Propensity score weighting for a continuous exposure with multilevel data

Long-term PM1 exposure and hypertension hospitalization: A causal inference study on a large community-based cohort in South China

Limited evidence exists on the effect of submicronic particulate matter (PM1) on hypertension hospitalization. Evidence based on causal inference and large cohorts is even more scarce. In 2015, 36,271 participants were enrolled in South China and followed up through 2020. Each participant was assigned single-year, lag0-1, and lag0-2 moving average concentration of PM1 and fine inhalable particulate matter (PM2.5) simulated based on satellite data at a 1-km resolution. We used an inverse probability weighting approach to balance confounders and utilized a marginal structural Cox model to evaluate the underlying causal links between PM1 exposure and hypertension hospitalization, with PM2.5-hypertension association for comparison. Several sensitivity studies and the analyses of effect modification were also conducted. We found that a higher hospitalization risk from both overall (HR: 1.13, 95% CI: 1.05-1.22) and essential hypertension (HR: 1.15, 95% CI: 1.06-1.25) was linked to each 1 µg/m3 increase in the yearly average PM1 concentration. At lag0-1 and lag0-2, we observed a 17%-21% higher risk of hypertension associated with PM1. The effect of PM1 was 6%-11% higher compared with PM2.5. Linear concentration-exposure associations between PM1 exposure and hypertension were identified, without safety thresholds. Women and participants that engaged in physical exercise exhibited higher susceptibility, with 4%-22% greater risk than their counterparts. This large cohort study identified a detrimental relationship between chronic PM1 exposure and hypertension hospitalization, which was more pronounced compared with PM2.5 and among certain groups.

Use of Healthcare Claims Data to Generate Real-World Evidence on Patients With Drug-Resistant Epilepsy: Practical Considerations for Research

Objectives: Regulatory bodies, health technology assessment agencies, payers, physicians, and other decision-makers increasingly recognize the importance of real-world evidence (RWE) to provide important and relevant insights on treatment patterns, burden/cost of illness, product safety, and long-term and comparative effectiveness. However, RWE generation requires a careful approach to ensure rigorous analysis and interpretation. There are limited examples of comprehensive methodology for the generation of RWE on patients who have undergone neuromodulation for drug-resistant epilepsy (DRE). This is likely due, at least in part, to the many challenges inherent in using real-world data to define DRE, neuromodulation (including type implanted), and related outcomes of interest. We sought to provide recommendations to enable generation of robust RWE that can increase knowledge of "real-world" patients with DRE and help inform the difficult decisions regarding treatment choices and reimbursement for this particularly vulnerable population. Methods: We drew upon our collective decades of experience in RWE generation and relevant disciplines (epidemiology, health economics, and biostatistics) to describe challenges inherent to this therapeutic area and to provide potential solutions thereto within healthcare claims databases. Several examples were provided from our experiences in DRE to further illustrate our recommendations for generation of robust RWE in this therapeutic area. Results: Our recommendations focus on considerations for the selection of an appropriate data source, development of a study timeline, exposure allotment (specifically, neuromodulation implantation for patients with DRE), and ascertainment of relevant outcomes. Conclusions: The need for RWE to inform healthcare decisions has never been greater and continues to grow in importance to regulators, payers, physicians, and other key stakeholders. However, as real-world data sources used to generate RWE are typically generated for reasons other than research, rigorous methodology is required to minimize bias and fully unlock their value.

Propensity Score Weighting with Missing Data on Covariates and Clustered Data Structure

Propensity score (PS) analyses are increasingly popular in behavioral sciences. Two issues often add complexities to PS analyses, including missing data in observed covariates and clustered data structure. In previous research, methods for conducting PS analyses with considering either issue alone were examined. In practice, the two issues often co-occur; but the performance of methods for PS analyses in the presence of both issues has not been evaluated previously. In this study, we consider PS weighting analysis when data are clustered and observed covariates have missing values. A simulation study is conducted to evaluate the performance of different missing data handling methods (complete-case, single-level imputation, or multilevel imputation) combined with different multilevel PS weighting methods (fixed- or random-effects PS models, inverse-propensity-weighting or the clustered weighting, weighted single-level or multilevel outcome models). The results suggest that the bias in average treatment effect estimation can be reduced, by better accounting for clustering in both the missing data handling stage (such as with the multilevel imputation) and the PS analysis stage (such as with the fixed-effects PS model, clustered weighting, and weighted multilevel outcome model). A real-data example is provided for illustration.

Multimodal digital phenotyping of diet, physical activity, and glycemia in Hispanic/Latino adults with or at risk of type 2 diabetes

Digital phenotyping refers to characterizing human bio-behavior through wearables, personal devices, and digital health technologies. Digital phenotyping in populations facing a disproportionate burden of type 2 diabetes (T2D) and health disparities continues to lag compared to other populations. Here, we report our study demonstrating the application of multimodal digital phenotyping, i.e., the simultaneous use of CGM, physical activity monitors, and meal tracking in Hispanic/Latino individuals with or at risk of T2D. For 14 days, 36 Hispanic/Latino adults (28 female, 14 with non-insulin treated T2D) wore a continuous glucose monitor (CGM) and a physical activity monitor (Actigraph) while simultaneously logging meals using the MyFitnessPal app. We model meal events and daily digital biomarkers representing diet, physical activity choices, and corresponding glycemic response. We develop a digital biomarker for meal events that differentiates meal events into normal and elevated categories. We examine the contribution of daily digital biomarkers of elevated meal event count and step count on daily time-in-range 54-140 mg/dL (TIR54-140) and average glucose. After adjusting for step count, a change in elevated meal event count from zero to two decreases TIR54-140 by 4.0% (p = 0.003). An increase in 1000 steps in post-meal step count also reduces the meal event glucose response by 641 min mg/dL (p = 0.0006) and reduces the odds of an elevated meal event by 55% (p < 0.0001). The proposed meal event digital biomarkers may provide an opportunity for non-pharmacologic interventions for Hispanic/Latino adults facing a disproportionate burden of T2D.

Inverse Probability Weights for Quasicontinuous Ordinal Exposures With a Binary Outcome: Method Comparison and Case Study

Inverse probability weighting (IPW), a well-established method of controlling for confounding in observational studies with binary exposures, has been extended to analyses with continuous exposures. Methods developed for continuous exposures may not apply when the exposure is quasicontinuous because of irregular exposure distributions that violate key assumptions. We used simulations and cluster-randomized clinical trial data to assess 4 approaches developed for continuous exposures-ordinary least squares (OLS), covariate balancing generalized propensity scores (CBGPS), nonparametric covariate balancing generalized propensity scores (npCBGPS), and quantile binning (QB)-and a novel method, a cumulative probability model (CPM), in quasicontinuous exposure settings. We compared IPW stability, covariate balance, bias, mean squared error, and standard error estimation across 3,000 simulations with 6 different quasicontinuous exposures, varying in skewness and granularity. In general, CBGPS and npCBGPS resulted in excellent covariate balance, and npCBGPS was the least biased but the most variable. The QB and CPM approaches had the lowest mean squared error, particularly with marginally skewed exposures. We then successfully applied the IPW approaches, together with missing-data techniques, to assess how session attendance (out of a possible 15) in a partners-based clustered intervention among pregnant couples living with human immunodeficiency virus in Mozambique (2017-2022) influenced postpartum contraceptive uptake.

Tuning Random Forests for Causal Inference under Cluster-Level Unmeasured Confounding

Recently, there has been growing interest in using machine learning methods for causal inference due to their automatic and flexible ability to model the propensity score and the outcome model. However, almost all the machine learning methods for causal inference have been studied under the assumption of no unmeasured confounding and there is little work on handling omitted/unmeasured variable bias. This paper focuses on a machine learning method based on random forests known as Causal Forests and presents five simple modifications for tuning Causal Forests so that they are robust to cluster-level unmeasured confounding. Our simulation study finds that adjusting the default tuning procedure with the propensity score from fixed effects logistic regression or using variables that are centered to their cluster means produces estimates that are more robust to cluster-level unmeasured confounding. Also, when these parametric propensity score models are mis-specified, our modified machine learning methods remain robust to bias from cluster-level unmeasured confounders compared to existing parametric approaches based on propensity score weighting. We conclude by demonstrating our proposals in a real data study concerning the effect of taking an eighth-grade algebra course on math achievement scores from the Early Childhood Longitudinal Study.

Validation of the Enhanced Recovery after Surgery (ERAS) society recommendations for liver surgery: a prospective, observational study

Background

Twenty-three recommendations were summarized by the Enhanced Recovery After Surgery (ERAS) society for liver surgery. The aim was to validate the protocol especially with regard to adherence and the impact on morbidity.

Methods

Using the ERAS Interactive Audit System (EIAS), ERAS items were evaluated in patients undergoing liver resection. Over a period of 26 months, 304 patients were prospectively enrolled in an observational study (DRKS00017229). Of those, 51 patients (non-ERAS) were enrolled before and 253 patients (ERAS) after the implementation of the ERAS protocol. Perioperative adherence and complications were compared between the two groups.

Results

Overall adherence increased from 45.2% in the non-ERAS group to 62.7% in the ERAS group (P<0.001). This was associated with significant improvements in the preoperative and postoperative phase (P<0.001), rather than in the outpatient and intraoperative phase (both P>0.05). Overall complications decreased from 41.2% (n=21) in the non-ERAS group to 26.5% (n=67) in the ERAS group (P=0.0423), which was mainly due to the reduction of grade 1-2 complications from 17.6% (n=9) to 7.6% (n=19) (P=0.0322). As for patients undergoing open surgery, implementation of ERAS lead to a reduction of overall complications in patients scheduled for minimally invasive liver surgery (MILS) (P=0.036).

Conclusions

Implementation of the ERAS protocol for liver surgery according to the ERAS guidelines of the ERAS Society reduced Clavien-Dindo grade 1-2 complications particularly in patients who underwent MILS. The ERAS guidelines are beneficial for the outcome, while adherence to the various items has not yet been satisfactorily defined.

The Role of Pre-Treatment Traumatic Stress Symptoms in Adolescent Substance Use Treatment Outcomes

Background: Prominent theories suggest that individuals with co-occurring traumatic stress symptoms (TSS) and substance use (SU) may be less responsive to SU treatment compared to those with SU only. However, empirical findings in adult samples are mixed, and there has been limited work among adolescents. This study assesses the association between TSS and SU treatment outcomes among trauma-exposed adolescents, using statistical methods to reduce potential confounding from important factors such as baseline SU severity. Method: 2,963 adolescents with lifetime history of victimization received evidence-based SU treatment in outpatient community settings. At baseline, 3- and 6-months, youth were assessed using the Global Appraisal of Individual Needs Traumatic Stress Scale and the Substance Frequency Scale. Propensity score weighting was used to mitigate potential confounding due to baseline differences in sociodemographic characteristics and SU across youth with varying levels of TSS. Results: Propensity score weighting successfully balanced baseline differences in sociodemographic factors and baseline SU across youth. Among all youth, mean SU was lower at both 3- and 6- month follow-up relative to baseline, indicating declining use. After adjusting for potential confounders, we observed no statistically significant relationship between TSS and SU at either 3- or 6-month follow-up. Conclusions: Based on this investigation, conducted among a large sample of trauma-exposed youth receiving evidence-based outpatient SU treatment, baseline TSS do not appear to be negatively associated with SU treatment outcomes. However, future research should examine whether youth with TSS achieve better outcomes through integrative treatment for both SU and TSS.

Random effects modelling versus logistic regression for the inclusion of cluster-level covariates in propensity score estimation: A Monte Carlo simulation and registry cohort analysis

Purpose: Surgeon and hospital-related features, such as volume, can be associated with treatment choices and outcomes. Accounting for these covariates with propensity score (PS) analysis can be challenging due to the clustered nature of the data. We studied six different PS estimation strategies for clustered data using random effects modelling (REM) compared with logistic regression. Methods: Monte Carlo simulations were used to generate variable cluster-level confounding intensity [odds ratio (OR) = 1.01-2.5] and cluster size (20-1,000 patients per cluster). The following PS estimation strategies were compared: i) logistic regression omitting cluster-level confounders; ii) logistic regression including cluster-level confounders; iii) the same as ii) but including cross-level interactions; iv), v), and vi), similar to i), ii), and iii), respectively, but using REM instead of logistic regression. The same strategies were tested in a trial emulation of partial versus total knee replacement (TKR) surgery, where observational versus trial-based estimates were compared as a proxy for bias. Performance metrics included bias and mean square error (MSE). Results: In most simulated scenarios, logistic regression, including cluster-level confounders, led to the lowest bias and MSE, for example, with 50 clusters × 200 individuals and confounding intensity OR = 1.5, a relative bias of 10%, and MSE of 0.003 for (i) compared to 32% and 0.010 for (iv). The results from the trial emulation also gave similar trends. Conclusion: Logistic regression, including patient and surgeon-/hospital-level confounders, appears to be the preferred strategy for PS estimation.

Causal Inference with Multilevel Data: A Comparison of Different Propensity Score Weighting Approaches

Propensity score methods are a widely recommended approach to adjust for confounding and to recover treatment effects with non-experimental, single-level data. This article reviews propensity score weighting estimators for multilevel data in which individuals (level 1) are nested in clusters (level 2) and nonrandomly assigned to either a treatment or control condition at level 1. We address the choice of a weighting strategy (inverse probability weights, trimming, overlap weights, calibration weights) and discuss key issues related to the specification of the propensity score model (fixed-effects model, multilevel random-effects model) in the context of multilevel data. In three simulation studies, we show that estimates based on calibration weights, which prioritize balancing the sample distribution of level-1 and (unmeasured) level-2 covariates, should be preferred under many scenarios (i.e., treatment effect heterogeneity, presence of strong level-2 confounding) and can accommodate covariate-by-cluster interactions. However, when level-1 covariate effects vary strongly across clusters (i.e., under random slopes), and this variation is present in both the treatment and outcome data-generating mechanisms, large cluster sizes are needed to obtain accurate estimates of the treatment effect. We also discuss the implementation of survey weights and present a real-data example that illustrates the different methods.

Strategic complements: Poverty-targeted subsidy programs show additive benefits on household toilet purchases in rural Cambodia when coupled with sanitation marketing

While poverty-targeted subsidies have shown promise as a means of reducing financial constraints on low-income populations to invest in new latrines, concerns have been raised about whether they may reduce demand for new latrines among non-eligible, non-poor populations, especially in geographically limited or closed markets. Using quasi experimental methods, we investigate the interaction effects of the "CHOBA" subsidy, a partial poverty-targeted monetary incentive to build a toilet, and a sanitation marketing program (SanMark) on new latrine uptake among households from different income segments in 110 rural villages across six Cambodian provinces. These programs were implemented either jointly with or independently. Overall, we find strong complementarity of the CHOBA subsidy with SanMark where the coupled implementation of the programs increased latrine uptake across all households as compared to exclusive deployment of the programs independently. Additionally, the CHOBA subsidy alone resulted in higher gains among the poor compared to SanMark suggesting that financial constraint is indeed a significant demand barrier for new latrines. The presence of the poverty-targeted subsidies did not reduce demand for new latrine purchases among ineligible households. Instead, we find some evidence for a positive spillover effect of subsidies on uptake of latrines among ineligible households in villages where both programs were implemented indicating that the presence of sanitation subsidies and the decision to purchase latrines among non-beneficiaries can be viewed as complements. We employ multivariate logistic regressions as well as further robustness checks to estimate the effects of the different interventions, with qualitatively consistent results.

An introduction to inverse probability of treatment weighting in observational research

In this article we introduce the concept of inverse probability of treatment weighting (IPTW) and describe how this method can be applied to adjust for measured confounding in observational research, illustrated by a clinical example from nephrology. IPTW involves two main steps. First, the probability—or propensity—of being exposed to the risk factor or intervention of interest is calculated, given an individual’s characteristics (i.e. propensity score). Second, weights are calculated as the inverse of the propensity score. The application of these weights to the study population creates a pseudopopulation in which confounders are equally distributed across exposed and unexposed groups. We also elaborate on how weighting can be applied in longitudinal studies to deal with informative censoring and time-dependent confounding in the setting of treatment-confounder feedback.

Propensity score stratification methods for continuous treatments

Continuous treatments propensity scoring remains understudied as the majority of methods are focused on the binary treatment setting. Current propensity score methods for continuous treatments typically rely on weighting in order to produce causal estimates. It has been shown that in some continuous treatment settings, weighting methods can result in worse covariate balance than had no adjustments been made to the data. Furthermore, weighting is not always stable, and resultant estimates may be unreliable due to extreme weights. These issues motivate the current development of novel propensity score stratification techniques to be used with continuous treatments. Specifically, the generalized propensity score cumulative distribution function (GPS-CDF) and the nonparametric GPS-CDF approaches are introduced. Empirical CDFs are used to stratify subjects based on pretreatment confounders in order to produce causal estimates. A detailed simulation study shows superiority of these new stratification methods based on the empirical CDF, when compared with standard weighting techniques. The proposed methods are applied to the "Mexican-American Tobacco use in Children" study to determine the causal relationship between continuous exposure to smoking imagery in movies, and smoking behavior among Mexican-American adolescents. These promising results provide investigators with new options for implementing continuous treatment propensity scoring.

Analysis of racial differences in hospital stays in the presence of geographic confounding

Using recent methods for spatial propensity score modeling, we examine differences in hospital stays between non-Hispanic black and non-Hispanic white veterans with type 2 diabetes. We augment a traditional patient-level propensity score model with a spatial random effect to create a matched sample based on the estimated propensity score. We then use a spatial negative binomial hurdle model to estimate differences in both hospital admissions and inpatient days. We demonstrate that in the presence of unmeasured geographic confounding, spatial propensity score matching in addition to the spatial negative binomial hurdle outcome model yields improved performance compared to the outcome model alone. In the motivating application, we construct three estimates of racial differences in hospitalizations: the risk difference in admission, the mean difference in number of inpatient days among those hospitalized, and the mean difference in number of inpatient days across all patients (hospitalized and non-hospitalized). Results indicate that non-Hispanic black veterans with type 2 diabetes have a lower risk of hospital admission and a greater number of inpatient days on average. The latter result is especially important considering that we observed much smaller effect sizes in analyses that did not incorporate spatial matching. These results emphasize the need to address geographic confounding in health disparity studies.

Effects of Exposure to the Communities That Care Prevention System on Youth Problem Behaviors in a Community-Randomized Trial: Employing an Inverse Probability Weighting Approach

Earlier intention-to-treat (ITT) findings from a community-randomized trial demonstrated effects of the Communities That Care (CTC) prevention system on reducing problem behaviors among youth. In ITT analyses, youth were analyzed according to their original study community's randomized condition even if they moved away from the community over the course of follow-up and received little to no exposure to intervention activities. Using inverse probability weights (IPWs), this study estimated effects of CTC in the same randomized trial among youth who remained in their original study communities throughout follow-up. Data were from the Community Youth Development Study, a community-randomized trial of 24 small towns in the United States. A cohort of 4,407 youth was followed from fifth grade (prior to CTC implementation) to eighth grade. IPWs for one's own moving status were calculated using fifth- and sixth-grade covariates. Results from inverse probability weighted multilevel models indicated larger effects for youth who remained in their study community for the first 2 years of CTC intervention implementation compared to ITT estimates. These effects included reduced likelihood of alcohol use, binge drinking, smokeless tobacco use, and delinquent behavior. These findings strengthen support for CTC as an efficacious system for preventing youth problem behaviors.