Semiparametric Estimation in General Repeated Measures Problems

This paper considers a wide class of semiparametric problems with a parametric part for some covariate effects and repeated evaluations of a nonparametric function. Special cases in our approach include marginal models for longitudinal/clustered data, conditional logistic regression for matched case-control studies, multivariate measurement error models, generalized linear mixed models with a semiparametric component, and many others. We propose profile-kernel and backfitting estimation methods for these problems, derive their asymptotic distributions, and show that in likelihood problems the methods are semiparametric efficient. While generally not true, with our methods profiling and backfitting are asymptotically equivalent. We also consider pseudolikelihood methods where some nuisance parameters are estimated from a different algorithm. The proposed methods are evaluated using simulation studies and applied to the Kenya hemoglobin data.

Semiparametric Normal Transformation Models for Spatially Correlated Survival Data

There is an emerging interest in modeling spatially correlated survival data in biomedical and epidemiological studies. In this paper, we propose a new class of semiparametric normal transformation models for right censored spatially correlated survival data. This class of models assumes that survival outcomes marginally follow a Cox proportional hazard model with unspecified baseline hazard, and their joint distribution is obtained by transforming survival outcomes to normal random variables, whose joint distribution is assumed to be multivariate normal with a spatial correlation structure. A key feature of the class of semiparametric normal transformation models is that it provides a rich class of spatial survival models where regression coefficients have population average interpretation and the spatial dependence of survival times is conveniently modeled using the transformed variables by flexible normal random fields. We study the relationship of the spatial correlation structure of the transformed normal variables and the dependence measures of the original survival times. Direct nonparametric maximum likelihood estimation in such models is practically prohibited due to the high dimensional intractable integration of the likelihood function and the infinite dimensional nuisance baseline hazard parameter. We hence develop a class of spatial semiparametric estimating equations, which conveniently estimate the population-level regression coefficients and the dependence parameters simultaneously. We study the asymptotic properties of the proposed estimators, and show that they are consistent and asymptotically normal. The proposed method is illustrated with an analysis of data from the East Boston Ashma Study and its performance is evaluated using simulations.

Fasting plasma glucose predicts survival and angiographic progression in high-risk postmenopausal women with coronary artery disease

BACKGROUND: We studied the association of baseline fasting plasma glucose (FPG) levels with survival and coronary artery disease (CAD) progression among postmenopausal women without unstable angina. METHODS: Women were recruited from seven centers in the Women's Angiographic Vitamin and Estrogen Trial (WAVE) (n = 423). Event follow-up was available for 400 women (65.1 +/- 8.5 years, 66% white, 92% hypertensive, 19% smokers, 67% hypercholesterolemic). Thirty-eight percent of the women had diabetes or FPG > 125 mg/dL, and 21% had a fasting glucose 100-125 mg/dL. Follow-up angiography was performed in 304 women. Cox regression was used to model survival from a composite outcome of death or myocardial infarction (D/MI, 26 events; median follow-up 2.4 years). Angiographic progression was analyzed quantitatively using linear regression accounting for baseline minimum lumen diameter (MLD), follow-up time, and intrasubject correlations using generalized estimating equations. Regression analyses were adjusted for follow-up time, baseline age, treatment assignment, and Framingham risk (excluding diabetes). RESULTS: Women with impaired fasting glucose/diabetes mellitus (IFG/DM) had a relative risk (RR) of D/MI of 4.2 ( p = 0.009). In all women, each 10 mg/dL increase in FPG was associated with an 11% increase ( p < 0.001) in the hazard of D/MI. Each 10 mg/dL increase in FPG was associated with a 6.8 mum decrease in MLD over the follow-up period ( p = 0.005). CONCLUSIONS: Higher FPG is associated with increased risk of D/MI and greater narrowing of the coronary lumen in women with CAD. Aggressive monitoring of glucose levels may be beneficial for secondary CAD prevention.

Physical long-term health problems and mental comorbidity: evidence from Vietnam

OBJECTIVE: The mental health of children living in low-income countries remains a neglected research area despite the high burden of disease. This study is one of the first that examines the effects of long-term physical health problems on child mental health disorders in a low-income country and investigates whether this association is modified by the socio-economic status of the child's family. METHODS: Community-based cross-sectional survey of 975 eight-year-old children from 20 sites in Vietnam. Long-term physical health problems were measured by a caregiver report and included conditions such as anaemia, congenital malformation, physical disability and skin problems. Child mental disorders were assessed using the strengths and difficulties questionnaire (SDQ). Generalised estimating equations models were fitted to explore the association between long-term physical health problems and child mental disorders. RESULTS: Vietnamese children who suffer from long-term physical health problems have odds 2:1 times greater than children without long-term physical health problems of having a mental disorder (95% CI 1.2 to 3.6, p = 0.006). No significant interaction with socio-economic status was found. CONCLUSIONS: This study showed a high burden of mental disorders among physically ill children, re-enforcing the idea that there is "no health without mental health". While this association needs to be explored longitudinally, children with long-term health problems may be a visible group for targeted mental-health interventions.

Uptake of and virological response to antiretroviral therapy among HIV-infected former and current injecting drug users and persons in an opiate substitution treatment programme: the Swiss HIV Cohort Study

OBJECTIVES: The aim of the study was to investigate the influence of continued injecting drug use, enrolment in an opiate substitution treatment programme (OSTP), or cessation of injecting drug use on the uptake and course of antiretroviral therapy (ART). Design A prospective observational study of all participants in the Swiss HIV Cohort Study followed between 1997 and 2006 was carried out. METHODS: We distinguished four groups of former or current injecting drug users (IDUs): (i) abstinent former IDUs; (ii) persons in OSTPs without concomitant injecting drug use; (iii) persons in OSTPs with concomitant injecting drug use; (vi) current IDUs. These groups were compared with a group of patients who had never been IDUs. Factors related to ART uptake and virological endpoints were analysed using logistic generalized estimating equations. RESULTS: We followed 8660 participants for 48 477 person-years; 29.7% were in the IDU HIV transmission group. The likelihood of being on ART at biannual visits was lower among individuals in OSTPs with concomitant injecting drug use [odds ratio (OR) 0.79; 95% confidence interval (CI) 0.71-0.89] and current IDUs (OR 0.80; 95% CI 0.67-0.96), compared with those who had never been IDUs (reference), abstinent former IDUs (OR 1.13; 95% CI 1.02-1.25) and individuals in OSTPs without injecting drug use (OR 1.18; 95% CI 1.06-1.31). The likelihood of suppressed viral replication on ART was similar among those who had never been IDUs, abstinent former IDUs and individuals in an OSTP without injecting drug use, and lower among those in OSTPs with concomitant drug use (OR 0.82; 95% CI 0.72-0.93) and current IDUs (OR 0.81; 0.65-1.00). Adherence to ART was decreased among persons with continued injecting drug use, and correlated with virological outcome. CONCLUSIONS: Uptake of and virological response to ART were improved among abstinent former IDUs and persons in OSTPs without concomitant injecting drug use, compared with persons with continued injecting drug use.

How reliable is an undetectable viral load?

OBJECTIVES: An article by the Swiss AIDS Commission states that patients with stably suppressed viraemia [i.e. several successive HIV-1 RNA plasma concentrations (viral loads, VL) below the limits of detection during 6 months or more of highly active antiretroviral therapy (HAART)] are unlikely to be infectious. Questions then arise: how reliable is the undetectability of the VL, given the history of measures? What factors determine reliability? METHODS: We assessed the probability (henceforth termed reliability) that the n+1 VL would exceed 50 or 1000 HIV-1 RNA copies/mL when the nth one had been <50 copies/mL in 6168 patients of the Swiss HIV Cohort Study who were continuing to take HAART between 2003 and 2007. General estimating equations were used to analyse potential factors of reliability. RESULTS: With a cut-off at 50 copies/mL, reliability was 84.5% (n=1), increasing to 94.5% (n=5). Compliance, the current type of HAART and the first antiretroviral therapy (ART) received (HAART or not) were predictive factors of reliability. With a cut-off at 1000 copies/mL, reliability was 97.5% (n=1), increasing to 99.1% (n=4). Chart review revealed that patients had stopped their treatment, admitted to major problems with compliance or were taking non-HAART ART in 72.2% of these cases. Viral escape caused by resistance was found in 5.6%. No explanation was found in the charts of 22.2% of cases. CONCLUSIONS: After several successive VLs at <50 copies/mL, reliability reaches approximately 94% with a cut-off of 50 copies/mL and approximately 99% with a cut-off at 1000 copies/mL. Compliance is the most important factor predicting reliability.

The Incidence of AIDS-Defining Illnesses at a Current CD4 Count >=200 Cells/ L in the Post-Combination Antiretroviral Therapy Era

Background. Few studies consider the incidence of individual AIDS-defining illnesses (ADIs) at higher CD4 counts, relevant on a population level for monitoring and resource allocation. Methods. Individuals from the Collaboration of Observational HIV Epidemiological Research Europe (COHERE) aged ≥14 years with ≥1 CD4 count of ≥200 µL between 1998 and 2010 were included. Incidence rates (per 1000 person-years of follow-up [PYFU]) were calculated for each ADI within different CD4 strata; Poisson regression, using generalized estimating equations and robust standard errors, was used to model rates of ADIs with current CD4 ≥500/µL. Results. A total of 12 135 ADIs occurred at a CD4 count of ≥200 cells/µL among 207 539 persons with 1 154 803 PYFU. Incidence rates declined from 20.5 per 1000 PYFU (95% confidence interval [CI], 20.0–21.1 per 1000 PYFU) with current CD4 200–349 cells/µL to 4.1 per 1000 PYFU (95% CI, 3.6–4.6 per 1000 PYFU) with current CD4 ≥ 1000 cells/µL. Persons with a current CD4 of 500–749 cells/µL had a significantly higher rate of ADIs (adjusted incidence rate ratio [aIRR], 1.20; 95% CI, 1.10–1.32), whereas those with a current CD4 of ≥1000 cells/µL had a similar rate (aIRR, 0.92; 95% CI, .79–1.07), compared to a current CD4 of 750–999 cells/µL. Results were consistent in persons with high or low viral load. Findings were stronger for malignant ADIs (aIRR, 1.52; 95% CI, 1.25–1.86) than for nonmalignant ADIs (aIRR, 1.12; 95% CI, 1.01–1.25), comparing persons with a current CD4 of 500–749 cells/µL to 750–999 cells/µL. Discussion. The incidence of ADIs was higher in individuals with a current CD4 count of 500–749 cells/µL compared to those with a CD4 count of 750–999 cells/µL, but did not decrease further at higher CD4 counts. Results were similar in patients virologically suppressed on combination antiretroviral therapy, suggesting that immune reconstitution is not complete until the CD4 increases to >750 cells/µL.

A comparison of Markov and generalized linear models of hospital mortality

This paper reports a comparison of three modeling strategies for the analysis of hospital mortality in a sample of general medicine inpatients in a Department of Veterans Affairs medical center. Logistic regression, a Markov chain model, and longitudinal logistic regression were evaluated on predictive performance as measured by the c-index and on accuracy of expected numbers of deaths compared to observed. The logistic regression used patient information collected at admission; the Markov model was comprised of two absorbing states for discharge and death and three transient states reflecting increasing severity of illness as measured by laboratory data collected during the hospital stay; longitudinal regression employed Generalized Estimating Equations (GEE) to model covariance structure for the repeated binary outcome. Results showed that the logistic regression predicted hospital mortality as well as the alternative methods but was limited in scope of application. The Markov chain provides insights into how day to day changes of illness severity lead to discharge or death. The longitudinal logistic regression showed that increasing illness trajectory is associated with hospital mortality. The conclusion is reached that for standard applications in modeling hospital mortality, logistic regression is adequate, but for new challenges facing health services research today, alternative methods are equally predictive, practical, and can provide new insights. ^

Elderly age is not a negative predictive factor for virological response to therapy with pegylated interferon-α and ribavirin in chronic hepatitis C virus patients

BACKGROUND & AIMS: Age is frequently discussed as negative host factor to achieve a sustained virological response (SVR) to antiviral therapy of chronic hepatitis C. However, elderly patients often show advanced fibrosis/cirrhosis as known negative predictive factor. The aim of this study was to assess age as an independent predictive factor during antiviral therapy. METHODS: Overall, 516 hepatitis C patients were treated with pegylated interferon-α and ribavirin, thereof 66 patients ≥60 years. We analysed the impact of host factors (age, gender, fibrosis, haemoglobin, previous hepatitis C treatment) and viral factors (genotype, viral load) on SVR per therapy course by performing a generalized estimating equations (GEE) regression modelling, a matched pair analysis and a classification tree analysis. RESULTS: Overall, SVR per therapy course was 42.9 and 26.1%, respectively, in young and elderly patients with hepatitis C virus (HCV) genotypes 1/4/6. The corresponding figures for HCV genotypes 2/3 were 74.4 and 84%. In the GEE model, age had no significant influence on achieving SVR. In matched pair analysis, SVR was not different in young and elderly patients (54.2 and 55.9% respectively; P = 0.795 in binominal test). In classification tree analysis, age was not a relevant splitting variable. CONCLUSIONS: Age is not a significant predictive factor for achieving SVR, when relevant confounders are taken into account. As life expectancy in Western Europe at age 60 is more than 20 years, it is reasonable to treat chronic hepatitis C in selected elderly patients with relevant fibrosis or cirrhosis but without major concomitant diseases, as SVR improves survival and reduces carcinogenesis.

Prediction of sepsis in the burn intensive care unit patient

Sepsis is a significant cause for multiple organ failure and death in the burn patient, yet identification in this population is confounded by chronic hypermetabolism and impaired immune function. The purpose of this study was twofold: 1) determine the ability of the systemic inflammatory response syndrome (SIRS) and American Burn Association (ABA) criteria to predict sepsis in the burn patient; and 2) develop a model representing the best combination of clinical predictors associated with sepsis in the same population. A retrospective, case-controlled, within-patient comparison of burn patients admitted to a single intensive care unit (ICU) was conducted for the period January 2005 to September 2010. Blood culture results were paired with clinical condition: "positive-sick"; "negative-sick", and "screening-not sick". Data were collected for the 72 hours prior to each blood culture. The most significant predictors were evaluated using logistic regression, Generalized Estimating Equations (GEE) and ROC area under the curve (AUC) analyses to assess model predictive ability. Bootstrapping methods were employed to evaluate potential model over-fitting. Fifty-nine subjects were included, representing 177 culture periods. SIRS criteria were not found to be associated with culture type, with an average of 98% of subjects meeting criteria in the 3 days prior. ABA sepsis criteria were significantly different among culture type only on the day prior (p = 0.004). The variables identified for the model included: heart rate>130 beats/min, mean blood pressure<60 mmHg, base deficit<-6 mEq/L, temperature>36°C, use of vasoactive medications, and glucose>150 mg/d1. The model was significant in predicting "positive culture-sick" and sepsis state, with AUC of 0.775 (p < 0.001) and 0.714 (p < .001), respectively; comparatively, the ABA criteria AUC was 0.619 (p = 0.028) and 0.597 (p = .035), respectively. SIRS criteria are not appropriate for identifying sepsis in the burn population. The ABA criteria perform better, but only for the day prior to positive blood culture results. The time period useful to diagnose sepsis using clinical criteria may be limited to 24 hours. A combination of predictors is superior to individual variable trends, yet algorithms or computer support will be necessary for the clinician to find such models useful. ^

Is H. pylori infection associated with diarrhea in a cohort of 3- to 8-year-old Colombian children?

The Estudio Comunitario sobre la Salud del Niño cohort study followed 326 3- to 8-year-old Colombian children for 4 years to observe the natural history of Helicobacter pylori infection and identify risk factors for acquisition, recurrence and persistence. Acute H. pylori infection during childhood may predispose to other enteric infections and therefore increase the risk of diarrheal disease. This dissertation aimed to estimate the effect of H. pylori infection on the occurrence of diarrhea and parasitic co-infections. The analysis used Generalized Estimating Equations to obtain odds ratios to estimate relative risks for diarrhea and the Zhang-Yu algorithm to estimate relative risks for on parasitic infections. Andersen-Gill models were used to estimate rate ratios for the effect of H. pylori status on the recurrence of parasitic infections. H. pylori status was classified for the entire follow-up duration in 1 of 3 categories: persistently positive, intermittently positive, and persistently negative. Multivariable models included child’s sex, age, symptoms, medication use, and socio-environmental factors. H. pylori infection was weakly and imprecisely associated with diarrheal disease, which occurred at an unexpectedly low frequency in this study. Persistently H. pylori-positive children had a somewhat higher incidence of reported diarrhea than intermittently positive or persistently negative children. Stratified analysis revealed that the presence of specific helminthes modified the effect of persistent H. pylori infection on diarrhea. The incidence of any parasitic infections was higher in children with persistent H. pylori infection relative to those with intermittent or persistently negative status, but this association did not hold when adjusted for the full set of selected covariates. The effects of H. pylori persistent status were similar for the occurrence or recurrence of Giardia duodenalis, Entamoeba histolytica, and Ascaris lumbricoides. These results show that H. pylori frequently co-exists with other parasites in Andean children and suggest that intermittently H. pylori–positive children might be at a lower risk of parasitic infections than persistently positive children. The relationship of H. pylori infection, helminthic infection and diarrheal disease should be further explored in studies that devote more intensive resources to accurate ascertainment of diarrhea.^

Prevalence of Staphylococcus aureus colonization among high-risk HIV populations

Staphylococcus aureus is a common microorganism in humans, typically colonizing the nasopharynx, skin and other mucosal surfaces. It is among the most frequent causes of clinically-significant bacterial infections accounting for increased morbidity and mortality among individuals with HIV/AIDS. Evidence of higher colonization rates among high-risk HIV populations have been observed however, prevalence estimates have varied. Additionally, behavioral, biological, and/or environmental factors that may account for these high colonization rates are not understood. Previous literature on clinic-based surveys were subject to considerable biases. Additionally, representative samples of high-risk HIV populations were difficult to obtain due in part to an underrepresentation of individuals who may not regularly obtain health care. ^ The main objective of this project is to determine the prevalence of methicillin-sensitive S. aureus (MSSA) and methicillin-resistant (MRSA) nasal colonization in two populations: 1) men who have sex with men (MSM) and 2) injection drug users (IDU). Both of these populations are included in the third round of the National HIV Behavioral Surveillance System (NHBS) in Houston, Texas. ^ In the NHBS-MSM3 study, logistic regression was used to report odds ratios and 95% confidence intervals (CI). For the NHBS-IDU3 study, to account for the lack of independence between samples, the method of generalized estimating equations was utilized to report adjusted odds ratios and 95% CI. The NHBS-MSM3 study enrolled 202 participants with a MSSA colonization rate of 26.7% and MRSA rate of 3%. In the NHBS-IDU3 study, 18.4% were nasally colonized with MSSA and 5.7% were nasally colonized with MRSA. Among the NHBS-MSM3 population, high-risk sexual practices were associated with colonization. For the NHBS-IDU3 population, age, marital status, employment status, and the presence of scabs, were associated with colonization status when controlling for size of recruitment network. In multivariate GEE analyses, the use of antiretroviral medications and age remained significantly associated with S. aureus nasal colonization when controlling for size of recruitment network and gender. In both studies, a significantly higher than expected S. aureus and MRSA colonization rate was observed as compared to colonization rates described for the general population. However, these estimates were moderate in comparison to reported clinic-based MSM and IDU S. aureus colonization findings. This study validates substantial prevalence differences and biases that may exist with data collected from clinic-based MSM and IDU. The prevalence of MSSA and MRSA nasal colonization did not differ significantly with respect to HIV status among NHBS-MSM3/NHBS-IDU3 participants. Continued examination on the effects of S. aureus colonization and infection should be examined longitudinally to confirm additional community-based determinants in populations that are disproportionately affected.^

A stochastic approximation algorithm with Markov chain Monte-Carlo method for incomplete data estimation problems

We propose a general procedure for solving incomplete data estimation problems. The procedure can be used to find the maximum likelihood estimate or to solve estimating equations in difficult cases such as estimation with the censored or truncated regression model, the nonlinear structural measurement error model, and the random effects model. The procedure is based on the general principle of stochastic approximation and the Markov chain Monte-Carlo method. Applying the theory on adaptive algorithms, we derive conditions under which the proposed procedure converges. Simulation studies also indicate that the proposed procedure consistently converges to the maximum likelihood estimate for the structural measurement error logistic regression model.

Putting teeth into the developmental origins hypothesis: A longitudinal study of early childhood malnutrition, enamel hypoplasia and adolescent health in Amazonian Bolivia

Thesis (Ph.D.)--University of Washington, 2016-06