Search

In a medical setting, observational studies commonly involve patients who initiate a particular treatment (e.g., medication therapy) and others who do not, and the goal is to draw causal inferences about the effect of treatment on a time-to-event outcome. A difficulty with such studies is that the notion of a treatment initiation time is not well-defined for the control group. In this paper, we propose a Bayesian approach to estimate treatment effects in longitudinal observational studies where treatment is given by indication and thereby the exact timing of treatment is only observed for treated units. We present a framework for conceptualizing an underlying randomized experiment in this setting based on separating the time of indication for treatment, which we model using a latent state-space process, from the mechanism that determines assignment to treatment versus control. Next, we develop a two-step inferential approach that uses Markov Chain Monte Carlo (MCMC) posterior sampling to (1) infer the unobserved indication times for units in the control group, and (2) estimate treatment effects based on inferential conclusions from Step 1. This approach allows us to incorporate uncertainty about the unobserved indication times which induces uncertainty in both the selection of the control group and the measurement of time-to-event outcomes for these controls. We demonstrate our approach to study the effects on mortality of inappropriately prescribing phosphodiesterase type 5 inhibitors (PDE5Is), a medication contraindicated for certain types of pulmonary hypertension, using data from the Veterans Affairs (VA) health care system. [ABSTRACT FROM AUTHOR]

Growing healthcare costs have been accompanied by increased policymakers' interest in the efficiency of healthcare systems. Network formation by hospitals as a vehicle for consolidation and achieving economies of scale has emerged as an important topic of conversation among academics and practitioners. Within networks, consolidation of particular specialties or entire campuses is expected and encouraged to take place. This paper describes the main findings of an effort to build gravity-type models to describe patient choices in inpatient and daycare hospital facilities. It analyzes the distance decay effects as a function of car travel times and great-circle distance, and it offers a method for inclusion of university hospitals. Additionally, it reviews the impact of driving and transit accessibility on hospital attraction and reviews the differences in distance decay for patient age groups and hospitalization types. In the described application, the best models achieve a Mean Absolute Percentage Error of around 10% in non-metropolitan areas, and 14.5% across different region types. Results in metropolitan areas suggest that latent factors unrelated to proximity and size have a significant role in determining hospital choices. Furthermore, the effects of relative driving and transit accessibility are found to be small or non-existent. [ABSTRACT FROM AUTHOR]

Analysis of public policy affecting dual eligibles requires accurate identification of survey respondents eligible for both Medicare and Medicaid. Doing so for Medicaid is particularly challenging given the complex eligibility rules tied to income and assets. In this paper we provide guidance on how to best identify eligible respondents in household surveys that have limited income or asset information, such as the National Health Interview Survey, American Community Survey, Current Population Survey, and Medical Expenditure Panel Survey. We show how two types of errors—false negative and false positive errors—are impacted by incorporating limited income or asset information, relative to the commonly-used approach of solely comparing total income to the income threshold. With the 2018 Health and Retirement Study, which has detailed income and asset information, we mimic the income and asset information available in those other household surveys and quantify how errors change when imposing income or asset tests with limited information. We show that incorporating all available income and asset data results in the lowest number of errors and the lowest overall error rates. We recommend that researchers adjust income and impose the asset test to the fullest extent possible when imputing Medicaid eligibility for Medicare enrollees. [ABSTRACT FROM AUTHOR]

Researchers are often faced with evaluating the effect of a policy or program that was simultaneously initiated across an entire population of units at a single point in time, and its effects over the targeted population can manifest at any time period afterwards. In the presence of data measured over time, Bayesian time series models have been used to impute what would have happened after the policy was initiated, had the policy not taken place, in order to estimate causal effects. However, the considerations regarding the definition of the target estimands, the underlying assumptions, the plausibility of such assumptions, and the choice of an appropriate model have not been thoroughly investigated. In this paper, we establish useful estimands for the evaluation of large-scale policies. We discuss that imputation of missing potential outcomes relies on an assumption which, even though untestable, can be partially evaluated using observed data. We illustrate an approach to evaluate this key causal assumption and facilitate model elicitation based on data from the time interval before policy initiation and using classic statistical techniques. As an illustration, we study the Hospital Readmissions Reduction Program (HRRP), a US federal intervention aiming to improve health outcomes for patients with pneumonia, acute myocardial infraction, or congestive failure admitted to a hospital. We evaluate the effect of the HRRP on population mortality among the elderly across the US and in four geographic subregions, and at different time windows. We find that the HRRP increased mortality from pneumonia and acute myocardial infraction across at least one geographical region and time horizon, and is likely to have had a detrimental effect on public health. [ABSTRACT FROM AUTHOR]

Exploitation of the ever-increasing volume of observational data has become a major challenge for real-world evidence (RWE). Causal inference can be viewed as emulations of clinical trials (ECTs) from real-world data (RWD). This requires in-depth knowledge of data management and modern statistical models, which evolve considerably. We have developed Plug-Stat® to facilitate these analyses. Plug-Stat® is a user-friendly graphical interface of R software implemented with JavaScript and VB.net. The implementation of Plug-Stat® for a specific database consists of two main tasks. The first concerns the automatic importation of data to link the software with the data and to avoid the repetition of data management activities for each study. The second is to define the outcomes and the associated possible confounders. To perform an ECT, we proposed the use of inverse probability weighting, flexible parametric estimation of the propensity score model with B-splines, and robust estimation of the variance. The statistical modeling steps are automatized to direct the user toward the correct methodology without data management. We illustrated the usefulness of Plug-Stat® with several recently published ECTs. In this paper, we presented the outcome comparison of preemptive deceased-donor kidney transplantations versus transplantations after dialysis. The interactive interfaces assist the user in conducting causal research based on ECT, and tutorials are available, offering self-training opportunities. Plug-Stat® reduces the time and, consequently, the cost necessary for different studies from the same cohort. [ABSTRACT FROM AUTHOR]

The nominal group technique (NGT) is a method to elicit healthcare priorities. Yet, there is variability on how to conduct the NGT, and limited guidance on how to analyse a diverse sample of multiple groups. This paper addresses some of this ambiguity, and explores whether different approaches to analysis provide the same outcome/s. Conceptual papers and empirical studies were identified via PubMed and informed an adapted version of the NGT. Twenty-six nominal groups were conducted, which provided in-depth knowledge on how to best conduct this method. Pilot group data were used to compare different analysis methods and to explore how this impacted on reported outcomes. Data analyses for large data-sets are complex; thematic analysis is needed to be able to conduct across group comparisons of participant priorities. Consideration should be given not just to the strength, i.e. sum of votes, or relative importance of the priority, but to the voting frequency, i.e. the popularity of the idea amongst participants; our case study demonstrated that this can affect priority rankings for those ideas with the same score. As a case study, this paper provides practical information on analysis for complex data sets. Researchers need to consider more than one analysis process to ensure that the results truly reflect participant priorities. A priority that has a high score may not necessarily reflect its popularity within the group; the voting frequency may also need to be considered. [ABSTRACT FROM AUTHOR]

Difference-in-difference (DD) methods are a common strategy for evaluating the effects of policies or programs that are instituted at a particular point in time, such as the implementation of a new law. The DD method compares changes over time in a group unaffected by the policy intervention to the changes over time in a group affected by the policy intervention, and attributes the 'difference-in-differences' to the effect of the policy. DD methods provide unbiased effect estimates if the trend over time would have been the same between the intervention and comparison groups in the absence of the intervention. However, a concern with DD models is that the program and intervention groups may differ in ways that would affect their trends over time, or their compositions may change over time. Propensity score methods are commonly used to handle this type of confounding in other non-experimental studies, but the particular considerations when using them in the context of a DD model have not been well investigated. In this paper, we describe the use of propensity scores in conjunction with DD models, in particular investigating a propensity score weighting strategy that weights the four groups (defined by time and intervention status) to be balanced on a set of characteristics. We discuss the conceptual issues associated with this approach, including the need for caution when selecting variables to include in the propensity score model, particularly given the multiple time point nature of the analysis. We illustrate the ideas and method with an application estimating the effects of a new payment and delivery system innovation (an accountable care organization model called the 'Alternative Quality Contract' (AQC) implemented by Blue Cross Blue Shield of Massachusetts) on health plan enrollee out-of-pocket mental health service expenditures. We find no evidence that the AQC affected out-of-pocket mental health service expenditures of enrollees. [ABSTRACT FROM AUTHOR]

Regression, propensity score (PS) and double-robust (DR) methods can reduce selection bias when estimating average treatment effects (ATEs). Economic evaluations of health care interventions exemplify complex data structures, in that the covariate–endpoint relationships tend to be highly non-linear, with highly skewed cost and health outcome endpoints. When either the regression or PS model is correct, DR methods can provide unbiased, efficient estimates of ATEs, but generally the specification of both models is unknown. Regression-adjusted matching can also protect against bias from model misspecification, but has not been compared to DR methods. This paper compares regression-adjusted matching to selected DR methods (weighted regression and augmented inverse probability of treatment weighting) as well as to regression and PS methods for addressing selection bias in cost-effectiveness analyses (CEA). We contrast the methods in a CEA of a pharmaceutical intervention, where there are extreme estimated PSs, hence unstable inverse probability of treatment (IPT) weights. The case study motivates a simulation which considers settings with functional form misspecification in the PS and endpoint regression models (e.g. cost model with log instead of identity link), stable and unstable PS weights. We find that in the realistic setting of unstable IPT weights and misspecifications to the PS and regression models, regression-adjusted matching reports less bias than DR methods. We conclude that regression-adjusted matching is a relatively robust method for estimating ATEs in applications with complex data structures exemplified by CEA. [ABSTRACT FROM AUTHOR]

The purpose of this study is to illustrate how matching approaches based on different balancing scores lead to variations in the treatment effect estimators. We introduced a case study evaluating the effect of anti-thrombotic agents on the severe progression of patients with intracerebral hemorrhage. We extracted subpopulations based on propensity and prognostic scores and then estimated the relative risk between the treatment groups. Furthermore, to illustrate the situation where the treatment effect estimates varied depending on employed balancing scores, we conducted a simulation experiment. In the case study, the matching using different balancing scores extracted subpopulations with different characteristics. Then, the estimated relative risk (95% confidence interval) was 1.27 (0.98–1.94) among the propensity score matched cohort, whereas it was 0.91 (0.76–1.08) among the prognostic score matched cohort. In the simulation experiments, the results indicated that the matching schemes based on different balancing scores created distinct matched cohorts, leading to varying estimates under treatment effect heterogeneity. Moreover, the variability of the estimated effect becomes substantial when there are covariates strongly related to the dependent variable of the nuisance functions. The difference in the selected subpopulation via matching based on different balancing scores is a thoughtful factor that can result in different estimates when there is effect heterogeneity. In practice, we recommend assessing the characteristics of the matched subpopulation and employing the balancing score that can estimate the treatment effect among the target population of interest in each study. Plain Language Summary: This study aimed to understand why different methods of matching patients in a study can lead to different estimates of how well a treatment works. We used a real example of a study where the treatment's effectiveness varied for different patients and compared how different methods of matching patients affected the results. We also did a series of experiments to see how the different methods affected estimates of treatment effect when the effectiveness varied among patients. Our study found that the different matching methods extracted the different patients. The difference in the matched patients led to very different results, especially when the treatment's effectiveness was not the same for all patients. Therefore, it is important to check the characteristics of patients after matching and consider whether the treatment's effectiveness is not the same for all patients. [ABSTRACT FROM AUTHOR]

This article clarifies how the biostatistical literature on time-varying treatments (TVT) can provide tools for dealing with time-varying confounding in difference-in-differences (DiD) studies. I use a simulation study to compare the bias and standard error of inverse probability weighting estimators from the TVT framework, a DiD framework, and hybrid approaches that combine ideas from both frameworks. I simulated longitudinal data with treatment effect heterogeneity over multiple time points using linear and logistic models. Simulation settings looked at both time-invariant confounders and time-varying confounders affected by prior treatment. Estimators that combined ideas from both frameworks had lower bias than standard TVT and DiD estimators when assumptions were unmet. The TVT framework provides estimation tools that can complement DiD tools in a wide range of applied settings. It also provides alternate estimands for consideration in policy settings. [ABSTRACT FROM AUTHOR]

In this paper, we introduce several statistical methods to evaluate the uncertainty in the concentration index ( C) for measuring socioeconomic equality in health and health care using aggregated total population register data. The C is a widely used index when measuring socioeconomic inequality, but previous studies have mainly focused on developing statistical inference for sampled data from population surveys. While data from large population-based or national registers provide complete coverage, registration comprises several sources of error. We simulate confidence intervals for the C with different Monte Carlo approaches, which take into account the nature of the population data. As an empirical example, we have an extensive dataset from the Finnish cause-of-death register on mortality amenable to health care interventions between 1996 and 2008. Amenable mortality has been often used as a tool to capture the effectiveness of health care. Thus, inequality in amenable mortality provides evidence on weaknesses in health care performance between socioeconomic groups. Our study shows using several approaches with different parametric assumptions that previously introduced methods to estimate the uncertainty of the C for sampled data are too conservative for aggregated population register data. Consequently, we recommend that inequality indices based on the register data should be presented together with an approximation of the uncertainty and suggest using a simulation approach we propose. The approach can also be adapted to other measures of equality in health. [ABSTRACT FROM AUTHOR]

In this paper we advance a dual-prong method for evaluating whether data from a natural experiment can be leveraged to draw convincing implications about causal effects of neighborhood context on health. In particular, we probe a natural experiment's ability to surmount potentially confounding measurement issues arising from geographic selection bias. We offer methods for assessing the degree of this potential bias both immediately after households are assigned to neighborhoods and over longer durations. We then illustrate these methods using data from a natural experiment involving scattered-site public housing in Denver and test whether neighborhood context affects the hazard of children's asthma diagnoses. We find that, conditioned on ethnicity, the tenant allocation process used by the Denver Housing Authority produced a quasi-random initial assignment across neighborhood characteristics and that the range of bias arising subsequently through selective out-migration can be narrowly delineated by replicating analyses for alternative samples defined by period of residence in public housing. We further find support for the proposition that observing no correlation between neighborhood and observed household characteristics is tantamount to observing no correlation between neighborhood and (typically) unobserved household characteristics. There appears to be little bias reintroduced through selective out-migration and staying in initially assigned neighborhood. These findings suggest that, after employing the evaluation strategy suggested here, natural experiments can indeed be a powerful tool for assessing neighborhood effects on health. [ABSTRACT FROM AUTHOR]

Change point estimation is recognized as an essential tool of root cause analyses within quality control programs as it enables clinical experts to search for potential causes of change in hospital outcomes more effectively. In this paper, we consider estimation of the time when a linear trend disturbance has occurred in survival time following an in-control clinical intervention in the presence of variable patient mix. To model the process and change point, a linear trend in the survival time of patients who underwent cardiac surgery is formulated using hierarchical models in a Bayesian framework. The data are right censored since the monitoring is conducted over a limited follow-up period. We capture the effect of risk factors prior to the surgery using a Weibull accelerated failure time regression model. We use Markov Chain Monte Carlo to obtain posterior distributions of the change point parameters including the location and the slope size of the trend and also corresponding probabilistic intervals and inferences. The performance of the Bayesian estimator is investigated through simulations and the result shows that precise estimates can be obtained when they are used in conjunction with the risk-adjusted survival time cumulative sum control chart (CUSUM) control charts for different trend scenarios. In comparison with the alternatives, step change point model and built-in CUSUM estimator, more accurate and precise estimates are obtained by the proposed Bayesian estimator over linear trends. These superiorities are enhanced when probability quantification, flexibility and generalizability of the Bayesian change point detection model are also considered. [ABSTRACT FROM AUTHOR]

Classic instrumental variable techniques involve the use of structural equation modeling or other forms of parameterized modeling. In this paper we use a nonparametric, matching-based instrumental variable methodology that is based on a study design approach. Similar to propensity score matching, though unlike classic instrumental variable approaches, near/far matching is capable of estimating causal effects when the outcome is not continuous. Unlike propensity score matching, though similar to instrumental variable techniques, near/far matching is also capable of estimating causal effects even when unmeasured covariates produce selection bias. We illustrate near/far matching by using Medicare data to compare the effectiveness of carotid arterial stents with cerebral protection versus carotid endarterectomy for the treatment of carotid stenosis. [ABSTRACT FROM AUTHOR]

Instrumental variables (IVs) enable causal estimates in observational studies to be obtained in the presence of unmeasured confounders. In practice, a diverse range of models and IV specifications can be brought to bear on a problem, particularly with longitudinal data where treatment effects can be estimated for various functions of current and past treatment. However, in practice the empirical consequences of different assumptions are seldom examined, despite the fact that IV analyses make strong assumptions that cannot be conclusively tested by the data. In this paper, we consider several longitudinal models and specifications of IVs. Methods are applied to data from a 7-year study of mental health costs of atypical and conventional antipsychotics whose purpose was to evaluate whether the newer and more expensive atypical antipsychotic medications lead to a reduction in overall mental health costs. [ABSTRACT FROM AUTHOR]

The quality of propensity scores is traditionally measured by assessing how well they make the distributions of covariates in the treatment and control groups match, which we refer to as 'good balance'. Good balance guarantees less biased estimates of the treatment effect. However, the cost of achieving good balance is that the variance of the estimates increases due to a reduction in effective sample size, either through the introduction of propensity score weights or dropping cases when propensity score matching. In this paper, we investigate whether it is best to optimize the balance or to settle for a less than optimal balance and use double robust estimation to adjust for remaining differences. We compare treatment effect estimates from regression, propensity score weighting, and double robust estimation with varying levels of effort expended to achieve balance using data from a study about the differences in outcomes by HIV status in heterosexually active homeless men residing in Los Angeles. Because of how costly data collection efforts are for this population, it is important to find an alternative estimation method that does not reduce effective sample size as much as methods that aggressively aim to optimize balance. Results from a simulation study suggest that there are instances in which we can obtain more precise treatment effect estimates without increasing bias too much by using a combination of regression and propensity score weights that achieve a less than optimal balance. There is a bias-variance tradeoff at work in propensity score estimation; every step toward better balance usually means an increase in variance and at some point a marginal decrease in bias may not be worth the associated increase in variance. [ABSTRACT FROM AUTHOR]

Two major statistical issues confronting comparative analyses of hospital outcomes are adequacy of case-mix adjustment and proper accounting for random variation. Hierarchical modeling has been proposed to improve precision and reduce the impact of random variation but becomes difficult to implement when there are numerous case-mix factors to control. In this paper we formulate the problem of hospital performance comparisons within the framework of potential outcomes and illustrate an approach to hospital comparisons which combines multiple category propensity score methods for the control of case-mix variations with hierarchical Bayesian modeling of case-mix adjusted summaries. The approach is similar to that proposed by Huang et al. (Health Serv Res 40:253-278, ) but extends their approach by using a Bayesian model to accommodate hospital level attributes and to facilitate joint modeling of performance for multiple outcomes. The analytical approach is illustrated by a comparison of 30 day post admission mortality risks for patients treated for acute myocardial infarction, pneumonia or stroke in 34 New Zealand public hospitals. In a small simulation study, reported in electronic supplementary material, hierarchical models outperformed non-hierarchical models, achieving both better credible interval coverage and shorter average interval lengths for measures of between hospital variation based on contrasts between the 90th and 10th percentiles of the mortality risk distribution. Simulation performance of hierarchical and non-hierarchical models in detecting unusual performance was similar. [ABSTRACT FROM AUTHOR]

Continuous variable dichotomization is a popular technique used in the estimation of the effect of risk factors on health outcomes in multivariate regression settings. Researchers follow this practice in order to simplify data analysis, which it unquestionably does. However thresholds used to dichotomize those variables are usually ad-hoc, based on expert opinions, or mean, median or quantile splits and can add bias to the effect of the risk factors on specific outcomes and underestimate such effect. In this paper, we suggest the use of a semi-parametric method and visualization for improvement of the threshold selection in variable dichotomization while accounting for mixture distributions in the outcome of interest and adjusting for covariates. For clinicians, these empirically based thresholds of risk factors, if they exist, could be informative in terms of the highest or lowest point of a risk factor beyond which no additional impact on the outcome should be expected. [ABSTRACT FROM AUTHOR]

Simulation studies were performed to investigate for which conditions of sample size of patients (n) and number of repeated measurements (k) (e.g., raters) the optimal (i.e., balance between precise and efficient) estimations of intraclass correlation coefficients (ICCs) and standard error of measurements (SEMs) can be achieved. Subsequently, we developed an online application that shows the implications for decisions about sample sizes in reliability studies. We simulated scores for repeated measurements of patients, based on different conditions of n, k, the correlation between scores on repeated measurements (r), the variance between patients' test scores (v), and the presence of systematic differences within k. The performance of the reliability parameters (based on one-way and two-way effects models) was determined by the calculation of bias, mean squared error (MSE), and coverage and width of the confidence intervals (CI). We showed that the gain in precision (i.e., largest change in MSE) of the ICC and SEM parameters diminishes at larger values of n or k. Next, we showed that the correlation and the presence of systematic differences have most influence on the MSE values, the coverage and the CI width. This influence differed between the models. As measurements can be expensive and burdensome for patients and professionals, we recommend to use an efficient design, in terms of the sample size and number of repeated measurements to come to precise ICC and SEM estimates. Utilizing the results, a user-friendly online application is developed to decide upon the optimal design, as 'one size fits all' doesn't hold. [ABSTRACT FROM AUTHOR]

Few rigorous studies provide a clear description of the methodological approach of developing an evidence-based implementation intervention, prior to implementation at scale. This study describes the development, mapping, rating, and review of the implementation strategies for the Care to Quit smoking cessation trial, prior to application in nine cancer services across Australia. Key stakeholders were engaged in the process from conception through to rating, reviewing and refinement of strategies and principles. An initial scoping review identified 21 barriers to provision of evidence-based smoking cessation care to patients with cancer, which were mapped to the Theoretical Domains Framework and Behaviour Change Wheel (BCW) to identify relevant intervention functions. The mapping identified 26 relevant behaviour change techniques, summarised into 11 implementation strategies. The implementation strategies were rated and reviewed against the BCW Affordability, Practicality, Effectiveness and cost-effectiveness, Acceptability, Side-effects/safety, and Equity criteria by key stakeholders during two interactive workshops to facilitate a focus on feasible interventions likely to resonate with clinical staff. The implementation strategies and associated intervention tools were then collated by form and function to provide a practical guide for implementing the intervention. This study illustrates the rigorous use of theories and frameworks to arrive at a practical intervention guide, with potential to inform future replication and scalability of evidence-based implementation across a range of health service settings. [ABSTRACT FROM AUTHOR]

Understanding how best to estimate state-level policy effects is important, and several unanswered questions remain, particularly about the ability of statistical models to disentangle the effects of concurrently enacted policies. In practice, many policy evaluation studies do not attempt to control for effects of co-occurring policies, and this issue has not received extensive attention in the methodological literature to date. In this study, we utilized Monte Carlo simulations to assess the impact of co-occurring policies on the performance of commonly-used statistical models in state policy evaluations. Simulation conditions varied effect sizes of the co-occurring policies and length of time between policy enactment dates, among other factors. Outcome data (annual state-specific opioid mortality rate per 100,000) were obtained from 1999 to 2016 National Vital Statistics System (NVSS) Multiple Cause of Death mortality files, thus yielding longitudinal annual state-level data over 18 years from 50 states. When co-occurring policies are ignored (i.e., omitted from the analytic model), our results demonstrated that high relative bias (> 82%) arises, particularly when policies are enacted in rapid succession. Moreover, as expected, controlling for all co-occurring policies will effectively mitigate the threat of confounding bias; however, effect estimates may be relatively imprecise (i.e., larger variance) when policies are enacted in near succession. Our findings highlight several key methodological issues regarding co-occurring policies in the context of opioid-policy research yet also generalize more broadly to evaluation of other state-level policies, such as policies related to firearms or COVID-19, showcasing the need to think critically about co-occurring policies that are likely to influence the outcome when specifying analytic models. [ABSTRACT FROM AUTHOR]

In addition to applications in meta-analysis, funnel plots have emerged as an effective graphical tool for visualizing the detection of health care providers with unusual performance. Although there already exist a variety of approaches to producing funnel plots in the literature of provider profiling, limited attention has been paid to elucidating the critical relationship between funnel plots and hypothesis testing. Within the framework of generalized linear models, here we establish methodological guidelines for creating funnel plots specific to the statistical tests of interest. Moreover, we show that the test-specific funnel plots can be created merely leveraging summary statistics instead of individual-level information. This appealing feature inhibits the leak of protected health information and reduces the cost of inter-institutional data transmission. Two data examples, one for surgical patients from Michigan hospitals and the other for Medicare-certified dialysis facilities, demonstrate the applicability to different types of providers and outcomes with either individual- or summary-level information. [ABSTRACT FROM AUTHOR]

The Consent-to-Contact (C2C) registry at the University of California, Irvine collects data from community participants to aid in the recruitment to clinical research studies. Self-selection into the C2C likely leads to bias due in part to enrollees having more years of education relative to the US general population. Salazar et al. (Alzheimer's Dementia Transl Res Clin Interv 6(1):e120023, 2020, https://doi.org/10.1002/trc2.12023) recently used the C2C to examine associations of race/ethnicity with participant willingness to be contacted about research studies. To obtain representative estimates from C2C we use weighted estimation of associations of interest where the weights are related to the probability of self-selection into the convenience sample. The selection probabilities are estimated using data from the National Health and Nutrition Examination Survey (NHANES). We create a combined dataset of C2C and NHANES subjects and evaluate the trade-offs of different approaches (logistic regression, covariate balancing propensity score, entropy balancing, and random forest) for estimating the probability of membership in C2C relative to NHANES. We further propose methods to estimate the variance of parameter estimates that account for uncertainty that arises from estimating propensity weights. Simulation studies explore the impact of propensity weight estimation on uncertainty. We demonstrate the approach by repeating the analysis by Salazar et al. (Alzheimer's Dementia Transl Res Clin Interv 6(1):e120023, 2020, https://doi.org/10.1002/trc2.12023) with the deduced propensity weights for the C2C subjects and contrast the results of the two analyses. This method can be implemented using our estweight package in R available on GitHub. [ABSTRACT FROM AUTHOR]

Shared savings arrangements are designed to financially reward provider groups that reduce healthcare spending through improved care coordination. A major concern with these arrangements is that annual changes in spending are subject to a variety of random factors that are unrelated to care coordination efforts. As a result, resources can be misallocated if providers who are unsuccessful at controlling spending are inappropriately rewarded and providers who are successful are inappropriately denied rewards. This paper provides a systematic analysis of the role of random variation using a general statistical model based on shared savings arrangements that are currently evolving in the public and private sectors. The model focuses specifically on the variance of the average savings rate (ASR), which is the quantity used to determine whether and by how much a provider group will be rewarded. Variance in the ASR is a major driver of the probabilities of Type I error (i.e., inappropriately rewarding providers) and Type II error (i.e., inappropriately failing to reward providers), which can lead to major resource misallocations. We find that the probabilities of Type I and Type II errors associated with common approaches to savings measurement can be quite high, often exceeding 10 or 25 %, respectively. We also find that the likelihood of both types of errors can be substantially reduced through careful planning and design of savings measurement schemes before payers and providers enter into shared savings agreements. [ABSTRACT FROM AUTHOR]

Marital status is recognized as an important social determinant of health, income, and social support, but is rarely available in administrative data. We assessed the feasibility of using exact address data and zip code history to identify cohabiting couples using the 2018 Medicare Vital Status file and ZIP codes in the 2011–2014 Master Beneficiary Summary Files. Medicare beneficiaries meeting our algorithm displayed characteristics consistent with assortative mating and resembled known married couples in the Health and Retirement Study linked to Medicare claims. Address information represents a promising strategy for identifying cohabiting couples in administrative data including healthcare claims and other data types. [ABSTRACT FROM AUTHOR]

Recurrent event data is frequently encountered in biomedical research. Estimators describing the time to the next event occurrence could be useful information in healthcare settings to deliver targeted, preemptive, patient-specific care. Distribution estimation of recurrence times is varied based on both the underlying assumptions of the estimator (e.g., dependence vs. independence of recurrence times within individual) and estimator construct. In an example among Veterans with spinal cord injury, distributional changes in recurrence times of health complications subsequent lower extremity (LE) fracture are explored. Veterans with LE fracture (cases) were matched to Veterans without LE fracture (controls) on demographic (age, race), level of injury (paraplegia vs. tetraplegia), extent of injury, Veterans Affairs connected service status, and comorbidities using Mahalanobis metric matching. Stratified distributional estimates of recurrence times between successive morbidity outcomes are compared between Veterans with/without LE fracture using three estimators: independent identically distributed product limit estimator, Wang-Chang product limit estimator, and a gamma frailty maximum likelihood estimator. It can be seen that the estimator selection can provide a very different showcasing of a recurrence time distribution. There is a change in the time-to-recurrence of recurrent urinary tract infections and pressure ulcers for fracture cases directly following LE fracture, however testing if this difference is statistically significant remains unclear. Causal inference of gap time analyses in observational data with recurrent events is considered and a call for methods in this area is much warranted. [ABSTRACT FROM AUTHOR]

A bivariate mixed-effects location-scale model is proposed for estimation of means, variances, and covariances of two continuous outcomes measured concurrently in time and repeatedly over subjects. Modeling the two outcomes jointly allows examination of BS and WS association between the outcomes and whether the associations are related to covariates. The variance-covariance matrices of the BS and WS effects are modeled in terms of covariates, explaining BS and WS heterogeneity. The proposed model relaxes assumptions on the homogeneity of the within-subject (WS) and between-subject (BS) variances. Furthermore, the WS variance models are extended by including random scale effects. Data from a natural history study on adolescent smoking are used for illustration. 461 students, from 9th and 10th grades, reported on their mood at random prompts during seven consecutive days. This resulted in 14,105 prompts with an average of 30 responses per student. The two outcomes considered were a subject's positive affect and a measure of how tired and bored they were feeling. Results showed that the WS association of the outcomes was negative and significantly associated with several covariates. The BS and WS variances were heterogeneous for both outcomes, and the variance of the random scale effects were significantly different from zero. [ABSTRACT FROM AUTHOR]

In most nonrandomized observational studies, differences between treatment groups may arise not only due to the treatment but also because of the effect of confounders. Therefore, causal inference regarding the treatment effect is not as straightforward as in a randomized trial. To adjust for confounding due to measured covariates, the average treatment effect is often estimated by using propensity scores. Typically, propensity scores are estimated by logistic regression. More recent suggestions have been to employ nonparametric classification algorithms from machine learning. In this article, we propose a weighted estimator combining parametric and nonparametric models. Some theoretical results regarding consistency of the procedure are given. Simulation studies are used to assess the performance of the newly proposed methods relative to existing methods, and a data analysis example from the Surveillance, Epidemiology and End Results database is presented. [ABSTRACT FROM AUTHOR]

The estimation of treatment effects is one of the primary goals of statistics in medicine. Estimation based on observational studies is subject to confounding. Statistical methods for controlling bias due to confounding include regression adjustment, propensity scores and inverse probability weighted estimators. These methods require that all confounders are recorded in the data. The method of instrumental variables (IVs) can eliminate bias in observational studies even in the absence of information on confounders. We propose a method for integrating IVs within the framework of Cox's proportional hazards model and demonstrate the conditions under which it recovers the causal effect of treatment. The methodology is based on the approximate orthogonality of an instrument with unobserved confounders among those at risk. We derive an estimator as the solution to an estimating equation that resembles the score equation of the partial likelihood in much the same way as the traditional IV estimator resembles the normal equations. To justify this IV estimator for a Cox model we perform simulations to evaluate its operating characteristics. Finally, we apply the estimator to an observational study of the effect of coronary catheterization on survival. [ABSTRACT FROM AUTHOR]

We apply three separate panel data estimation methods to examine the diffusion of technologies at the state-level. These methods include the Hausman-Taylor random effects model, the fixed effects vector decomposition, and generalized estimating equations. We discuss the assumptions required of each and assess the stability of our policy results across the three models for a longitudinal study of the diffusion of newer psychotropic technologies. We find a reasonable level of consistency among marginal effects for time varying independent variables between our three estimation methods but some discrepancy in the estimated measure of precision in our empirical application. We find a number of policy conclusions are quite stable across estimation methods and may be of interest to state-level mental health policy decision makers. [ABSTRACT FROM AUTHOR]

The common approach to modeling healthcare cost data is to use aggregated total cost from multiple categories or sources (e.g. inpatient, outpatient, prescriptions, etc.) as the dependent variable. However, this approach could hide the differential impact of covariates on the different cost categories. An alternative is to model each cost category separately. This could also lead to wrong conclusions due to failure to account for the interdependence among the multiple cost outcomes. Therefore, we propose a multivariate generalized linear mixed model (mGLMM) that allows for joint modeling of longitudinal cost data from multiple sources. We assessed four different approaches, (1) shared random intercept, (2) shared random intercept and slope, (3) separate random intercepts from a joint multivariate distribution, and (4) separate random intercepts and slopes from a joint multivariate distribution. Each of these approaches differs in the way they account for the correlation among the multiple cost outcomes. Comparison was made via goodness of fit measures and residual plots. Longitudinal cost data from a national cohort of 740,195 veterans with diabetes (followed from 2002-2006) was used to demonstrate joint modeling. Among examined models, the separate random intercept approach exhibited the lowest AIC/BIC in both log-normal and gamma GLMMs. However, for our data example, the shared random intercept approach seemed to be sufficient as the more complex models did not lead to qualitatively different conclusions. [ABSTRACT FROM AUTHOR]

Provider profiling as a means to describe and compare the performance of health care professionals has gained momentum in the past decade. As a key component of pay-for-performance programs profiling has been increasingly used to identify top-performing providers. However, rigorous examination of the performance of statistical methods for profiling when used to classify top-performing providers is lacking. The objective of this study was to compare the classification accuracy of three methods for identifying providers exceeding performance thresholds and to analyze data on satisfaction with mental health care providers at Group Health Cooperative using these methods. Questionnaire data on patient satisfaction with mental health care providers at Group Health Cooperative was collected between April 2008 and January 2010. Simulated data were used to compare the classification accuracy of alternative statistical methods. We evaluated sensitivity, specificity, and root mean squared error of alternative statistical methods using simulated data. For Group Health providers, we compared agreement of alternative approaches to classification. We found that when between-provider variability in performance was low, all three methods exhibited poor classification accuracy. When used to evaluate mental health care provider performance, we found substantial uncertainty in the estimates and poor agreement across methods. Based on these findings, we recommend providing uncertainty estimates for provider rankings and caution against the use of any classification method when between-provider variability is low. [ABSTRACT FROM AUTHOR]

We examined the relationship between the proportion of time spent in the mammography exam room relative to the total visit time and the rate of repeat visits across mammography facilities. We hypothesized that the means for these proportions would be greater among facilities with high repeat rates compared to those with low repeat rates. The beta distribution was used to model the proportions. However, not all of the observations were observed as some were (right- or interval-) censored due to the inability to observe complete information (e.g., time of arrival at the facility) to calculate the proportions for some observations. Using a maximum likelihood approach, we estimated the mean proportion of time spent during appointments on the actual exams. To analyze these data we utilized survival analysis methods, which typically are applied to data on time scales, to analyze these data restricted to the range (0, 1). We found that facilities with high repeat mammography rates on average had a greater proportion of their visits spent receiving the exams than the low repeat rate facilities ( p = 0.0003). Women from low repeat rate facilities spent on average 45 % of their appointment time engaging in the exam; thus, these women averaged more time waiting for the exam to begin than receiving the exam. Conversely, those from high repeat rate facilities averaged more of their appointment time engaging in the mammographic exams. These findings were consistent with our hypothesis that women would have greater satisfaction with their mammography appointments-and thus be more likely to return for routine screening-in facilities where their appointments were more efficient. [ABSTRACT FROM AUTHOR]

Evaluating treatment effects in non-randomized studies is challenging due to the potential unmeasured confounding and complex form of observed confounding. Propensity score based approaches, such as matching or weighting, are commonly used to handle observed confounding variables. The instrumental variable (IV) method is known to guard against unmeasured confounding if a good instrument can be identified. We propose to combine both methods to estimate the long-term treatment effect in a longitudinal psychiatric study. The NIMH collaborative Multi-site Treatment study of children with Attention-deficit/hyperactivity disorder (ADHD) compared different treatment strategies for children diagnosed with ADHD (known as MTA study). The first 14 months is a randomized study and the participants are allowed to choose their desired treatment strategies afterwards. Follow-up measurements are taken at 24, 36 and 72 months. Randomization is often considered as a good instrument since it is not associated with any covariate, observed or unobserved. We first apply a randomization based IV method to estimate the self-selected medication effect on outcome at the end of 72 months. However this approach yields results with huge standard errors due to randomization's weak relationship with later treatment selection. We then consider the self-selection right after the randomization as an instrument, because it is associated with later treatment selection and it is unlikely to affect the outcome directly given the 5-year time lapse. To better control the confounding due to observed factors, propensity score matching is used to create a subpopulation with comparable covariate distributions across different self-selected treatments. Using MTA data, matching-enhanced IV estimation yields the most sensible result, while other estimation strategies tend to imply a spurious significant effect. Also, our simulation study shows that the matching-enhanced IV estimation outperforms non-matched methods in terms of relative bias. [ABSTRACT FROM AUTHOR]

As the amount of data generated continues to increase, consideration of individuals' privacy is a growing concern. As a result, there has been a vast quantity of research done on methods of statistical disclosure control. Some of these methods propose to release a randomized version of the data rather than the actual data. While methods of this type certainly offer some layer of protection, there is still the potential for private information to be disclosed. Quantifying the level of privacy provided by these methods is often difficult. In the past, a method for assessing privacy using the receiver operating characteristic (ROC) curve based on ideas related to differential privacy was proposed. However, the method was only demonstrated for univariate randomized releases. Here, the ROC-based privacy measure is extended to the release of randomized vectors. [ABSTRACT FROM AUTHOR]

We develop a new tool for assessing the sensitivity of findings on treatment effectiveness to differential follow-up rates in the two treatment conditions being compared. The method censors the group with the higher response rate to create a synthetic respondent group that is then compared with the observed cases in the other condition to estimate a treatment effect. Censoring is done under various assumptions about the strength of the relationship between follow-up and outcomes to determine how informative differential dropout can alter inferences relative to estimates from models that assume the data are Missing at Random. The method provides an intuitive measure for understanding the strength of the association between outcomes and dropout that would be required to alter inferences about treatment effects. Our approach is motivated by translational research in which treatments found to be effective under experimental conditions are tested in standard treatment settings. In such applications, follow-up rates in the experimental setting are likely to be substantially higher than in the standard setting, especially when observational data are used in the evaluation. We test the method on a case study evaluation of the effectiveness of an evidence-supported adolescent substance abuse treatment program (Motivational Enhancement Therapy/Cognitive Behavioral Therapy-5) delivered by community-based treatment providers relative to its performance in a controlled research trial. In this case study, follow-up rates in the community-based settings were extremely low (54 %) compared to the experimental setting (95 %) giving raise to concerns about non-ignorable drop-out. [ABSTRACT FROM AUTHOR]

Joint modeling of longitudinal and survival data can provide more efficient and less biased estimates of treatment effects through accounting for the associations between these two data types. Sponsors of oncology clinical trials routinely and increasingly include patient-reported outcome (PRO) instruments to evaluate the effect of treatment on symptoms, functioning, and quality of life. Known publications of these trials typically do not include jointly modeled analyses and results. We formulated several joint models based on a latent growth model for longitudinal PRO data and a Cox proportional hazards model for survival data. The longitudinal and survival components were linked through either a latent growth trajectory or shared random effects. We applied these models to data from a randomized phase III oncology clinical trial in mesothelioma. We compared the results derived under different model specifications and showed that the use of joint modeling may result in improved estimates of the overall treatment effect. [ABSTRACT FROM AUTHOR]

Community colleges have increased post-secondary educational access for disadvantaged youth, but it is unknown how community college degrees fit into the educational gradient of health status disparities. Using data from high school graduates in the National Longitudinal Study of Adolescent Health, we compared young adults ages 26-31 whose highest degrees were high school diplomas ( n = 5,584), sub-baccalaureate credentials (sub-BAs include community college certificates and associate's degrees) ( n = 2,415), and baccalaureate degrees (BAs) ( n = 3,303) on measures of hypertension, obesity, smoking, sleep problems, dyslipidemia, and depression. Comparisons used multivariate Poisson regression with robust standard errors after exact and nearest-neighbor Mahalanobis matching within propensity score calipers on 23 baseline factors measured in 1995. High school graduates and sub-BAs differed significantly on 3 of 23 baseline factors. After matching, sub-BAs were 16 % less likely to smoke daily than if they had only a high school diploma but did not differ in other health status measures. Sub-BAs and BAs differed significantly on 14 of 23 baseline factors. After matching, BAs were 60 % less likely to smoke daily, 14 % less likely to be obese, and 38 % less likely to have been diagnosed with depression. Sub-BA degrees are accessible to high school graduates irrespective of academic backgrounds and predict lower smoking prevalence. BAs are less accessible to high school graduates and predict lower chances of smoking, depression, and obesity. [ABSTRACT FROM AUTHOR]

In observational studies of treatments or interventions, propensity score (PS) adjustment is often useful for controlling bias in estimation of treatment effects. Regression on PS is used most often and can be highly efficient, but it can lead to biased results when model assumptions are violated. The validity of stratification on PS depends on fewer model assumptions, but this approach is less efficient than regression adjustment when the regression assumptions hold. To investigate these issues, we compare stratification and regression adjustments in a Monte Carlo simulation study. We consider two stratification approaches: equal frequency strata and an approach that attempts to choose strata that minimize the mean squared error (MSE) of the treatment effect estimate. The regression approach that we consider is a generalized additive model (GAM) that estimates treatment effect controlling for a potentially nonlinear association between PS and outcome. We find that under a wide range of plausible data generating distributions the GAM approach outperforms stratification in treatment effect estimation with respect to bias, variance, and thereby MSE. We illustrate each approach in an analysis of insurance plan choice and its relation to satisfaction with asthma care. [ABSTRACT FROM AUTHOR]

Standard approaches for determining weights when calculating a composite measure of health care quality from individual quality indicators (QIs) include equal weighting, opportunity-based weights, and judgment-based weights. Benefit-of-the-doubt approaches have not been used in the health services area, though one has been used to calculate composite measures for profiling countries. Underlying these approaches is the assumption that relative performance on a set of indicators is, at least to some extent, a revealed preference by the organizational unit about the relative importance of the indicators. A benefit-of-the-doubt approach recognizes these revealed preferences by assigning higher weights to indicators on which performance is better and lower weights to indicators on which performance is poorer. We consider two benefit-of-the-doubt approaches. The first uses simple linear programming (LP) models; the second uses data envelopment analysis (DEA), the way in which the benefit-of-the-doubt approach has been previously implemented. In both cases, constraints are added to limit weight adjustments to some percentage of policy-determined baseline weights. Using both standard and benefit-of-the-doubt approaches, composite scores are calculated from data on five QIs from 32 Department of Veterans Affairs (VA) nursing homes. We examine the tradeoff between the level of allowable weight adjustment and impact on facility rankings. If weights are constrained to be within 75% of baseline weights, all approaches identify pretty much the same high performing facilities. Weights from benefit-of-the-doubt approaches, because they are able to reflect local preferences and conditions, should be attractive to facilities and, in a collaborative environment, to policy makers. [ABSTRACT FROM AUTHOR]

In order to create an empirically derived parsimonious typology of physician financial incentives that will be useful for future research, we used data from the nationally representative 2004–2005 Community Tracking Study Physician Survey (N = 6,628). Linear regression analyses informed by economic theory were used to identify the combinations of incentives associated with an overall financial incentive to expand services to individual patients. The approach was validated using two nonparametric methods (CART analysis and data mining techniques) and by examining the relationship between the resulting typology and other measures of physician behavior including hours worked, visit volume, and specialty-adjusted income. Of the 6,628 physicians surveyed, approximately 25% (1,605) reported an overall incentive to increase services and 75% (5,023) reported either neutral incentives or incentives to decrease services. Men, who were approximately 75% of respondents, were slightly more likely to report incentives to increase services ( P < 0.05). There were no differences in reported incentives according to specialty. We created two typologies (one with eleven categories and the other with a collapsed set of six categories) based on combinations of variables measuring ownership, base compensation methods, and financial incentives. The percentage with an overall incentive to increase services ranges from 6% for employed physicians compensated via fixed salary to 36.7% for owners in low capitation environments with either individual or practice level productivity incentives. The criterion validity of the typology was established by examining the relationship with adjusted physician income, hours worked, and visit volume, which showed generally consistent relationships in the expected direction. A parsimonious typology consisting of six mutually exclusive groups reasonably captures the continuum of incentives to increase service delivery experienced by physicians. [ABSTRACT FROM AUTHOR]