<%BANNER%>

Analytical Approaches to Achieve Quasi-Randomization in Retrospective Database Analysis.

www.ispor.org ( Publisher's URL )
University of Florida Institutional Repository
Permanent Link: http://ufdc.ufl.edu/IR00000600/00001

Material Information

Title: Analytical Approaches to Achieve Quasi-Randomization in Retrospective Database Analysis.
Series Title: ISPOR CONNECTIONS. March/April 2011;17(2):10-11.
Physical Description: Journal Article
Creator: Ayad K Ali
Publisher: ISPOR
Place of Publication: USA
Publication Date: April 10, 2011

Notes

Acquisition: Collected for University of Florida's Institutional Repository by the UFIR Self-Submittal tool. Submitted by Ayad Ali.
Publication Status: Published

Record Information

Source Institution: University of Florida Institutional Repository
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution.
System ID: IR00000600:00001

Permanent Link: http://ufdc.ufl.edu/IR00000600/00001

Material Information

Title: Analytical Approaches to Achieve Quasi-Randomization in Retrospective Database Analysis.
Series Title: ISPOR CONNECTIONS. March/April 2011;17(2):10-11.
Physical Description: Journal Article
Creator: Ayad K Ali
Publisher: ISPOR
Place of Publication: USA
Publication Date: April 10, 2011

Notes

Acquisition: Collected for University of Florida's Institutional Repository by the UFIR Self-Submittal tool. Submitted by Ayad Ali.
Publication Status: Published

Record Information

Source Institution: University of Florida Institutional Repository
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution.
System ID: IR00000600:00001


This item is only available as the following downloads:


Full Text

PAGE 1

10 March/April 2011 ISPOR CONNECTIONS References 1 Pateriya LP, Jha A, Munjal S. Application of information technology to overcome noncompliance of drugs.https://www.indianjournals.com/glogift2k6-1/theme_1article%201.htm. [Accessed April 4, 2009].2 Cost of Patient Non-Compliance. Available from: http://www.alignmap.com/wp-content/uploads/2006/03/Costs%20Of%20Patient%20Noncompliance.pdf. [Accessed August 7, 2010].3 Weinstein MC, Fineberg HV. Clinical decision analysis. Philadelphia: WB Saunders Co., 1980.4 Van Dongen. Are you patient intelligent? Pharmaceutical Marketing Journal 2010;1:42-3.5 Van Dongen Nadine. Let's be effective, let the patients talk; does patient intelligence have an effect on improvements in quality in the healthcare environment? Dovepress, 2009.6 Jariath N, Weinstein J. The Delphi methodology (part two): a useful administrative approach. CJONA 1994;7:7-20.7 Jones J, Hunter D. Consensus methods for medical and health services research. BMJ 1995;311:376-80.8 Evans C. The use of consensus methods and expert panels in pharmacoeconomic studies: practical applications and method-ological shortcomings. Pharmacoeconomics 1997;12:121-9.9 Nuijten MJ. The selection of data sources for use in modelling studies. Pharmacoeconomics 1998;13:305-16. P erhaps randomization is one of the characteristics that granted controlled, clinical trials (RCTs) the gold standard label for causal inference of exposure effect. By randomly assigning patients to exposure groups, randomization is the only approach that ensures an equalor nearly equaldistribution of measured and unmeasured confounders across exposure groups. Lack of randomization in observational epidemio-logical designsincluding retrospective database studiespredisposes them to a myriad of limitations that affect the interpretation of ndings from such studies [1]. Considering the ethical and logistical weaknesses of RCTs, well designed and conducted observational studies are efcient in exploring exposure safety and effectiveness. Although comparability across exposure groups is rarely guaranteed in observational studies, multiple analytical approaches are utilized to achieve a quasi-randomization state [2], which results in unbiased estimates of association between the exposure and the outcome of interest. Propensity scoring, instrumental variable, and disease severity measures are some of these approaches (Table 1). Propensity Scoring (PS) Propensity scoring adjusts for the likelihood of a patient being exposed given a set of measured confounders [3]. This method is mainly used to account for selection bias and confounding by indication. Although PS balances confounders across exposure groups, it does not balance unmeasured confounders [4]. Propensity score is dened as the probability of receiving an exposure given baseline information (i.e. confounders). On average, groups with similar scores are expected to have similar baseline information. There are four applications of the PS technique:t Propensity Score Matching: Patients can ostensibly be treated as if they were randomly assigned to exposures by matching exposure groups with similar scores based upon a PS scalar/caliber. One limitati on of this approach is restricting generalizability of ex posure effect to a subset of patients with overlapp ing scores across groups [5].t Propensity Score Stratication: Exposure groups can be grouped into strata based on the score, where a stratum-specic treatment effect can be estimated [6]. Both applications can be used to evaluate the comparability of exposure groups by checking for overlap across scores. Residual confounding, however, could occur as a result of score categorization [7].t Propensity Score Regression: Incorporating the score as a covariate in the regression model will simplify the nal model in terms of the number of covariates [6]. Compared to the earlier two methods, the effect of some important covariates on the outcome can be elucidated in regression models. Model misspecication, however, can be a limitation in this method [7].t Inverse Probability of Exposure Weighting (IPW): The inverse of the propensity score can be used to create a weighted average of the exposure effect. IPW is de ned as the inverse of the probability of receiving the exposure actually received [8]. Compared to the PS matching technique, the IPTW uses the whole sample data for analysis, which enhances generalizability. Additionally, IPW extends to multicategory exposure groups, and timevarying exposures [7]. Weight instability, however, occurs when some exposure groups become uncommon [4,7]. Instrumental Variable (IV) As mentioned above, PS methods do not account for unmeasured variables. In case of unavailability of an important confounding factor, using a variable, den oted an instrument can account for this problem. Three conditions, however, ought to be met in a variable in order to be an IV (Fig. 1) [9]. The variable (I):1. Is associated with exposure (E). Either (I) causes (E) (solid gray arrow), or both share a common cause (C). STUDENT CORNER Analytical Approaches to Achieve Quasi-Randomization in Retrospective Database Analysis Ayad K. Ali, MSPharm, Fulbright Scholar, PhD Candid ate, ISPOR Student Chapter President, Pharmaceutical Outcomes and Policy, Universit y of Florida, Gainesville, FL, USA Figure 1. Characteristics of an instrumental variable ISPOR STUDENT TRAVEL GRANTS AND ISPOR MEETING TRAVEL SCHOLARSHIP AWARDS ISPOR Student Travel Grants and ISPOR Meeting Travel Scholarship Awards are now available for the ISPOR 3rd Latin America Conference in Mexico City, Mexico and 14th Annual European Congress in Madrid, Spain. Students and Fellows may apply for an ISPOR Student Travel Grant Details and submission can be found at: http://www.ispor.org/student/Travel/grant_info.asp. Non-Students may apply for an ISPOR Meeting Travel Scholarship Details and submission can be found at: http://www.ispor.org/awards/MeetingTravelScholarship.asp. APPLICATION DEADLINES: 3rd Latin America Conference, Mexico City, Mexico (8-10 September 2011): July 15, 2011 14th Annual European Congress, Madrid, Spain (5 8 November 2011): September 12, 2011

PAGE 2

March/April 2011 ISPOR CONNECTIONS 112. Affects the outcome (O) only through the exposure (I E O).3. Doesn't share a common cause (B) with the outcome (dashed gray arrows). Although this approach is imperative in accounting for unmeasured confounders and minimizing residual confounding, it is very difcult to identify and measure a good IV that meets the above-mentioned conditions and ts the research question in hand. Disease Severity Measures When patients with high disease severity are preferentially prescribed specic medications, confounding by disease severity is said to occur [2]. This problem can be accounted for by identifying proxy variables that implicitly depict the disease state, and include them in the covariate pool of the PS technique. The disadvantage of this approach goes in tandem with the limitation of the PS mentioned above (i.e. failure to adjust for unmeasured severity-related variables). Yet, using rich databases will account for much of these measures. Acknowledgement Gratitude is expressed to Prof. Abraham Hartzema for his supervisory efforts. References 1 B!gaud B. Dictionary of Pharmacoepidemiology. West Sussex, UK: John Wiley & Sons, 2008.2 Hudson M, Suissa S. Avoiding common pitfalls in the analysis of observational studies of new treatments for rheumatoid arthritis. Arthritis Care Res 2010;62:805-10.3 Rosenbaum P, Rubin D. The central role of the propensity score in observational studies for causal effects. Biometrika 1983;70:41-55.4 Gelman A, Hill J. Causal inference using more advanced models. in: gelman a, hill j. data analysis using regression and multilevel/hierarchical models. New York: Cambridge University Press, 2007.5 Rosenbaum P, Rubin D. Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. Am Stat 1985;39:33-8.6 D'Agostino R, Jr. Propensity Score Methods for Bias Reduction in the Comparison of a Treatment to a Non-Randomized Control Group. Stat Med 1998;17:2265-81.7 Greenland S. Introduction to Regression Modeling. In: Rothman K, Greenland S, Lash T, eds., Modern Epidemiology (3rd ed.). Philadelphia, PA: Lippincott Williams & Wilkins, 2008.8 Rosenbaum P. Model-Based Direct Adjustment. J Am Statist Assoc 1987;82:387-94.9 Hernan M, Robins J. Instruments for causal inference. An epidemiologist's dream? Epidemiology 2006;17:360-72. Table 1. Problems that can be controlled by different analytical approaches IC