Heterogeneous Treatment Effects Research Articles

The Heterogeneous Effects of College Education on Outcomes Related to Deaths of Despair.

College education features prominently in research on determinants of deaths from substance use disorders and self-harm-outcomes collectively referred to as "deaths of despair" (DoD). Limited attention has been given to whether the protective effects of college education on indicators of despair vary by individuals' likelihood of college completion. We use data from the National Longitudinal Study of Adolescent to Adult Health for 6,145 individuals to test whether the protective effects of college completion on precursors to DoD vary according to individuals' propensity to attain a college degree. Understanding whether the benefits of college education differ depending on the propensity to complete it is important for designing effective educational policies. Using the heterogeneous treatment effects approach, we find that individuals with a relatively low propensity for graduating from college but who complete it have a lower likelihood of depressive symptoms, binge drinking, prescription drug abuse, and hard drug use.

Abstract 4118122: Effect of Beta Blockers on Exercise Capacity, Diastolic Function, and Quality of Life in Patients with Heart Failure with Preserved Ejection Fraction

Introduction: In recent landmark HFpEF trials, 75-80% of patients are taking beta blockers (BB) without clear evidence of clinical or symptomatic benefit. Patients with HFpEF commonly have comorbidities that have been classically managed with BB (AF, CAD, HTN), though the utility of this medication class for these conditions is being reconsidered. It remains unclear what impact BB use, in patients with HFpEF, has on exercise capacity, diastolic function, and quality of life. Hypothesis: We hypothesized that BB use is associated with worse exercise capacity, diastolic function, and quality of life in HFpEF patients. Methods: This retrospective cohort study was completed using deidentified patient data from the INDIE-HFpEF and RELAX clinical trials (two trials of HFpEF patients). Patients were divided into two groups based on BB use at trial enrollment. The primary outcomes were exercise capacity, diastolic function, and quality of life at baseline evaluation. A multivariate logistic regression model was used to assess for statistically significant differences after controlling for potentially confounding variables including age, sex, BMI, AF, prior MI, COPD, DM, HTN, serum creatinine, and NYHA functional class. Results: A total of 320 patients were included in this analysis, with 230 on BB (72%). Baseline demographics are shown in Table 1. Patients studied were older (68 years old), white (90%), obese (average BMI 34 kg/m 2 ) individuals with an average LVEF of 61%. Outcomes and multivariate linear regression results are shown in Table 2. After controlling for potential confounding variables, there were no differences in baseline exercise capacity [peak VO2 (mL/kg/min): 12.1 vs 13.2, P=0.233], diastolic function [average E/e’: 14 vs 12, P=0.125; LAVI (mL/m 2 ): 42 vs 37, P = 0.665; peak TR velocity (m/s): 2.8 vs 2.7, P = 0.165; PASP (mmHg): 41 vs 36, P = 0.357], quality of life [KCCQ: 54 vs 58, P = 0.206; MLHFQ 43 vs 46, P = 0.762], or NT-proBNP (633 vs 342; pg/mL, P = 0.208) in HFpEF patients on BB compared to those not on BB. Conclusion: There was no association with BB use and worse exercise capacity, diastolic function, or quality of life in HFpEF patients. This may be due to heterogenous treatment effects, with differing impact of BB based upon the underlying pathophysiology of unique HFpEF phenotypes. To definitively determine the impact of BB use in patients with HFpEF, both on clinical outcomes and symptoms, randomized controlled trials are necessary.

Heterogeneous Treatment Effects of High-Frequency Oscillatory Ventilation for Acute Respiratory Distress Syndrome: A Post Hoc Analysis of the Oscillation for Acute Respiratory Distress Syndrome Treated Early (OSCILLATE) Trial.

We sought to evaluate whether different subgroups of adults with acute respiratory distress syndrome (ARDS) respond differently to high-frequency oscillatory ventilation (HFOV). The Oscillation for ARDS Treated Early (OSCILLATE) trial was a randomized controlled trial of HFOV vs. conventional ventilation that found an increased risk of in-hospital mortality (primary outcome) with HFOV. In a post hoc analysis, we applied three different approaches to evaluate heterogeneity of treatment effect for in-hospital mortality: 1) subgroup analyses based on baseline Pao2:Fio2 ratio and oxygenation index (OI); 2) a risk-based approach using a multivariable outcome prediction model; and 3) a clustering approach via multivariable latent class analysis. We used multivariable logistic regression models to assess for interaction. Thirty-nine ICUs, five countries. Five hundred forty-eight adults with moderate to severe ARDS. HFOV vs. conventional mechanical ventilation with low tidal volume and higher positive end-expiratory pressure. The effect of HFOV on in-hospital mortality was consistent across categories of Pao2:Fio2 ratio (adjusted odds ratio [aOR], 2.04; 95% CI, 1.32-3.17 and aOR, 1.16; 95% CI, 0.49-2.75 for groups with Pao2:Fio2 above or equal to 80, vs. below 80, respectively; interaction p = 0.23) and OI (aOR, 1.78; 95% CI, 0.67-4.70; aOR, 3.19; 95% CI, 1.44-7.09; aOR, 1.73; 95% CI, 0.82-3.65; and aOR, 1.33; 95% CI, 0.61-2.90 for quartiles of baseline OI, respectively; interaction p = 0.44). Point estimates for the effect of HFOV were consistent across risk categories (aOR, 2.44; 95% CI, 0.40-14.83; aOR, 1.69; 95% CI, 0.75-3.85; and aOR, 2.10; 95% CI, 0.59-7.54 for the lowest, moderate, and highest risk categories, respectively; interaction p = 0.32). Using a clustering approach, point estimates for HFOV were also consistent (cluster 1: aOR, 1.85; 95% CI, 1.15-3.00 and cluster 2: aOR, 1.75; 95% CI, 0.91-3.38; interaction p = 0.75). We did not identify heterogeneity in the effect of HFOV across different subgroups of patients with ARDS.

Isotonic subgroup selection

Abstract Given a sample of covariate–response pairs, we consider the subgroup selection problem of identifying a subset of the covariate domain where the regression function exceeds a predetermined threshold. We introduce a computationally feasible approach for subgroup selection in the context of multivariate isotonic regression based on martingale tests and multiple testing procedures for logically structured hypotheses. Our proposed procedure satisfies a non-asymptotic, uniform Type I error rate guarantee with power that attains the minimax optimal rate up to poly-logarithmic factors. Extensions cover classification, isotonic quantile regression, and heterogeneous treatment effect settings. Numerical studies on both simulated and real data confirm the practical effectiveness of our proposal, which is implemented in the R package ISS.

Tele-collaborative outreach to rural patients with chronic pain: pragmatic effectiveness trial protocol for the CORPs study.

Despite the increased availability of evidence-based treatments for chronic pain, many patients in rural areas experience poor access to services. Patients receiving care through the VA may also need to navigate multiple systems of care. To examine the effectiveness of a remotely delivered collaborative care intervention for improving pain interference among veterans with high-impact chronic pain living in rural areas. We will conduct a four-site pragmatic effectiveness trial of remotely delivered collaborative care for high-impact chronic pain. Participants (n = 608) will be randomized to the Tele-Collaborative Outreach to Rural Patients (CORPs) intervention or to minimally enhanced usual care (MEUC). Participants randomized to CORPs will complete a biopsychosocial assessment and five follow-up sessions with a nurse care manager (NCM), who will collaborate with a consulting clinician to provide personalized recommendations and care management. CORP participants will also be invited to a virtual 6-session pain education group class. Participants randomized to MEUC will receive a one-time education session with the NCM to review available pain services. All participants will complete quarterly research assessments for one year. The primary study outcome is pain interference. This trial will oversample veterans of female birth sex and minoritized race or ethnicity to test heterogeneity of treatment effects across these patient characteristics. We will conduct an implementation process evaluation and incremental cost-effectiveness analysis. This pragmatic trial will test the real-world effectiveness of a remotely delivered collaborative care intervention for chronic pain. Study findings will inform future implementation efforts to support potential uptake of the intervention.

Estimates of intra-cluster correlation coefficients from 2018 USA Medicare data to inform the design of cluster randomized trials in Alzheimer’s and related dementias

BackgroundCluster randomized trials (CRTs) are increasingly important for evaluating interventions embedded in health care systems. An essential parameter in sample size calculation to detect both overall and heterogeneous treatment effects for CRTs is the intra-cluster correlation coefficient (ICC) of both outcome and covariates of interest. However, obtaining advance estimates for the ICC can be challenging. When trial outcomes will be obtained from routinely collected data sources, there is an opportunity to obtain reliable ICC estimates in advance of the trial. Using USA national Medicare data, we estimated ICCs for a range of outcomes to inform the design of CRTs for people living with Alzheimer’s and related dementias (ADRD).MethodData from 2018 Medicare Fee-for-Service beneficiaries, specifically, 1,898,812 individuals (≥ 65 years) with diagnosis of ADRD within 3436 hospital service areas (treated as clusters) and 306 hospital referral regions (treated as fixed strata), were used to calculate unadjusted and adjusted ICC estimates for three outcomes: death, any hospitalizations, and any emergency department (ED) visits and three covariates: age, race and sex. We present both overall and stratum-specific ICC estimates. We illustrate their use in sample size calculations for overall treatment effects as well as detecting treatment effect heterogeneity.ResultsThe unadjusted overall ICCs for death, hospitalizations, and ED visits were 0.001, 0.010, and 0.017 respectively. Stratum-specific ICCs varied widely across the 306 HRRs: median 0.001, 0.010 and 0.025 for death, hospitalizations, and ED visits respectively and 0.007, 0.001, and 0.080 for age, sex and race. An interactive R Shiny app is provided that allows users to retrieve estimates overlayed on a map of the USA.ConclusionsWe presented both adjusted and unadjusted ICCs for outcomes as well as unadjusted ICCs for covariates of potential interest from population-level data in the USA and demonstrated how the estimates may be used in sample size calculations for CRTs in ADRD.

Safer and targeted use of antipsychotics in youth: an embedded, pragmatic randomized trial

BackgroundAntipsychotic medications (AP) are inappropriately prescribed to young people. The goal of this pragmatic trial was to test a four‐component approach to improved targeting of antipsychotic prescribing to people aged ≥3 and <18 years.MethodsClinicians in four health systems were cluster randomized by the number of previous AP orders and service line – specialty mental health and all others. Intervention arm clinicians received a best practice alert and child psychiatrist consultation and feedback. Families received system navigation and expedited access to psychotherapy. Primary outcomes were total days' supply of AP medication and proportion of youth with any AP supply at 6 months. We estimated the log‐odds of AP use at 6 months and the relative rate of AP over 6 months. The Safer and Targeted Use of Antipsychotics in Youth (SUAY) trial took place between 3/2018 and 12/2020.ResultsThe trial enrolled 733 patients. The odds ratio (OR) comparing use at 6 months was 0.75 (95% CI: 0.52, 1.09). The mean number of days using AP was 118.5 for intervention patients and 128.2 for control patients (relative risk [RR] = 0.92; 95% CI: 0.81–1.04). Exploratory heterogeneity of treatment effects (HTE) was not detected in groups defined by age, gender, provider specialty, and insurance type. HTE by race/ethnicity was present: among youth of color, mean days' supply was 103.2 for intervention arm and 131.2 for the control arm (RR 0.79, 95% CI: 0.67–0.93). Among secondary outcomes, only new psychotherapy referrals differed with 44.3% (n = 154) of intervention participants having a new order for psychotherapy compared to 33.5% (n = 129) in the control arm (OR 1.47: 95% CI: 1.01–2.14).ConclusionsThis intervention did not result in less AP use at 6 months or a reduction in the days' supply of AP medication, although psychotherapy orders increased. The intervention may be effective for some subgroups.

Assessment of treatment effect heterogeneity for multiregional randomized clinical trials

Multiregional clinical trials (MRCTs) have become increasingly common in recent years. Detecting underlying regional heterogeneity is a critical issue for these trials. Existing methods for assessing treatment effect heterogeneity across regions have ignored the incomparability of baseline extrinsic risk factors of the randomized patients from different regions. In this paper, a calibration weighting method is proposed to calibrate the distribution of these extrinsic risk factors between multiple regions. We establish the consistency and the asymptotic normality of the calibration weighting estimator. Simulation studies confirm the finite sample properties of the proposed estimator as well as its superior performance over naive methods and the inverse probability weighting method. The proposed method is illustrated using a randomized clinical trial of adjuvant chemotherapy for resected non-small-cell lung cancer.

Icosapent ethyl by baseline small dense low-density lipoprotein cholesterol: an analysis of REDUCE-IT

Abstract Background Small dense low-density lipoprotein cholesterol (sdLDL-C) is associated with cardiovascular (CV) risk. Purpose We assessed if baseline sdLDL-C modified the effect of icosapent ethyl among patients in REDUCE-IT. Methods REDUCE-IT randomized patients to icosapent ethyl vs placebo. In this post hoc analysis, patients ≥45 years with CV disease or ≥50 years with diabetes and CV risk factors were included. Patients were required to have triglycerides of 135-499 mg/dL and low-density lipoprotein cholesterol (LDL-C) of 41-100 mg/dL. Exclusion criteria included severe heart failure, acute severe liver disease, or hemoglobin A1c >10%. Patients were stratified by baseline calculated sdLDL-C. A threshold of 46 mg/dL was chosen because sdLDL-C greater than this cutoff has predicted CV risk despite well-controlled LDL-C. The primary outcome included a composite of CV death, nonfatal myocardial infarction (MI), nonfatal stroke, coronary revascularization, or unstable angina. The key secondary composite outcome included CV death, nonfatal MI, or nonfatal stroke. Outcomes were assessed with Cox proportional hazard models and interaction analyses were conducted to assess heterogeneity in treatment effect by baseline sdLDL-C. Results REDUCE-IT included 8179 patients, with 8157 (99.7%) patients having sdLDL-C data. Among these patients, 7698 (94.4%) had sdLDL-C <46 mg/dL and 459 (5.6%) had sdLDL-C ≥46 mg/dL. Median sdLDL-C was 34.2 mg/dL (interquartile range [IQR]: 30.3, 38.4 mg/dL) in the lower sdLDL-C group and 48.4 mg/dL (IQR: 47.1, 50.6 mg/dL) in the higher sdLDL-C group (Table). The placebo group had a greater event rate in the higher sdLDL-C group compared to the lower sdLDL-C group (72.0 vs 56.4 events per 1000 p-y), though event rates in the icosapent ethyl group were similar by sdLDL-C group (44.1 vs 43.3 events per 1000 p-y, respectively). Icosapent ethyl reduced the rate of the primary outcome compared with placebo (17.2% vs 22.0%; hazard ratio [HR]: 0.75 [95% CI: 0.68, 0.83]; P<0.0001). The number needed to treat [NNT] to avoid one primary endpoint event was 21. In the lower sdLDL-C group, icosapent ethyl decreased rates of the primary outcome (43.3 events per 1000 patient-years [p-y]) compared with the placebo group (56.4 events per 1000 p-y) (17.3% vs 21.7%; HR: 0.77 [95% CI: 0.69, 0.85]; NNT: 22). Similarly, in the higher sdLDL-C group, icosapent ethyl decreased the rate of the primary outcome (44.1 events per 1000 p-y) compared with the placebo group (72.0 events per 1000 p-y) (16.6% vs 26.4%; HR: 0.58 [95% CI: 0.38, 0.88]; NNT: 10). There was no significant interaction between treatment benefit and baseline sdLDL-C for the primary (Pinteraction = 0.22) and key secondary outcome (Pinteraction = 0.82). Findings were overall similar for the other secondary outcomes (Figure). Conclusion Icosapent ethyl reduced CV outcomes in patients with high CV risk and elevated triglycerides irrespective of low or high sdLDL-C.

Blood pressure-lowering and risk of cancer: an individual participant-level data meta-analysis and mendelian randomisation studies

Abstract Background The current body of randomised evidence concerning the effect of blood pressure lowering on the risk of cancer remains limited. Purpose We leveraged the strengths of both randomised controlled trials (RCTs) and genomic information to examine this question. Methods Individual-level data from 314,016 participants from 42 RCTs were pooled to investigate the effect of blood pressure lowering on the risk of cancer through one-stage individual participant data (IPD) meta-analysis. The primary outcome was incident cancer, defined as the first occurrence of any cancer diagnosed after randomisation. Pre-specified subgroup analyses were conducted to assess heterogeneity in effect by follow-up time and baseline age groups, sex, body mass index categories, smoking status and previous use of antihypertensive drugs. Secondary outcomes were cause-specific cancer death and site-specific cancer risk comprising breast, colorectal, kidney, lung, prostate, and skin. Cox proportional hazard regression, stratified by trials, were used for statistical analyses. For site-specific cancers, analyses were complemented with Mendelian randomisation analyses using naturally randomised genetic variants associated with blood pressure lowering, retrieved from genome-wide association studies involving participants of European ancestry. Results Over a median of 4 years (interquartile range 2), 17,954 participants in RCTs were diagnosed with cancer of any type and 4,878 participants were reported to have died with cancer as the cause of death. In the IPD meta-analysis that compared the treatment arm with the comparator, no associations were identified between reductions in either systolic or diastolic blood pressure and the risk of incident cancer (hazard ratios [HRs] per 5 mmHg reduction in systolic and per 3 mmHg diastolic blood pressure were 1.03 (95% confidence interval [CI] 0.99-1.06) and 1.03 [0.98-1.07], respectively). We also found no pattern of increasing or decreasing relative risk for incident cancer during the follow-up period, nor was there evidence of heterogeneity in treatment effects across baseline subgroups. No effect on the risk of cause-specific cancer death was identified for either systolic or diastolic blood pressure lowering. Considering site-specific cancers, we also found no evidence of effect, except for a possible link with lung cancer risk, with HRs of 1.17 [99.5% CI 1.02-1.32] for systolic blood pressure and 1.17 [99.5% CI 0.98-1.36] for diastolic blood pressure lowering. In Mendelian randomisation studies, no association was observed for genetically determined systolic and diastolic blood pressure and all the considered site-specific cancers, including overall lung cancer and its subtypes. Conclusion We found no consistent randomised evidence to suggest that blood pressure-lowering has a substantial effect, whether increasing or decreasing, on incident cancer, cause-specific cancer death, or selected site-specific cancers.

Menopausal status and clinical outcomes in women with heart failure with reduced ejection fraction: the GALACTIC-HF trial

Abstract Introduction Mechanisms driving disease progression and potential responsiveness to investigational therapies may theoretically differ among women with heart failure (HF) based on menopausal status. However, few data are available describing the clinical course of premenopausal and postmenopausal women with HF. We investigated differences in baseline characteristics, clinical outcomes, and response to omecamtiv mecarbil in premenopausal and postmenopausal women participants of the GALACTIC-HF trial. Methods The GALACTIC-HF trial randomized patients with symptomatic HFrEF with LVEF of 35% or less to omecamtiv mecarbil or placebo. Menopausal status at randomization was collected. When menopausal status was unknown/unreported, menopausal status was inferred based on the expected distribution of age at menopause in the general population. The risk of the primary endpoint, worsening HF event or cardiovascular death, was compared among women by menopausal status using Cox regression models adjusted for clinical covariates, including age and baseline EF. Treatment effect heterogeneity was evaluated by menopausal status. Results Of the 8,232 patients enrolled, 1,749 (21.2%) were women. Menopausal status was available at baseline in 1,483 (84.8%). One hundred women reported being potentially childbearing (pre-menopausal) and 1,383 women reported being post-menopausal. Among the 266 patients that had unknown menopausal status, those younger than 47 years were additionally classified as pre-menopausal and those older than 56 years as post-menopausal based on estimated 10th /90th percentiles extrapolated from the general population. Compared to post-menopausal women, pre-menopausal women were younger, had higher eGFR, and had fewer comorbidities. Premenopausal women also had slightly higher BMI and lower LVEF, but lower NTproBNP and troponin at baseline compared to postmenopausal women. Premenopausal and postmenopausal women had similarly high burden of symptoms as assessed by the KCCQ-total symptoms score (median 63.5 vs. 64.6; P=0.79). During a median follow-up of 22 months, premenopausal and postmenopausal women experienced similar rates of primary and secondary outcomes (Table 1). Omecamtiv mecarbil was beneficial across the entire female cohort with no effect modification by menopausal status (Pinteraction=0.39). Sensitivity analysis excluding all women with unknown menopausal status yielded similar results. Conclusions In this high-risk HFrEF cohort, premenopausal women experienced similar rates of adverse cardiovascular outcomes as post-menopausal women despite younger age, fewer comorbidities, and lower cardiac biomarkers. Additional research is necessary to understand how risk factors unique to females might influence outcomes in those with heart failure. The benefit of omecamtiv mecarbil on reducing cardiovascular risk was consistent regardless of menopausal status.Primary/secondary outcomes by menopausal

Digital health technology to support patient-centered shared decision making at point of care for juvenile idiopathic arthritis.

Despite availability of multiple FDA approved therapies, many children with juvenile idiopathic arthritis (JIA) suffer pain and disability due to uncontrolled disease. The term JIA includes a heterogeneous set of conditions unified by chronic inflammatory arthritis, collectively affecting 1:1,000 children. When reviewing treatment options with families the rheumatologist currently refers to the experience of the average patient in relatively small controlled clinical trials, to consensus-based treatment plans, or increasingly the choice is dictated by the formulary restrictions of insurance payers. The current paradigm for treatment selection does not incorporate real-world evidence of treatment effectiveness centered to the individual patients with whom decisions are to be made. Treatment decisions based on the evidence of the average patient are not optimized to reflect the unique clinical characteristics of an individual with JIA and their disease course, nor does it account for heterogeneous treatment effects. To guide treatment choices centered around each patient, we describe a novel concept of utilizing digital health technology to bring patient-centered information into shared decision-making discussions based on comparative effectiveness analysis of electronic health record or observational clinical registry data of patients with similar characteristics. The envisioned digital tool will organize and present data relevant to the individual patient and enable evidence-based individualized treatment decision making when used in a collaborative manner with the patient family and rheumatologist. Capabilities in digital health technology, data capturing, and analytical methodologies are ripe for this endeavor. This brings the concept of a learning health system directly to the point of care.

Heterogeneity in the Effect of Early Goal-Directed Therapy for Septic Shock: A Secondary Analysis of Two Multicenter International Trials.

The optimal approach for resuscitation in septic shock remains unclear despite multiple randomized controlled trials (RCTs). Our objective was to investigate whether previously uncharacterized variation across individuals in their response to resuscitation strategies may contribute to conflicting average treatment effects in prior RCTs. We randomly split study sites from the Australian Resuscitation of Sepsis Evaluation (ARISE) and Protocolized Care for Early Septic Shock (ProCESS) trials into derivation and validation cohorts. We trained machine learning models to predict individual absolute risk differences (iARDs) in 90-day mortality in derivation cohorts and tested for heterogeneity of treatment effect (HTE) in validation cohorts and swapped these cohorts in sensitivity analyses. We fit the best-performing model in a combined dataset to explore roles of patient characteristics and individual components of early goal-directed therapy (EGDT) to determine treatment responses. Eighty-one sites in Australia, New Zealand, Hong Kong, Finland, Republic of Ireland, and the United States. Adult patients presenting to the emergency department with severe sepsis or septic shock. EGDT vs. usual care. A local-linear random forest model performed best in predicting iARDs. In the validation cohort, HTE was confirmed, evidenced by an interaction between iARD prediction and treatment (p < 0.001). When patients were grouped based on predicted iARDs, treatment response increased from the lowest to the highest quintiles (absolute risk difference [95% CI], -8% [-19% to 4%] and relative risk reduction, 1.34 [0.89-2.01] in quintile 1 suggesting harm from EGDT, and 12% [1-23%] and 0.64 [0.42-0.96] in quintile 5 suggesting benefit). Sensitivity analyses showed similar findings. Pre-intervention albumin contributed the most to HTE. Analyses of individual EGDT components were inconclusive. Treatment response to EGDT varied across patients in two multicenter RCTs with large benefits for some patients while others were harmed. Patient characteristics, including albumin, were most important in identifying HTE.

The effect of grandparental involvement on grandchildren's school grades: Heterogeneity by the extended family characteristics

As the early years are crucial for individuals' lifelong socioeconomic success, extensive research has examined the impact of non-maternal childcare on children's development. This study aims to enhance the understanding of the relationship between grandparental involvement (defined as grandparent childcare, frequency of contact, and financial support) and grandchildren's school grades, exploring a mechanism of positive selection: children from extended families with specific socioeconomic characteristics are more likely to spend time with grandparents and benefit the most from this involvement.We utilize data from the German Pairfam survey, which uniquely provides rich information on three family generations. By conducting a heterogeneous treatment effect analysis, we account for confounding factors associated with grandparental involvement and school performance that could bias our findings. For example, children from advantaged families could be more likely to spend time with grandparents and achieve better school grades. Additionally, this approach examines whether the effect of grandparental involvement systematically varies across children based on the extended family's characteristics. For example, children from advantaged families may benefit the most from spending time with grandparents who possess social, cultural, and cognitive resources conducive to their development.After accounting for confounding factors and heterogeneity, our analyses do not reveal a statistically significant effect of grandparental investment on children's school grades. The study concludes by discussing possible reasons for this result and highlighting the implications for the intergenerational transmission of inequality.

An Approach for Combining Clinical Judgment with Machine Learning to Inform Medical Decision Making: Analysis of Nonemergency Surgery Strategies for Acute Appendicitis in Patients with Multiple Long-Term Conditions.

Machine learning (ML) methods can identify complex patterns of treatment effect heterogeneity. However, before ML can help to personalize decision making, transparent approaches must be developed that draw on clinical judgment. We develop an approach that combines clinical judgment with ML to generate appropriate comparative effectiveness evidence for informing decision making. We motivate this approach in evaluating the effectiveness of nonemergency surgery (NES) strategies, such as antibiotic therapy, for people with acute appendicitis who have multiple long-term conditions (MLTCs) compared with emergency surgery (ES). Our 4-stage approach 1) draws on clinical judgment about which patient characteristics and morbidities modify the relative effectiveness of NES; 2) selects additional covariates from a high-dimensional covariate space (P > 500) by applying an ML approach, least absolute shrinkage and selection operator (LASSO), to large-scale administrative data (N = 24,312); 3) generates estimates of comparative effectiveness for relevant subgroups; and 4) presents evidence in a suitable form for decision making. This approach provides useful evidence for clinically relevant subgroups. We found that overall NES strategies led to increases in the mean number of days alive and out-of-hospital compared with ES, but estimates differed across subgroups, ranging from 21.2 (95% confidence interval: 1.8 to 40.5) for patients with chronic heart failure and chronic kidney disease to -10.4 (-29.8 to 9.1) for patients with cancer and hypertension. Our interactive tool for visualizing ML output allows for findings to be customized according to the specific needs of the clinical decision maker. This principled approach of combining clinical judgment with an ML approach can improve trust, relevance, and usefulness of the evidence generated for clinical decision making. Machine learning (ML) methods have many potential applications in medical decision making, but the lack of model interpretability and usability constitutes an important barrier for the wider adoption of ML evidence in practice.We develop a 4-stage approach for integrating clinical judgment into the way an ML approach is used to estimate and report comparative effectiveness.We illustrate the approach in undertaking an evaluation of nonemergency surgery (NES) strategies for acute appendicitis in patients with multiple long-term conditions and find that NES strategies lead to better outcomes compared with emergency surgery and that the effects differ across subgroups.We develop an interactive tool for visualizing the results of this study that allows findings to be customized according to the user's preferences.

The application of Bayesian method in estimating heterogeneity of treatment effect

It is well recognized that the benefit of medical interventions may not be distributed evenly in the target population due to patient heterogeneity, and conclusions based on traditional randomized clinical trials do not apply to everyone. Considering the increasing cost of randomized trials and difficulties in recruiting patients, there is an urgent need to develop analytical methods to estimate treatment effects in subgroups. In particular, due to limited sample sizes of subgroups, standard analysis tends to yield wide confidence intervals for the treatment effect, which are often uninformative. In this article, the superiority of the empirical Bayes estimate of the conditional average treatment effect is demonstrated by using a linear regression model to specifically represent treatment-covariate interactions. To obtain more accurate prior distributions and posterior estimates of treatment effects in subgroups, we utilize the covariate-dependent precision parameter across subgroups. Based on subgroup data obtained from clinical trials, the covariate-dependent precision parameter is used to assess the similarity of treatment effects across subgroups, which is an approach to obtain more reasonable prior information. To verify the validity of the empirical Bayes method, we evaluate it by a real clinical trial comparing insulin and metformin.

Design-Based Theory for Lasso Adjustment in Randomized Block Experiments and Rerandomized Experiments

Blocking, a special case of rerandomization, is routinely implemented in the design stage of randomized experiments to balance the baseline covariates. This study proposes a regression adjustment method based on the least absolute shrinkage and selection operator (Lasso) to efficiently estimate the average treatment effect in randomized block experiments with high-dimensional covariates. We derive the asymptotic properties of the proposed estimator and outline the conditions under which this estimator is more efficient than the unadjusted one. We provide a conservative variance estimator to facilitate valid inferences. Our framework allows one treated or control unit in some blocks and heterogeneous propensity scores across blocks, thus including paired experiments and finely stratified experiments as special cases. We further accommodate rerandomized experiments and a combination of blocking and rerandomization. Moreover, our analysis allows both the number of blocks and block sizes to tend to infinity, as well as heterogeneous treatment effects across blocks without assuming a true outcome data-generating model. Simulation studies and two real-data analyses demonstrate the advantages of the proposed method.

Quantile‐Based Test for Heterogeneous Treatment Effects

ABSTRACTWe introduce a permutation test for heterogeneous treatment effects based on the quantile process. However, tests based on the quantile process often suffer from estimated nuisance parameters that jeopardize their validity, even in large samples. To overcome this problem, we use Khmaladze's martingale transformation. We show that the permutation test based on the transformed statistic controls size asymptotically. Numerical evidence asserts the good size and power performance of our test procedure compared to other popular quantile‐based tests. We discuss a fast implementation algorithm and illustrate our method using experimental data from a welfare reform.

Rectifying local governments’ strategic environmental enforcement: Can refined air pollution monitoring enhance local air quality in China?

Insufficient local environmental enforcement has become a pivotal economic and political concern in China. In 2017, the Chinese central government introduced the Air Pollution Hot-Spot Grid (APHG) monitoring program. However, little is known about the consequences. This study investigates in detail how this monitoring program affects city-level air pollution by applying the difference-in-differences method. This program was responsible for a notable decrease in air pollution at the city level. This decrease was mostly the result of stronger local environmental governance, advances in green innovation, and an upgraded industrial structure. Additionally, an examination of heterogeneous treatment effects across different cities revealed that the most substantial effects were observed in large, non-resource-based, and eastern cities. Additional analysis reveals the presence of a spillover effect, whereby the adoption of this program in one city helps reduce air pollution in nearby cities. These results highlight the significance of refined air pollution monitoring for addressing imperfect local environmental enforcement.

Considerations for using tree-based machine learning to assess causation between demographic and environmental risk factors and health outcomes.

Evaluation of the heterogeneous treatment effect (HTE) allows for the assessment of the causal effect of a therapy or intervention while considering heterogeneity in individual factors within a population. Machine learning (ML) methods have previously been employed for HTE evaluation, addressing the limitations associated with modelling complex systems. In this work, three tree-based ML algorithms, causal random forest (CRF), causal Bayesian additive regression trees (CBART), and causal rule ensemble (CRE), are used to analyze the potential causation of benzene exposure to cause childhood acute myeloid leukemia (AML). Data for this analysis is generated by drawing samples from a previously developed model that estimates AML probability given as input demographic information and benzene exposure. Comparison is drawn between the three tree-based algorithms in terms of the predicted average treatment effect (ATE), the regression coefficient of determination, and the computational time of each algorithm. Minimal difference is reported between the three tree-based algorithms in terms of the ATE, as well as the regression coefficient of determination. However, CRF outperforms CBART in terms of algorithm computational time. Moreover, CRF allows for both continuous and binary treatment variables, as opposed to CBART and CRE, making it better suited to environmental health studies, where exposure levels of pollutants shall be considered continuous. Following the comparison of all three algorithms, the influence of adding Gaussian noise to the treatment and outcome variables, as well as outliers, is investigated using CRF. A set of considerations is drawn to guide researchers in using these algorithms. These considerations detail the simulation settings, applications, and results interpretation and aim to provide prompt information in decision-making surrounding the establishment of pollutant exposure thresholds in environmental risk assessments.