-18]). Few cancer treatments are
free of risks and most survivors face long-term adverse
consequences of treatment [3]. Despite advances in our
current cancer survivorship research, there are still serious

SCorrespondence hlogan@dental ufl edu
1Department of Community Dentistry and Behavioral Science, University of
Florida, 1329 SW 16th Street Room 5174, Gainesville, FL 32610-3628, USA
Full list of author information is available at the end of the article


gaps in our knowledge of late effects, especially in under-
studied cancers of older individuals [19].
Investigation of older cancer survivors to identify and
document the prevalence of these late and long-term
effects requires careful attention to the selection of an
appropriate comparison group [20]. For instance, as
people age, medical comorbidities become more com-
mon. Among individuals with high lifetime exposure to
alcohol and tobacco, the proportion developing comor-
bidities, including cardiovascular disease, is believed to
be greater than in the general population [21]. In study-
ing the rate at which late effects from cancer treatment
occur, including conditions such as cardiovascular dis-
eases or secondary tumors, researchers must be careful
that the effect of the cancer treatment is not being con-
founded with the effects of aging or "at-risk" health
behaviors.


0 2012 Logan et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative
SBiologi Central Commons Attribution License (http//creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.







Logan et al. BMC Medical Research Methodology 2012, 12:63
http://www.biomedcentral.com/1471-2288/12/63




The study of late effects of treatment among oral and
pharyngeal cancer (OPC) survivors presents challenges
similar to those of other cancers in older individuals
[22]. Many survivors of OPC face disfiguring surgery,
damage to oral function, and increased acute and late
toxicities resulting from more aggressive multimodal
treatment regimens [23]. In addition, these cancers dis-
proportionately affects older ethnoracially diverse men
for whom accurate incidence data on specific negative
health outcomes may not be available [24]; there are also
no normative data for frequently used assessment scales
[25]. In the past, few studies of OPC survivors included
a control group matching the target group on relevant
lifestyle factors, such as tobacco and alcohol use [20].
This limitation in otherwise informative studies makes it
problematic to draw conclusions about the long-term
effects of cancer therapy [25-28].
Several methods to produce comparison data have been
proposed, ranging from community-based to within-subject
designs [29]. One major problem with population-based
comparison groups is that tobacco and alcohol lifetime
exposure may be lower than those of cancer survivors. This
is a particular problem in the case of OPC because tobacco
and alcohol use are putative risk factors and independently
produce negative health outcomes. Therefore, matching on
tobacco use and alcohol use is desirable if the effects of
treatment are to be separated from those associated with
these "at-risk" behaviors [30].
A peer-matched strategy is believed to offer advantages
[31-33]. Compared to the case-survivors, friends may use
the healthcare system in similar ways and may be of
similar socioeconomic background and lifestyle [34].
Tobacco and alcohol use may also be similar. Older sur-
vivors, however, may not be able to nominate a peer,
especially when they have experienced debilitating late
effects and have become socially isolated [35]. Concern
is often raised that using friend-nominated controls may
result in overmatching, a phenomenon in which match-
ing occurs on an intermediate variable in a causal path-
way resulting in bias [36]. When the risk factors for the
disease are age and lifestyle-related and we seek to exam-
ine the consequences of cancer treatment, it may be ne-
cessary to homogenize the group [37,38].
The purpose of this investigation was to compare and
contrast two methods of recruiting a comparison group
for five-year survivors of OPC-a peer-nominated con-
trol group and a listed sample-on quality of the match,
costs, and feasibility. Match was defined as similarity in
sex, race, age, smoking and alcohol use, employment,
and education level [39].

Methods
A professional call center conducted a 20-minute tele-
phone interview with all participants in this project. The


Page 2 of 8


overall goal of the interview was to assess pain levels of
a matched non-cancer comparison group to 5-year sur-
vivors of head and neck cancer and to test predictive
models of oral pain among 5-year survivors. The survey
and methodology received prior approval from the
Institutional Review Board at the University of Florida.
Items were drawn from published instruments and are
reported elsewhere [40]. Participants received a $25 gift
card for completing the survey.
A target of 100 survivors, 100 friends, and 100 from
the listed sample had been set in advance. Best practices
(to meet the target numbers) were used by the profes-
sional call center to contact participants for each of the
three groups, cancer survivor group, peer-nominated
comparison group, and listed sample comparison group.


Selecting survivors
Survivors were drawn from 378 individuals treated for
OPC at the institution's radiation oncology clinic and who
had survived five years (plus or minus three months). After
verification of status and contact information, the 356
remaining individuals were sent a letter describing the
study. (See Figure 1.) A toll-free number was provided for
those who did not wish to participate in the study, and
seven individuals requested not to be called for the survey.


378 5-years survivors alive at last
contact to which letters were mailed
with toll-free telephone number*






20 refusals (no reason given)

83 no response or repeated requests
to call back
102 bad telephone numbers (fax
line, disconnected, non-working)

14 refused because mentally or
physically ill

6 deceased

30 non-eligible respondents






100 completed telephone
surveys


7 refused *

6 bad addresses

Phone list of 365 remaining
survivors generated















4 partial completes

6 telephone numbers not called
because target number reached


Figure 1 Flowchart of survivor recruitment.







Logan et a. BMC Medical Research Methodology 2012, 12:63
http://www.biomedcentral.com/1471-2288/12/63




The subsequent list was turned over to the call center who
conducted the survey.


Selecting peer-nominated group
Individuals from the survivor group were contacted first
and were asked to provide the name and telephone num-
ber of two peers who were similar to themselves, who
had not had cancer, and who might participate. Peers
were screened to exclude respondents with a history of
cancer. (See Figure 2)


Selecting listed sample methodology
A commercial list was purchased that matched the survi-
vors by age (within five years), sex, and geographical loca-
tion. Ten people of the same sex and age (+/- five years)
were selected from each survivor's zip code. Numbers
were chosen randomly from that list until one individual


91 peers nominated by 61 of
the 5-years survivors


Page 3 of 8


completed the survey. Each participant was screened to
exclude those with a history of cancer. (See Figure 3)



Data analysis
Descriptive statistics were calculated as percentages,
means, and standard deviations. To calculate the cost in-
curred, we divided the base cost of $25/hour, which
included both fixed and variable costs for the survey, by
the number of surveys completed per hour for each
group respectively. This base cost may have differed re-
gionally but provided general guidance for cost compari-
son. That amount was multiplied by the number of
completed surveys for each method, giving a measure of
total costs per group.
We evaluated the matching properties of the samples
onsex, race, age, education, employment status, and
cigarette and alcohol use. Alcohol use was derived from
these questions: "How often do you drink?" "How much
do you drink when you drink?" and "How often do you
drink at least six drinks in one day?" Duration of smok-
ing was calculated as the difference between age at the
first cigarette and age at last cigarette. For current smo-
kers, current age was used as age at last cigarette.


98 strata (with 10 telephone
numbers per strata) were matched
by gender, age, and zip code to the
5-years survivors.




117 refusals (no reason given)

188 no response or repeated
requests to call back

40 bad telephone numbers (fax line,
disconnected, non-working)

16 refused because mentally or
physically ill

18 reported eligible person
deceased

60 non-eligible respondents


101 completed telephone surveys


List of 980 telephone numbers
generated. Phone calls were made
in each stratum until a completed
survey was achieved.


4 partial completes

436 telephone numbers not
called when completed survey
achieved in stratum.


Figure 2 Flowchart of peer nominee recruitment.


Figure 3 Flowchart of listed sample recruitment.







Logan et al. BMC Medical Research Methodology 2012, 12:63
http://www.biomedcentral.com/1471-2288/12/63




Two analytic strategies were used. First, we tested for
marginal group (ii .. n. .. We then tested for the qual-
ity of the match at an individual pair level.
To test for marginal group i,1 i. 1 .. in each variable,
we used two-sided p-values for testing Ho: "the marginal
' i.:,, 1. of survivors is the same as that of peers" vs.
Ha: "the marginal distributions are Il i ...i '- nn, ,
we used two-sided p-values for testing Ho: "the marginal
distribution of survivors is the same as that of listed
sample" vs. Ha: "the two marginal i,. ..' *,,' .1, are differ-
ent"; the smaller the p-value, the larger the difference be-
tween the two marginal, i i..i .,! ....
To compare the quality of match between the survivor
and the peer to that between the survivor and the listed
sample, we calculated the absolute l*,:; '' between
each pair on each variable. We used a Wilcoxon rank
sum test to determine whether the two groups of differ-
ences were different. One-sided p-values were reported for
testing Ho: "the '," ,lih of matching between peers and
survivors is the same as that between listed sample and
survivors" vs. Ha: "the quality of matching between peers
and survivors is better than that between listed sample
and survivors." If p-value < 0.05, then we would claim that
the quality of matching between peers and survivors
was better than that between listed sample and survivors;
if p-value > 0.95, then vice versa. If p-value < 0.5, then
there was a trend that the quality of matching between
peers and survivors was better than that between listed
sample and survivors; if p-value > 0.5, then vice versa.
We used the Wilcoxon rank sum test to compare the
locations of two :.Li.,. ...... Because the data con-
tained paired and unpaired observations, the assumption
of independence between observations did not hold. The
method proposed by I,.Il I...I and colleagues [41] was
adapted for correlated data. P-values were obtained from
testing the group difference by the permutation method
described in Dallas and Rao [42].
We compared both marginal group differences and ab-
solute iil,. i .!.. between pairs in order to provide a
comprehensive picture to evaluate each method. For in-
stance, we expected the match on age and sex to be ex-
cellent for the listed sample and survivors because the
telephone screening ... 1,,I this match. If the quality of
the match was similar for the peer group and survivors
on these variables, then we could conclude that either
approach was adequate. In contrast, we might expect al-
cohol and tobacco use to be more similar between the
individuals in the peer group and the survivors than be-
tween those matched in the listed sample. The compari-
son of two marginal distributions may, however, be
misleading. For example, the paired data (1,10), (10,1),
(1,10), (10,1) show no ii.i. .... between the two mar-
ginal .li..i-.I...i,. (means equal 5.5), but the quality of
match at the individual level is quite poor.


Page 4 of 8


Results
Characteristics
Table 1 shows subject characteristics divided by selection
method. Table 2 shows the means and Table 3 shows the
p-values for matching variables (including sex, age, edu-
cation, employment, smoking and alcohol history).


Recruitment and retention
Figures 1, 2, and 3 provide flowcharts of the recruitment
patterns for each of the three groups. The interviews
were conducted over a 6-month time period and the
same interviewers were used for each group.
Figure 1 shows the disposition of the recruitment. Eli-
gible survivors were based on those alive at last contact
from the Department of Radiation Oncology and con-
ii .. *. through the Tumor Registry Shands at University
of Florida. In response to the letter that was mailed,
seven survivors called the toll-free number requesting
not to be called. After multiple attempts to obtain a cor-
rected phone number, six names had to be dropped from
the list. When the calls were made by the professional
call center, 20 individuals refused to participate and 83
either '" i- .,.*1; requested we call back or the call was
never answered. Six survivors had died and 14 were ei-
ther too mentally or i;.' -..1:*!. compromised to partici-
pate in the survey. Four i' Ill, completed the survey
but were never .I.-.1,- to finish the o,, 1I I III .. All

Table 1 Subject characteristics by selection methods
Characteristic Survivor Peer Listed Sample
N=100 N=44 N=101
Sex of subject


43% 71%
57% 29%


Womer
ice
White


Other
Education
Less than HS
High School grad
or equivalent
Post HS education
Employment Status
Currently employed
Retired
Other


94.0% 96.0% 93.0%
2.0% 2.0% 4.0%
4.0% 2.0% 3.0%


7.0% 6.8% 10.9%


21.3% 25.0% 21.8%
71.6% 68.2% 67.4%


32.0% 50.0% 43.0%


68.0% 47.0%
3.0%


Disabled 19.0%
Married 72.0%
*Prevalence of current smoking 15.0%
*Proportion who have smoked in the last 30 days.


4.8%
77.3%
27.0%


57.0%


10.3%
79.2%
10.0%







Logan et al. BMC Medical Research Methodology 2012, 12:63
http://www.biomedcentral.com/1471-2288/12/63




Table 2 Means for matching variables


Group
Variable
Age (yrs)"
Days of Smok


Survivor

100


the past 30


Lifetime duration of smoking
How often do you drink alcohol?
(during month)
How much alcohol do you drink on
a typical day when you are drinking?
(number of drinks)
How often do you drink six drinks of
alcohol in one day? (during month)


Sample Size"
Peer

44
44
27
44

42


34


Listed

96
96
53
95

95


95


Survivor

64.90 (10.09)
4.25 (10.3)
36.60 (15.0)
1.50 (1.63)

0.66 (0.70)


0.78 (0.95)


Mean (S.D.)
Peer

57.40 (13.9)
6.29 (11.7)
30.40(14.2)
1.50 (1.30)

0.98 (0.87)


1.03 (1.31)


Page 5 of 8


Listed

64.4 (10.08)
2.54 (8.13)
28.90 (15.1)
1.64 (1.64)

0.77 (0.74)


0.82 (0.82)


*: Sample sizes vary due to missing data.
**: Median ages (range) in survivor group, peer group, and listed group were 65.5 (33-90), 59.5 (29-80), and 64.0 (32-90), respectively.


but six telephone numbers from the original 378 five-
year survivors were used to complete the 100 surveys
(100/372=- "
F.'..'. 2 shows data n.1.. recruitment of the peer-
nominated group. Of the 100 survivors who completed
the questionnaires, 61 recommended at least one peer:
thirty-one survivors provided one peer and 30 survivors
nominated two peers, yielding a total of 91 nominees.
Because we had fallen short of our target of 100 peer
nominees, multiple attempts were made to contact all of
the nominees; 12 refused and six telephone numbers
were non-working. Fourteen nominees reported that
they were not l,,i.l, either because they had a history
of cancer or were outside the age range. Not shown in


the figure, ...1 i, ..... analysis showed that the percentage
of men and women survivors who nominated a peer
were roughly the same. Seventy-two percent of the men
were able to nominate at least one peer, as did 79% of
the women survivors. Of those peers nominated, we
were able to contact 38% of the men nominees compared
to 64% of the women. The observed success rate for con-
tacting the peer-nominated group was 44/91 = 48.3%.
The yield for the listed sample as shown in Figure 3
was 101/544= 19%. Out of the original 980 telephone
numbers purchased, all but 436 were used to complete
the sample. One hundred seventeen refused with no rea-
son given, and another 188 either i i-, *. asked the
interviewer to call back or there was no response after


Table 3 P-values for matching variables


Group
Variable
Sex of subject'
Age (yrs)"
Education'
Employment'
Days of smokir


Survivor


the past 30


Lifetime duration of smoking
How often do you drink alcohol?
(during month)
How much alcohol do you drink
on a typical day when you are
drinking? (number of drinks)
How often do you drink six drinks
of alcohol in one day? (during month)
*: sample sizes vary due to missing data.


Sample Size"
Peer

44
44
44
44
44
27
44

42


34


Listed

96
96
96
96
96
53
95

95


95


Survivor
vs. Peer#
0.47
0.017
0.75
0.18
0.07
0.32
0.77

0.40


0.51


P-Value
Survivor
vs. Listed'
1.00
0.79
0.41
0.25
0.23
0.011
0.44

0.26


0.42


Quality of
Matching^
0.99
1.00
0.018
0.28
0.77
0.18
0.23

0.53


0.25


#: Two-sided p-values for testing Ho: "the marginal distribution of survivors is the same as that of peers" vs. Ha: "the two marginal distributions are different." The
smaller the p-value, the larger the difference between the two marginal distributions.
$: Two-sided p-values for testing Ho: "the marginal distribution of survivors is the same as that of listed sample" vs. Ha: "the two marginal distributions are
different." The smaller the p-value, the larger the difference between the two marginal distributions.
AA: One-sided p-values for testing Ho: "the quality of matching between peers and survivors is the same as that between listed samples and survivors" vs. Ha: "the
quality of matching between peers and survivors is better than that between listed samples and survivors."







Logan et al. BMC Medical Research Methodology 2012, 12:63
http://www.biomedcentral.com/1471-2288/12/63



multiple attempts. Analysis showed that to complete a
sample of women required 4.6 of 10 numbers to be used;
men required 5.9. The average of telephone numbers
from the strata used to reach the group under 65 years
was 5.6 and for those 65 years or older was 5.5.

Cost of completed calls
The maximum number of telephone call attempts for
the survivor, peer nominee, and listed sample individual
were 11, 19, and 15, respectively. The rate of completed
calls per hour ranged from a high of .65 for the survivors
to a rate of .40 for the peer-nominated group (See
Table 4). For each completed call in the survivor group,
the cost, which included the f I....' charged by the
call center, was $38.46, compared to $41.66 for the listed
sample and $62.50 for the peer-nominated sample.

Matching
Table 2 shows the means, standard deviations, and
Table 3 shows the p-values for the comparison between
the survivor group versus the peer-nominated and listed
sample groups, respectively. Column five in Table 3,
titled Survivor vs. Peer, shows the two-sided p-values for
testing whether the marginal distribution of the variable
in column 1 for survivors is the same as or different than
that for the peer-nominated sample. Column six in
Table 3, titled Survivor vs. Listed, shows the two-sided
p-value for testing whether the marginal .1 i!.,,... of
the variable in column 1 for survivors is the same as or
(lif, .... r than that for the listed sample. For both columns
five and six, the smaller the given p-value, the larger the
ill ii .* is between the two marginal .-1 n d-I '.. The
final column in Table 3 shows the one-sided p-value for
the test of the overall quality of match. The p-values were
obtained by testing the null hypothesis, "the quality of
matching between peers and survivors is the same as that
between listed samples and survivors" vs. the alternative
hypothesis, "the quality of matching between peers and
survivors is better than that between listed samples and
survivors."
The peers smoked more in the 30 days prior to the
survey (average smoking days 6.29) compared to both the
survivors and the listed sample. The I in past 30-
day smoking between survivors and the peer group yielded
a two-sided p-value of .07, whereas the two-sided p-value


Page 6 of 8


for the survivors and the listed sample was .23. Table 3
shows, however, that the quality of the match at the indi-
vidual pair level did not differ on days of smoking in the
past 30 days (p = 0.77). That is, we could not reject the null
hypothesis. In terms of lifetime duration of smoking, there
was a trend toward the peer-nominated group yielding a
better match with the survivors than the listed sample
(p = 0.18). The survivors had significantly greater duration
on smoking than the listed sample (36.60 vs. 28.90,
p =0 ii'i. whereas survivors and peers '1I vs. 30.40,
p 0.32) were more similar. No differences approached
!i.m, i.... for alcohol use.
Overall, we observed that the ,p ,ilr, of the matching
was not significantly different between the two methods
except on age, sex, and education level. Specifically, at the
individual pair level, the listed sample produced a better
match on age and sex than the peer-nominated group
(p=0.99 and 1.00, respectively). Age (64.90 vs. 64.40)
(Table 2) and sex ,[.*i,!.... (72% men vs. 71% men)
(Table 2) were nearly identical for the survivors and listed
sample, respectively. The peer-nominated group produced
a better match on education than the listed sample (e.g.,
6.8% less than high school for peers vs. 7.0% less than high
school for survivors) (Table 1). Thus, as we show in
Table 3, we reject the null hypothesis (p = 0.018).



Conclusions
The key i:li.-.. from this study are (1) The quality of
matching between the two methods of choosing a com-
parison group for five-year cancer survivors did not dif-
fer except for education; the peer method provided a
-;i-;_;f...:.], better match for the survivors than the
listed sample, and the listed sample, as expected, pro-
duced a match on age and sex based on the criteria used
to purchase the list; (2) The yield for the listed sample
method was greater than for the peer-matching method;
(3) The cost per completed interview was greater for the
peer-matching method than the listed sample. Previous
studies have documented important changes in patients'
ability to function :..II. ... cancer treatment but the
goal of survivorship research has expanded to include
!.....i.-, I .. i._ the more subtle effects of treatment [25].
To accomplish this goal, longer term studies with appro-
priate comparison groups are needed [43].


Table 4 Cost per selection method


Recruitment
methodology
Total costs per group

Completes per hour
Costs per completed
telephone interview


Survivor
N=100
$3884.46

0.65
$25/0.65- $38.46


Peer
N=44
$2750.00

0.40
$25/0.40- $62.50


Listed Sample
N=101
$4207.66 (includes purchase of
telephone list in hourly rate)
0.60
$25/0.60- $41.66







Logan et al. BMC Medical Research Methodology 2012, 12:63
http://www.biomedcentral.com/1471-2288/12/63




Our results suggest that both methods of selecting a
comparison group yield similar matches to the survivor
group with the exception of education level. The out-
come of our study is somewhat :itt r.... I from that
reported by Kaplan et al. [35]. They found that peer
nominees were better educated than the target group
and --...' 1 patients from the same catchment area. We
found that the peer group was less educated. This is an
issue that may affect some survivor studies, especially
where socioeconomic status is a consideration. We
found a trend toward a better match on lifetime duration
of smoking for the peer-nominated and survivor groups
that may interest some investigators. Smoking exposure
is known to carry its own health risks, and when the
cancer survivors and controls are not matched on this
variable, it may be problematic to separate the effects of
cancer treatment from those of smoking exposure.
Kaplan found [35] that older patients were less likely
to nominate a peer for the comparison group than
younger patients. Our < ilii .. was not in the nomin-
ation of a peer but in our ability to contact that nominee,
which ::: I the yield of participants and the cost per
completion of the survey. Reducing the delay by telephon-
ing the peers immediately after their number is obtained
might reduce the amount of nonworking numbers. A sec-
ond alternative could be to mail a letter to the nominee
asking them to verify their telephone number. These
methods would allow immediate corrective actions to any
bias in men and women survivors having correct contact
information. Our success with the toll-free number for the
survivors to call for information suggests that we should
offer one to nominees in the future. In short, i ':::: the
methodology may provide a closer match on sex for the
peer-nominated and survivor group while also lowering
the cost per completion. Overall, 91 peers were nomi-
nated, but after 19 attempts, we were only able to reach 44
peers. Of those 44 peer respondents, we noted that 19 of
the women peers were nominated by women survivors,
while 6 were nominated by men survivors. Our instruc-
tions to the survivors did not specify that the nominees be
the same sex. Visual inspection of the data suggests that
the telephone numbers of the male nominees were less ac-
curate or less likely to be answered after multiple attempts
than for the women nominees. This added to the cost per
completion and may have contributed to the poorer match
on sex.
The results of the test for match quality showed the
two methods yielded similar results. This must be con-
sidered in light of the greater cost per successful comple-
tion for the peer-nominated method and its low yield in
this study. We maintain that our experience is important
for any researcher hoping to investigate the late effects
of cancer or cancer treatment among older adults [25].
We found that the survivors nominated peers, but the


Page 7 of 8


.!I:,!*!,: was in contacting those nominees. Readers
should be aware that the listed sample was purchased for
$715, and this was a part of the variable cost figured into
the survey center's hourly rate. Mailings to the initial 378
survivors added to our recruitment costs, but are not
included in the survey center's hourly rate.
In summary, we tested the methods using both a test of
the marginal distributions by group and the .la. i .... at
the individual pairs in an attempt to provide methodo-
logical evidence for comparison group selection. Our
results indicate that I... ; ini., of the late effects of
cancer treatment should give careful consideration to the
methods for comparison group selection. We believe that
many of the lessons learned from this study generalize to
comparison group selection for the study of other adult-
cancer survivors.


Conflict of interest
This research was partially funded by NC grant CA111593 and
1U54DE019261-01 awarded to H Loga (P) The authors have
interest with the findings From th:s research


Authors' contributions
HL was responsible for the design and execution


closely wth th
assisted in the
closely with ot
assisted in the
oversaw the ar
and he worked
proofing of thi:
eligibility, and
and proofing c


DCR grant
o conflict of


)f survey, and worked


ter contributors in the composition of ths manuscript ST
exception, design, and execution of the study, and worked
contributors in the composition of this manuscript MC
:eptua: design, provided the data analysis plan, and
SGT was responsible for managing participant eligibility,
sey with other contributors in the composition and
nuscript WM was responsible for managing participant
worked closely with other contributors in the composition
s manuscript CR was responsible for managing


participant eligibility, and he worked closely with other contributors in the
composition and proofing of this manuscript Al authors read and approved
the final manuscript

Acknowledgements
This project was partially funded by NCI grant CA111593 and NIDCR grant
1U54D0E19261-01 Pubhcation of th:s art:de was funded in part by the
University of F oda Open-Access Publishing Fund

Author details
'Department of Community Dentistry and Behavioral Science, University of
Florida, 1329 SW 16th Street Room 5174, Gainesville, FL 32610-3628, USA
2Department of Biostatistics, University of Florida, Gainesville, FL, USA
Department of Prosthodontics, University of Florida, Gainesville, FL, USA
4Department of Radiation Oncology, University of Florida, Gainesville, FL, USA
Department of Medicine, University of Florida, Gainesville, FL, USA

Received: 8 August 2011 Accepted: 16 April 2012
Published: 2 May 2012

References
1 Cancer survivorship research I I
2 Cancer facts and figures 2006 : .. .. /docroot/S T/content/
STT1 xC ncer_Facs _Fiiures _2 00asp
3 Alfano CM, Rowland JH Recovery issues in cancer survivorship: A new
challenge for supportive care. Cancer i 2006, 12(51)432-443
4 Wingo PA, Howe HI, Thun Ma BallardBarbash R Ward E, Brown ML, Sylvester J,
Friedell GH, Alley L Rowland JH, et al A national framework for cancer
surveillance in the United States. Cancer Causs Control 2005, 16(2)151-170
5 Givens DJ, Karnell l H Gup;a AK, Clamon GH, Pagedar NA, Chang KE, an
Daele DJ, Funk GF Adverse events associated with concurrent
chemoradiation therapy in patients with head and neck cancer. Archives
of otolacrvnoloav-heajd & neck suro er 2009 135(12) 1209-1217








Logan et a. BMC Medical Research Methodology 2012, 12:63
http://www.biomedcentral.com/1471-2288/12/63


6 Clegg LX, Li FP, Hankey BF, Chu K, Edwards BK Cancer Survival Among US
Whites and Minorities: A SEER (Surveillance, Epidemiology, and End
Results) Program Population-Based Study. Arch Intern Med 2002,
162(17)1985-1993
7 Pollack LA, Greer GE, Rowland JH, Miller A, Doneski D, Coughlin 55, 5tovall E,
Ulman D Cancer survivorship: a new challenge in comprehensive cancer
control. Cancer Causes Control 2005, 16(Suppl 1)'51-59
8 Jung BF, Herrmann D, Griggs J, Oaklander AL, Dworkin RH Neuropathic
pain associated with non-surgical treatment of breast cancer. Pain 2005,
118(1-2)'10-14
9 Aziz NM Cancer survivorship research: challenge and opportunity. J Nutr
2002, 132(11)'34945-35035
10 Curtis R, Freedman D, Ron E, Ries L, Hacker D, Edwards B, Tucker M,
Faumeni JJ New Malignancies Among Cancer Survivors SEER Cancer
Registries, 1973-2000, vol No 05-5302 NIH Publication Bethesda, MD; 2006
11 5enkus-Konefka E, Jassem J Complications of breast-cancer radiotherapy.
hin Oncol 2006, 18(3)229-235
12 Butler RW, Haser JK Neurocognitive effects of treatment for childhood
cancer. Merit Retard Dev Disabii Res Rev 2006, 12(3)184-191
13 Peterson DE, Keefe DM, Hutchins RD, Schubert MM Alimentary tract
mucositis in cancer patients: impact of terminology and assessment on
research and clinical practice. Support Care Cancer 2006, 14(6)499-504
14 Jordan VC Improvements in tumor targeting, survivorship, and
chemoprevention pioneered by tamoxifen A personal perspective.
Oncology New York 2006, 20(6)'553-562
15 Lockwood-Rayermann 5 Survivorship issues in ovarian cancer: A review.
Oncol Nurs Forum 2006, 33(3)'553-562
16 5ugimura H, Yang P Long-term survivorship in lung cancer A review.
Chest 2006, 129(4) 1088-1097
17 Van Poznak C, 5auter NP Clinical management of osteoporosis in women
with a history of breast carcinoma. Cancer 2005, 104(3)443-456
18 Wefel J5, 5aleeba AK Buzdar AU, Meyers CA Acute and late onset
cognitive dysfunction associated with chemotherapy in women with
breast cancer. Cancer, 116(14) 3348-3356
19 Rowland JH, Hewitt M, Ganz PA Cancer survivorship: a new challenge in
delivering quality cancer care. J Cin Oncol 2006, 24(32)'5101-5104
20 5tolley PD, 5chlesselman JJ Planning and conducting a study In In Case
Control Studies Edited by 5chlesselman JJ New York Oxford University
Press; 198276-85
21 2004 surgeon general's report links more cancers to smoking. CA Cancer
J Cin 2004, 54(5) 243-244 doi 10 3322/canjclin 545 243
22 Feuerstein M Defining cancer survivorship. Journal of Cancer Survivorship
2007, 1(1)5-7
23 Murphy BA, Gilbert J, Ridner 5H Systemic and global toxicities of head
and neck treatment. Expert Rev Anticancer Ther 2007, 7(7)'1043-1053
24 Aziz NM, Rowland JH Cancer survivorship research among ethnic
minority and medically underserved groups. Oncol Nurs Forum 2002,
29(5)789-801
25 Bond 5M' Dietch MS Murphy BA Neurocognitive function in head and
neck cancer patients prior to treatment Supportive Care in Cancer; 2011
26 Abendstein H, Nordgren M, Boysen M, Jannert M, 5ilander E, Ahlner-Elmqvist
M, Hammerlid E, Bjordal K Quality of life and head and neck cancer:
A 5 year prospective study. Laryngoscope 2005, 115(12) 2183-2192
27 Extermann M Measurement and impact of comorbidity in older cancer
patients. Crt Rev Oncol Hematol 2000, 35(3) 181-200
28 Nordgren M, Abendstein H, Jannert M, Boysen M, Ahlner-Elmqvist M,
5ilander E, Bjordal K, Hammerlid E Health-related quality of life five years
after diagnosis of laryngeal carcinoma. Int J Radiat Oncol Biol Phys 2003,
56(5)'1333-1343
29 Wacholder 5, 5ilverman DT, McLaughlin JK, Mandel J' Selection of controls
in case-control studies II. Types of controls. 1041 1992, 135(9) 1029
30 Ribeiro KC, Kowalski LP, Latorre MR Impact of comorbidity, symptoms, and
patients' characteristics on the prognosis of oral carcinomas. Archives of
otolaryngology-head & neck surgery 2000, 126(9)'1079-1085
31 Shaw GL, Tucker MA, Kase RG, Hoover RN Problems ascertaining friend
controls in a case-control study of lung-cancer. Am J Epidemiol 1991,
133(1) 63-66
32 Zhou W, Liu G, Miller DP, Thurston SW, Xu LL, Wain JC, Lynch TJ, 5u L,
Christiani DC Gene-environment interaction for the ERCC2
polymorphisms and cumulative cigarette smoking exposure in lung
cancer. Cancer Res 2002, 62(5)'1377-1381


33 Zhou W, Liu G, Thurston SW, Xu LL, Miller DP, Wain JC, Lynch TJ, Su L,
Christiani DC Genetic polymorphisms in N-acetyltransferase-2 and
microsomal epoxide hydrolase, cumulative cigarette smoking, and lung
cancer. Cancer Epidemiol Biomarkers Prev 2002, 11(1) 15-21
34 Tchen N, Bedard P, Yi QL, Klein M, Cella D, Eremenco 5, Tannock IF Quality
of life and understanding of disease status among cancer patients of
different ethnic origin. Br J Cancer 2003, 89(4)'641-647
35 Kaplan 5, Novikov I, Modan B A methodolgical note on the selection of
friends as controls. Int J Epidemiol 1998, 27(4)727-729
36 Rothman KJ, Greenland Matching In Modern Epidemiology Volume 1, 2nd
edn Edited by Rothman KJ, Greenland S Philadelphia Lippincott Williams &
William; 1998 147-161
37 Breslow NE, Day NE The need for the construction of special comparison
groups In Statistical Methods in Cancer Research edn Lyon International
Agency for Research on Cancer; 1987 32-46
38 Earle CC Long term planning for cancer survivors: a health services
research agenda. Journal of Cancer Survivorship 2007, 1(1) 64-74
39 Silverman S Demographics and occurrence of oral and pharyngeal
cancers The outcomes, the trends, the challenge. J Am Dent Assoc 2001,
132:75-115
40 Logan HL, Bartoshuk LM, Fillingim RB, Tomar SL, Mendenhall WM Metallic
taste phantom predicts oral pain among 5-year survivors of head and
neck cancer. Pain 2008, 140(2)'323-331
41 Hollander M, Pledger G, Lin P Robustness on the Wilcoxon Test to a
Certain Dependency between Samples. Ann Stat 1974, 2(1) 177-181
42 Dallas M, Rao P Testing Equality of Survival Functions Based on both
Paired and Unpaired Censored Data. Biometrics 2000, 56:154-159
43 Feuerstein M Optimizing cancer survivorship. Journal of Cancer
Survivorship Research and Practice 2007, 1(1) 1

doi:10.1186/1471-2288-12-63
Cite this article as: Logan et al Selecting a comparison group for 5-year
oral and pharyngeal cancer survivors: Two methods. BMC Medical
Research Methodology 2012 12'63


Page 8 of 8


Submit your next manuscript to BioMed Central
and take full advantage of:

* Convenient online submission
* Thorough peer review
* No space constraints or color figure charges
* Immediate publication on acceptance
* Inclusion in PubMed, CAS, Scopus and Google Scholar
* Research which is freely available for redistribution


Submit your manuscript at
www.biomedcentral.com/submit 0y) BIIleUd Central




Full Text
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EQKPSLCIN_ACHGDM INGEST_TIME 2013-01-22T15:14:29Z PACKAGE AA00013223_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES


!DOCTYPE art SYSTEM 'http:www.biomedcentral.comxmlarticle.dtd'
ui 1471-2288-12-63
ji 1471-2288
fm
dochead Research article
bibl
title
p Selecting a comparison group for 5-year oral and pharyngeal cancer survivors: Two methods
aug
au id A1 ca yes snm Loganmi Lfnm Henriettainsr iid I1 email hlogan@dental.ufl.edu
A2 TomarLScottstomar@dental.ufl.edu
A3 ChangMyronI2 mchang@cog.ufl.edu
A4 TurnerEGlennI3 gturner@dental.ufl.edu
A5 MendenhallMWilliamI4 mendwm@shands.ufl.edu
A6 RiggsECharlessuf JrI5 Charles.Riggs@medicine.ufl.edu
insg
ins Department of Community Dentistry and Behavioral Science, University of Florida, 1329 SW 16th Street Room 5174, Gainesville, FL, 32610-3628, USA
Department of Biostatistics, University of Florida, Gainesville, FL, USA
Department of Prosthodontics, University of Florida, Gainesville, FL, USA
Department of Radiation Oncology, University of Florida, Gainesville, FL, USA
Department of Medicine, University of Florida, Gainesville, FL, USA
source BMC Medical Research Methodology
issn 1471-2288
pubdate 2012
volume 12
issue 1
fpage 63
url http://www.biomedcentral.com/1471-2288/12/63
xrefbib pubidlist pubid idtype doi 10.1186/1471-2288-12-63pmpid 22551236
history rec date day 8month 8year 2011acc 1642012pub 252012
cpyrt 2012collab Logan et al; licensee BioMed Central Ltd.note This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
abs
sec
st
Abstract
Background
To assess potential long-term consequences of cancer treatment, studies that include comparison groups are needed. These comparison groups should be selected in a way that allows the subtle long-range effects of cancer therapy to be detected and distinguishes them from the effects of aging and other risk factors. The purpose of this investigation was to test two methods of recruiting a comparison group for 5-year oral and pharyngeal cancer survivors (peer-nominated and listed sample) with emphasis on feasibility and the quality of the match.
Methods
Participants were drawn from a pool of 5-year survivors treated at a large Southeastern hospital. A peer-nominated sample was solicited from the survivors. A listed sample matched on sex, age, and zip code was purchased. Telephone interviews were conducted by a professional call center.
Results
The following represent our key findings: The quality of matching between survivors and listed sample was better than that between survivors and peer-nominated group in age and sex. The quality of matching between the two methods on other key variables did not differ except for education, with the peer method providing a better match for the survivors than the listed sample. The yield for the listed sample method was greater than for the peer-nominated method. The cost per completed interview was greater for the peer-nominated method than the listed sample.
Conclusion
This study not only documents the methodological challenges in selecting a comparison group for studies examining the late effects of cancer treatment among older individuals but also documents challenges in matching groups that potentially have disproportionate levels of comorbidities and at-risk health behaviors.
bdy
Background
Individuals are surviving cancer for longer time periods abbrgrp
abbr bid B1 1
B2 2
. However, survivors face treatment protocols that may produce late effects that negatively affect quality of life
B3 3
B4 4
B5 5
. Some effects appear transient and end with treatment
B6 6
, but many may persist for months and even years
3
B7 7
B8 8
B9 9
; others develop later
B10 10
B11 11
. The greatest collection of work on long-term treatment effects is related to childhood cancers but evidence on adult-onset malignancies is accumulating (e.g.,
3
4
B12 12
B13 13
B14 14
B15 15
B16 16
B17 17
B18 18
). Few cancer treatments are free of risks and most survivors face long-term adverse consequences of treatment
3
. Despite advances in our current cancer survivorship research, there are still serious gaps in our knowledge of late effects, especially in understudied cancers of older individuals
B19 19
.Investigation of older cancer survivors to identify and document the prevalence of these late and long-term effects requires careful attention to the selection of an appropriate comparison group
B20 20
. For instance, as people age, medical comorbidities become more common. Among individuals with high lifetime exposure to alcohol and tobacco, the proportion developing comorbidities, including cardiovascular disease, is believed to be greater than in the general population
B21 21
. In studying the rate at which late effects from cancer treatment occur, including conditions such as cardiovascular diseases or secondary tumors, researchers must be careful that the effect of the cancer treatment is not being confounded with the effects of aging or “at-risk” health behaviors.The study of late effects of treatment among oral and pharyngeal cancer (OPC) survivors presents challenges similar to those of other cancers in older individuals
B22 22
. Many survivors of OPC face disfiguring surgery, damage to oral function, and increased acute and late toxicities resulting from more aggressive multimodal treatment regimens
B23 23
. In addition, these cancers disproportionately affects older ethnoracially diverse men for whom accurate incidence data on specific negative health outcomes may not be available
B24 24
; there are also no normative data for frequently used assessment scales
B25 25
. In the past, few studies of OPC survivors included a control group matching the target group on relevant lifestyle factors, such as tobacco and alcohol use
20
. This limitation in otherwise informative studies makes it problematic to draw conclusions about the long-term effects of cancer therapy
25
B26 26
B27 27
B28 28
.Several methods to produce comparison data have been proposed, ranging from community-based to within-subject designs
B29 29
. One major problem with population-based comparison groups is that tobacco and alcohol lifetime exposure may be lower than those of cancer survivors. This is a particular problem in the case of OPC because tobacco and alcohol use are putative risk factors and independently produce negative health outcomes. Therefore, matching on tobacco use and alcohol use is desirable if the effects of treatment are to be separated from those associated with these “at-risk” behaviors
B30 30
.A peer-matched strategy is believed to offer advantages
B31 31
B32 32
B33 33
. Compared to the case-survivors, friends may use the healthcare system in similar ways and may be of similar socioeconomic background and lifestyle
B34 34
. Tobacco and alcohol use may also be similar. Older survivors, however, may not be able to nominate a peer, especially when they have experienced debilitating late effects and have become socially isolated
B35 35
. Concern is often raised that using friend-nominated controls may result in overmatching, a phenomenon in which matching occurs on an intermediate variable in a causal pathway resulting in bias
B36 36
. When the risk factors for the disease are age and lifestyle-related and we seek to examine the consequences of cancer treatment, it may be necessary to homogenize the group
B37 37
B38 38
.The purpose of this investigation was to compare and contrast two methods of recruiting a comparison group for five-year survivors of OPC—a peer-nominated control group and a listed sample—on quality of the match, costs, and feasibility. Match was defined as similarity in sex, race, age, smoking and alcohol use, employment, and education level
B39 39
.
Methods
A professional call center conducted a 20-minute telephone interview with all participants in this project. The overall goal of the interview was to assess pain levels of a matched non-cancer comparison group to 5-year survivors of head and neck cancer and to test predictive models of oral pain among 5-year survivors. The survey and methodology received prior approval from the Institutional Review Board at the University of Florida. Items were drawn from published instruments and are reported elsewhere
B40 40
. Participants received a $25 gift card for completing the survey.A target of 100 survivors, 100 friends, and 100 from the listed sample had been set in advance. Best practices (to meet the target numbers) were used by the professional call center to contact participants for each of the three groups, cancer survivor group, peer-nominated comparison group, and listed sample comparison group.
Selecting survivors
Survivors were drawn from 378 individuals treated for OPC at the institution’s radiation oncology clinic and who had survived five years (plus or minus three months). After verification of status and contact information, the 356 remaining individuals were sent a letter describing the study. (See Figure figr fid F1 1.) A toll-free number was provided for those who did not wish to participate in the study, and seven individuals requested not to be called for the survey. The subsequent list was turned over to the call center who conducted the survey.
fig Figure 1caption Flowchart of survivor recruitmenttext
b Flowchart of survivor recruitment.
graphic file 1471-2288-12-63-1
Selecting peer-nominated group
Individuals from the survivor group were contacted first and were asked to provide the name and telephone number of two peers who were similar to themselves, who had not had cancer, and who might participate. Peers were screened to exclude respondents with a history of cancer. (See Figure F2 2)
Figure 2Flowchart of peer nominee recruitment
Flowchart of peer nominee recruitment.
1471-2288-12-63-2
Selecting listed sample methodology
A commercial list was purchased that matched the survivors by age (within five years), sex, and geographical location. Ten people of the same sex and age (+/− five years) were selected from each survivor’s zip code. Numbers were chosen randomly from that list until one individual completed the survey. Each participant was screened to exclude those with a history of cancer. (See Figure F3 3)
Figure 3Flowchart of listed sample recruitment
Flowchart of listed sample recruitment.
1471-2288-12-63-3
Data analysis
Descriptive statistics were calculated as percentages, means, and standard deviations. To calculate the cost incurred, we divided the base cost of $25/hour, which included both fixed and variable costs for the survey, by the number of surveys completed per hour for each group respectively. This base cost may have differed regionally but provided general guidance for cost comparison. That amount was multiplied by the number of completed surveys for each method, giving a measure of total costs per group.We evaluated the matching properties of the samples onsex, race, age, education, employment status, and cigarette and alcohol use. Alcohol use was derived from these questions: “How often do you drink?” “How much do you drink when you drink?” and “How often do you drink at least six drinks in one day?” Duration of smoking was calculated as the difference between age at the first cigarette and age at last cigarette. For current smokers, current age was used as age at last cigarette.Two analytic strategies were used. First, we tested for marginal group differences. We then tested for the quality of the match at an individual pair level.To test for marginal group differences in each variable, we used two-sided p-values for testing Hsub o: “the marginal distribution of survivors is the same as that of peers” vs. Ha: “the marginal distributions are different.” Similarly, we used two-sided p-values for testing Ho: “the marginal distribution of survivors is the same as that of listed sample” vs. Ha: “the two marginal distributions are different”; the smaller the p-value, the larger the difference between the two marginal distributions.To compare the quality of match between the survivor and the peer to that between the survivor and the listed sample, we calculated the absolute difference between each pair on each variable. We used a Wilcoxon rank sum test to determine whether the two groups of differences were different. One-sided p-values were reported for testing Ho: “the quality of matching between peers and survivors is the same as that between listed sample and survivors” vs. Ha: “the quality of matching between peers and survivors is better than that between listed sample and survivors.” If p-value < 0.05, then we would claim that the quality of matching between peers and survivors was better than that between listed sample and survivors; if p-value > 0.95, then vice versa. If p-value < 0.5, then there was a trend that the quality of matching between peers and survivors was better than that between listed sample and survivors; if p-value > 0.5, then vice versa.We used the Wilcoxon rank sum test to compare the locations of two distributions. Because the data contained paired and unpaired observations, the assumption of independence between observations did not hold. The method proposed by Hollander and colleagues
B41 41
was adapted for correlated data. P-values were obtained from testing the group difference by the permutation method described in Dallas and Rao
B42 42
.We compared both marginal group differences and absolute differences between pairs in order to provide a comprehensive picture to evaluate each method. For instance, we expected the match on age and sex to be excellent for the listed sample and survivors because the telephone screening specified this match. If the quality of the match was similar for the peer group and survivors on these variables, then we could conclude that either approach was adequate. In contrast, we might expect alcohol and tobacco use to be more similar between the individuals in the peer group and the survivors than between those matched in the listed sample. The comparison of two marginal distributions may, however, be misleading. For example, the paired data (1,10), (10,1), (1,10), (10,1) show no difference between the two marginal distributions (means equal 5.5), but the quality of match at the individual level is quite poor.
Results
Characteristics
Table tblr tid T1 1 shows subject characteristics divided by selection method. Table T2 2 shows the means and Table T3 3 shows the p-values for matching variables (including sex, age, education, employment, smoking and alcohol history).
table
Table 1
Subject characteristics by selection methods
tgroup align left cols 4
colspec colname c1 colnum 1 colwidth 1*
c2 2
c3 3
c4
thead valign top
row rowsep
entry
Characteristic
Survivor N = 100
Peer N = 44
Listed Sample N = 101
tfoot
*Proportion who have smoked in the last 30 days.
tbody
nameend namest
Sex of subject
  Men
72%
43%
71%
  Women
28%
57%
29%
Race
  White
94.0%
96.0%
93.0%
  Black
2.0%
2.0%
4.0%
  Other
4.0%
2.0%
3.0%
Education
  Less than HS
7.0%
6.8%
10.9%
  High School grad
  or equivalent
21.3%
25.0%
21.8%
  Post HS education
71.6%
68.2%
67.4%
Employment Status
  Currently employed
32.0%
50.0%
43.0%
  Retired
68.0%
47.0%
57.0%
  Other
-
3.0%
-
Disabled
19.0%
4.8%
10.3%
Married
72.0%
77.3%
79.2%
*Prevalence of current smoking
15.0%
27.0%
10.0%
Table 2
Means for matching variables
7
center
c5 5
c6 6
c7
Sample Sizesup *
Mean (S.D.)
Group
Survivor
Peer
Listed
Survivor
Peer
Listed
Variable
*: Sample sizes vary due to missing data.**: Median ages (range) in survivor group, peer group, and listed group were 65.5 (33-90), 59.5 (29-80), and 64.0 (32-90), respectively.
Age (yrs)**
100
44
96
64.90 (10.09)
57.40 (13.9)
64.4 (10.08)
Days of Smoking in the past 30
100
44
96
4.25 (10.3)
6.29 (11.7)
2.54 (8.13)
Lifetime duration of smoking
60
27
53
36.60 (15.0)
30.40(14.2)
28.90 (15.1)
How often do you drink alcohol? (during month)
100
44
95
1.50 (1.63)
1.50 (1.30)
1.64 (1.64)
How much alcohol do you drink on a typical day when you are drinking? (number of drinks)
99
42
95
0.66 (0.70)
0.98 (0.87)
0.77 (0.74)
How often do you drink six drinks of alcohol in one day? (during month)
100
34
95
0.78 (0.95)
1.03 (1.31)
0.82 (0.82)
Table 3
P-values for matching variables
Sample Size*
P-Value
Group
Survivor
Peer
Listed
Survivor vs. Peer#
Survivor vs. Listed$
Quality of Matching^^
Variable
*: sample sizes vary due to missing data.#: Two-sided p-values for testing Ho: “the marginal distribution of survivors is the same as that of peers” vs. Ha: ”the two marginal distributions are different”. The smaller the p-value, the larger the difference between the two marginal distributions.$: Two-sided p-values for testing Ho: ”the marginal distribution of survivors is the same as that of listed sample” vs. Ha: ”the two marginal distributions are different”. The smaller the p-value, the larger the difference between the two marginal distributions.^^: One-sided p-values for testing Ho: ”the quality of matching between peers and survivors is the same as that between listed samples and survivors” vs. Ha: ”the quality of matching between peers and survivors is better than that between listed samples and survivors”.
Sex of subject!
100
44
96
0.47
1.00
0.99
Age (yrs)!!
100
44
96
0.017
0.79
1.00
Education!
99
44
96
0.75
0.41
0.018
Employment!
100
44
96
0.18
0.25
0.28
Days of smoking in the past 30
100
44
96
0.07
0.23
0.77
Lifetime duration of smoking
60
27
53
0.32
0.011
0.18
How often do you drink alcohol? (during month)
100
44
95
0.77
0.44
0.23
How much alcohol do you drink on a typical day when you are drinking? (number of drinks)
99
42
95
0.40
0.26
0.53
How often do you drink six drinks of alcohol in one day? (during month)
100
34
95
0.51
0.42
0.25
Recruitment and retention
Figures 1, 2, and 3 provide flowcharts of the recruitment patterns for each of the three groups. The interviews were conducted over a 6-month time period and the same interviewers were used for each group.Figure 1 shows the disposition of the recruitment. Eligible survivors were based on those alive at last contact from the Department of Radiation Oncology and confirmed through the Tumor Registry Shands at University of Florida. In response to the letter that was mailed, seven survivors called the toll-free number requesting not to be called. After multiple attempts to obtain a corrected phone number, six names had to be dropped from the list. When the calls were made by the professional call center, 20 individuals refused to participate and 83 either repeatedly requested we call back or the call was never answered. Six survivors had died and 14 were either too mentally or physically compromised to participate in the survey. Four partially completed the survey but were never available to finish the questionnaire. All but six telephone numbers from the original 378 five-year survivors were used to complete the 100 surveys (100/372 = 27%).Figure 2 shows data regarding recruitment of the peer-nominated group. Of the 100 survivors who completed the questionnaires, 61 recommended at least one peer: thirty-one survivors provided one peer and 30 survivors nominated two peers, yielding a total of 91 nominees. Because we had fallen short of our target of 100 peer nominees, multiple attempts were made to contact all of the nominees; 12 refused and six telephone numbers were non-working. Fourteen nominees reported that they were not eligible, either because they had a history of cancer or were outside the age range. Not shown in the figure, additional analysis showed that the percentage of men and women survivors who nominated a peer were roughly the same. Seventy-two percent of the men were able to nominate at least one peer, as did 79% of the women survivors. Of those peers nominated, we were able to contact 38% of the men nominees compared to 64% of the women. The observed success rate for contacting the peer-nominated group was 44/91 = 48.3%.The yield for the listed sample as shown in Figure 3 was 101/544 = 19%. Out of the original 980 telephone numbers purchased, all but 436 were used to complete the sample. One hundred seventeen refused with no reason given, and another 188 either repeatedly asked the interviewer to call back or there was no response after multiple attempts. Analysis showed that to complete a sample of women required 4.6 of 10 numbers to be used; men required 5.9. The average of telephone numbers from the strata used to reach the group under 65 years was 5.6 and for those 65 years or older was 5.5.
Cost of completed calls
The maximum number of telephone call attempts for the survivor, peer nominee, and listed sample individual were 11, 19, and 15, respectively. The rate of completed calls per hour ranged from a high of .65 for the survivors to a rate of .40 for the peer-nominated group (See Table T4 4). For each completed call in the survivor group, the cost, which included the $25/hour charged by the call center, was $38.46, compared to $41.66 for the listed sample and $62.50 for the peer-nominated sample.
Table 4
Cost per selection method
Recruitment methodology
Survivor N = 100
Peer N = 44
Listed Sample N = 101
Total costs per group
$3884.46
$2750.00
$4207.66 (includes purchase of telephone list in hourly rate)
Completes per hour
0.65
0.40
0.60
Costs per completed telephone interview
$25/0.65 = $38.46
$25/0.40 = $62.50
$25/0.60 = $41.66
Matching
Table 2 shows the means, standard deviations, and Table 3 shows the p-values for the comparison between the survivor group versus the peer-nominated and listed sample groups, respectively. Column five in Table 3, titled Survivor vs. Peer, shows the two-sided p-values for testing whether the marginal distribution of the variable in column 1 for survivors is the same as or different than that for the peer-nominated sample. Column six in Table 3, titled Survivor vs. Listed, shows the two-sided p-value for testing whether the marginal distribution of the variable in column 1 for survivors is the same as or different than that for the listed sample. For both columns five and six, the smaller the given p-value, the larger the difference is between the two marginal distributions. The final column in Table 3 shows the one-sided p-value for the test of the overall quality of match. The p-values were obtained by testing the null hypothesis, “the quality of matching between peers and survivors is the same as that between listed samples and survivors” vs. the alternative hypothesis, “the quality of matching between peers and survivors is better than that between listed samples and survivors.”The peers smoked more in the 30 days prior to the survey (average smoking days 6.29) compared to both the survivors and the listed sample. The difference in past 30-day smoking between survivors and the peer group yielded a two-sided p-value of .07, whereas the two-sided p-value for the survivors and the listed sample was .23. Table 3 shows, however, that the quality of the match at the individual pair level did not differ on days of smoking in the past 30 days (p = 0.77). That is, we could not reject the null hypothesis. In terms of lifetime duration of smoking, there was a trend toward the peer-nominated group yielding a better match with the survivors than the listed sample (p = 0.18). The survivors had significantly greater duration on smoking than the listed sample (36.60 vs. 28.90, p = 0.011), whereas survivors and peers (36.60 vs. 30.40, p = 0.32) were more similar. No differences approached significance for alcohol use.Overall, we observed that the quality of the matching was not significantly different between the two methods except on age, sex, and education level. Specifically, at the individual pair level, the listed sample produced a better match on age and sex than the peer-nominated group (p = 0.99 and 1.00, respectively). Age (64.90 vs. 64.40) (Table 2) and sex distribution (72% men vs. 71% men) (Table 2) were nearly identical for the survivors and listed sample, respectively. The peer-nominated group produced a better match on education than the listed sample (e.g., 6.8% less than high school for peers vs. 7.0% less than high school for survivors) (Table 1). Thus, as we show in Table 3, we reject the null hypothesis (p = 0.018).
Conclusions
The key findings from this study are (1) The quality of matching between the two methods of choosing a comparison group for five-year cancer survivors did not differ except for education; the peer method provided a significantly better match for the survivors than the listed sample, and the listed sample, as expected, produced a match on age and sex based on the criteria used to purchase the list; (2) The yield for the listed sample method was greater than for the peer-matching method; (3) The cost per completed interview was greater for the peer-matching method than the listed sample. Previous studies have documented important changes in patients’ ability to function following cancer treatment but the goal of survivorship research has expanded to include understanding the more subtle effects of treatment
25
. To accomplish this goal, longer term studies with appropriate comparison groups are needed
B43 43
.Our results suggest that both methods of selecting a comparison group yield similar matches to the survivor group with the exception of education level. The outcome of our study is somewhat different from that reported by Kaplan et al.
35
. They found that peer nominees were better educated than the target group and hospital patients from the same catchment area. We found that the peer group was less educated. This is an issue that may affect some survivor studies, especially where socioeconomic status is a consideration. We found a trend toward a better match on lifetime duration of smoking for the peer-nominated and survivor groups that may interest some investigators. Smoking exposure is known to carry its own health risks, and when the cancer survivors and controls are ul not matched on this variable, it may be problematic to separate the effects of cancer treatment from those of smoking exposure.Kaplan found
35
that older patients were less likely to nominate a peer for the comparison group than younger patients. Our challenge was not in the nomination of a peer but in our ability to contact that nominee, which affected the yield of participants and the cost per completion of the survey. Reducing the delay by telephoning the peers immediately after their number is obtained might reduce the amount of nonworking numbers. A second alternative could be to mail a letter to the nominee asking them to verify their telephone number. These methods would allow immediate corrective actions to any bias in men and women survivors having correct contact information. Our success with the toll-free number for the survivors to call for information suggests that we should offer one to nominees in the future. In short, refining the methodology may provide a closer match on sex for the peer-nominated and survivor group while also lowering the cost per completion. Overall, 91 peers were nominated, but after 19 attempts, we were only able to reach 44 peers. Of those 44 peer respondents, we noted that 19 of the women peers were nominated by women survivors, while 6 were nominated by men survivors. Our instructions to the survivors did not specify that the nominees be the same sex. Visual inspection of the data suggests that the telephone numbers of the male nominees were less accurate or less likely to be answered after multiple attempts than for the women nominees. This added to the cost per completion and may have contributed to the poorer match on sex.The results of the test for match quality showed the two methods yielded similar results. This must be considered in light of the greater cost per successful completion for the peer-nominated method and its low yield in this study. We maintain that our experience is important for any researcher hoping to investigate the late effects of cancer or cancer treatment among older adults
25
. We found that the survivors nominated peers, but the challenge was in contacting those nominees. Readers should be aware that the listed sample was purchased for $715, and this was a part of the variable cost figured into the survey center’s hourly rate. Mailings to the initial 378 survivors added to our recruitment costs, but are not included in the survey center’s hourly rate.In summary, we tested the methods using both a test of the marginal distributions by group and the differences at the individual pairs in an attempt to provide methodological evidence for comparison group selection. Our results indicate that investigators of the late effects of cancer treatment should give careful consideration to the methods for comparison group selection. We believe that many of the lessons learned from this study generalize to comparison group selection for the study of other adult-cancer survivors.
Conflict of interest
This research was partially funded by NCI grant CA111593 and NIDCR grant 1U54DEO19261-01 awarded to H. Logan (PI). The authors have no conflict of interest with the findings from this research.
Authors’ contributions
HL was responsible for the design and execution of survey, and worked closely with the other contributors in the composition of this manuscript. ST assisted in the conception, design, and execution of the study, and worked closely with other contributors in the composition of this manuscript. MC assisted in the conceptual design, provided the data analysis plan, and oversaw the analysis. GT was responsible for managing participant eligibility, and he worked closely with other contributors in the composition and proofing of this manuscript. WM was responsible for managing participant eligibility, and he worked closely with other contributors in the composition and proofing of this manuscript. CR was responsible for managing participant eligibility, and he worked closely with other contributors in the composition and proofing of this manuscript. All authors read and approved the final manuscript.
bm
ack
Acknowledgements
This project was partially funded by NCI grant CA111593 and NIDCR grant 1U54DEO19261-01. Publication of this article was funded in part by the University of Florida Open-Access Publishing Fund.
refgrp Cancer survivorship research
http://dccps.nci.nih.gov/ocs/index.html
Cancer facts and figures2006
http://www.cancer.org/docroot/STT/content/STT_1x_Cancer_Facts__Figures_2006.asp
Recovery issues in cancer survivorship: A new challenge for supportive careAlfanoCMRowlandJHCancer J2006125432lpage 44310.1097/00130404-200609000-0001217034679A national framework for cancer surveillance in the United StatesWingoPAHoweHLThunMJBallard-BarbashRWardEBrownMLSylvesterJFriedellGHAlleyLRowlandJHetal Cancer Causes Control200516215117010.1007/s10552-004-3487-5link fulltext 15868456Adverse events associated with concurrent chemoradiation therapy in patients with head and neck cancerGivensDJKarnellLHGuptaAKClamonGHPagedarNAChangKEVan DaeleDJFunkGFArchives of otolaryngology--head & neck surgery2009135121209121710.1001/archoto.2009.17422710503Cancer Survival Among US Whites and Minorities: A SEER (Surveillance, Epidemiology, and End Results) Program Population-Based StudyCleggLXLiFPHankeyBFChuKEdwardsBKArch Intern Med2002162171985199310.1001/archinte.162.17.198512230422Cancer survivorship: a new challenge in comprehensive cancer controlPollackLAGreerGERowlandJHMillerADoneskiDCoughlinSSStovallEUlmanDCancer Causes Control200516Suppl 1515916208574Neuropathic pain associated with non-surgical treatment of breast cancerJungBFHerrmannDGriggsJOaklanderALDworkinRHPain20051181–2101416213086Cancer survivorship research: challenge and opportunityAzizNMJ Nutr2002132113494S3503S12421876CurtisRFreedmanDRonERiesLHackerDEdwardsBTuckerMFaumeniJJpublisher New Malignancies Among Cancer Survivors: SEER Cancer Registries, 1973–2000, vol. No. 05–5302: NIH Publication Bethesda, MD2006Complications of breast-cancer radiotherapySenkus-KonefkaEJassemJClin Oncol200618322923510.1016/j.clon.2005.11.004Neurocognitive effects of treatment for childhood cancerButlerRWHaserJKMent Retard Dev Disabil Res Rev200612318419110.1002/mrdd.2011017061287Alimentary tract mucositis in cancer patients: impact of terminology and assessment on research and clinical practicePetersonDEKeefeDMHutchinsRDSchubertMMSupport Care Cancer200614649950410.1007/s00520-006-0057-216775646Improvements in tumor targeting, survivorship, and chemoprevention pioneered by tamoxifen A personal perspectiveJordanVCOncology-New York2006206553562Survivorship issues in ovarian cancer: A reviewLockwood-RayermannSOncol Nurs Forum200633355356210.1188/06.ONF.553-56216676012Long-term survivorship in lung cancer A reviewSugimuraHYangPChest200612941088109710.1378/chest.129.4.108816608961Clinical management of osteoporosis in women with a history of breast carcinomaVan PoznakCSauterNPCancer2005104344345610.1002/cncr.2120115968687Acute and late onset cognitive dysfunction associated with chemotherapy in women with breast cancerWefelJSSaleebaAKBuzdarAUMeyersCACancer1161433483356Cancer survivorship: a new challenge in delivering quality cancer careRowlandJHHewittMGanzPAJ Clin Oncol200624325101510410.1200/JCO.2006.09.270017093270Planning and conducting a studyStolleyPDSchlesselmanJJIn: Case Control StudiesNew York: Oxford University Presseditor Schlesselman JJ198276852004 surgeon general's report links more cancers to smokingCA Cancer J Clin2004545243244Defining cancer survivorshipFeuersteinMJournal of Cancer Survivorship2007115710.1007/s11764-006-0002-x18648939Systemic and global toxicities of head and neck treatmentMurphyBAGilbertJRidnerSHExpert Rev Anticancer Ther2007771043105310.1586/14737140.7.7.104317627463Cancer survivorship research among ethnic minority and medically underserved groupsAzizNMRowlandJHOncol Nurs Forum200229578980110.1188/02.ONF.789-80112058154BondSMDietrich MSMurphy BA: Neurocognitive function in head and neck cancer patients prior to treatment. Supportive Care in Cancer2011Quality of life and head and neck cancer: A 5 year prospective studyAbendsteinHNordgrenMBoysenMJannertMSilanderEAhlner-ElmqvistMHammerlidEBjordalKLaryngoscope2005115122183219210.1097/01.MLG.0000181507.69620.1416369164Measurement and impact of comorbidity in older cancer patientsExtermannMCrit Rev Oncol Hematol200035318120010.1016/S1040-8428(00)00090-110960800Health-related quality of life five years after diagnosis of laryngeal carcinomaNordgrenMAbendsteinHJannertMBoysenMAhlner-ElmqvistMSilanderEBjordalKHammerlidEInt J Radiat Oncol Biol Phys20035651333134310.1016/S0360-3016(03)00292-X12873678Selection of controls in case–control studies II. Types of controlsWacholderSSilvermanDTMcLaughlinJKMandelJS1041199213591029Impact of comorbidity, symptoms, and patients' characteristics on the prognosis of oral carcinomasRibeiroKCKowalskiLPLatorreMRArchives of otolaryngology--head & neck surgery200012691079108522710503Problems ascertaining friend controls in a case–control study of lung-cancerShawGLTuckerMAKaseRGHooverRNAm J Epidemiol1991133163661845760Gene-environment interaction for the ERCC2 polymorphisms and cumulative cigarette smoking exposure in lung cancerZhouWLiuGMillerDPThurstonSWXuLLWainJCLynchTJSuLChristianiDCCancer Res20026251377138111888908Genetic polymorphisms in N-acetyltransferase-2 and microsomal epoxide hydrolase, cumulative cigarette smoking, and lung cancerZhouWLiuGThurstonSWXuLLMillerDPWainJCLynchTJSuLChristianiDCCancer Epidemiol Biomarkers Prev2002111152111815396Quality of life and understanding of disease status among cancer patients of different ethnic originTchenNBedardPYiQLKleinMCellaDEremencoSTannockIFBr J Cancer200389464164710.1038/sj.bjc.6601159pmcid 237691212915871A methodolgical note on the selection of friends as controlsKaplanSNovikovIModanBInt J Epidemiol199827472772910.1093/ije/27.4.7279758132MatchingRothmanKJGreenlandSModern EpidemiologyPhiladelphia: Lippincott Williams & WilliamRothman KJ, Greenland Sedition 1, 21998147161The need for the construction of special comparison groupsBreslowNEDayNEStatistical Methods in Cancer ResearchInternational Agency for Research on CancerLyon 19873246Long term planning for cancer survivors: a health services research agendaEarleCCJournal of Cancer Survivorship2007116410.1007/s11764-006-0003-918648946Demographics and occurrence of oral and pharyngeal cancers The outcomes, the trends, the challengeSilvermanSJ Am Dent Assoc20011327S11S11803655Metallic taste phantom predicts oral pain among 5-year survivors of head and neck cancerLoganHLBartoshukLMFillingimRBTomarSLMendenhallWMPain2008140232333110.1016/j.pain.2008.09.004264578618845396Robustness on the Wilcoxon Test to a Certain Dependency between SamplesHollanderMPledgerGLinPAnn Stat19742117718110.1214/aos/1176342623Testing Equality of Survival Functions Based on both Paired and Unpaired Censored DataDallasMRaoPBiometrics20005615415910.1111/j.0006-341X.2000.00154.x10783790Optimizing cancer survivorshipFeuersteinMJournal of Cancer Survivorship: Research and Practice200711110.1007/s11764-006-0001-y
Pre-publication historyThe pre-publication history for this paper can be accessed here:http://www.biomedcentral.com/1471-2288/12/63/prepub


xml version 1.0 encoding utf-8 standalone no
mets ID sort-mets_mets OBJID sword-mets LABEL DSpace SWORD Item PROFILE METS SIP Profile xmlns http:www.loc.govMETS
xmlns:xlink http:www.w3.org1999xlink xmlns:xsi http:www.w3.org2001XMLSchema-instance
xsi:schemaLocation http:www.loc.govstandardsmetsmets.xsd
metsHdr CREATEDATE 2012-10-08T12:00:48
agent ROLE CUSTODIAN TYPE ORGANIZATION
name BioMed Central
dmdSec sword-mets-dmd-1 GROUPID sword-mets-dmd-1_group-1
mdWrap SWAP Metadata MDTYPE OTHER OTHERMDTYPE EPDCX MIMETYPE textxml
xmlData
epdcx:descriptionSet xmlns:epdcx http:purl.orgeprintepdcx2006-11-16 xmlns:MIOJAVI
http:purl.orgeprintepdcxxsd2006-11-16epdcx.xsd
epdcx:description epdcx:resourceId sword-mets-epdcx-1
epdcx:statement epdcx:propertyURI http:purl.orgdcelements1.1type epdcx:valueURI http:purl.orgeprintentityTypeScholarlyWork
http:purl.orgdcelements1.1title
epdcx:valueString Selecting a comparison group for 5-year oral and
pharyngeal cancer survivors: Two methods
http:purl.orgdctermsabstract
Abstract
Background
To assess potential long-term consequences of cancer treatment, studies that include comparison groups are needed. These comparison groups should be selected in a way that allows the subtle long-range effects of cancer therapy to be detected and distinguishes them from the effects of aging and other risk factors. The purpose of this investigation was to test two methods of recruiting a comparison group for 5-year oral and pharyngeal cancer survivors (peer-nominated and listed sample) with emphasis on feasibility and the quality of the match.
Methods
Participants were drawn from a pool of 5-year survivors treated at a large Southeastern hospital. A peer-nominated sample was solicited from the survivors. A listed sample matched on sex, age, and zip code was purchased. Telephone interviews were conducted by a professional call center.
Results
The following represent our key findings: The quality of matching between survivors and listed sample was better than that between survivors and peer-nominated group in age and sex. The quality of matching between the two methods on other key variables did not differ except for education, with the peer method providing a better match for the survivors than the listed sample. The yield for the listed sample method was greater than for the peer-nominated method. The cost per completed interview was greater for the peer-nominated method than the listed sample.
Conclusion
This study not only documents the methodological challenges in selecting a comparison group for studies examining the late effects of cancer treatment among older individuals but also documents challenges in matching groups that potentially have disproportionate levels of comorbidities and at-risk health behaviors.
http:purl.orgdcelements1.1creator
Logan, Henrietta L
Tomar, Scott L
Chang, Myron
Turner, Glenn E
Mendenhall, William M
Riggs, Charles E Jr
http:purl.orgeprinttermsisExpressedAs epdcx:valueRef sword-mets-expr-1
http:purl.orgeprintentityTypeExpression
http:purl.orgdcelements1.1language epdcx:vesURI http:purl.orgdctermsRFC3066
en
http:purl.orgeprinttermsType
http:purl.orgeprinttypeJournalArticle
http:purl.orgdctermsavailable
epdcx:sesURI http:purl.orgdctermsW3CDTF 2012-05-02
http:purl.orgdcelements1.1publisher
BioMed Central Ltd
http:purl.orgeprinttermsstatus http:purl.orgeprinttermsStatus
http:purl.orgeprintstatusPeerReviewed
http:purl.orgeprinttermscopyrightHolder
Henrietta L Logan et al.; licensee BioMed Central Ltd.
http:purl.orgdctermslicense
http://creativecommons.org/licenses/by/2.0
http:purl.orgdctermsaccessRights http:purl.orgeprinttermsAccessRights
http:purl.orgeprintaccessRightsOpenAccess
http:purl.orgeprinttermsbibliographicCitation
BMC Medical Research Methodology. 2012 May 02;12(1):63
http:purl.orgdcelements1.1identifier
http:purl.orgdctermsURI http://dx.doi.org/10.1186/1471-2288-12-63
fileSec
fileGrp sword-mets-fgrp-1 USE CONTENT
file sword-mets-fgid-0 sword-mets-file-1
FLocat LOCTYPE URL xlink:href 1471-2288-12-63.xml
sword-mets-fgid-1 sword-mets-file-2 applicationpdf
1471-2288-12-63.pdf
structMap sword-mets-struct-1 structure LOGICAL
div sword-mets-div-1 DMDID Object
sword-mets-div-2 File
fptr FILEID
sword-mets-div-3



PAGE 1

RESEARCHARTICLEOpenAccessSelectingacomparisongroupfor5-yearoraland pharyngealcancersurvivors:TwomethodsHenriettaLLogan1*,ScottLTomar1,MyronChang2,GlennETurner3,WilliamMMendenhall4and CharlesERiggsJr5AbstractBackground: Toassesspotentiallong-termconsequencesofcancertreatment,studiesthatincludecomparison groupsareneeded.Thesecomparisongroupsshouldbeselectedinawaythatallowsthesubtlelong-rangeeffects ofcancertherapytobedetectedanddistinguishesthemfromtheeffectsofagingandotherriskfactors.The purposeofthisinvestigationwastotesttwomethodsofrecruitingacomparisongroupfor5-yearoraland pharyngealcancersurvivors(peer-nominatedandlistedsample)withemphasisonfeasibilityandthequalityofthe match. Methods: Participantsweredrawnfromapoolof5-yearsurvivorstreatedatalargeSoutheasternhospital.Apeernominatedsamplewassolicitedfromthesurvivors.Alistedsamplematchedonsex,age,andzipcodewas purchased.Telephoneinterviewswereconductedbyaprofessionalcallcenter. Results: Thefollowingrepresentourkeyfindings:Thequalityofmatchingbetweensurvivorsandlistedsamplewas betterthanthatbetweensurvivorsandpeer-nominatedgroupinageandsex.Thequalityofmatchingbetween thetwomethodsonotherkeyvariablesdidnotdifferexceptforeducation,withthepeermethodprovidinga bettermatchforthesurvivorsthanthelistedsample.Theyieldforthelistedsamplemethodwasgreaterthanfor thepeer-nominatedmethod.Thecostpercompletedinterviewwasgreaterforthepeer-nominatedmethodthan thelistedsample. Conclusion: Thisstudynotonlydocumentsthemethodologicalchallengesinselectingacomparisongroupfor studiesexaminingthelateeffectsofcancertreatmentamongolderindividualsbutalsodocumentschallengesin matchinggroupsthatpotentiallyhavedisproportionatelevelsofcomorbiditiesandat-riskhealthbehaviors.BackgroundIndividualsaresurvivingcancerforlongertimeperiods [1,2].However,survivorsfacetreatmentprotocolsthat mayproducelateeffectsthatnegativelyaffectqualityof life[3-5].Someeffectsappeartransientandendwithtreatment[6],butmanymaypersistformonthsandevenyears [3,7-9];othersdeveloplater[10,11].Thegreatestcollection ofworkonlong-termtreatmenteffectsisrelatedtochildhoodcancersbutevidenceonadult-onsetmalignanciesis accumulating(e.g.,[3,4,12-18]).Fewcancertreatmentsare freeofrisksandmostsurvivorsfacelong-termadverse consequencesoftreatment[3].Despiteadvancesinour currentcancersurvivorshipresearch,therearestillserious gapsinourknowledgeoflateeffects,especiallyinunderstudiedcancersofolderindividuals[19]. Investigationofoldercancersurvivorstoidentifyand documenttheprevalenceoftheselateandlong-term effectsrequirescarefulattentiontotheselectionofan appropriatecomparisongroup[20].Forinstance,as peopleage,medicalcomorbiditiesbecomemorecommon.Amongindividualswithhighlifetimeexposureto alcoholandtobacco,theproportiondevelopingcomorbidities,includingcardiovasculardisease,isbelievedto begreaterthaninthegeneralpopulation[21].Instudyingtherateatwhichlateeffectsfromcancertreatment occur,includingconditionssuchascardiovasculardiseasesorsecondarytumors,researchersmustbecareful thattheeffectofthecancertreatmentisnotbeingconfoundedwiththeeffectsofagingor “ at-risk ” health behaviors. *Correspondence: hlogan@dental.ufl.edu1DepartmentofCommunityDentistryandBehavioralScience,Universityof Florida,1329SW16thStreetRoom5174,Gainesville,FL32610-3628,USA Fulllistofauthorinformationisavailableattheendofthearticle 2012Loganetal;licenseeBioMedCentralLtd.ThisisanOpenAccessarticledistributedunderthetermsoftheCreative CommonsAttributionLicense(http://creativecommons.org/licenses/by/2.0),whichpermitsunrestricteduse,distribution,and reproductioninanymedium,providedtheoriginalworkisproperlycited.Logan etal.BMCMedicalResearchMethodology 2012, 12 :63 http://www.biomedcentral.com/1471-2288/12/63

PAGE 2

Thestudyoflateeffectsoftreatmentamongoraland pharyngealcancer(OPC)survivorspresentschallenges similartothoseofothercancersinolderindividuals [22].ManysurvivorsofOPCfacedisfiguringsurgery, damagetooralfunction,andincreasedacuteandlate toxicitiesresultingfrommoreaggressivemultimodal treatmentregimens[23].Inaddition,thesecancersdisproportionatelyaffectsolderethnoraciallydiversemen forwhomaccurateincidencedataonspecificnegative healthoutcomesmaynotbeavailable[24];therearealso nonormativedataforfrequentlyusedassessmentscales [25].Inthepast,fewstudiesofOPCsurvivorsincluded acontrolgroupmatchingthetargetgrouponrelevant lifestylefactors,suchastobaccoandalcoholuse[20]. Thislimitationinotherwiseinformativestudiesmakesit problematictodrawconclusionsaboutthelong-term effectsofcancertherapy[25-28]. Severalmethodstoproducecomparisondatahavebeen proposed,rangingfromcommunity-basedtowithin-subject designs[29].Onemajorproblemwithpopulation-based comparisongroupsisthattobaccoandalcohollifetime exposuremaybelowerthanthoseofcancersurvivors.This isaparticularprobleminthecaseofOPCbecausetobacco andalcoholuseareputativeriskfactorsandindependently producenegativehealthoutcomes.Therefore,matchingon tobaccouseandalcoholuseisdesirableiftheeffectsof treatmentaretobeseparatedfromthoseassociatedwith these “ at-risk ” behaviors[30]. Apeer-matchedstrategyisbelievedtoofferadvantages [31-33].Comparedtothecase-survivors,friendsmayuse thehealthcaresysteminsimilarwaysandmaybeof similarsocioeconomicbackgroundandlifestyle[34]. Tobaccoandalcoholusemayalsobesimilar.Oldersurvivors,however,maynotbeabletonominateapeer, especiallywhentheyhaveexperienceddebilitatinglate effectsandhavebecomesociallyisolated[35].Concern isoftenraisedthatusingfriend-nominatedcontrolsmay resultinovermatching,aphenomenoninwhichmatchingoccursonanintermediatevariableinacausalpathwayresultinginbias[36].Whentheriskfactorsforthe diseaseareageandlifestyle-relatedandweseektoexaminetheconsequencesofcancertreatment,itmaybenecessarytohomogenizethegroup[37,38]. Thepurposeofthisinvestigationwastocompareand contrasttwomethodsofrecruitingacomparisongroup forfive-yearsurvivorsofOPC — apeer-nominatedcontrolgroupandalistedsample — onqualityofthematch, costs,andfeasibility.Matchwasdefinedassimilarityin sex,race,age,smokingandalcoholuse,employment, andeducationlevel[39].MethodsAprofessionalcallcenterconducteda20-minutetelephoneinterviewwithallparticipantsinthisproject.The overallgoaloftheinterviewwastoassesspainlevelsof amatchednon-cancercomparisongroupto5-yearsurvivorsofheadandneckcancerandtotestpredictive modelsoforalpainamong5-yearsurvivors.Thesurvey andmethodologyreceivedpriorapprovalfromthe InstitutionalReviewBoardattheUniversityofFlorida. Itemsweredrawnfrompublishedinstrumentsandare reportedelsewhere[40].Participantsreceiveda$25gift cardforcompletingthesurvey. Atargetof100survivors,100friends,and100from thelistedsamplehadbeensetinadvance.Bestpractices (tomeetthetargetnumbers)wereusedbytheprofessionalcallcentertocontactparticipantsforeachofthe threegroups,cancersurvivorgroup,peer-nominated comparisongroup,andlistedsamplecomparisongroup.SelectingsurvivorsSurvivorsweredrawnfrom378individualstreatedfor OPCattheinstitution ’ sradiationoncologyclinicandwho hadsurvivedfiveyears(plusorminusthreemonths).After verificationofstatusandcontactinformation,the356 remainingindividualsweresentaletterdescribingthe study.(SeeFigure1.)Atoll-freenumberwasprovidedfor thosewhodidnotwishtoparticipateinthestudy,and sevenindividualsrequestednottobecalledforthesurvey. 378 5-years survivors alive at last contact to which letters were mailed with toll-free telephone number* 100 completed telephone surve y s 7 refused 20 refusals (no reason given) Phone list of 365 remaining survivorsgenerated 83 no response or repeated requests tocallback 6 bad addresses 102 bad telephone numbers (fax line, disconnected, non-working) 4 partial completes 14 refused because mentally or physically ill 6 deceased 30 non-eligible respondents 6 telephone numbers not called because target number reached Figure1 Flowchartofsurvivorrecruitment. Logan etal.BMCMedicalResearchMethodology 2012, 12 :63Page2of8 http://www.biomedcentral.com/1471-2288/12/63

PAGE 3

Thesubsequentlistwasturnedovertothecallcenterwho conductedthesurvey.Selectingpeer-nominatedgroupIndividualsfromthesurvivorgroupwerecontactedfirst andwereaskedtoprovidethenameandtelephonenumberoftwopeerswhoweresimilartothemselves,who hadnothadcancer,andwhomightparticipate.Peers werescreenedtoexcluderespondentswithahistoryof cancer.(SeeFigure2)SelectinglistedsamplemethodologyAcommerciallistwaspurchasedthatmatchedthesurvivorsbyage(withinfiveyears),sex,andgeographicallocation.Tenpeopleofthesamesexandage(+/ fiveyears) wereselectedfromeachsurvivor ’ szipcode.Numbers werechosenrandomlyfromthatlistuntiloneindividual completedthesurvey.Eachparticipantwasscreenedto excludethosewithahistoryofcancer.(SeeFigure3)DataanalysisDescriptivestatisticswerecalculatedaspercentages, means,andstandarddeviations.Tocalculatethecostincurred,wedividedthebasecostof$25/hour,which includedbothfixedandvariablecostsforthesurvey,by thenumberofsurveyscompletedperhourforeach grouprespectively.Thisbasecostmayhavedifferedregionallybutprovidedgeneralguidanceforcostcomparison.Thatamountwasmultipliedbythenumberof completedsurveysforeachmethod,givingameasureof totalcostspergroup. Weevaluatedthematchingpropertiesofthesamples onsex,race,age,education,employmentstatus,and cigaretteandalcoholuse.Alcoholusewasderivedfrom thesequestions: “ Howoftendoyoudrink? ”“ Howmuch doyoudrinkwhenyoudrink? ” and “ Howoftendoyou drinkatleastsixdrinksinoneday? ” Durationofsmokingwascalculatedasthedifferencebetweenageatthe firstcigaretteandageatlastcigarette.Forcurrentsmokers,currentagewasusedasageatlastcigarette. 91 peers nominated by 61 of the 5-years survivors 44 completed surveys 31 survivors nominated 1 peer 12 refusals 15 no response or repeated requests tocallback 30 survivors nominated 2 peers 6 bad telephone numbers (disconnected and non-working) 14 non-eligible respondents Figure2 Flowchartofpeernomineerecruitment. 98 strata (with 10 telephone numbers per strata) were matched by gender, age, and zip code to the 5-years survivors. 101 completed telephone surveys 117 refusals (no reason given) List of 980 telephone numbers generated. Phone calls were made in each stratum until a completed survey was achieved. 188 no response or repeated requests to call back 40 bad telephone numbers (fax line, disconnected, non-working) 4 partial completes 16 refused because mentally or physically ill 18 reported eligible person deceased 60 non-eligible respondents 436 telephone numbers not called when completed survey achieved in stratum. Figure3 Flowchartoflistedsamplerecruitment. Logan etal.BMCMedicalResearchMethodology 2012, 12 :63Page3of8 http://www.biomedcentral.com/1471-2288/12/63

PAGE 4

Twoanalyticstrategieswereused.First,wetestedfor marginalgroupdifferences.Wethentestedforthequalityofthematchatanindividualpairlevel. Totestformarginalgroupdifferencesineachvariable, weusedtwo-sidedp-valuesfortestingHo: “ themarginal distributionofsurvivorsisthesameasthatofpeers ” vs. Ha: “ themarginaldistributionsaredifferent. ” Similarly, weusedtwo-sidedp-valuesfortestingHo: “ themarginal distributionofsurvivorsisthesameasthatoflisted sample ” vs.Ha: “ thetwomarginaldistributionsaredifferent ” ;thesmallerthep-value,thelargerthedifferencebetweenthetwomarginaldistributions. Tocomparethequalityofmatchbetweenthesurvivor andthepeertothatbetweenthesurvivorandthelisted sample,wecalculatedtheabsolutedifferencebetween eachpaironeachvariable.WeusedaWilcoxonrank sumtesttodeterminewhetherthetwogroupsofdifferencesweredifferent.One-sidedp-valueswerereportedfor testingHo: “ thequalityofmatchingbetweenpeersand survivorsisthesameasthatbetweenlistedsampleand survivors ” vs.Ha: “ thequalityofmatchingbetweenpeers andsurvivorsisbetterthanthatbetweenlistedsample andsurvivors. ” Ifp-value < 0.05,thenwewouldclaimthat thequalityofmatchingbetweenpeersandsurvivors wasbetterthanthatbetweenlistedsampleandsurvivors; ifp-value > 0.95,thenviceversa.Ifp-value < 0.5,then therewasatrendthatthequalityofmatchingbetween peersandsurvivorswasbetterthanthatbetweenlisted sampleandsurvivors;ifp-value > 0.5,thenviceversa. WeusedtheWilcoxonranksumtesttocomparethe locationsoftwodistributions.Becausethedatacontainedpairedandunpairedobservations,theassumption ofindependencebetweenobservationsdidnothold.The methodproposedbyHollanderandcolleagues[41]was adaptedforcorrelateddata.P-valueswereobtainedfrom testingthegroupdifferencebythepermutationmethod describedinDallasandRao[42]. Wecomparedbothmarginalgroupdifferencesandabsolutedifferencesbetweenpairsinordertoprovidea comprehensivepicturetoevaluateeachmethod.Forinstance,weexpectedthematchonageandsextobeexcellentforthelistedsampleandsurvivorsbecausethe telephonescreeningspecifiedthismatch.Ifthequalityof thematchwassimilarforthepeergroupandsurvivors onthesevariables,thenwecouldconcludethateither approachwasadequate.Incontrast,wemightexpectalcoholandtobaccousetobemoresimilarbetweenthe individualsinthepeergroupandthesurvivorsthanbetweenthosematchedinthelistedsample.Thecomparisonoftwomarginaldistributionsmay,however,be misleading.Forexample,thepaireddata(1,10),(10,1), (1,10),(10,1)shownodifferencebetweenthetwomarginaldistributions(meansequal5.5),butthequalityof matchattheindividuallevelisquitepoor.ResultsCharacteristicsTable1showssubjectcharacteristicsdividedbyselection method.Table2showsthemeansandTable3showsthe p-valuesformatchingvariables(includingsex,age,education,employment,smokingandalcoholhistory).RecruitmentandretentionFigures1,2,and3provideflowchartsoftherecruitment patternsforeachofthethreegroups.Theinterviews wereconductedovera6-monthtimeperiodandthe sameinterviewerswereusedforeachgroup. Figure1showsthedispositionoftherecruitment.Eligiblesurvivorswerebasedonthosealiveatlastcontact fromtheDepartmentofRadiationOncologyandconfirmedthroughtheTumorRegistryShandsatUniversity ofFlorida.Inresponsetotheletterthatwasmailed, sevensurvivorscalledthetoll-freenumberrequesting nottobecalled.Aftermultipleattemptstoobtainacorrectedphonenumber,sixnameshadtobedroppedfrom thelist.Whenthecallsweremadebytheprofessional callcenter,20individualsrefusedtoparticipateand83 eitherrepeatedlyrequestedwecallbackorthecallwas neveranswered.Sixsurvivorshaddiedand14wereeithertoomentallyorphysicallycompromisedtoparticipateinthesurvey.Fourpartiallycompletedthesurvey butwereneveravailabletofinishthequestionnaire.All Table1SubjectcharacteristicsbyselectionmethodsCharacteristicSurvivor N=100 Peer N=44 ListedSample N=101 Sexofsubject Men72%43%71% Women28%57%29% Race White94.0%96.0%93.0% Black2.0%2.0%4.0% Other4.0%2.0%3.0% Education LessthanHS7.0%6.8%10.9% HighSchoolgrad orequivalent21.3%25.0%21.8% PostHSeducation71.6%68.2%67.4% EmploymentStatus Currentlyemployed32.0%50.0%43.0% Retired68.0%47.0%57.0% Other-3.0%Disabled19.0%4.8%10.3% Married72.0%77.3%79.2% *Prevalenceofcurrentsmoking15.0%27.0%10.0%*Proportionwhohavesmokedinthelast30days.Logan etal.BMCMedicalResearchMethodology 2012, 12 :63Page4of8 http://www.biomedcentral.com/1471-2288/12/63

PAGE 5

butsixtelephonenumbersfromtheoriginal378fiveyearsurvivorswereusedtocompletethe100surveys (100/372=27%). Figure2showsdataregardingrecruitmentofthepeernominatedgroup.Ofthe100survivorswhocompleted thequestionnaires,61recommendedatleastonepeer: thirty-onesurvivorsprovidedonepeerand30survivors nominatedtwopeers,yieldingatotalof91nominees. Becausewehadfallenshortofourtargetof100peer nominees,multipleattemptsweremadetocontactallof thenominees;12refusedandsixtelephonenumbers werenon-working.Fourteennomineesreportedthat theywerenoteligible,eitherbecausetheyhadahistory ofcancerorwereoutsidetheagerange.Notshownin thefigure,additionalanalysisshowedthatthepercentage ofmenandwomensurvivorswhonominatedapeer wereroughlythesame.Seventy-twopercentofthemen wereabletonominateatleastonepeer,asdid79%of thewomensurvivors.Ofthosepeersnominated,we wereabletocontact38%ofthemennomineescompared to64%ofthewomen.Theobservedsuccessrateforcontactingthepeer-nominatedgroupwas44/91=48.3%. TheyieldforthelistedsampleasshowninFigure3 was101/544=19%.Outoftheoriginal980telephone numberspurchased,allbut436wereusedtocomplete thesample.Onehundredseventeenrefusedwithnoreasongiven,andanother188eitherrepeatedlyaskedthe interviewertocallbackortherewasnoresponseafter Table2MeansformatchingvariablesSampleSize*Mean(S.D.) GroupSurvivorPeerListedSurvivorPeerListed Variable Age(yrs)**100449664.90(10.09)57.40(13.9)64.4(10.08) DaysofSmokinginthepast3010044964.25(10.3)6.29(11.7)2.54(8.13) Lifetimedurationofsmoking60275336.60(15.0)30.40(14.2)28.90(15.1) Howoftendoyoudrinkalcohol? (duringmonth) 10044951.50(1.63)1.50(1.30)1.64(1.64) Howmuchalcoholdoyoudrinkon atypicaldaywhenyouaredrinking? (numberofdrinks) 9942950.66(0.70)0.98(0.87)0.77(0.74) Howoftendoyoudrinksixdrinksof alcoholinoneday?(duringmonth) 10034950.78(0.95)1.03(1.31)0.82(0.82)*:Samplesizesvaryduetomissingdata. **:Medianages(range)insurvivorgroup,peergroup,andlistedgroupwere65.5(33-90),59.5(29-80),and64.0(32-90),respectively. Table3P-valuesformatchingvariablesSampleSize*P-Value GroupSurvivorPeerListedSurvivor vs.Peer#Survivor vs.Listed$Qualityof Matching^^Variable Sexofsubject!10044960.471.000.99 Age(yrs)!!10044960.0170.791.00 Education!9944960.750.410.018 Employment!10044960.180.250.28 Daysofsmokinginthepast3010044960.070.230.77 Lifetimedurationofsmoking6027530.320.0110.18 Howoftendoyoudrinkalcohol? (duringmonth) 10044950.770.440.23 Howmuchalcoholdoyoudrink onatypicaldaywhenyouare drinking?(numberofdrinks) 9942950.400.260.53 Howoftendoyoudrinksixdrinks ofalcoholinoneday?(duringmonth) 10034950.510.420.25*:samplesizesvaryduetomissingdata. #:Two-sidedp-valuesfortestingHo: “ themarginaldistributionofsurvivorsisthesameasthatofpeers ” vs.Ha: “ thetwomarginaldistributionsaredifferent. ” The smallerthep-value,thelargerthedifferencebetweenthetwomarginaldistributions. $:Two-sidedp-valuesfortestingHo: “ themarginaldistributionofsurvivorsisthesameasthatoflistedsample ” vs.Ha: “ thetwomarginaldistributionsare different. ” Thesmallerthep-value,thelargerthedifferencebetweenthetwomarginaldistributions. ^^:One-sidedp-valuesfortestingHo: “ thequalityofmatchingbetweenpeersandsurvivorsisthesameasthatbetweenlistedsamplesandsurvivors ” vs.Ha: “ the qualityofmatchingbetweenpeersandsurvivorsisbetterthanthatbetweenlistedsamplesandsurvivors. ”Logan etal.BMCMedicalResearchMethodology 2012, 12 :63Page5of8 http://www.biomedcentral.com/1471-2288/12/63

PAGE 6

multipleattempts.Analysisshowedthattocompletea sampleofwomenrequired4.6of10numberstobeused; menrequired5.9.Theaverageoftelephonenumbers fromthestratausedtoreachthegroupunder65years was5.6andforthose65yearsorolderwas5.5.CostofcompletedcallsThemaximumnumberoftelephonecallattemptsfor thesurvivor,peernominee,andlistedsampleindividual were11,19,and15,respectively.Therateofcompleted callsperhourrangedfromahighof.65forthesurvivors toarateof.40forthepeer-nominatedgroup(See Table4).Foreachcompletedcallinthesurvivorgroup, thecost,whichincludedthe$25/hourchargedbythe callcenter,was$38.46,comparedto$41.66forthelisted sampleand$62.50forthepeer-nominatedsample.MatchingTable2showsthemeans,standarddeviations,and Table3showsthep-valuesforthecomparisonbetween thesurvivorgroupversusthepeer-nominatedandlisted samplegroups,respectively.ColumnfiveinTable3, titledSurvivorvs.Peer,showsthetwo-sidedp-valuesfor testingwhetherthemarginaldistributionofthevariable incolumn1forsurvivorsisthesameasordifferentthan thatforthepeer-nominatedsample.Columnsixin Table3,titledSurvivorvs.Listed,showsthetwo-sided p-valuefortestingwhetherthemarginaldistributionof thevariableincolumn1forsurvivorsisthesameasor differentthanthatforthelistedsample.Forbothcolumns fiveandsix,thesmallerthegivenp-value,thelargerthe differenceisbetweenthetwomarginaldistributions.The finalcolumninTable3showstheone-sidedp-valuefor thetestoftheoverallqualityofmatch.Thep-valueswere obtainedbytestingthenullhypothesis, “ thequalityof matchingbetweenpeersandsurvivorsisthesameasthat betweenlistedsamplesandsurvivors ” vs.thealternative hypothesis, “ thequalityofmatchingbetweenpeersand survivorsisbetterthanthatbetweenlistedsamplesand survivors. ” Thepeerssmokedmoreinthe30dayspriortothe survey(averagesmokingdays6.29)comparedtoboththe survivorsandthelistedsample.Thedifferenceinpast30daysmokingbetweensurvivorsandthepeergroupyielded atwo-sidedp-valueof.07,whereasthetwo-sidedp-value forthesurvivorsandthelistedsamplewas.23.Table3 shows,however,thatthequalityofthematchattheindividualpairleveldidnotdifferondaysofsmokinginthe past30days(p=0.77).Thatis,wecouldnotrejectthenull hypothesis.Intermsoflifetimedurationofsmoking,there wasatrendtowardthepeer-nominatedgroupyieldinga bettermatchwiththesurvivorsthanthelistedsample (p=0.18).Thesurvivorshadsignificantlygreaterduration onsmokingthanthelistedsample(36.60vs.28.90, p=0.011),whereassurvivorsandpeers(36.60vs.30.40, p=0.32)weremoresimilar.Nodifferencesapproached significanceforalcoholuse. Overall,weobservedthatthequalityofthematching wasnotsignificantlydifferentbetweenthetwomethods exceptonage,sex,andeducationlevel.Specifically,atthe individualpairlevel,thelistedsampleproducedabetter matchonageandsexthanthepeer-nominatedgroup (p=0.99and1.00,respectively).Age(64.90vs.64.40) (Table2)andsexdistribution(72%menvs.71%men) (Table2)werenearlyidenticalforthesurvivorsandlisted sample,respectively.Thepeer-nominatedgroupproduced abettermatchoneducationthanthelistedsample(e.g., 6.8%lessthanhighschoolforpeersvs.7.0%lessthanhigh schoolforsurvivors)(Table1).Thus,asweshowin Table3,werejectthenullhypothesis(p=0.018).ConclusionsThekeyfindingsfromthisstudyare(1)Thequalityof matchingbetweenthetwomethodsofchoosingacomparisongroupforfive-yearcancersurvivorsdidnotdifferexceptforeducation;thepeermethodprovideda significantlybettermatchforthesurvivorsthanthe listedsample,andthelistedsample,asexpected,producedamatchonageandsexbasedonthecriteriaused topurchasethelist;(2)Theyieldforthelistedsample methodwasgreaterthanforthepeer-matchingmethod; (3)Thecostpercompletedinterviewwasgreaterforthe peer-matchingmethodthanthelistedsample.Previous studieshavedocumentedimportantchangesinpatients ’ abilitytofunctionfollowingcancertreatmentbutthe goalofsurvivorshipresearchhasexpandedtoinclude understandingthemoresubtleeffectsoftreatment[25]. Toaccomplishthisgoal,longertermstudieswithappropriatecomparisongroupsareneeded[43]. Table4CostperselectionmethodRecruitment methodology Survivor N=100 Peer N=44 ListedSample N=101 Totalcostspergroup$3884.46$2750.00$4207.66(includespurchaseof telephonelistinhourlyrate) Completesperhour0.650.400.60 Costspercompleted telephoneinterview $25/0.65=$38.46$25/0.40=$62.50$25/0.60=$41.66 Logan etal.BMCMedicalResearchMethodology 2012, 12 :63Page6of8 http://www.biomedcentral.com/1471-2288/12/63

PAGE 7

Ourresultssuggestthatbothmethodsofselectinga comparisongroupyieldsimilarmatchestothesurvivor groupwiththeexceptionofeducationlevel.Theoutcomeofourstudyissomewhatdifferentfromthat reportedbyKaplanetal.[35].Theyfoundthatpeer nomineeswerebettereducatedthanthetargetgroup andhospitalpatientsfromthesamecatchmentarea.We foundthatthepeergroupwaslesseducated.Thisisan issuethatmayaffectsomesurvivorstudies,especially wheresocioeconomicstatusisaconsideration.We foundatrendtowardabettermatchonlifetimeduration ofsmokingforthepeer-nominatedandsurvivorgroups thatmayinterestsomeinvestigators.Smokingexposure isknowntocarryitsownhealthrisks,andwhenthe cancersurvivorsandcontrolsare notmatchedonthis variable,itmaybeproblematictoseparatetheeffectsof cancertreatmentfromthoseofsmokingexposure. Kaplanfound[35]thatolderpatientswerelesslikely tonominateapeerforthecomparisongroupthan youngerpatients.Ourchallengewasnotinthenominationofapeerbutinourabilitytocontactthatnominee, whichaffectedtheyieldofparticipantsandthecostper completionofthesurvey.Reducingthedelaybytelephoningthepeersimmediatelyaftertheirnumberisobtained mightreducetheamountofnonworkingnumbers.Asecondalternativecouldbetomailalettertothenominee askingthemtoverifytheirtelephonenumber.These methodswouldallowimmediatecorrectiveactionstoany biasinmenandwomensurvivorshavingcorrectcontact information.Oursuccesswiththetoll-freenumberforthe survivorstocallforinformationsuggeststhatweshould offeronetonomineesinthefuture.Inshort,refiningthe methodologymayprovideaclosermatchonsexforthe peer-nominatedandsurvivorgroupwhilealsolowering thecostpercompletion.Overall,91peerswerenominated,butafter19attempts,wewereonlyabletoreach44 peers.Ofthose44peerrespondents,wenotedthat19of thewomenpeerswerenominatedbywomensurvivors, while6werenominatedbymensurvivors.Ourinstructionstothesurvivorsdidnotspecifythatthenomineesbe thesamesex.Visualinspectionofthedatasuggeststhat thetelephonenumbersofthemalenomineeswerelessaccurateorlesslikelytobeansweredaftermultipleattempts thanforthewomennominees.Thisaddedtothecostper completionandmayhavecontributedtothepoorermatch onsex. Theresultsofthetestformatchqualityshowedthe twomethodsyieldedsimilarresults.Thismustbeconsideredinlightofthegreatercostpersuccessfulcompletionforthepeer-nominatedmethodanditslowyieldin thisstudy.Wemaintainthatourexperienceisimportant foranyresearcherhopingtoinvestigatethelateeffects ofcancerorcancertreatmentamongolderadults[25]. Wefoundthatthesurvivorsnominatedpeers,butthe challengewasincontactingthosenominees.Readers shouldbeawarethatthelistedsamplewaspurchasedfor $715,andthiswasapartofthevariablecostfiguredinto thesurveycenter ’ shourlyrate.Mailingstotheinitial378 survivorsaddedtoourrecruitmentcosts,butarenot includedinthesurveycenter ’ shourlyrate. Insummary,wetestedthemethodsusingbothatestof themarginaldistributionsbygroupandthedifferencesat theindividualpairsinanattempttoprovidemethodologicalevidenceforcomparisongroupselection.Our resultsindicatethatinvestigatorsofthelateeffectsof cancertreatmentshouldgivecarefulconsiderationtothe methodsforcomparisongroupselection.Webelievethat manyofthelessonslearnedfromthisstudygeneralizeto comparisongroupselectionforthestudyofotheradultcancersurvivors.Conflictofinterest ThisresearchwaspartiallyfundedbyNCIgrantCA111593andNIDCRgrant 1U54DEO19261-01awardedtoH.Logan(PI).Theauthorshavenoconflictof interestwiththefindingsfromthisresearch. Authors ’ contributions HLwasresponsibleforthedesignandexecutionofsurvey,andworked closelywiththeothercontributorsinthecompositionofthismanuscript.ST assistedintheconception,design,andexecutionofthestudy,andworked closelywithothercontributorsinthecompositionofthismanuscript.MC assistedintheconceptualdesign,providedthedataanalysisplan,and oversawtheanalysis.GTwasresponsibleformanagingparticipanteligibility, andheworkedcloselywithothercontributorsinthecompositionand proofingofthismanuscript.WMwasresponsibleformanagingparticipant eligibility,andheworkedcloselywithothercontributorsinthecomposition andproofingofthismanuscript.CRwasresponsibleformanaging participanteligibility,andheworkedcloselywithothercontributorsinthe compositionandproofingofthismanuscript.Allauthorsreadandapproved thefinalmanuscript. Acknowledgements ThisprojectwaspartiallyfundedbyNCIgrantCA111593andNIDCRgrant 1U54DEO19261-01.Publicationofthisarticlewasfundedinpartbythe UniversityofFloridaOpen-AccessPublishingFund. Authordetails1DepartmentofCommunityDentistryandBehavioralScience,Universityof Florida,1329SW16thStreetRoom5174,Gainesville,FL32610-3628,USA.2DepartmentofBiostatistics,UniversityofFlorida,Gainesville,FL,USA.3DepartmentofProsthodontics,UniversityofFlorida,Gainesville,FL,USA.4DepartmentofRadiationOncology,UniversityofFlorida,Gainesville,FL,USA.5DepartmentofMedicine,UniversityofFlorida,Gainesville,FL,USA. Received:8August2011Accepted:16April2012 Published:2May2012 References1.Cancersurvivorshipresearch[http://dccps.nci.nih.gov/ocs/index.html] 2.Cancerfactsandfigures2006[http://www.cancer.org/docroot/STT/content/ STT_1x_Cancer_Facts__Figures_2006.asp] 3.AlfanoCM,RowlandJH: Recoveryissuesincancersurvivorship:Anew challengeforsupportivecare. CancerJ 2006, 12 (5):432 – 443. 4.WingoPA,HoweHL,ThunMJ,Ballard-BarbashR,WardE,BrownML,SylvesterJ, FriedellGH,AlleyL,RowlandJH, etal : Anationalframeworkforcancer surveillanceintheUnitedStates. CancerCausesControl 2005, 16 (2):151 – 170. 5.GivensDJ,KarnellLH,GuptaAK,ClamonGH,PagedarNA,ChangKE,Van DaeleDJ,FunkGF: Adverseeventsassociatedwithconcurrent chemoradiationtherapyinpatientswithheadandneckcancer. Archives ofotolaryngology – head&necksurgery 2009, 135 (12):1209 – 1217.Logan etal.BMCMedicalResearchMethodology 2012, 12 :63Page7of8 http://www.biomedcentral.com/1471-2288/12/63

PAGE 8

6.CleggLX,LiFP,HankeyBF,ChuK,EdwardsBK: CancerSurvivalAmongUS WhitesandMinorities:ASEER(Surveillance,Epidemiology,andEnd Results)ProgramPopulation-BasedStudy. ArchInternMed 2002, 162 (17):1985 – 1993. 7.PollackLA,GreerGE,RowlandJH,MillerA,DoneskiD,CoughlinSS,StovallE, UlmanD: Cancersurvivorship:anewchallengeincomprehensivecancer control. CancerCausesControl 2005, 16 (Suppl1):51 – 59. 8.JungBF,HerrmannD,GriggsJ,OaklanderAL,DworkinRH: Neuropathic painassociatedwithnon-surgicaltreatmentofbreastcancer. Pain 2005, 118 (1 – 2):10 – 14. 9.AzizNM: Cancersurvivorshipresearch:challengeandopportunity. JNutr 2002, 132 (11):3494S – 3503S. 10.CurtisR,FreedmanD,RonE,RiesL,HackerD,EdwardsB,TuckerM, FaumeniJJ:NewMalignanciesAmongCancerSurvivors:SEERCancer Registries,1973 – 2000,vol.No.05 – 5302:NIHPublicationBethesda,MD;2006. 11.Senkus-KonefkaE,JassemJ: Complicationsofbreast-cancerradiotherapy. ClinOncol 2006, 18 (3):229 – 235. 12.ButlerRW,HaserJK: Neurocognitiveeffectsoftreatmentforchildhood cancer. MentRetardDevDisabilResRev 2006, 12 (3):184 – 191. 13.PetersonDE,KeefeDM,HutchinsRD,SchubertMM: Alimentarytract mucositisincancerpatients:impactofterminologyandassessmenton researchandclinicalpractice. SupportCareCancer 2006, 14 (6):499 – 504. 14.JordanVC: Improvementsintumortargeting,survivorship,and chemopreventionpioneeredbytamoxifen-Apersonalperspective. Oncology-NewYork 2006, 20 (6):553 – 562. 15.Lockwood-RayermannS: Survivorshipissuesinovariancancer:Areview. OncolNursForum 2006, 33 (3):553 – 562. 16.SugimuraH,YangP: Long-termsurvivorshipinlungcancer-Areview. Chest 2006, 129 (4):1088 – 1097. 17.VanPoznakC,SauterNP: Clinicalmanagementofosteoporosisinwomen withahistoryofbreastcarcinoma. Cancer 2005, 104 (3):443– 456. 18.WefelJS,SaleebaAK,BuzdarAU,MeyersCA: Acuteandlateonset cognitivedysfunctionassociatedwithchemotherapyinwomenwith breastcancer. Cancer 116 (14):3348 – 3356. 19.RowlandJH,HewittM,GanzPA: Cancersurvivorship:anewchallengein deliveringqualitycancercare. JClinOncol 2006, 24 (32):5101 – 5104. 20.StolleyPD,SchlesselmanJJ: Planningandconductingastudy .In In:Case ControlStudies .EditedbySchlesselmanJJ.NewYork:OxfordUniversity Press;1982:76 – 85. 21. 2004surgeongeneral'sreportlinksmorecancerstosmoking. CACancer JClin 2004, 54 (5):243 – 244.doi:10.3322/canjclin.54.5.243. 22.FeuersteinM: Definingcancersurvivorship. JournalofCancerSurvivorship 2007, 1 (1):5 – 7. 23.MurphyBA,GilbertJ,RidnerSH: Systemicandglobaltoxicitiesofhead andnecktreatment. ExpertRevAnticancerTher 2007, 7 (7):1043 – 1053. 24.AzizNM,RowlandJH: Cancersurvivorshipresearchamongethnic minorityandmedicallyunderservedgroups. OncolNursForum 2002, 29 (5):789 – 801. 25.BondSM: DietrichMS .MurphyBA:Neurocognitivefunctioninheadand neckcancerpatientspriortotreatment.SupportiveCareinCancer;2011. 26.AbendsteinH,NordgrenM,BoysenM,JannertM,SilanderE,Ahlner-Elmqvist M,HammerlidE,BjordalK: Qualityoflifeandheadandneckcancer: A5yearprospectivestudy. Laryngoscope 2005, 115 (12):2183 – 2192. 27.ExtermannM: Measurementandimpactofcomorbidityinoldercancer patients. CritRevOncolHematol 2000, 35 (3):181 – 200. 28.NordgrenM,AbendsteinH,JannertM,BoysenM,Ahlner-ElmqvistM, SilanderE,BjordalK,HammerlidE: Health-relatedqualityoflifefiveyears afterdiagnosisoflaryngealcarcinoma. IntJRadiatOncolBiolPhys 2003, 56 (5):1333 – 1343. 29.WacholderS,SilvermanDT,McLaughlinJK,MandelJS: Selectionofcontrols incase –controlstudiesII.Typesofcontrols. 1041 1992, 135 (9):1029. 30.RibeiroKC,KowalskiLP,LatorreMR: Impactofcomorbidity,symptoms,and patients'characteristicsontheprognosisoforalcarcinomas. Archivesof otolaryngology – head&necksurgery 2000, 126 (9):1079 – 1085. 31.ShawGL,TuckerMA,KaseRG,HooverRN: Problemsascertainingfriend controlsinacase – controlstudyoflung-cancer. AmJEpidemiol 1991, 133 (1):63 – 66. 32.ZhouW,LiuG,MillerDP,ThurstonSW,XuLL,WainJC,LynchTJ,SuL, ChristianiDC: Gene-environmentinteractionfortheERCC2 polymorphismsandcumulativecigarettesmokingexposureinlung cancer. CancerRes 2002, 62 (5):1377 – 1381. 33.ZhouW,LiuG,ThurstonSW,XuLL,MillerDP,WainJC,LynchTJ,SuL, ChristianiDC: GeneticpolymorphismsinN-acetyltransferase-2and microsomalepoxidehydrolase,cumulativecigarettesmoking,andlung cancer. CancerEpidemiolBiomarkersPrev 2002, 11 (1):15 – 21. 34.TchenN,BedardP,YiQL,KleinM,CellaD,EremencoS,TannockIF: Quality oflifeandunderstandingofdiseasestatusamongcancerpatientsof differentethnicorigin. BrJCancer 2003, 89 (4):641 – 647. 35.KaplanS,NovikovI,ModanB: Amethodolgicalnoteontheselectionof friendsascontrols. IntJEpidemiol 1998, 27 (4):727 – 729. 36.RothmanKJ,GreenlandS:Matching.In: ModernEpidemiology.Volume1 ,2nd edn.EditedbyRothmanKJ,GreenlandS.Philadelphia:LippincottWilliams& William;1998:147 – 161. 37.BreslowNE,DayNE:Theneedfortheconstructionofspecialcomparison groups.In: StatisticalMethodsinCancerResearch edn.Lyon:International AgencyforResearchonCancer;1987:32 – 46. 38.EarleCC: Longtermplanningforcancersurvivors:ahealthservices researchagenda. JournalofCancerSurvivorship 2007, 1 (1):64 – 74. 39.SilvermanS: Demographicsandoccurrenceoforalandpharyngeal cancers-Theoutcomes,thetrends,thechallenge. JAmDentAssoc 2001, 132: 7S – 11S. 40.LoganHL,BartoshukLM,FillingimRB,TomarSL,MendenhallWM: Metallic tastephantompredictsoralpainamong5-yearsurvivorsofheadandneckcancer. Pain 2008, 140 (2):323 – 331. 41.HollanderM,PledgerG,LinP: RobustnessontheWilcoxonTesttoa CertainDependencybetweenSamples. AnnStat 1974, 2 (1):177 – 181. 42.DallasM,RaoP: TestingEqualityofSurvivalFunctionsBasedonboth PairedandUnpairedCensoredData. Biometrics 2000, 56: 154 – 159. 43.FeuersteinM: Optimizingcancersurvivorship. JournalofCancer Survivorship:ResearchandPractice 2007, 1 (1):1.doi:10.1186/1471-2288-12-63 Citethisarticleas: Logan etal. : Selectingacomparisongroupfor5-year oralandpharyngealcancersurvivors:Twomethods. BMCMedical ResearchMethodology 2012 12 :63. Submit your next manuscript to BioMed Central and take full advantage of: € Convenient online submission € Thorough peer review € No space constraints or color “gure charges € Immediate publication on acceptance € Inclusion in PubMed, CAS, Scopus and Google Scholar € Research which is freely available for redistribution Submit your manuscript at www.biomedcentral.com/submit Logan etal.BMCMedicalResearchMethodology 2012, 12 :63Page8of8 http://www.biomedcentral.com/1471-2288/12/63