Public Health Department Dissertations Collection

Latent variable modeling for biomarker analysis

Liu, Yan — Mon, 24 Sep 2012 11:00:39 PDT

Characterizing associations among multiple single-nucleotide polymorphisms (SNPs) within and across genes, and measures of disease progression or disease status will potentially offer new insight into disease etiology and disease progression. However, several analytical challenges arise due to the existence of multiple potentially informative genetic loci, as well as environmental and demographic factors, and the generally uncharacterized and complex relationships among them. Latent variable modeling offers a natural framework for data arising from these population-based association studies to uncover simultaneous effects of multiple biomarkers. In the first chapter, we describe applications and performance of two such latent variable methods, namely structural equation models (SEMs) and mixed effects models (MEMs), and highlight their theoretical overlap. The relative advantages of each paradigm are investigated through simulation studies and an application to data arising from a study of anti-retroviral-associated dyslipidemia in HIV-1 infected individuals is provided for illustration. In the second chapter, we address a prediction-based classification (PBC) method that allows the use of repeatedly measured biomarkers for CD4⁺ T cell outcome prediction through first-stage of fitting MEMs and subsequent classification based on clinical relevant thresholds ( CD4⁺ T cell count 200 or 350 cells/mm³). Then we apply this PBC approach to a prospective cohort of HIV-1 infected subjects (n=3357) monitored upon anti-retroviral therapy initiation in 7 clinical sites with distinct geographical and socio-economic settings. ^

Understanding maternal health-care seeking behavior in low-income communities in Accra, Ghana

Anafi, Patricia — Mon, 24 Sep 2012 11:00:18 PDT

This study sought to examine health care decisions and choices that women make during pregnancy and childbirth in selected low-income and poor urban communities in Ghana. Specifically, it examined women's and community members' knowledge and perceptions about pregnancy and childbirth; existing forms of health care available to women during pregnancy and childbirth; and factors that influence preference for the type of health care that women use during pregnancy and childbirth. The study employed a two-phased data collection strategy involving in-depth interviews and focus group discussions to examine maternal health care seeking behavior of the target population. ^ The findings revealed that the poor urban women have a wide range of perceptions and knowledge about pregnancy including knowledge about what constitutes a successful pregnancy and risk factors of pregnancy and childbirth complications. The study found that three major forms of health care exist for pregnant women: biomedical care; herbal-traditional birth attendant care; and spiritual care. While some women use or prefer to use either solely medical care or herbal-traditional birth attendant care for their pregnancy and delivery, others combine two or all the three forms of health care. Pregnant women seek traditional birth attendants (TBAs) and spiritual care for spiritual protection against death, due to affection and cultural attachment to TBAs, fears about medical care and health facilities, and due to cost of seeking medical care. ^ Long waiting time and early reporting time at antenatal clinic were identified as partly limiting the use of medical care during pregnancy. Intimate partners of pregnant women were identified as negative normative influence since most of them do not support their wives during pregnancy. Quality and safety of care were the major reasons why pregnant mothers seek biomedical care other than other forms of care. However, majority of women who seek biomedical care do not seek timely antenatal care. Only 42 percent made their first antenatal visit in the first trimester. These findings have implications for policies and programs that are likely to help increase the use of skilled attendance and improve maternal health outcomes in the study population and other similar low-income urban communities in Ghana.^

Classification and risk factors of sporadic anovulation in a longitudinal evaluation of menstrual cycle hormone patterns

Lynch, Kristine E — Mon, 16 Apr 2012 14:13:54 PDT

There are 8 commonly used algorithms for classifying the ovulatory status of menstrual cycles, however a gold standard algorithm has not been identified. Disagreement between algorithms may influence study results and interpretations. Excessive alcohol consumption and physical activity can have negative effects on menstrual cycle function, including ovulation. Less clear however are the effects of moderate alcohol consumption and moderate amounts of physical activity. The purpose of this dissertation was to evaluate different methods for classifying ovulation status, and to explore modifiable risk factors, including alcohol consumption and physical activity, for anovulation and menstrual cycle dysfunction. We used data from a large prospective cohort study, The BioCycle Study, which followed 259 healthy premenopausal women for 1-2 menstrual cycles. Estradiol, progesterone, luteinizing hormone, follicle-stimulating hormone, and sex hormone binding globulin (SHBG) were measured in serum up to 8 times per cycle, timed using fertility monitors to capture relevant cycle phase data. Fiber intake was assessed with multiple 24-h dietary recalls. Alcohol and physical activity were assessed with daily diaries and baseline questionnaires. We calculated the prevalence of anovulation and assessed the effect of fiber consumption on anovulation using each of the 8 algorithms and compared odds ratios (ORs) and 95% confidence intervals (CI). We used linear mixed models to determine the effect of alcohol and physical activity on menstrual cycle hormone values, and generalized linear mixed models to evaluate their effect on anovulation. The prevalence of anovulation ranged from 3.3% to 17.5% across the 8 algorithms. In multivariate analyses of fiber consumption and risk of anovulation, fiber intake was positively associated with anovulation across all algorithms, however precision varied widely; when comparing first to forth quartiles, odds ratios ranged from 3.39 (95% CI: 1.01-11.46) to 7.52 (95% CI: 0.77-73.25). Both physical activity and alcohol were associated with decreased levels of SHBG, but only in the follicular and luteal phases, and had no effect on other hormones or ovulation. The results of this study suggest that moderate levels of physical activity do not have substantial effects on reproductive hormone levels or ovulation status, aside from a slight decrease in SHBG levels.^

Maternal and Fetal Factors Associated with Labor and Delivery Complications

Gawade, Prasad L. — Tue, 06 Mar 2012 10:02:28 PST

Prolonged second stage of labor, excessive gestational weight gain and cesarean delivery has been associated with adverse maternal and fetal outcomes. Physical activity during pregnancy is a modifiable risk factor which has never been studied among Hispanic women. Gestational weight gain, another modifiable risk factor has only been evaluated as a risk factor for cesarean delivery in two studies among women induced for labor. To date, no study has examined the effect of duration of second stage of labor on intra-ventricular hemorrhage in very preterm births. We examined these maternal risk factors for prolonged second stage of labor, rate of cesarean delivery and fetal outcomes. The first study evaluated the association between physical activity and duration of second stage of labor. Prior studies regarding physical activity and duration of second stage of labor have been conflicting and none have examined the Hispanic population. During pregnancy, activities such as household chores, childcare, sports and women's occupation constitute a significant proportion of physical activity but have not been considered in prior studies. We examined the association between total physical activity (occupational, sport/exercise, household/care giving, and active living) during pre, early and mid-pregnancy and duration of second stage of labor in a prospective cohort of 1,231 Hispanic participants. Physical activity was quantified using the Kaiser Physical Activity Survey administered during pregnancy. Using multivariate linear regression we did not find statistically significant association between pre, early and mid-pregnancy physical activity and duration of second stage of labor. The second study focused on the effect of gestational weight gain on the cesarean delivery rate after induction of labor. The rate of induction of labor (IOL) has more than doubled from 9.5% in 1990 to 22.5% in 2006. Cesarean delivery usually follows a failed IOL and is associated with maternal and fetal morbidity. One of the two studies evaluating the effect of gestational weight gain on the rate of cesarean section in patients undergoing IOL was restricted to women with normal Body Mass Index (BMI) and the other was subjected to bias because more than half of the patients were missing BMI data. Therefore, we evaluated the effect of gestational weight gain on the rate of cesarean delivery after labor induction. In a retrospective cohort study design, using data from May 2005 to June 2008 and a multivariate logistic regression we found a 13% increase in risk of cesarean delivery with 5 kg increase in gestational weight gain. Finally, we evaluated the effect of mode of delivery and duration of second stage of labor on intra-ventricular hemorrhage (IVH) among early preterm births. IVH is a serious complication associated with preterm birth and important predictors of cerebral palsy and neurodevelopmental delays. Prior studies on this relationship in early preterm births are sparse. In a retrospective cohort study of newborns born less than 30 weeks or less than 1500 g between May 2003 and August 2008, we found an increase in risk of IVH after vaginal delivery. However, duration of second stage of labor had no significant effect on risk of IVH.

Tree-based methods and a mixed ridge estimator for analyzing longitudinal data with correlated predictors

Eliot, Melissa — Wed, 18 Jan 2012 12:11:29 PST

Due to recent advances in technology that facilitate acquisition of multi-parameter defined phenotypes, new opportunities have arisen for predicting patient outcomes based on individual specific cell subset changes. The data resulting from these trials can be a challenge to analyze, as predictors may be highly correlated with each other or related to outcome within levels of other predictor variables. As a result, applying traditional methods like simple linear models and univariate approaches such as odds ratios may be insufficient. In this dissertation, we describe potential solutions including tree-based methods, ridge regression, mixed modeling, and a new estimator called a mixed ridge estimator with expectation-maximization (EM) algorithm. Data examples are provided. ^ In particular, flow cytometry is a method of measuring a large number of particle counts at once by suspending them in a fluid and shining a beam of light onto the fluid. This is specifically relevant in the context of studying human immunodeficiency virus (HIV), where there exists a great potential to draw from the rich array of data on host cell-mediated response to infection and drug exposures, to inform and discover patient level determinants of disease progression and/or response to anti-retroviral therapy (ART). The data sets collected are often high dimensional with correlated columns, which can be challenging to analyze. We demonstrate the application and comparative interpretations of three tree-based algorithms for the analysis of data arising from flow cytometry in the first chapter of this manuscript. Specifically, we consider the question of what best predicts CD4 T-cell recovery in HIV-1 infected persons starting antiretroviral therapy with CD4 count between 200-350 cell/μl. The tree-based approaches, namely, classification and regression trees (CART), random forests (RF) and logic regression (LR), were designed specifically to uncover complex structure in high dimensional data settings. While contingency table analysis and RFs provide information on the importance of each potential predictor variable, CART and LR offer additional insight into the combinations of variables that together are predictive of the outcome. Specifically, application of tree-based methods to our data suggest that a combination of baseline immune activation states, with emphasis on CD8 T cell activation, may be a better predictor than any single T cell/innate cell subset analyzed. In the following chapter, tree-based methods are compared to each other via a simulation study. Each has its merits in particular circumstances; for example, RF is able to identify the order of importance of predictors regardless of whether there is a tree-like structure. It is able to adjust for correlation among predictors by using a machine learning algorithm, analyzing subsets of predictors and subjects over a number of iterations. CART is useful when variables are predictive of outcome within levels of other variables, and is able to find the most parsimonious model using pruning. LR also identifies structure within the set of predictor variables, and nicely illustrates relationship among variables. However, due to the vast number of combinations of predictor variables that would need to be analyzed in order to find the single best LR tree, an algorithm is used that only searches a subset of potential combinations of predictors. Therefore, results may be different each time the algorithm is used on the same data set.^ Next we use a regression approach to analyzing data with correlated predictors. Ridge regression is a method of accounting for correlated data by adding a shrinkage component to the estimators for a linear model. We perform a simulation study to compare ridge regression to linear regression over various correlation coefficients and find that ridge regression outperforms linear regression as correlation increases. To account for collinearity among the predictors along with longitudinal data, a new estimator that combines the applicability of ridge regression and mixed models using an EM algorithm is developed and compared to the mixed model. We find from a simulation study comparing our mixed ridge (MR) approach with a traditional mixed model that our new mixed ridge estimator is able to handle collinearity of predictor variables better than the mixed model, while accounting for random within-subject effects that regular ridge regression does not take into account. As correlation among predictors increases, power decreases more quickly for the mixed model than MR. Additionally, type I error rate is not significantly elevated when the MR approach is taken. The MR estimator gives us new insight into flow cytometry data and other data sets with correlated predictor variables that our tree-based methods could not give us. These methods all provide unique insight into our data that more traditional methods of analysis do not offer.^

Micronutrient Intake and Premenstrual Syndrome

Chocano-Bedoya, Patricia O. — Mon, 05 Dec 2011 10:48:06 PST

Premenstrual syndrome (PMS) is characterized by the presence of physical and psychological symptoms restricted to the late luteal phase of the menstrual cycle and associated with substantial impairment in life activities. In the U.S. about 8 to 15% of women of reproductive age suffer from PMS. Many micronutrients are potentially involved in the development of this disorder due to their role in the synthesis of neurotransmitters and hormones or in their regulation, but few previous studies have evaluated the effects of micronutrients on PMS. The first study examined the association between B vitamin intakes, and PMS development among women participating in the Nurses' Health Study 2 (NHS2). We found that high thiamin and high riboflavin intake from food sources were associated with lower risk of PMS. There were not significant associations between niacin, vitamin B6, folate, and vitamin B12 dietary intake and incident PMS. Intakes of B vitamins from supplements were not associated with lower risk of PMS. The second study evaluated the association between selected mineral intakes and PMS development in the NHS2. In this study, high iron intakes were associated with lower risk of PMS. Although there was no association between zinc and PMS risk, high intake of zinc relative to copper was associated with lower risk of PMS. There were no associations between of magnesium, copper, and manganese intakes and PMS. We observed a significantly higher risk of PMS in women with high intakes of potassium from food sources. The third study focused on the association between dietary intakes of B vitamins, zinc, magnesium, iron, potassium, and sodium and some biomarkers and PMS prevalence among younger women. In this study, we found an association between zinc intake and lower prevalence of PMS. Each 1 mg/d increase in vitamin B6 from foods was associated with a lower PMS symptom score. Blood magnesium levels were higher in women with PMS compared to women without PMS. We observed that intakes of some micronutrients were associated with lower risk of PMS, but further studies should be conducted. This dissertation contributes to the research on modifiable risk factors for PMS.

Factors associated with genital tract trauma at spontaneous vaginal delivery

DeJoy, Susan A — Wed, 19 Oct 2011 12:50:43 PDT

It has long been traditional for nurse-midwives (CNMs) to keep a log or database of the deliveries they attend. Because CNMs now conduct approximately 400,000 or 10% of all births in the US, these databases are a potentially rich source of research information on the birth outcomes of women cared for by CNMs. ^ The first study assessed the validity of 3,133 deliveries in the electronic Baystate Midwifery and Women’s Health (BMWH) Delivery Database from 2001 to 2008, using the patient’s electronic medical record as the ‘gold standard’. There have been only four validation studies of nurse-midwifery delivery databases, and these have been limited by small sample sizes, sparse presentation of results and inadequate statistical methodology. Results from these analyses demonstrated excellent overall agreement and a range of agreement by individual variable; agreement among CNM clinicians, by years of CNM clinical experience, and by early versus late study periods was also excellent. ^ Genital tract trauma is defined as episiotomy and/or genital tract lacerations and is a complication in more than 50% of all vaginal births in the US. Many factors influence the incidence of maternal genital tract trauma. ^ The second study examined the relationship between provider type, gender and years of clinical experience and the risk of major genital tract trauma among 19,261 spontaneous vaginal births from 2001 to 2008 at Baystate Medical Center. Significantly less major genital tract trauma was associated with later time period, CNM versus physician provider type, and greater than five years of clinical experience. Provider gender did not influence risk of maternal major genital tract trauma. ^ Finally, we evaluated the relationship between maternal back position, maternal hip flexion, and four derived maternal back and hip flexion positions, and the risk of major genital tract trauma in a cohort of 2,513 vaginal births occurring in 2008 at Baystate Medical Center. Sitting positions were associated with a statistically significant decrease in major genital tract trauma among births attended by CNMs but not physicians. No significant associations were found for hip flexion or derived maternal positions. ^

Developing best linear unbiased estimator in finite population accounting for measurement error due to interviewer

Zhang, Ruitao — Tue, 21 Dec 2010 10:21:36 PST

Godambe (1955) give a general finite population sampling model and proved that a best linear unbiased estimator (BLUE) of population total does not exist when there is no measurement error. In this research, Godambe’s linear estimator is expanded to include two types of measurement errors and their mixture. We check Godambe’s non-existence theorem and explore the method to develop the best linear unbiased estimator of the latent population total by using individual unbiased constraints and average unbiased constraints, respectively. We start from Godambe’s general framework and then reduce to two probability models which are less general than Godambe’s. The model is developed under unequal probability sampling without replacement. As a special case, the model under simple random sampling without replacement can be derived directly based on the results. The traditional definition of inclusion probability is extended and applied to the unequal probability sampling. These results connect the traditional sampling method and the design-based method using random permutation models based on the work of Royall (1976) as proposed by Stanek, Singer and Lencina (2004). We also examine the relationship among Godambe’s general finite sampling model, the expanded model finite population model and the finite population mixed model. Also, we are able to give another set of solutions by giving a distribution to the sample latent values. The research can serve as the basis for extensions to multi-stage sampling or other complex sampling designs.^

Determinants of peak bone mass in young premenopausal women

Zagarins, Sofija E — Tue, 25 May 2010 11:01:08 PDT

Peak bone mass is defined as the highest level of bone mass attained during life through normal growth, and small increases in peak bone mass may prevent osteoporotic fractures later in life. Non-modifiable genetic factors account for as much as 50-70% of the variation in peak bone mass, but to achieve full genetic potential an individual must work to optimize the modifiable environmental and lifestyle factors that also contribute to the variation in peak bone mass. Thus, understanding these modifiable factors affecting peak bone mass is important for osteoporosis prevention. ^ The first paper of this dissertation examines the association between peak bone mass and body composition. Previous studies of this association are based on size-dependent bone mass measures, so the true association between body composition and peak bone mass has remained unclear. In size-adjusted analyses, we found that lean mass was positively associated peak bone mass, while fat mass measures were inversely associated with peak bone mass. This analysis clarifies that lean mass, and not fat mass, is important to peak bone mass in young premenopausal women. Women in this group should maintain a healthy body fat percentage to ensure attainment and maintenance of optimum peak bone mass.^ The second paper of this dissertation examines the association between overall diet quality and peak bone mass using established indices of diet quality. We found no association between these established diet quality scores and bone mass, suggesting these scores may not be appropriate for use in studies related to bone. Because scores measuring overall diet are important for epidemiologic research, a score specifically tailored to reflect bone-specific dietary components would benefit future research. ^ The third paper of this dissertation evaluates the relative contribution of bone-related clinical and lifestyle factors to peak bone mass. We found that waist circumference explained the most variation in peak bone mass, and that physical activity measures, dietary factors, and age at menarche were also important. These findings inform osteoporosis prevention, and factors identified in this analysis are appropriate for use in a clinical setting as part of a prescreening tool for low peak bone mass.^

Predictors of treatment means for a one factor completely randomized design

Xu, Bo — Tue, 25 May 2010 11:01:07 PDT

A one factor experimental design is developed based on the potential observable outcome framework, sampling from finite population of units and random allocation of treatments. No assumptions are made about the distribution of units and distribution of treatments. We introduce sampling and treatment assignment random variables to represent the joint permutation of the potentially observable population. The joint roles of sampling and treatment assignments are considered. The predictors for treatment means, presented by the observed values in the sample and unobserved values of the remainder, are obtained by Royall’s (1976) prediction theory. We take three cases into account: the latent values correspond to a no interaction model and there are no response errors; the latent values correspond to a model with interaction and there are no response errors; the latent values correspond to a model with interaction and response errors. The predictors with the property of being “shrunk” towards the overall mean are similar to realized random effects in the one way random effect model. If the treatment is not selected in the sample, the predictors correspond to the overall mean. The model is based solely on the population sampling and treatment assignments, and provides a design based framework for inference of linear combinations of the treatment means. The population of treatments can be of small size and up to all of the treatments can be assigned to the samples. This model extended the randomization model developed by Kempthorne (1952) via introducing random allocation of the treatments. Theoretically the predictors provide smaller MSEs than using the linear combination of sample means. When the variance components are unknown, the empirical predictors are considered. The confidence intervals for the empirical predictors are calculated using bootstrapping methods. Several bootstrap methods, such as bootstrapping with replacement (BWR) or bootstrapping without replacement (BWO) are introduced. Each bootstrap method is developed to account for the sampling from the finite population of units and random allocation of treatments. Comparisons of the different bootstrapping methods are made methodologically, and via simulation. We discuss these comparisons, and recommend an appropriate bootstrapping method for statistical inference in the one factor experimental design.^