Multiplicity-adjusted semiparametric benefiting subgroup identification in clinical trials.

Background A recent focus in the health sciences has been the development of personalized medicine, which includes determining the population for which a given treatment is effective. Due to limited data, identifying the true benefiting population is a challenging task. To tackle this difficulty, the credible subgroups approach provides a pair of bounding subgroups for the true benefiting subgroup, constructed so that one is contained by the benefiting subgroup while the other contains the benefiting subgroup with high probability. However, the method has so far only been developed for parametric linear models. Methods In this article, we develop the details required to follow the credible subgroups approach in more realistic settings by considering nonlinear and semiparametric regression models, supported for regulatory science by conditional power simulations. We also present an improved multiple testing approach using a step-down procedure. We evaluate our approach via simulations and apply it to data from four trials of Alzheimer's disease treatments carried out by AbbVie. Results Semiparametric modeling yields credible subgroups that are more robust to violations of linear treatment effect assumptions, and careful choice of the population of interest as well as the step-down multiple testing procedure result in a higher rate of detection of benefiting types of patients. The approach allows us to identify types of patients that benefit from treatment in the Alzheimer's disease trials. Conclusion Attempts to identify benefiting subgroups of patients in clinical trials are often met with skepticism due to a lack of multiplicity control and unrealistically restrictive assumptions. Our proposed approach merges two techniques, credible subgroups, and semiparametric regression, which avoids these problems and makes benefiting subgroup identification practical and reliable.

Design and analysis of Bayesian adaptive crossover trials for evaluating contact lens safety and efficacy.

A crossover study, also referred to as a crossover trial, is a form of longitudinal study. Subjects are randomly assigned to different arms of the study and receive different treatments sequentially. While there are many frequentist methods to analyze data from a crossover study, random effects models for longitudinal data are perhaps most naturally modeled within a Bayesian framework. In this article, we introduce a Bayesian adaptive approach to crossover studies for both efficacy and safety endpoints using Gibbs sampling. Using simulation, we find our approach can detect a true difference between two treatments with a specific false-positive rate that we can readily control via the standard equal-tail posterior credible interval. We then illustrate our Bayesian approaches using real data from Johnson & Johnson Vision Care, Inc. contact lens studies. We then design a variety of Bayesian adaptive predictive probability crossover studies for single and multiple continuous efficacy endpoints, indicate their extension to binary safety endpoints, and investigate their frequentist operating characteristics via simulation. The Bayesian adaptive approach emerges as a crossover trials tool that is useful yet surprisingly overlooked to date, particularly in contact lens development.

A practical Bayesian stepped wedge design for community-based cluster-randomized clinical trials: The British Columbia Telehealth Trial.

BACKGROUND: Many clinical trial designs are impractical for community-based clinical intervention trials. Stepped wedge trial designs provide practical advantages, but few descriptions exist of their clinical implementational features, statistical design efficiencies, and limitations. OBJECTIVES: Enhance efficiency of stepped wedge trial designs by evaluating the impact of design characteristics on statistical power for the British Columbia Telehealth Trial. METHODS: The British Columbia Telehealth Trial is a community-based, cluster-randomized, controlled clinical trial in rural and urban British Columbia. To determine the effect of an Internet-based telehealth intervention on healthcare utilization, 1000 subjects with an existing diagnosis of congestive heart failure or type 2 diabetes will be enrolled from 50 clinical practices. Hospital utilization is measured using a composite of disease-specific hospital admissions and emergency visits. The intervention comprises online telehealth data collection and counseling provided to support a disease-specific action plan developed by the primary care provider. The planned intervention is sequentially introduced across all participating practices. We adopt a fully Bayesian, Markov chain Monte Carlo-driven statistical approach, wherein we use simulation to determine the effect of cluster size, sample size, and crossover interval choice on type I error and power to evaluate differences in hospital utilization. RESULTS: For our Bayesian stepped wedge trial design, simulations suggest moderate decreases in power when crossover intervals from control to intervention are reduced from every 3 to 2 weeks, and dramatic decreases in power as the numbers of clusters decrease. Power and type I error performance were not notably affected by the addition of nonzero cluster effects or a temporal trend in hospitalization intensity. CONCLUSION/LIMITATIONS: Stepped wedge trial designs that intervene in small clusters across longer periods can provide enhanced power to evaluate comparative effectiveness, while offering practical implementation advantages in geographic stratification, temporal change, use of existing data, and resource distribution. Current population estimates were used; however, models may not reflect actual event rates during the trial. In addition, temporal or spatial heterogeneity can bias treatment effect estimates.

Do alcohol compliance checks decrease underage sales at neighboring establishments?

OBJECTIVE: Underage alcohol compliance checks conducted by law enforcement agencies can reduce the likelihood of illegal alcohol sales at checked alcohol establishments, and theory suggests that an alcohol establishment that is checked may warn nearby establishments that compliance checks are being conducted in the area. In this study, we examined whether the effects of compliance checks diffuse to neighboring establishments. METHOD: We used data from the Complying with the Minimum Drinking Age trial, which included more than 2,000 compliance checks conducted at more than 900 alcohol establishments. The primary outcome was the sale of alcohol to a pseudo-underage buyer without the need for age identification. A multilevel logistic regression was used to model the effect of a compliance check at each establishment as well as the effect of compliance checks at neighboring establishments within 500 m (stratified into four equal-radius concentric rings), after buyer, license, establishment, and community-level variables were controlled for. RESULTS: We observed a decrease in the likelihood of establishments selling alcohol to underage youth after they had been checked by law enforcement, but these effects quickly decayed over time. Establishments that had a close neighbor (within 125 m) checked in the past 90 days were also less likely to sell alcohol to young-appearing buyers. The spatial effect of compliance checks on other establishments decayed rapidly with increasing distance. CONCLUSIONS: Results confirm the hypothesis that the effects of police compliance checks do spill over to neighboring establishments. These findings have implications for the development of an optimal schedule of police compliance checks.

A randomized controlled trial comparing health and quality of life of lung transplant recipients following nurse and computer-based triage utilizing home spirometry monitoring.

BACKGROUND: Lung transplantation is now a standard intervention for patients with advanced lung disease. Home monitoring of pulmonary function and symptoms has been used to follow the progress of lung transplant recipients in an effort to improve care and clinical status. The study objective was to determine the relative performance of a computer-based Bayesian algorithm compared with a manual nurse decision process for triaging clinical intervention in lung transplant recipients participating in a home monitoring program. MATERIALS AND METHODS: This randomized controlled trial had 65 lung transplant recipients assigned to either the Bayesian or nurse triage study arm. Subjects monitored and transmitted spirometry and respiratory symptoms daily to the data center using an electronic spirometer/diary device. Subjects completed the Short Form-36 (SF-36) survey at baseline and after 1 year. End points were change from baseline after 1 year in forced expiratory volume at 1 s (FEV1) and quality of life (SF-36 scales) within and between each study arm. RESULTS: There were no statistically significant differences between groups in FEV1 or SF-36 scales at baseline or after 1 year.: Results were comparable between nurse and Bayesian system for detecting changes in spirometry and symptoms, providing support for using computer-based triage support systems as remote monitoring triage programs become more widely available. CONCLUSIONS: The feasibility of monitoring critical patient data with a computer-based decision system is especially important given the likely economic constraints on the growth in the nurse workforce capable of providing these early detection triage services.

The association between density of alcohol establishments and violent crime within urban neighborhoods.

BACKGROUND: Numerous studies have found that areas with higher alcohol establishment density are more likely to have higher violent crime rates, but many of these studies did not assess the differential effects of type of establishments or the effects on multiple categories of crime. In this study, we assess whether alcohol establishment density is associated with 4 categories of violent crime and whether the strength of the associations varies by type of violent crime and by on-premise establishments (e.g., bars, restaurants) versus off-premise establishments (e.g., liquor and convenience stores). METHODS: Data come from the city of Minneapolis, Minnesota in 2009 and were aggregated and analyzed at the neighborhood level. Across the 83 neighborhoods in Minneapolis, we examined 4 categories of violent crime: assault, rape, robbery, and total violent crime. We used a Bayesian hierarchical inference approach to model the data, accounting for spatial auto-correlation and controlling for relevant neighborhood demographics. Models were estimated for total alcohol establishment density as well as separately for on-premise establishments and off-premise establishments. RESULTS: Positive, statistically significant associations were observed for total alcohol establishment density and each of the violent crime outcomes. We estimate that a 3.9 to 4.3% increase across crime categories would result from a 20% increase in neighborhood establishment density. The associations between on-premise density and each of the individual violent crime outcomes were also all positive and significant and similar in strength as for total establishment density. The relationships between off-premise density and the crime outcomes were all positive but not significant for rape or total violent crime, and the strength of the associations was weaker than those for total and on-premise density. CONCLUSIONS: Results of this study, combined with earlier findings, provide more evidence that community leaders should be cautious about increasing the density of alcohol establishments within their neighborhoods.

Is the density of alcohol establishments related to nonviolent crime?

OBJECTIVE: We examined the associations between the density of alcohol establishments and five types of nonviolent crime across urban neighborhoods. METHOD: Data from the city of Minneapolis, MN, in 2009 were aggregated and analyzed at the neighborhood level. We examined the association between alcohol establishment density and five categories of nonviolent crime: vandalism, nuisance crime, public alcohol consumption, driving while intoxicated, and underage alcohol possession/consumption. A Bayesian approach was used for model estimation accounting for spatial auto-correlation and controlling for relevant neighborhood demographics. Models were estimated for total alcohol establishment density and then separately for off-premise establishments (e.g., liquor and convenience stores) and on-premise establishments (e.g., bars and restaurants). RESULTS: We found positive associations between density and each crime category. The association was strongest for public consumption and weakest for vandalism. We estimated that a 3.3%-10.9% increase across crime categories would result from a 20% increase in neighborhood establishment density. Similar results were seen for on- and off-premise establishments, although the strength of the associations was lower for off-premise density. CONCLUSIONS: Our results indicate that communities should consider the potential increase in nonviolent crime associated with an increase in the number of alcohol establishments within neighborhoods.

Bayesian hierarchical modeling for detecting safety signals in clinical trials.

Detection of safety signals from clinical trial adverse event data is critical in drug development, but carries a challenging statistical multiplicity problem. Bayesian hierarchical mixture modeling is appealing for its ability to borrow strength across subgroups in the data, as well as moderate extreme findings most likely due merely to chance. We implement such a model for subject incidence (Berry and Berry, 2004 ) using a binomial likelihood, and extend it to subject-year adjusted incidence rate estimation under a Poisson likelihood. We use simulation to choose a signal detection threshold, and illustrate some effective graphics for displaying the flagged signals.

Hierarchical and joint site-edge methods for medicare hospice service region boundary analysis.

Hospice service offers a convenient and ethically preferable health-care option for terminally ill patients. However, this option is unavailable to patients in remote areas not served by any hospice system. In this article, we seek to determine the service areas of two particular cancer hospice systems in northeastern Minnesota based only on death counts abstracted from Medicare billing records. The problem is one of spatial boundary analysis, a field that appears statistically underdeveloped for irregular areal (lattice) data, even though most publicly available human health data are of this type. In this article, we suggest a variety of hierarchical models for areal boundary analysis that hierarchically or jointly parameterize both the areas and the edge segments. This leads to conceptually appealing solutions for our data that remain computationally feasible. While our approaches parallel similar developments in statistical image restoration using Markov random fields, important differences arise due to the irregular nature of our lattices, the sparseness and high variability of our data, the existence of important covariate information, and most importantly, our desire for full posterior inference on the boundary. Our results successfully delineate service areas for our two Minnesota hospice systems that sometimes conflict with the hospices' self-reported service areas. We also obtain boundaries for the spatial residuals from our fits, separating regions that differ for reasons yet unaccounted for by our model.

Bayesian wombling for spatial point processes.

In many applications involving geographically indexed data, interest focuses on identifying regions of rapid change in the spatial surface, or the related problem of the construction or testing of boundaries separating regions with markedly different observed values of the spatial variable. This process is often referred to in the literature as boundary analysis or wombling. Recent developments in hierarchical models for point-referenced (geostatistical) and areal (lattice) data have led to corresponding statistical wombling methods, but there does not appear to be any literature on the subject in the point-process case, where the locations themselves are assumed to be random and likelihood evaluation is notoriously difficult. We extend existing point-level and areal wombling tools to this case, obtaining full posterior inference for multivariate spatial random effects that, when mapped, can help suggest spatial covariates still missing from the model. In the areal case we can also construct wombled maps showing significant boundaries in the fitted intensity surface, while the point-referenced formulation permits testing the significance of a postulated boundary. In the computationally demanding point-referenced case, our algorithm combines Monte Carlo approximants to the likelihood with a predictive process step to reduce the dimension of the problem to a manageable size. We apply these techniques to an analysis of colorectal and prostate cancer data from the northern half of Minnesota, where a key substantive concern is possible similarities in their spatial patterns, and whether they are affected by each patient's distance to facilities likely to offer helpful cancer screening options.

Modeling and detecting potentially ruinous streaks in health expenditures.

The mean of a distribution of medical expenditures in an insured population can be affected significantly by the occurrence of a few high cost cases. This fact leads some organizations that hold the primary risk for the population (e.g., health plans or self-insured employers) to seek reinsurance arrangements that spread the risk of high cost cases across a broader pool. Recently, the private reinsurance market has experienced some difficulties, attributable to information asymmetries between primary risk holders and reinsurers. The disproportionate effect of a few high cost cases also has generated interest in the development of "risk-adjustment" systems that attempt to reduce the difference in health plans' unreimbursed costs either to endogenous management decisions or random chance. We discuss these issues in light of a well-known statistical result regarding the probability of "streaks" in random data. We illustrate problems that can arise and suggest methods to distinguish random streaks from systematic trends.

Insurance status and access to urgent ambulatory care follow-up appointments.

CONTEXT: There is growing pressure to avoid hospitalizing emergency department patients who can be treated safely as outpatients, but this strategy depends on timely access to follow-up care. OBJECTIVE: To determine the association between reported insurance status and access to follow-up appointments for serious conditions that are commonly identified during an emergency department visit. DESIGN, SETTING, AND PARTICIPANTS: Eight research assistants called 499 randomly selected ambulatory clinics in 9 US cities (May 2002-February 2003) and identified themselves as new patients who had been seen in an emergency department and needed an urgent follow-up appointment (within 1 week) for 1 of 3 clinical vignettes (pneumonia, hypertension, or possible ectopic pregnancy). The same person called each clinic twice using the same clinical vignette but different insurance status. MAIN OUTCOME MEASURE: Proportion of callers who were offered an appointment within a week. RESULTS: Of 499 clinics contacted in the final sample, 430 completed the study protocol. Four hundred six (47.2%) of 860 total callers and 277 (64.4%) of 430 privately insured callers were offered appointments within a week. Callers who claimed to have private insurance were more likely to receive appointments than those who claimed to have Medicaid coverage (63.6% [147/231] vs 34.2% [79/231]; difference, 29.4 percentage points; 95% confidence interval, 21.2-37.6; P<.001). Callers reporting private insurance coverage had higher appointment rates than callers who reported that they were uninsured but offered to pay 20 dollars and arrange payment of the balance (65.3% [130/199] vs 25.1% [50/199]; difference, 40.2; 95% confidence interval, 31.4-49.1; P<.001). There were no differences in appointment rates between callers who claimed to have private insurance coverage and those who reportedly were uninsured but willing to pay cash for the entire visit fee (66.3% [132/199] vs 62.8% [125/199]; difference, 3.5; 95% confidence interval -3.7 to 10.8; P = .31). The median charge was 100 dollars (range, 25 dollars-600 dollars). Seventy-two percent of clinics did not attempt to determine the severity of the caller's condition. CONCLUSIONS: Reported insurance status is associated with access to timely follow-up ambulatory care for potentially serious conditions. Having private insurance and being willing to pay cash may not eliminate the difficulty in obtaining urgent follow-up appointments.

Multivariate parametric spatiotemporal models for county level breast cancer survival data.

In clustered survival settings where the clusters correspond to geographic regions, biostatisticians are increasingly turning to models with spatially distributed random effects. These models begin with spatially oriented frailty terms, but may also include further region-level terms in the parametrization of the baseline hazards or various covariate effects (as in a spatially-varying coefficients model). In this paper, we propose a multivariate conditionally autoregressive (MCAR) model as a mixing distribution for these random effects, as a way of capturing correlation across both the regions and the elements of the random effect vector for any particular region. We then extend this model to permit analysis of temporal cohort effects, where we use the term "temporal cohort" to mean a group of subjects all of whom were diagnosed with the disease of interest (and thus, entered the study) during the same time period (say, calendar year). We show how our spatiotemporal model may be efficiently fit in a hierarchical Bayesian framework implemented using Markov chain Monte Carlo (MCMC) computational techniques. We illustrate our approach in the context of county-level breast cancer data from 22 annual cohorts of women living in the state of Iowa, as recorded by the Surveillance, Epidemiology, and End Results (SEER) database. Hierarchical model comparison using the Deviance Information Criterion (DIC), as well as maps of the fitted county-level effects, reveal the benefit of our approach.

Parametric spatial cure rate models for interval-censored time-to-relapse data.

Several recent papers (e.g., Chen, Ibrahim, and Sinha, 1999, Journal of the American Statistical Association 94, 909-919; Ibrahim, Chen, and Sinha, 2001a, Biometrics 57, 383-388) have described statistical methods for use with time-to-event data featuring a surviving fraction (i.e., a proportion of the population that never experiences the event). Such cure rate models and their multivariate generalizations are quite useful in studies of multiple diseases to which an individual may never succumb, or from which an individual may reasonably be expected to recover following treatment (e.g., various types of cancer). In this article we extend these models to allow for spatial correlation (estimable via zip code identifiers for the subjects) as well as interval censoring. Our approach is Bayesian, where posterior summaries are obtained via a hybrid Markov chain Monte Carlo algorithm. We compare across a broad collection of rather high-dimensional hierarchical models using the deviance information criterion, a tool recently developed for just this purpose. We apply our approach to the analysis of a smoking cessation study where the subjects reside in 53 southeastern Minnesota zip codes. In addition to the usual posterior estimates, our approach yields smoothed zip code level maps of model parameters related to the relapse rates over time and the ultimate proportion of quitters (the cure rates).

Comparison of Bayesian, classical, and heuristic approaches in identifying acute disease events in lung transplant recipients.

This study compares a typical heuristic algorithm with classical and Bayesian regression models in ascertaining the presence of acute bronchopulmonary disease events in lung transplant recipients. These models attempt to predict whether an epoch will end in an event, based on the preceding two weeks of data. The data consist of 150 two-week epochs of daily to biweekly spirometry and symptom covariates for 30 subjects over 60 subject-years. Seventy-five 'event' epochs end on a day when an acute bronchopulmonary disease event is documented in the medical record; 75 randomly selected 'non-event' epochs end on a day when no event is documented. The data are partitioned by randomly assigning 15 subjects for training and the remaining 15 subjects for testing. For cross-validation, a second random partition is generated from the same data set. The statistical models are trained and tested on both partitions. For the heuristic algorithm, its historical event classifications on the same test cases are used. Classification performance on both partitions of all models is compared using receiver operating characteristic curves, sensitivity and specificity, and a Shannon information score. Data partition did not appreciably affect statistical model performance. All statistical models, unlike the heuristic algorithm, performed significantly different than chance (family significance < 0.05, Pearson independence chi-square, Bonferroni multiple correction), and better than the heuristic algorithm. The best models were Bayesian changepoint models. Through a clinically oriented discussion, a case classified by all of these algorithms is presented, suggesting the clinical usefulness of the Bayesian approach compared with the classical and heuristic approaches.

On the precision of the conditionally autoregressive prior in spatial models.

Bayesian analyses of spatial data often use a conditionally autoregressive (CAR) prior, which can be written as the kernel of an improper density that depends on a precision parameter tau that is typically unknown. To include tau in the Bayesian analysis, the kernel must be multiplied by tau(k) for some k. This article rigorously derives k = (n - I)/2 for the L2 norm CAR prior (also called a Gaussian Markov random field model) and k = n - I for the L1 norm CAR prior, where n is the number of regions and I the number of "islands" (disconnected groups of regions) in the spatial map. Since I = 1 for a spatial structure defining a connected graph, this supports Knorr-Held's (2002, in Highly Structured Stochastic Systems, 260-264) suggestion that k = (n - 1)/2 in the L2 norm case, instead of the more common k = n/2. We illustrate the practical significance of our results using a periodontal example.

Using hierarchical spatial models for cancer control planning in Minnesota (United States).

OBJECTIVE: Region-specific maps of cancer incidence, mortality, late detection rates, and screening rates can be very helpful in the planning, targeting, and coordination of cancer control activities. Unfortunately, past efforts in this area have been few, and have not used appropriate statistical models that account for the correlation of rates across both neighboring regions and different cancer types. In this article we develop such models, and apply them to the problem of cancer control in the counties of Minnesota during the period 1993-1997. METHODS: We use hierarchical Bayesian spatial statistical methods, implemented using modern Markov chain Monte Carlo computing techniques and software. RESULTS: Our approach results in spatially smoothed maps emphasizing either cancer prevention or cancer outcome for breast, colorectal, and lung cancer, as well as an overall map which combines results from these three individual cancers. CONCLUSIONS: Our methods enable us to produce a more statistically accurate picture of the geographic distribution of important cancer prevention and outcome variables in Minnesota, and appear useful for making decisions regarding targeting cancer control resources within the state.