A novel method for evaluating an interaction effect of correlated continuous covariates in a linear model

Interaction effect is an important scientific interest for many areas of research. Common approach for investigating the interaction effect of two continuous covariates on a response variable is through a cross-product term in multiple linear regression. In epidemiological studies, the two-way analysis of variance (ANOVA) type of method has also been utilized to examine the interaction effect by replacing the continuous covariates with their discretized levels. However, the implications of model assumptions of either approach have not been examined and the statistical validation has only focused on the general method, not specifically for the interaction effect.^ In this dissertation, we investigated the validity of both approaches based on the mathematical assumptions for non-skewed data. We showed that linear regression may not be an appropriate model when the interaction effect exists because it implies a highly skewed distribution for the response variable. We also showed that the normality and constant variance assumptions required by ANOVA are not satisfied in the model where the continuous covariates are replaced with their discretized levels. Therefore, naïve application of ANOVA method may lead to an incorrect conclusion. ^ Given the problems identified above, we proposed a novel method modifying from the traditional ANOVA approach to rigorously evaluate the interaction effect. The analytical expression of the interaction effect was derived based on the conditional distribution of the response variable given the discretized continuous covariates. A testing procedure that combines the p-values from each level of the discretized covariates was developed to test the overall significance of the interaction effect. According to the simulation study, the proposed method is more powerful then the least squares regression and the ANOVA method in detecting the interaction effect when data comes from a trivariate normal distribution. The proposed method was applied to a dataset from the National Institute of Neurological Disorders and Stroke (NINDS) tissue plasminogen activator (t-PA) stroke trial, and baseline age-by-weight interaction effect was found significant in predicting the change from baseline in NIHSS at Month-3 among patients received t-PA therapy.^

A cross-sectional analysis of the association between polycyclic aromatic hydrocarbons and diesel particulate matter exposures and hypertension among individuals of Mexican origin residing in Houston, TX

Background. Research has shown that elevations of only 10 mmHg diastolic blood pressure (BP) and 5 mmHg systolic BP are associated with substantial (as large as 50%) increases in risks for cardiovascular disease, a leading cause of death, worldwide. Epidemiological studies have found that particulate matter (PM) increases blood pressure (BP) and many biological mechanisms which may suggest that the organic matter of PM contributes to the increase in BP. To understand components of PM which may contribute to the increase in BP, this study focuses on diesel particulate matter (DPM) and polycyclic aromatic hydrocarbons (PAHs). To our knowledge, there have been only four epidemiological studies on BP and DPM, and no epidemiological studies on BP and PAHs. ^ Objective. Our objective was to evaluate the association between prevalent hypertension and two ambient exposures: DPM and PAHs amongst the Mano a Mano cohort. ^ Methods. The Mano a Mano cohort which was established by the M.D. Anderson Cancer Center in 2001, is comprised of individuals of Mexican origin residing in Houston, TX. Using geographical information systems, we linked modeled annual estimates of PAHs and DPM at the census track level from the U.S. Environmental Protection Agency's National-Scale Air Toxics Assessment to residential addresses of cohort members. Mixed-effects logistic regression models were applied to determine associations between DPM and PAHs and hypertension while adjusting for confounders. ^ Results. Ambient levels of DPM, categorized into quartiles, were not statistically associated with hypertension and did not indicate a dose response relationship. Ambient levels of PAHs, categorized into quartiles, were not associated with hypertension, but did indicate a dose response relationship in multiple models (for example: Q2: OR = 0.98; 95% CI, 0.73–1.31, Q3: OR = 1.08; 95% CI, 0.82–1.41, Q4: OR = 1.26; 95% CI, 0.94–1.70). ^ Conclusion. This is the first assessment to analyze the relationship between ambient levels of PAHs and hypertension and it is amongst a few studies investigating the association between ambient levels of DPM and hypertension. Future analyses are warranted to explore the effects DPM and PAHs using different categorizations in order to clarify their relationships with hypertension.^

Impact of traffic pollution on telomere length and chronic diseases: Results from the multi-ethnic study of atherosclerosis and the Mexican-American cohort study

Exposure to air pollutants in urban locales has been associated with increased risk for chronic diseases including cardiovascular disease (CVD) and pulmonary diseases in epidemiological studies. The exact mechanism explaining how air pollution affects chronic disease is still unknown. However, oxidative stress and inflammatory pathways have been posited as likely mechanisms. ^ Data from the Multi-Ethnic Study of Atherosclerosis (MESA) and the Mexican-American Cohort Study (2003-2009) were used to examine the following aims, respectively: 1) to evaluate the association between long-term exposure to ambient particulate matter (PM) (PM10 and PM2.5) and nitrogen oxides (NO x) and telomere length (TL) among approximately 1,000 participants within MESA; and 2) to evaluate the association between traffic-related air pollution with self-reported asthma, diabetes, and hypertension among Mexican-Americans in Houston, Texas. ^ Our results from MESA were inconsistent regarding associations between long-term exposure to air pollution and shorter telomere length based on whether the participants came from New York (NY) or Los Angeles (LA). Although not statistically significant, we observed a negative association between long-term air pollution exposure and mean telomere length for NY participants, which was consistent with our hypothesis. Positive (statistically insignificant) associations were observed for LA participants. It is possible that our findings were more influenced by both outcome and exposure misclassification than by the absence of a relationship between pollution and TL. Future studies are needed that include longitudinal measures of telomere length as well as focus on effects of specific constituents of PM and other pollutant exposures on changes in telomere length over time. ^ This research provides support that Mexican-American adults who live near a major roadway or in close proximity to a dense street network have a higher prevalence of asthma. There was a non-significant trend towards an increased prevalence of adult asthma with increasing residential traffic exposure especially for residents who lived three or more years at their baseline address. Even though the prevalence of asthma is low in the Mexican-origin population, it is the fastest growing minority group in the U.S. and we would expect a growing number of Mexican-Americans who suffer from asthma in the future. Future studies are needed to better characterize risks for asthma associated with air pollution in this population.^

The epidemiology of left ventricular outflow tract obstruction defects in Texas, 2002-2008: An update and assessment of effect heterogeneity

Left ventricular outflow tract (LVOT) defects are an important group of congenital heart defects (CHDs) because of their associated mortality and long-term complications. LVOT defects include aortic valve stenosis (AVS), coarctation of aorta (CoA), and hypoplastic left heart syndrome (HLHS). Despite their clinical significance, their etiology is not completely understood. Even though the individual component phenotypes (AVS, CoA, and HLHS) may have different etiologies, they are often "lumped" together in epidemiological studies. Though "lumping" of component phenotypes may improve the power to detect associations, it may also lead to ambiguous findings if these defects are etiologically distinct. This is due to potential for effect heterogeneity across component phenotypes. ^ This study had two aims: (1) to identify the association between various risk factors and both the component (i.e., split) and composite (i.e., lumped) LVOT phenotypes, and (2) to assess the effect heterogeneity of risk factors across component phenotypes of LVOT defects. ^ This study was a secondary data analysis. Primary data were obtained from the Texas Birth Defect Registry (TBDR). TBDR uses an active surveillance method to ascertain birth defects in Texas. All cases of non complex LVOT defects which met our inclusion criteria during the period of 2002–2008 were included in the study. The comparison groups included all unaffected live births for the same period (2002–2008). Data from vital statistics were used to evaluate associations. Statistical associations between selected risk factors and LVOT defects was determined by calculating crude and adjusted prevalence ratio using Poisson regression analysis. Effect heterogeneity was evaluated using polytomous logistic regression. ^ There were a total of 2,353 cases of LVOT defects among 2,730,035 live births during the study period. There were a total of 1,311 definite cases of non-complex LVOT defects for analysis after excluding "complex" cardiac cases and cases associated with syndromes (n=168). Among infant characteristics, males were at a significantly higher risk of developing LVOT defects compared to females. Among maternal characteristics, significant associations were seen with maternal age > 40 years (compared to maternal age 20–24 years) and maternal residence in Texas-Mexico border (compared to non-border residence). Among birth characteristics, significant associations were seen with preterm birth and small for gestation age LVOT defects. ^ When evaluating effect heterogeneity, the following variables had significantly different effects among the component LVOT defect phenotypes: infant sex, plurality, maternal age, maternal race/ethnicity, and Texas-Mexico border residence. ^ This study found significant associations between various demographic factors and LVOT defects. While many findings from this study were consistent with results from previous studies, we also identified new factors associated with LVOT defects. Additionally, this study was the first to assess effect heterogeneity across LVOT defect component phenotypes. These findings contribute to a growing body of literature on characteristics associated with LVOT defects. ^

Maternal exposure to hazardous air pollutants and oral clefts among offspring: Texas, 1999-2008

Birth defects are a leading cause of infant mortality in the United States. About one in 33 births in the United States is diagnosed with birth defects. Common birth defects include neural tube defects, Down syndrome and oral clefts. The present study focused on oral clefts. ^ Oral clefts refer to the malformation of lip, mouth or both. Birth prevalence of oral clefts in Texas is about 11 per 10,000 births. Etiologically, oral clefts have been classified into two groups, cleft lip with or without cleft palate (CL±P) and isolated cleft palate (CP). In spite of their high prevalence and clinical significance, the etiology of oral clefts in humans has not been well understood. Though a number of risk factors have been identified in epidemiological studies, most of them do not explain the majority of the cases. The need to identify novel risk factors associated with oral clefts provided the motivation for this study. The present study focused on maternal exposure to several hazardous air pollutants. A common subgroup of hazardous air pollutants is the volatile organic compounds found in petroleum derivatives. Four important hydrocarbons in this group are benzene, toluene, ethyl benzene and xylenes (BTEX). ^ The specific aim of this study was to evaluate the association between maternal exposure to environmental levels of BTEX and oral clefts among offspring in Texas for the period 1999-2008. ^ A case-control study design was used to assess if maternal exposure to BTEX increased the risk of oral clefts. The Texas Birth Defects Registry provided data on cases of non-syndromic oral clefts delivered in Texas during the period 1999-2008. Census tract level maternal exposure to BTEX concentrations were obtained from the Hazardous Air Pollutant Exposure Model (HAPEM) developed by the U.S. Environmental Protection Agency. Unconditional logistic regression was used to assess the relationship between maternal exposure to BTEX levels and risk of oral clefts in offspring. ^ In the selected population, mothers who had high estimated exposure to any of the BTEX compounds were not more likely to deliver an offspring with oral clefts. Future research efforts will focus on additional birth defects and thorough assessment of additional potential confounders.^

Adolescent tobacco use: Studies of prevalence, factorial validity, and factorial invariance between African Americans and Whites

In the last thirty years, increasing efforts have been made to reduce the prevalence of adolescent tobacco use in the United States. Although the prevalence has declined dramatically over the past decade, there are still sharp differences in adolescent smoking-initiation rates across racial/ethnic groups. Large-scale surveys frequently assess smoking-related attitudes, self-efficacy, and intentions to explain the differences in smoking rates between African Americans and Whites. However, there is little agreement about which constructs are significant. Moreover, the psychometric properties of smoking-related attitude, self-efficacy, and intention constructs have not been fully examined. More studies are needed to understand existing patterns of tobacco use and to validate and fully exploit the constructs' relationship to adolescent smoking initiation across racial/ethnic groups. ^ This dissertation reports on a secondary analysis of data from a large multi-ethnic convenience sample of sixth- through eighth-grade students in 22 schools in East Texas and the city of Houston. The specific aims of this dissertation were to (1) describe smoking and alternate tobacco product use rates by race/ethnicity, gender, age, and grade level (Article 1); (2) test the factorial validity of smoking-related attitudes, self-efficacy, and intentions using confirmatory factor analysis techniques (Article 2); and (3) test the factorial invariance of smoking-related attitudes, self-efficacy, and intentions between African Americans and Whites (Article 3). ^ The prevalence findings confirm the disparities in tobacco use among African American, Hispanic, and White adolescents that other surveys have reported (Article 1). This study also demonstrates the usefulness of examining use patterns of not only cigarettes but also alternative tobacco products in younger multiethnic populations, as well as of providing epidemiological data estimates about different phases of smoking. The confirmatory factor analysis provides evidence of construct validity of attitude, self-efficacy, and intention scales for the multiethnic sample (Article 2). Finally, the factorial invariance analyses indicates that some measures representing smoking-related attitudes, self-efficacy, and intentions may not be appropriate for use among both African Americans and Whites (Article 3). Additional research is needed to further our understanding of the patterns and predictors of youth tobacco use initiation. ^

The association between diabetes and hepatitis C: A systematic review of epidemiological evidence

Background. Several studies have proposed a link between type 2 Diabetes mellitus (DM2) and Hepatitis C infection (HCV) with conflicting results. Since DM2 and HCV have high prevalence, establishing a link between the two may guide further studies aimed at DM2 prevention. A systematic review was conducted to estimate the magnitude and direction of association between DM2 and HCV. Temporality was assessed from cohort studies and case-control studies where such information was available. ^ Methods. MEDLINE searches were conducted for studies that provided risk estimates and fulfill criteria regarding the definition of exposure (HCV) and outcomes (DM2). HCV was defined in terms of method of diagnosis, laboratory technique and method of data collection; DM2 was defined in terms of the classification [World Health Organization (WHO) and American Diabetes Association (ADA)] 1-3 used for diagnosis, laboratory technique and method of data collection. Standardized searches and data abstraction for construction of tables was performed. Unadjusted or adjusted measures of association for individual studies were obtained or calculated from the full text of the studies. Template designed by Dr. David Ramsey. ^ Results. Forty-six studies out of one hundred and nine potentially eligible articles finally met the inclusion and exclusion criteria and were classified separately based on the study design as cross-sectional (twenty four), case-control (fifteen) or cohort studies (seven). The cohort studies showed a three-fold high (confidence interval 1.66–6.29) occurrence of DM2 in individuals with HCV compared to those who were unexposed to HCV and cross sectional studies had a summary odds ratio of 2.53 (1.96, 3.25). In case control studies, the summary odds ratio for studies done in subjects with DM2 was 3.61 (1.93, 6.74); in HCV, it was 2.30 (1.56, 3.38); and all fifteen studies, together, yielded an odds ratio of 2.60 (1.82, 3.73). ^ Conclusion. The above results support the hypothesis that there is an association between DM and HCV. The temporal relationship evident from cohort studies and proposed pathogenic mechanisms also suggest that HCV predisposes patients to development of DM2. Further cohort or prospective studies are needed, however, to determine whether treatment of HCV infections prevents development of DM2.^

Clinical and epidemiological characteristics of breast cancer metastases

Introduction. Breast cancer is a highly variable disease, and long-term outcomes for individual patients are difficult to predict. We evaluated a retrospective cohort of early stage breast cancer (ESBC) patients based on a variety of clinical and epidemiological factors, specifically looking at the distribution of metastasis and associations with these clinical and epidemiological factors. ^ Methods. Patients were derived from the Early Stage Breast Cancer Repository (ESBCR) with a breast cancer diagnosed between 1985 and 2000. We conducted univariate and multivariate analysis of the data to evaluate associations between characteristics and occurrence of overall, visceral, and bone metastasis. Visceral metastasis was defined as lung, liver, peritoneal, lymph node (thoracic, abdominal, pelvis), and contralateral breast cancer. ^ Results. Overall, 394 (16%) patients developed a metastasis. Of these, 83% were visceral and 17% were bone. Multivariate analyses identified the following variables to be associated with metastasis: Any metastasis: age at diagnosis, stage, ER/PR status, hormone treatment, and type of surgery (1)Visceral metastasis: age at diagnosis, stage, hormone treatment, and type of surgery (2) Bone metastasis –Alcohol use, stage, and ER/PR status ^ Discussion/conclusion. ER-/PR- status has previously been found to be associated with bone metastasis, as we confirm in our cohort. We report an association between alcohol use and bone metastasis whereas previous studies find an association with recurrence. Distribution of metastases varies from previous studies. Typically, previous studies reported bone metastasis >30%, yet our study found 17%. Previous studies varied in design, and definition of visceral metastasis. Future research is needed to further elucidate prognostic factors associated with specific metastases A more thorough understanding of what might predict which ESBC patients will develop metastases can help direct future treatment. Future studies of this nature could include the Perou intrinsic subtypes, biomarkers like Ki-67, and genetic analyses such as Oncotype DX or MammaPrint.^

Epidemiological trends of influenza A positive patients hospitalized at St. Luke's Episcopal Hospital, Houston, Texas September, 2009--January, 2010

This study was a descriptive analysis of 437 influenza A positive inpatients and outpatients during the five month period between September, 2009 and January, 2010. The objective of the study was to describe the epidemiological trends of the total influenza A positive population and more specifically the clinical features of patients hospitalized with influenza A at St. Luke's Episcopal Hospital in Houston, Texas from September 2009 through January 2010. Eligible cases were included if they tested positive for influenza A test using the rapid antigen test and/or rRT-PCR. Hospitalized cases were included based on the laboratory confirmation of influenza A as well as hospital admission for at least 24 hours. Data was collected from medical record abstraction and included patient demographics, clinical history and history of chronic disease. Clinical findings in the differential diagnosis that led to laboratory-confirmation of influenza A as well as course of treatment during the hospital admission were summarized. Finally, co-morbid conditions charted during the hospital visit were reviewed and evaluated for associations with influenza A complications. During the study period, forty-eight patients were included in the study of which 27 tested positive for the H1N1 subtype. Females were more likely to be hospitalized than men. The median age of all patients admitted to St. Luke's Episcopal Hospital with influenza A was 42. The distribution for admitted cases was 15 White, 15 Black, and 18 Hispanic. Patients with co-morbid disease constituted 81% of the admissions for Influenza A. The presence of an underlying medical condition remains a risk factor for both seasonal and H1N1 influenza. Although respiratory conditions such as asthma and COPD are commonly associated with complications of seasonal influenza, patients with metabolic disorders such as kidney disease and/or diabetes were admitted more frequently (58%) during the study period. The patients in the study also of a much younger age than the age that is usually associated with complications of influenza infection, i.e. no patients greater than 65 years of age were admitted with a diagnosis of influenza A. Lower infection rates among elderly populations were similarly reported in other studies of influenza during the same time period. Older patient populations may benefit from antibodies to previous H1N1 strains that have circulated during the twentieth century, whereas younger age groups lack these exposures.^