This item is only available as the following downloads:
University of Florida | Journal of Undergraduate Research | Volume 15 Issue 3 | Summer 2014 1 Photoplethysmography and Heart Rate Variability for the Prediction of Preeclampsia Lauren M. Silva1, Shalom Darmanjian, PhD2, Tammy Y. Euliano, MD1 1College of Medicine, University of Florida, 2Convergent Engineering, Gainesville, FL The focus of this study was to develop a noninvasive, inexpensive bedside test that can accurately predict which women are at risk for developing preeclampsia. Healthy controls and high risk participants were monitored intermittently throughout their pregnancy (beginning before 24 weeks gestation). Three lead electrocardiogram (ECG) recordings from the maternal chest and pulse oximetry waveforms (photoplethysmography, PPG) from the middle finger were obtai ned for 30 minutes with the patient at rest at each prenatal visit. Upon delivery, their preeclampsia status was recorded. The data were utilized as an indirect measurement of arterial compliance and used to train an Artificial Neural Network (ANN) to pred ict preeclampsia at delivery. Of the 20 subjects studied prior to the 3rd trimester, the system accurately predicted 5/6 that developed preeclampsia. Overall, the system had a sensitivity of 0.91 and specificity of 0.77. INTRODUCTION Preeclampsia is a hypertensive disorder that occurs in 3 5% of all pregnancies1, 2, 3 and can have serious consequences for both maternal and fetal health. Symptoms of the disorder include hypertension, proteinuria, ed ema, and neurological symptom s such as severe headaches, light sensitivity nausea, and even seizures (ec lampsia) Women are at higher risk for preeclampsia if they possess one or more of the following preexisting conditions: chronic hypertens ion, obesity, multiple gestation diabetes, gestational diabetes, preeclampsia in a previous pregnancy, liver disease and/or kidney disease2,4. According to the Internat ional Society for the Study of Hypertension in Pregnancy5, a previously normotensive woman is determined t o have preeclampsia if she has one sys tolic two successive hour of protein in the urine in a 24hour urine collection. The exact causes of pree clampsia are unclear, making it difficult to develop a treatment or cure. Studies have demonstrated placental endothelial dysfunction due to oxidative stress and irregular placental development as potential underlying causes4,6. Current ly, the only available cure for preeclampsia is delivery2. When a wom an is admitted to a hospital with preeclamptic signs and symptoms physicians may administer magnesium sulfate to prevent seizures and steroids to mature the fetal lungs Physicians are sometimes able to avoid preterm delivery if a woman is admitted before her symptoms become severe and is responsive to blood pressure control therapy. Preeclampsia in the Third World In underdeveloped countries, the prevalence of preeclampsia is approximately seven times greater than in the developed areas of North America and Europe7. One factor contributing to the increased incidence is lack of education expectant mothers are unaware of the signs and symptoms indicating that medical attention must be sought. Additionally, receiving medical attention is unaffordable and access is very limited8. The majority of the population is located in remote areas and transporta tion is either impractical or unavailable. For example, in Nigeria 50% of women live more than 5 kilometers from the closest hospital6. Additionally, if a woman is actually able to travel to the hospital, the quality of care provided may be extremely poor8. Prevention of the complications of preeclampsia requires the ability to predict which women are at high risk for developing the disorder7. A simple, cost effective, and reliable test that can predict these patients in underdeveloped countries could allow direction of education and resources to this most vulnerable group. Thus far, researchers have not found a test that can fulfill this need. If a womans relative risk of developing preeclampsia can be estimated, the total number of fetal and maternal deaths can be dramatically reduced, especially in the Third World6,7. Preeclampsia Predi c t ion Presently, there are two commonly studied methods of predicting preeclampsia before the onset of symptoms. Serum biomarkers have been studied based on their
LAUREN M. SILVA, SHALOM DARMANJIAN, TAMMY Y. EULIANO University of Florida | Journal of Undergraduate Research | Volume 15 Issue 3 | Summer 2014 2 involvement in plac ental dysfunction, specifically the concentration ratio of placental growth factor (PlGF) to soluble fms like tyrosine kinase (sFlt 1)3. Uterine artery Doppler ultrasonography has also been considered by noninvasively searching for indic ations of placental dysfunction, such as resistance to blood flow in the uterine arteries and decreased placental volume6,9. Although studies have shown that the two methods function more efficiently when used in conjunction, these tests remain very expens ive and unable to be utilized outside of a clinical setting. In recent years, researchers have attempted to study preeclampsia by noninvasive applanation tonometry via a device placed on the wrist to extract the radial arterial pressure waveform, which a lternately applies pressure to the artery and examines the reflecting waves. Although applanation tonometry is expensive, requires training, and suffers from reproducibility issues, the studies provide useful insight into the physiology: preeclamptic patie nts demonstrate an increase in arterial stiffness10,11. Features of the photoplethysmogram have been correlated with arterial stiffness/compliance12, therefore our study was interested in investigating the use of a pulse oximeter and an electrocardiogram f or the detection of these vascular changes. Our findings illustrate the potential to accurately determine a womans risk of developing preeclampsia through a single test in a more economical and less invasive manner. METHODS After written, informed consent, 20 women receiving prenatal care at Womens Health at Shan ds Medical Plaza, a high risk prenatal clinic, participated in this observational prospective pilot study. Inclusion criteria consist ed of women prior to 25 weeks of gestation with multi f etal gestation, chronic hypertension, pre gestational diabetes, and/ or history of pree clampsia in a prior pregnancy. Three lead electrocardiogram ( ECG ) recordings and pulse oximetry waveforms (photoplethysmogra phy, PPG ) were collected from the subjects for 30minutes at their prenatal visits and again when they pr esent ed for delivery, if possible. Prenatal visit collections were performed at 2 4 w eek intervals depending on the womens pren atal visit frequencies. Data were stored for subsequent analysis in light of delivery outcome. To date, the 20 women have delivered: 6 pre eclamptics and 14 normotensives. The maternal ECG was recorded using Convergent E ngineerings amplifier and s oftware system. The ECG data were synchr onized with PPG signals extracted in real time from a commercial pulse oximeter and logged with PC user interface ( UI ) software. These recordings were all saved to databases along with patient information using the UI. As with any signal processing approach, the first st ep involved understanding the under lying features exhibited by pre eclampsia. Using a variety of techniques to extract the important fe atures (described below), the features were refined base d on simple clustering methods to reduce complexity given the size of the datasets. As more data were recorded, models with higher degrees of freedom were employed to model these data. Data Collection Protocol 1. Four ECG electrodes were attached to the maternal chest for continuous ECG recording. 2. Pulse oximetry probe w as attached to middle finger. 3. ECG and pulse oximetry waveforms were collected for 30minutes after a stable tracing was achieved. 4. Blood pressure, weight and proteinuria (if available) measurements were recorded. Feature Extraction The current feature set consists of parameters from three different physiologic classes: A) heart rate, B) pulse transit time (PTT, correlates with blood pressure) and C) augmentation indices. Multiple parameters from each class capture different representations of the fundamental data (e.g. heart rate or PTT variability), and combinations of parameters are also derived (e.g. change in PTT per change in heart rate). Using the different covariates a high dimensional feature vector was assembled as input into our classifier. More importantly, performance of classifiers is often best improved by ensuring the features adequately represent the desired information in all possible situations. Refinement of the feature extraction software is very important to ensure that the model parameters do not contain excess noise or outliers during poor data collection conditions. Classifier Although traditionally a black box modeling tool, artificial neural networks (ANN) afford an increase in the degrees of freedom to model the aforementioned data nonlinearly. For this mode l, LevenbergMarquardt training was used as a compromise between GaussN ewton and gradient descent (for faster more robust error minimization). We used the logistic function (which we determined empirically) for the sigmoid function of our processing element (PE) and an 80/20 percent mixture of training/testing sets. The ANN w as trained and test ed with 1000 different trials using data acquired in a parallel study of the same equipment analyzing patients whose preeclampsia status was known (preeclampsia vs normotensive).
PHOTOPLETHYSMOGRAPHY AND HEART RATE VARIABILITY FOR THE PREDICTION OF PREECLAMPSIA University of Florida | Journal of Undergraduate Research | Volume 15 Issue 3 | Summer 2014 3 RESULTS Figure 1 shows two features [ rate variabi lity (Y axis) (X axis ) ] changing through the time of gestation for three patients. Darker colors indicate more observations of the particular variability measures. Looking at Patient #3, heart rate variability and pulse transit tim e variability did not change significantly as gestational age progressed. Ultimately, Patient #3 did not develop preeclampsia. When analyzing the figures for Patient #1 and Patient #2 at 27 and 26 weeks of gestation, respectively, a decrease in heart rate variability is clearly observed (reduce spread in the Y axis) Both patients were diagnosed with preeclampsia at 38 weeks of gestation. We see from the figure that more than 10 12 weeks before a diagnosis is made there are fundamental changes that can be observed from the preeclamptics features (patient #1 and #2) vs. the patient without preeclampsia (patient #3). Figure 1 : Preeclampsia feature changes through time. Of the 20 subjects the system accurately predicted 5 of the 6 patients that developed preeclampsia. Overall across the 1000 randomized trials, the system had a sensitivity of 0.91 and specificity of 0.77. D ISCUSSION The ability to predict preeclampsia before the onset of symptoms i s not an innovative undertaking but has proven challenging. The two most studied methods have been serum biomarkers and ute rine artery Doppler ultrasonography6. S erum biochemical markers have been studied by researchers as potential predictors of preecl ampsia based on their concentrations during pregnancy (i.e., first, second, or third trimester)24,6 and their involvement in placental dysfunction. Unfortunately, no single bio marker has been identified as a reliable indicator of predicting preeclampsia; rather a combination of biomarkers has been observed to be more dependable4. The most promising combination is the concentration ratio of placental growth factor (PlGF) to soluble fms like tyrosine kinase (sFlt 1)3. Pl GF is an angiogenic factor that is resp onsible for the development of new blood ve ssels from existing endothelium and is essential for normal placental development 2. SFlt 1 is an anti angiogenic factor that binds to and inhibits the function of PlGF2. I t has been determined that the levels of PlGF decrease and the levels of sFlt 1 increase24, 9 in women who have been diagnosed with preeclampsia In Germany, a study4 was performed by Verlohren, et al. that tested the ability of the concentration ratio of PlGF to sFlt 1 to predict the occurrence of preeclampsia in 351 pregnant women after 20 weeks gestation. Overall the test had a sensitivity of 82% and a specificity of 95%4. Although these results indicate PlGF:sFlt1 to be a reliable predictor of preeclampsia, the expense of running the test is not cost effective, it can only be performed in a professional medical setting, and it is an invasive procedure. Researchers have also studied Doppler ultrasonography as a method of predicting preeclampsia by noninvasively searching for indications of placental dysfunction, such as resistance to blood flow in the uterine arteries and decreased placental volume6,10. One such study recruited 17,480 women and studied their uterine arteries via transvaginal sonography at 23 weeks of gestation. Ultimately, the test had a high falsepositive rate of 25% and a low detection rate of 63.1%9. Doppler ultrasonography alone has not prove n to be a reliable predictor of preeclampsia. Other studies, however, have combined Doppler with other clinical tests and have observed signific ant improvements in sensitivity, specificity and positive predictive values. These multiparameter tests include various combinations of Doppler ultrasonography, serum biomarkers, and maternal history6, 911. Although the statistics have improved, these tests remain expensive, invasive and impractical for our intended application4,12. The results of our s tudy do not vary significantly from those of the multipa rameter studies mentioned above. Our findings, however, illustrate the potential to accurately determine a womans risk of developing preeclampsia through a single test in a more economical and less invasive manner More importantly, it provides an avenue for developing a device that can be used outside of the hospital se tting and in areas with limited resources12. T he majority of expec tant mothers in underdeveloped countries do not have immediate access to medical attention because the hospitals and facilities are located at great distances and transportation is either unavailable or limited. Sadly, most mothers will never even see a physician during their pregn ancies. These impoverished living conditions demand an efficient method for predicting a womans risk of developing pr eeclampsia3. Collecting heart rate and pulse oximetry data is noninvasive and inexpensive data collectors require very little training, while the equipment used is compact easily
LAUREN M. SILVA, SHALOM DARMANJIAN, TAMMY Y. EULIANO University of Florida | Journal of Undergraduate Research | Volume 15 Issue 3 | Summer 2014 4 transported and reu sable. Currently, volunteers and technicians are dedicating their time providing minimal healthcare in underdeveloped countries through organizations such as the W orld H ealth Organization Having a simple device that can quickly estimate a womans risk for developing preeclampsia will allow individuals such as these volunteers, t he opportunity t o not only educate women about the dangers of preeclampsia, but also inform them of their risk of developing the disorder. Possessing the ability to determi ne a womans risk early in her pregnancy will also give her the extra time to travel and seek professi onal medical attention before symptoms emerge or worsen7. Ultimately, the number of fetal and maternal deaths will be dramatically reduced. In addition to its Third World applications, our studys findings can also benefit the First World. In countries such as the United States, the majority of expec ting mothers has immediate access to a hospital and receives prenatal care on a timely ba sis throughout their pregnancy ; therefore, medical ac c ess is not a prevalent problem. The real issue is the time and money4 spent on admitting women for monitoring at hospitals. The simplicity of our test will potentially allow a woman to be monitored outside of the hospital and only be admitted if she is considered at high risk for preeclampsia. Additionally, it will permit researchers to conduct more efficient studies by only enrolling women who will show explicit signs and symptoms Thus t he ability to determine a womans risk of developing preeclampsia will decrease the amount of money spent on healthcare and ensure a more efficient allocation of research funding. FUTURE WORK The studys results demonstrate that economical, non invasive and reusable ECG and PPG technology have potential as reliable tools in predicting preeclampsia. The eventual goal is to develop a small, inexpensive, portable device that employs smart phone technology to predict a womans risk of developing preeclampsia and appropriately recommend necessary level of care for delivery. Further data are necessary and data collection continues under a NIH STTR grant. ACKNOWLEDGMENTS I wo uld like to thank my mentor, Tammy Y. Euliano, M. D., for her guidance and encouragement throughout the course of our study. It has been and always will be an honor to collaborate with such a wonderful instructor I would also like to thank Convergent Engineering for providing all the technology and equipment necessary to the study and for extrapolating the raw data. Additionally, I owe a special thanks to the University Scholars Program for t heir assistance and financial support throughout the research process. Finally, thank you to the Bill and Melinda Gates Foundation for funding our study CONFLICT OF INTEREST Tammy Y. Euliano, M D ., is the study principal investigator and is married to Neil Euliano, Ph.D ., the president of the sponsoring company, Convergent Engineering. Any resulting patents and products could have financial implications. ENDNOTES 1Walther T, Wessel N, Malb erg H, Voss A, Stepan H, Faber R A combined technique for predicting preeclampsia: concurrent measurement of uterine perfusion and analysis of heart rate and blood pressure variability, J. Hypertension. 24, 747-750 (2006). 2Carty DM, Delles C, Dominiczak A F, Novel Biomarkers for Predicting Preeclampsia, Trends Cardio. Med. 18, 186 -194 (2008). 3Verlohren S, Galindo A, Schlemback D, Zeisler H, Herraiz I, Moertl MG, Pape J, Dudenhausen JW, Denk B, Stepan H, An automated method for the determination of the s Flt1/PIGF ratio in the assessment of preeclampsia, Am. J. Obstet. Gynecol. 202, 161.e1 161.e11 (2010). 4Anderson UD, Olsson MG, Kristensen KH, Akerstrom B, Hansson SR, Review: Biochemical markers to predict preeclampsia, Placenta. 33, S42-S47 (2012). 5National High Blood Pressure Education Program Working Group on High Blood Pressure in Pregnancy: Report of the National High Blood Pressure Education Program Working Group on High Blood Pressure in Pregnancy, Am. J. Obstet. Gynecol. 183, 1 -22 (2000). 6T uuli MG, Odibo AO, The Role of Serum Markers and Uterine Artery Doppler in Identifying At Risk Pregnancies, Clin. Perinatol. 38, 1 -19 (2011). 7Osungbade KO, Ige OK, Public Health Perspectives of Preeclampsia in Developing Countries: Implication for Heal th System Strengthening, J. Pregnancy. 2011, 481095 (2011). 8Richard F, Witter S De Brouwere V Innovative approaches to reducing financial barriers to obstetric care in low -income countries, Am J Public Health 100,1845 1852 (2010). 9Papageorghiou AT, Yu CKH, Erasmus IE, Cuckle HS, Nicolaides KH, Assessment of risk for the development of pre -eclampsia by maternal characteristics and uterine artery Doppler, BJOG: An International J. Obstet. Gynecol. 112, 703 709 (2005). 10Antsaklis A Daskalakis G, Uterine Artery Doppler in the Prediction of Preeclampsia and Adverse Pregnancy Outcome, Donald School Journal of Ultrasound in Obstetrics and Gynecology. 4(2),117122 (2010). 11Yu CKH, Khouri O, Onwudiwe N, Spiliopoulos Y, Nicolaides KH, Prediction of pre -eclampsia by uterine artery Doppler imaging: relationship to gestational age at delivery and small -for -gestational age, Ultrasound Obstet Gynecol. 31, 310313 (2008). 12Hadker N, Garg S, Costanzo C Miller JD, Foster T, van der Helm W, Creeden J, Financial impact of a novel pre-eclampsia diagnostic test versus standard practice: a decision -analytic modeling analysis from a UK healthcare payer perspective, J. Med. Econ. 13, 728 -737 (2010).