A flexible approach for causal inference with multiple treatments and clustered survival outcomes.

When drawing causal inferences about the effects of multiple treatments on clustered survival outcomes using observational data, we need to address implications of the multilevel data structure, multiple treatments, censoring, and unmeasured confounding for causal analyses. Few off-the-shelf causal inference tools are available to simultaneously tackle these issues. We develop a flexible random-intercept accelerated failure time model, in which we use Bayesian additive regression trees to capture arbitrarily complex relationships between censored survival times and pre-treatment covariates and use the random intercepts to capture cluster-specific main effects. We develop an efficient Markov chain Monte Carlo algorithm to draw posterior inferences about the population survival effects of multiple treatments and examine the variability in cluster-level effects. We further propose an interpretable sensitivity analysis approach to evaluate the sensitivity of drawn causal inferences about treatment effect to the potential magnitude of departure from the causal assumption of no unmeasured confounding. Expansive simulations empirically validate and demonstrate good practical operating characteristics of our proposed methods. Applying the proposed methods to a dataset on older high-risk localized prostate cancer patients drawn from the National Cancer Database, we evaluate the comparative effects of three treatment approaches on patient survival, and assess the ramifications of potential unmeasured confounding. The methods developed in this work are readily available in the R $$ \mathsf{R}\kern.15em $$ package riAFTBART $$ \mathsf{riAFTBART} $$ .

Initial Patterns of Prescription Opioid Supply and Risk of Mortality Among Insured Adults in the United States.

OBJECTIVE: To examine the association between initial patterns of prescription opioid supply (POS) and risk of all-cause mortality among an insured opioid-naïve patient population in the United States (US). METHODS: This retrospective observational cohort study used de-identified, administrative health care claims data from a large national insurer (Optum Clinformatics Data Mart) from 2010 to 2015. Participants included insured, cancer-free adults prescribed opioid analgesics. Prescription opioids received during the first 6 months of therapy were used to categorize initial patterns of POS as daily or nondaily. Cox regression was used to estimate the association of initial patterns of POS with all-cause mortality within one year of follow-up, adjusting for baseline covariates to control for confounding. RESULTS: A total of 4,054,417 patients were included, of which 2.75% had incident daily POS; 54.8% were female; median age was 50 years; mean Charlson comorbidity index (CCI) was 0.21 (standard deviation = 0.77); and mean daily morphine milligram equivalent was 34.61 (95% confidence intervals: 34.59, 34.63). There were 2068 more deaths per 100,000 person-years among patients who were prescribed opioids daily than nondaily. After adjusting for baseline covariates, the hazard of all-cause mortality among patients with incident daily POS was nearly twice that among those prescribed nondaily (hazard ratio [HR] = 1.94; 95% confidence intervals: 1.84, 2.04). CONCLUSIONS: Among insured adult patients with noncancer pain, incident chronic POS was associated with a significantly increased risk of all-cause mortality over at most 1 year of follow-up. Because these results may be susceptible to bias, more research is needed to establish causality.

Classification using ensemble learning under weighted misclassification loss.

Binary classification rules based on covariates typically depend on simple loss functions such as zero-one misclassification. Some cases may require more complex loss functions. For example, individual-level monitoring of HIV-infected individuals on antiretroviral therapy requires periodic assessment of treatment failure, defined as having a viral load (VL) value above a certain threshold. In some resource limited settings, VL tests may be limited by cost or technology, and diagnoses are based on other clinical markers. Depending on scenario, higher premium may be placed on avoiding false-positives, which brings greater cost and reduced treatment options. Here, the optimal rule is determined by minimizing a weighted misclassification loss/risk. We propose a method for finding and cross-validating optimal binary classification rules under weighted misclassification loss. We focus on rules comprising a prediction score and an associated threshold, where the score is derived using an ensemble learner. Simulations and examples show that our method, which derives the score and threshold jointly, more accurately estimates overall risk and has better operating characteristics compared with methods that derive the score first and the cutoff conditionally on the score especially for finite samples.

Evaluating the Population Impact on Racial/Ethnic Disparities in HIV in Adulthood of Intervening on Specific Targets: A Conceptual and Methodological Framework.

Reducing racial/ethnic disparities in human immunodeficiency virus (HIV) disease is a high priority. Reductions in HIV racial/ethnic disparities can potentially be achieved by intervening on important intermediate factors. The potential population impact of intervening on intermediates can be evaluated using observational data when certain conditions are met. However, using standard stratification-based approaches commonly employed in the observational HIV literature to estimate the potential population impact in this setting may yield results that do not accurately estimate quantities of interest. Here we describe a useful conceptual and methodological framework for using observational data to appropriately evaluate the impact on HIV racial/ethnic disparities of interventions. This framework reframes relevant scientific questions in terms of a controlled direct effect and estimates a corresponding proportion eliminated. We review methods and conditions sufficient for accurate estimation within the proposed framework. We use the framework to analyze data on 2,329 participants in the CFAR [Centers for AIDS Research] Network of Integrated Clinical Systems (2008-2014) to evaluate the potential impact of universal prescription of and ≥95% adherence to antiretroviral therapy on racial disparities in HIV virological suppression. We encourage the use of the described framework to appropriately evaluate the potential impact of targeted interventions in addressing HIV racial/ethnic disparities using observational data.

Mediterranean and Dietary Approaches to Stop Hypertension dietary patterns and risk of sudden cardiac death in postmenopausal women.

BACKGROUND: The Mediterranean and Dietary Approaches to Stop Hypertension (DASH) diets are characterized by higher intake of fruit, vegetables, whole grains, and unsaturated fatty acids. All of these foods and nutrients may affect cholesterol, inflammation, the development of atherosclerosis, and, therefore, risk of cardiac death. OBJECTIVE: Our objective was to examine the association between the Mediterranean and DASH dietary patterns and risk of sudden cardiac death (SCD) in women. DESIGN: We used a prospective cohort of 93,122 postmenopausal women enrolled in the Women's Health Initiative study between 1993 and 1998 and followed for an average of 10.5 y. Women completed a food-frequency questionnaire (FFQ) twice during follow-up. We scored their diets according to how closely the reported diet resembled each dietary pattern. SCD was defined as death that occurred within 1 h of symptom onset. RESULTS: A higher Mediterranean diet score was associated with lower risk of SCD (HR: 0.64; 95% CI: 0.43, 0.94) when women in the highest quintile were compared with women in the lowest quintile after adjustment for age, total energy, race, income, smoking, and physical activity. After adjustment for potential mediators, the association was similar (HR: 0.67; 95% CI: 0.46, 0.99). A higher DASH diet score was not associated with risk of SCD. However, sodium intake, which is a crucial component of the DASH dietary pattern, was not well characterized by the FFQ. CONCLUSION: The Mediterranean dietary pattern may be associated with lower risk of SCD in women. This trial was registered at clinicaltrials.gov as NCT00000611.

Long-term alcohol and caffeine intake and risk of sudden cardiac death in women.

BACKGROUND: Alcohol and caffeine intakes may play a role in the development of sudden cardiac death (SCD) because of their effects on cholesterol, blood pressure, heart rate variability, and inflammation. OBJECTIVE: Our objective was to examine the association between long-term alcohol and caffeine intakes and risk of SCD in women. DESIGN: We examined 93,676 postmenopausal women who participated in the Women's Health Initiative Observational Study. Women were enrolled between 1993 and 1998 and were followed until August 2009. Women completed a food-frequency questionnaire at baseline and again at year 3. We modeled exposure to alcohol 3 ways: by using baseline intake only, a cumulative average of baseline and year 3 intake, and the most recent reported intake (a simple time-varying analysis). RESULTS: Intake of 5-15 g alcohol/d (about one drink) was associated with a nonsignificantly reduced risk of SCD compared with 0.1-5 g/d of baseline intake (HR: 0.64; 95% CI: 0.40, 1.02), of cumulative average intake (HR: 0.69; 95% CI: 0.43, 1.11), and of most recent intake (HR: 0.58; 95% CI: 0.35, 0.96), with adjustment for age, race, income, smoking, body mass index, physical activity, hormone use, and total energy. No association was found between SCD and total caffeine intake (mg/d) or cups of caffeinated coffee, decaffeinated coffee, and caffeinated tea. CONCLUSIONS: Our results suggest that about one drink per day (or 5.1-15 g/d) may be associated with a reduced risk of SCD in this population; however, this association was only statistically significant for a model using the most recent alcohol intake. Total caffeine, regular coffee, decaffeinated coffee, and regular tea intake were not associated with the risk of SCD. This trial was registered at clinicaltrials.gov as NCT00000611.

Prenatal factors for childhood blood pressure mediated by intrauterine and/or childhood growth?

OBJECTIVE: Some prenatal factors may program an offspring's blood pressure, but existing evidence is inconclusive and mechanisms remain unclear. We examined the mediating roles of intrauterine and childhood growth in the associations between childhood systolic blood pressure (SBP) and 5 potentially modifiable prenatal factors: maternal smoking during pregnancy; prepregnancy BMI; pregnancy weight gain; chronic hypertension; and preeclampsia-eclampsia. METHODS: The sample contained 30 461 mother-child pairs in the Collaborative Perinatal Project. Prenatal data were extracted from obstetric forms, and children's SBP was measured at 7 years of age. Potential mediation by intrauterine growth restriction (IUGR) and childhood growth was examined by the causal step method. RESULTS: Heavy maternal smoking during pregnancy was significantly associated with higher offspring SBP (adjusted mean difference versus nonsmoking: 0.73 mm Hg [95% confidence interval (CI): 0.32-1.14]), which attenuated to null (0.13 [95% CI: -0.27-0.54]) after adjustment for changes in BMI from birth to 7 years of age. Prepregnancy overweight-obesity was significantly associated with higher offspring SBP (versus normal weight: 0.89 mm Hg [95% CI: 0.52-1.26]), which also attenuated to null (-0.04 mm Hg [95% CI: -0.40-0.31]) after adjustment for childhood BMI trajectory. Adjustment for BMI trajectory augmented the association between maternal pregnancy weight gain and offspring SBP. Adjustment for childhood weight trajectory similarly changed these associations. However, all these associations were independent of IUGR. CONCLUSIONS: Childhood BMI and weight trajectory, but not IUGR, may largely mediate the associations of maternal smoking during pregnancy and prepregnancy BMI with an offspring's SBP.

Birth weight and adult hypercholesterolemia: subgroups of small-for-gestational-age based on maternal smoking status during pregnancy.

BACKGROUND: Being born small-for-gestational-age (SGA) is associated with hypercholesterolemia in later life. It is possible that only certain subgroups of SGA are at elevated risk for hypercholesterolemia. We examined the associations between SGA subgroups based on levels of maternal smoking during pregnancy and adult hypercholesterolemia. METHODS: A subsample of 1625 adult offspring from the Collaborative Perinatal Project were followed at mean age 39 years. Subjects were classified by recorded fetal growth and maternal smoking status during pregnancy. Clinical diagnosis of hypercholesterolemia was obtained in interviews. RESULTS: Compared with the appropriate-for-gestational-age subgroup without maternal smoking during pregnancy, only SGA subgroups with maternal smoking during pregnancy had higher risk of hypercholesterolemia: for heavy smoking, adjusted hazard ratio = 2.5 (95% confidence interval = 1.4-4.3); moderate smoking, 1.7 (1.0-2.8); nonsmoking, 1.1 (0.5-2.1). CONCLUSION: Only SGA infants whose mothers smoked during pregnancy had elevated risk of hypercholesterolemia in adulthood.

Predictors of the initiation of HIV postexposure prophylaxis in Rhode Island emergency departments.

The objective of this study was to elucidate factors that predicted the initiation of HIV postexposure prophylaxis (PEP) for blood or body fluid exposures evaluated at Rhode Island emergency departments (EDs). The study involved a retrospective review of patient visits to all civilian Rhode Island EDs for these exposures from 1995 to mid-2001. Multivariate logistic regression models were created to evaluate predictors of the offering and the acceptance and receipt of HIV PEP from 1996 to 2001. The search identified 3622 patients who sustained a blood or body fluid exposure. Of these, 43.8% were health care workers (HCWs) and 57.2% were not HCWs. Most (52.0%) of the exposures were nonsexual. HIV PEP was offered to 21.0% and accepted and received by 9.4% of all patients. HIV PEP was offered more often after significant exposures, exposures to known HIV-infected sources, when time elapsed after the exposure was shorter, if the patients were HCWs, adults, presented to a teaching hospital, presented during the latter years of the study, or sustained nonsexual exposures. Once offered HIV PEP, patients who were male, adult, sustained a significant exposure, knew the source was HIV infected, sustained a nonsexual exposure, or were HCWs had a greater odds of accepting and receiving HIV PEP. Even when controlling for exposure significance, HIV status, and time elapsed since the exposure, several factors such as gender and type of hospital that are unrelated to the exposure appeared to influence the initiation of HIV PEP. ED providers should ensure that these factors do not inappropriately restrict its initiation.

Principal stratification with predictors of compliance for randomized trials with 2 active treatments.

In behavioral medicine trials, such as smoking cessation trials, 2 or more active treatments are often compared. Noncompliance by some subjects with their assigned treatment poses a challenge to the data analyst. The principal stratification framework permits inference about causal effects among subpopulations characterized by potential compliance. However, in the absence of prior information, there are 2 significant limitations: (1) the causal effects cannot be point identified for some strata and (2) individuals in the subpopulations (strata) cannot be identified. We propose to use additional information-compliance-predictive covariates-to help identify the causal effects and to help describe characteristics of the subpopulations. The probability of membership in each principal stratum is modeled as a function of these covariates. The model is constructed using marginal compliance models (which are identified) and a sensitivity parameter that captures the association between the 2 marginal distributions. We illustrate our methods in both a simulation study and an analysis of data from a smoking cessation trial.

Human papillomavirus capsid antibody response to natural infection and risk of subsequent HPV infection in HIV-positive and HIV-negative women.

The association between seropositivity to virus-like particles (VLP) of human papillomavirus (HPV) types 16, 18, 31, 35, or 45 and subsequent cervical HPV infection was examined in 829 women with HIV and 413 risk-matched HIV-negative women. We found no statistically significant differences between HPV-seropositive and HPV-seronegative women in the risk of a new infection with the homologous HPV type, with the exception of a reduced risk of HPV 45 infections 4.5 years beyond the baseline serology measurement in HIV-positive women [hazard ratio, 0.21; 95% confidence interval (CI), 0.05-0.89]. Among HIV-negative women, HPV seropositivity was not associated with a statistically significant reduced risk of infections with related viruses in the HPV 16, HPV 18, or "other" HPV groups. Among HIV-positive women, HPV seropositivity was associated with a slightly increased risk of infection with group-related viruses, but the differences were only statistically significant for infection with HPV 16 group viruses (hazard ratio, 1.6; 95% CI, 1.1-2.3) in HPV 18-seropositive women and for infections with "other" HPV group viruses in HPV 31-seropositive women (hazard ratio, 1.45; 95% CI, 1.0-2.0). The lack of a protective immune effect from natural infection is most likely due to the low level of antibody elicited by natural HPV infection and/or the potential for reactivation of HPV, especially in HIV-positive women.

Longitudinal study of the inception of perimenopause in relation to lifetime history of sexual or physical violence.

STUDY OBJECTIVE: To investigate of the extent to which violence over the life course accelerates the onset of perimenopause, as measured by menstrual changes. DESIGN: Prospective cohort study. SETTING: Boston, Massachusetts. PARTICIPANTS: 603 premenopausal women aged 36-45 years at baseline who completed a cross sectional survey on childhood and adult violence history. MAIN OUTCOME MEASURE: Time to perimenopause, defined as time in months from baseline interview to a woman's report of (1) an absolute change of at least seven days in menstrual cycle length from baseline or subjective report of menstrual irregularity; (2) a change in menstrual flow amount or duration; or (3) cessation of periods for at least three months, whichever came first. MAIN RESULTS: Experience of abuse was associated with delayed onset of menstrual changes indicative of onset of perimenopause. Women reporting childhood or adolescent abuse entered perimenopause about 35% slower than women who reported no abuse (IRR(adj) = 0.65, 95% CI 0.45 to 0.95) after adjusting for age, age at menarche, ever live birth, ability to maintain living standard, smoking, BMI, and depression. There was a similar association among women who reported first abuse during adulthood (IRR(adj) = 0.72, 95% CI 0.28 to 1.80). These findings persisted when the cohort was restricted to non-depressed women (childhood/adolescent IRR(adj) = 0.57, 95% CI 0.36 to 0.90; adulthood IRR(adj) = 0.63, 95% CI 0.23 to 1.77). CONCLUSIONS: This study is the first longitudinal analysis of the timing of perimenopause to show an association with a history of physical or sexual abuse. Further study of the relation between violence and reproductive aging is needed.

Serum estradiol positively predicts outcomes in patients undergoing in vitro fertilization.

In patients undergoing in vitro fertilization, the presence of higher E(2) levels at the time of hCG administration predict a greater likelihood of ongoing pregnancy.

Handling drop-out in longitudinal studies.

Drop-out is a prevalent complication in the analysis of data from longitudinal studies, and remains an active area of research for statisticians and other quantitative methodologists. This tutorial is designed to synthesize and illustrate the broad array of techniques that are used to address outcome-related drop-out, with emphasis on regression-based methods. We begin with a review of important assumptions underlying likelihood-based and semi-parametric models, followed by an overview of models and methods used to draw inferences from incomplete longitudinal data. The majority of the tutorial is devoted to detailed analysis of two studies with substantial rates of drop-out, designed to illustrate the use of effective methods that are relatively easy to apply: in the first example, we use both semi-parametric and fully parametric models to analyse repeated binary responses from a clinical trial of smoking cessation interventions; in the second, pattern mixture models are used to analyse longitudinal CD4 counts from an observational cohort study of HIV-infected women. In each example, we describe exploratory analyses, model formulation, estimation methodology and interpretation of results. Analyses of incomplete data requires making unverifiable assumptions, and these are discussed in detail within the context of each application. Relevant SAS code is provided.

Prevalence, incidence, and persistence or recurrence of trichomoniasis among human immunodeficiency virus (HIV)-positive women and among HIV-negative women at high risk for HIV infection.

Trichomoniasis has been implicated in the acquisition and transmission of human immunodeficiency virus (HIV) infection. The prevalence, incidence, and persistence or recurrence of trichomoniasis were assessed among HIV-positive women and among HIV-negative women at high risk for HIV infection. A total of 871 HIV-seropositive women and 439 HIV-seronegative women enrolled in the HIV Epidemiology Study (HERS) were seen biannually. The prevalence of trichomoniasis was 9.4%-29.5% among HIV-seropositive women and 8.2%-23.4% among HIV-seronegative women. Prevalence decreased over time, did not vary according to HIV status or CD4 cell count, and was higher among women who reported crack use (P=.02) or cigarette use (P=.02), women who had bacterial vaginosis (P=.02), and those who were black (compared with white women, P<.001). There were no differences, according to HIV status or CD4 cell count, in the adjusted incidence, unadjusted incidence, or persistence or recurrence of trichomoniasis. HIV infection does not make a woman more likely to have prevalent, incident, or persistent or recurrent trichomoniasis.

Predictors of quitting and dropout among women in a clinic-based smoking cessation program.

Most cessation studies assume that dropouts are smokers. Instead, the authors analyzed these outcomes separately using multinomial regression to model the relative risk of quitting versus continued smoking and dropping out. Female (N = 281) smokers were randomly assigned to a 12-week smoking cessation program plus either a 3-times-per-week exercise program or a contact control wellness program. Higher body mass index and longer prior quit attempts predicted cessation. Self-efficacy was associated with a lower likelihood of dropout. Greater nicotine dependence and lower education predicted continued smoking or dropout versus quitting among exercisers. Patterns of smoking, dropping out, and quitting between Weeks 5 and 12 were different between exercisers and controls. Dropouts should be considered as a separate category from smokers.