Inverse Probability Weighting Approach Research Articles

Assessing Etiologic Heterogeneity for Multinomial Outcome with Two-Phase Outcome-Dependent Sampling Design.

Etiologic heterogeneity occurs when distinct sets of events or exposures give rise to different subtypes of disease. Inference about subtype-specific exposure effects from two-phase outcome-dependent sampling data requires adjustment for both confounding and the sampling design. Common approaches to inference for these effects do not necessarily appropriately adjust for these sources of bias, or allow for formal comparisons of effects across different subtypes. Herein, using inverse probability weighting (IPW) to fit a multinomial model is shown to yield valid inference with this sampling design for subtype-specific exposure effects and contrasts thereof. The IPW approach is compared to common regression-based methods for assessing exposure effect heterogeneity using simulations. The methods are applied to estimate subtype-specific effects of various exposures on breast cancer risk in the Carolina Breast Cancer Study.

Why Are We Weighting? Understanding the Estimates From Propensity Score Weighting and Matching Methods.

Propensity score methods are used in observational studies to compensate for the lack of random allocation by balancing measured baseline characteristics between treated and untreated patients. We sought to explain the treatment effect estimates derived from different propensity score methods. We performed a retrospective analysis of long-term mortality after single internal mammary artery versus bilateral internal mammary artery (BIMA) conduit in 47 984 index isolated coronary artery bypass grafting procedures from 1992 to 2014 in the Northern New England Cardiovascular Disease Study Group registry using multivariable Cox regression, 1:1 propensity score matching, inverse probability weighting (IPW) among the treated, and IPW among the overall population treatment estimates. The mean duration of follow-up was 13.2 (interquartile range, 7.4-17.7) years. In multivariable Cox regression, the adjusted hazard ratio for mortality was 0.83 (95% CI, 0.75-0.92) in patients receiving BIMA compared with a single internal mammary artery. The 1:1 propensity matched (hazard ratio, 0.79 [95% CI, 0.69-0.91]) and IPW among the treated (hazard ratio, 0.83 [95% CI, 0.75-0.92]) estimates showed a protective treatment effect of BIMA use on mortality. However, the IPW estimate of treatment effect for the overall population showed an increased risk of mortality after BIMA that was not statistically significant (hazard ratio, 1.08 [95% CI, 0.94-1.24]). While the multivariable Cox regression, 1:1 propensity matching, and IPW treatment effect in the treated estimates demonstrate that BIMA was associated with a statistically significantly decreased risk of mortality, the IPW treatment effect in the average study population showed an increased risk of mortality associated with BIMA that was not statistically significant. This is attributed to the different populations (weighted to look like the overall study population versus treated group) represented by the 2 IPW approaches. Determining how the study population is balanced is a large driver of the treatment effect. Ultimately, the treatment effect estimate desired should drive the choice of the propensity score method.

Joint evaluation of placebo and treatment effects in cluster randomized trials by causal inference models

The term placebo effect refers to the psychobiological effect of a patient's knowledge or belief of being treated. A placebo effect is patient-driven, which makes it fundamentally different from the usual treatment effect resulting from external actions. In modern clinical research, the presence of a placebo effect is often treated as a nuisance issue, something to be “adjusted away” in estimating a treatment effect of primary interest. However, from a patient-centered perspective, we believe that a possible placebo produces substantial improvements in patient-centered outcomes. Understanding placebo effects is therefore an important part of patient-centered outcomes research. The available methods for estimating placebo effects are designed for individually randomized trials and are not directly applicable to cluster randomized trials (CRTs). There are several challenges in estimating placebo effects in CRTs. A major challenge is the possible presence of interference within clusters, in the sense that a subject's outcome may depend on the beliefs subjects in the same cluster about treatment assignment (mentality) and therefore possible correlation in outcome and mentality among subjects exists in the same cluster. In this article, we extend the previously developed causal inference framework to also encompass CRTs, using the G-Computation and inverse probability weighting (IPW) approaches. We also develop methodologies and further extend the G-Computation and IPW approaches to handle missingness for jointly evaluating placebo effect and treatment-specific effect, specifically in the context of CRTs. The proposed methods are demonstrated in simulation studies and a cluster randomized trial on effect of fermented dairy drink.

Vaccination against coronavirus disease 2019 in patients with pulmonary hypertension: A national prospective cohort study.

Coronavirus disease 2019 (COVID-19) has potential risks for both clinically worsening pulmonary hypertension (PH) and increasing mortality. However, the data regarding the protective role of vaccination in this population are still lacking. This study aimed to assess the safety of approved vaccination for patients with PH. In this national prospective cohort study, patients diagnosed with PH (World Health Organization [WHO] groups 1 and 4) were enrolled from October 2021 to April 2022. The primary outcome was the composite of PH-related major adverse events. We used an inverse probability weighting (IPW) approach to control for possible confounding factors in the baseline characteristics of patients. In total, 706 patients with PH participated in this study (mean age, 40.3 years; mean duration after diagnosis of PH, 8.2 years). All patients received standardized treatment for PH in accordance with guidelines for the diagnosis and treatment of PH in China. Among them, 278 patients did not receive vaccination, whereas 428 patients completed the vaccination series. None of the participants were infected with COVID-19 during our study period. Overall, 398 patients received inactivated severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) vaccine, whereas 30 received recombinant protein subunit vaccine. After adjusting for baseline covariates using the IPW approach, the odds of any adverse events due to PH in the vaccinated group did not statistically significantly increase (27/428 [6.3%] vs. 24/278 [8.6%], odds ratio = 0.72, P = 0.302). Approximately half of the vaccinated patients reported at least one post-vaccination side effects, most of which were mild, including pain at the injection site (159/428, 37.1%), fever (11/428, 2.6%), and fatigue (26/428, 6.1%). COVID-19 vaccination did not significantly augment the PH-related major adverse events for patients with WHO groups 1 and 4 PH, although there were some tolerable side effects. A large-scale randomized controlled trial is warranted to confirm this finding. The final approval of the COVID-19 vaccination for patients with PH as a public health strategy is promising.

Practical strategies for operationalizing optimal allocationin stratified cluster-based outcome-dependent sampling designs.

Cluster-based outcome-dependent sampling (ODS) has the potential to yield efficiency gains when the outcome of interest is relatively rare, and resource constraints allow only a certain number of clusters to be visited for data collection. Previous research has shown that when the intended analysis is inverse-probability weighted generalized estimating equations, and the number of clusters that can be sampled is fixed, optimal allocation of the (cluster-level) sample size across strata defined by auxiliary variables readily available at the design stage has the potential to increase efficiency in the estimation of the parameter(s) of interest. In such a setting, the optimal allocation formulae depend on quantities that are unknown in practice, currently making such designs difficult to implement. In this paper, we consider a two-wave adaptive sampling approach, in which data is collected from a first wave sample, and subsequently used to compute the optimal second wave stratum-specific sample sizes. We consider two strategies for estimating the necessary components using the first wave data: an inverse-probability weighting (IPW) approach and a multiple imputation (MI) approach. In a comprehensive simulation study, we show that the adaptive sampling approach performs well, and that the MI approach yields designs that are very near-optimal, regardless of the covariate type. The IPW approach, on the other hand, has mixed results. Finally, we illustrate the proposed adaptive sampling procedures with data on maternal characteristics and birth outcomes among women enrolled in the Safer Deliveries program in Zanzibar, Tanzania.

Age-Related Sex Disparities in Esophageal Cancer Survival: A Population-Based Study in the United States.

BackgroundThe association between sex and the survival of patients with esophageal cancer (EC) remains controversial. We sought to systematically investigate sex-based disparities in EC survival using the Surveillance, Epidemiology, and End Results (SEER) registry data from the United States.MethodsPatients with EC diagnosed from 2004 to 2015 registered in the SEER database were selected. The association between sex and cancer-specific survival (CSS) was evaluated using survival analysis. The Inverse Probability Weighting (IPW) approach was applied to reduce the observed bias between males and females. Subgroup analyses were used to investigate the robustness of the sex-based disparity and to explore potential interaction effects with other variables.ResultsOverall, 29,312 eligible EC patients were analyzed, of whom 5,781 were females, and 23,531 were males. Females had higher crude CSS compared to males (10-year CSS: 24.5 vs. 21.3%; P < 0.001). Similar results were obtained after adjusting for selection bias using the IPW approach and multivariate regression. Subgroup analyses confirmed the relative robustness of sex as a prognostic factor. However, significant interactions were observed between sex and other variables, such as age, race, tumor grade, histology, and treatment modality. In particular, there was no survival advantage for premenopausal females compared to their male counterparts, but the association between sex and EC survival was prominent in 46–55-year-old patients.ConclusionsFemale EC patients had better long-term survival than males. The association between sex and EC survival vary according to age, race, tumor grade, histology, and treatment modality. Sex-based disparity in EC-specific survival was age-related in the United States population.

Sex Differences in Extrahepatic Outcomes After Antiviral Treatment for Hepatitis C.

Despite recognized differences in the rates of cardiovascular and renal disease between men and women in the general population, studies of the downstream effects of antiviral treatment for hepatitis C (HCV) have not investigated differences in outcomes based on sex. We analyzed sex differences in risk of acute coronary syndrome (ACS), end-stage renal disease (ESRD), and ischemic stroke by treatment and response in a large US-based multisite cohort of HCV patients. Observation started at the HCV diagnosis date (untreated) or last antiviral treatment start (treated). Treatment selection bias was addressed using an inverse probability-weighting approach. We estimated the effect of treatment on the cumulative incidence of outcomes using the Fine-Gray method (subdistribution hazard ratios [sHR] and 95% confidence intervals [95% CI]). Death was a competing risk. Roughly 40% of 15,295 HCV patients were women. After controlling for other risk factors, sustained virological response (SVR) (interferon-based [IFN] or direct-acting antiviral [DAA]) significantly reduced risk of all outcomes, particularly among female patients. Female patients who achieved SVR after IFN-based treatment had significantly lower risk of ACS compared with male patients with SVR from either treatment type (sHR 0.45 [95% CI 0.35-0.59] vs 0.81 [95% CI 0.69-0.96, for DAA SVR] and sHR 0.72 [95% 0.62, 0.85, for IFN SVR]). Successful treatment seemed to be most protective against ESRD; female patients who achieved SVR were at 66%-68% lower risk than untreated patients (sHR 0.32 [95% CI 0.17-0.60 for DAA SVR] and 0.34 [95% CI 0.20-0.58 for IFN SVR]), whereas men were at 38%-42% lower risk (sHR 0.62 [95% CI 0.46-0.85 for DAA SVR] and 0.58 [95% CI 0.43-0.76 for IFN SVR]). IFN treatment failure significantly increased risk of all outcomes by 50%-100% among female patients. Compared with no treatment, female patients who experienced IFN treatment failure were at 63% increased risk of ACS (sHR 1.63 [95% CI 1.35-1.96]), almost twice the risk of ESRD (sHR 1.95 [95% CI 1.43-2.66]) and 51% increased risk of stroke (sHR 1.49 [95%CI 1.11-2.00]). SVR reduced the risk of extrahepatic complications, particularly in females. The significantly increased risk associated with IFN TF in women-a subset who represented roughly 10% of that group-underscores the importance of prioritizing these patients for DAA treatment irrespective of the fibrosis stage.

PNS216 Beyond the Statistical Methods: Design Strategies Impacting the Method to Compare Cohorts in Prospective Observational Studies

The comparison of outcomes from different cohorts in observational studies is subject to many biases, due in part to important imbalances between the characteristics of the cohorts. Many statistical methods exist to control confounding in this context, such as propensity score matching (PSM), inverse probability weighting (IPW), regression (R), and exact matching (EM). Most papers comparing methods either explain the statistical mechanism of each method or perform simulations. In real-word studies, most assumptions cannot be verified. Therefore, there is a need to provide clear guidance on planning the study design scenario for which each method would be optimal, not only from an analytic perspective but from a feasibility approach that considers the essential elements of the underlying research question. The authors present several popular methods to account for confounding variables in the context of a prospective comparative study and illustrate the pros and cons of those methods applied in typical prospective study scenarios. Several aspects were considered for the conceptualization of the study, such as data sources, data quality, data availability, overlap, and recruitment logistics. Statistical methods presented will focus on PSM, IPW, R, and EM. The methods will be investigated for external control arms, rare diseases, and prevalent disease studies. We illustrated through a short simulation [N=is 100, 1 binary treatment assignment, 6 other baseline covariates, treatment choice associated with 3 of the baseline covariates], that matching methods were not always preferred. Using PSM, p value=0.146, and with a doubly-robust IPW approach p-value 0.021. The statistical method choice is often influenced by many factors that are outside of the statistical considerations. There is no single statistical method that will work as a best strategy for addressing confounding in all prospective observational studies, and methods are influenced by the design, data quality and study feasibility.

Abstract 171: User Beware: Understanding the Adjusted Estimates of Propensity Score Methods

Background: Propensity score methods are used in observational studies to compensate for the lack of random allocation by balancing measured baseline characteristics between treated and untreated patients. Yet default implementations of propensity score methods may lead to contradictory findings. Our goal was to illustrate differences in treatment effect estimates produced using propensity score methods such as matching and inverse probability weighting (IPW). Methods: We used data from a retrospective analysis of the Northern New England Cardiovascular Disease Study Group registry that studied reintervention after single internal mammary artery (SIMA) versus bilateral internal mammary artery (BIMA) conduit use in 41,481 coronary artery bypass grafting (CABG) procedures from 1992-2014. Results: The mean duration of follow-up was 13.2 (IQR: 7.4-17.7) years. In our standard multivariable Cox regression analysis, the adjusted HR for reintervention was 0.83 (95% confidence interval (CI): 0.75-0.92) in patients receiving BIMA compared to SIMA (Table). The 1:1 propensity matched analysis (HR=0.79, 95% CI: 0.69-0.91) and IPW estimate among the treated (HR=0.83, 95% CI: 0.75-0.92) show a similar protective treatment effect of BIMA use on reintervention. However, the IPW approach for the overall population effect unusually showed no difference between BIMA and SIMA on reintervention-free survival (HR=1.08, 95% CI: 0.94-1.24). Conclusions and Relevance: The adjusted HR for reintervention after BIMA versus SIMA use in CABG was remarkably different for the two IPW estimates. This difference is attributed to the population represented by the two IPW approaches- the IPW among the treated estimate represents the effect in the average study population, whereas the IPW in the treated represents the effect in the treated population alone. Determining how the study population is balanced (weighted to look like the overall study population versus treated group) is a large driver of treatment effect, and a key element of propensity score methods. Ultimately, the treatment effect estimate desired should drive the choice of propensity score adjustment method.

IVF success corrected for drop-out: use of inverse probability weighting.

Does inverse probability weighting (IPW) provide a more valid estimate of the cumulative incidence of live birth after multiple cycles of IVF? IPW can provide a more accurate estimate of treatment success for counseling and decision-making regarding IVF. Different approaches have been used to define and calculate IVF success; however, many of these approaches have limitations and potentially violate statistical assumptions. IPW can address potential selection bias that arises when people do not continue IVF treatment after a failed cycle. Data were derived from a cohort study of women undergoing their first fresh embryo transfer IVF cycle at our institution between 1 January 1995 and 31 December 2014. All autologous cycles (fresh and frozen) were included, up to six total cycles. We identified 20 015 women who underwent 47 079 IVF cycles and had 10 031 live births during the study period. The cumulative incidence of live birth was calculated using three approaches. First, we used a standard Kaplan-Meier approach, 'the optimistic approach', censoring women when they dropped out of treatment. Second, we used a 'conservative' Kaplan-Meier approach that assumed women who dropped out of treatment did not achieve a live birth. Finally, we used IPW to calculate the probability of remaining in treatment, while accounting for differences in treatment drop out. IPW up-weights the data of those remaining under observation who resembled the women who dropped out of treatment, thereby decreasing the potential selection bias resulting from loss to follow-up. The IPW was incorporated into a Kaplan-Meier approach. The cumulative incidence of live birth was 72.1% (95% CI: 71.0-73.1%) for the optimistic approach, 50.1% (49.4-50.8%) for the conservative approach and 66.8% (65.5-68.1%) for the IPW approach. Among women < 38 years of age, the cumulative incidence of live birth calculated by the IPW was slightly higher than that calculated by the optimistic approach. For women 41-42 years of age, the IPW cumulative incidence of live birth was slightly lower. The IPW was similar to the optimistic approach for the other age groups. The conservative estimate was lowest for all age groups. Only clinical data recorded by the providers during an IVF cycle were used to generate weights for IPW. Covariates included: age, gravidity and year at the start of the cycle; primary infertility diagnosis; procedure type (i.e. whether a fresh or frozen embryo was transferred); number of mature oocytes retrieved; number of embryos transferred; cycle cancellation; pregnancy loss in the cycle; and insurance status. We were unable to determine exact reasons for treatment drop out (e.g. cessation of IVF treatment, transfer to another institution or spontaneous pregnancy). Our IPW model was moderately predictive based on the c-statistic from the calculation of the denominator of the weight; however, residual selection bias may remain due to the limited range of covariate data. IPW can be used in a variety of settings to address selection bias introduced by differential loss to follow up or treatment drop out. AMM was supported by National Institutes of Health (NIH) T32 HD052458-Boston University Reproductive, Perinatal and Pediatric Epidemiology Training Program. The authors report no conflicts of interest. N/A.

Abstract P200: Estimating the Receiver Operating Characteristics (ROC) Curve and its Associated AUC in Matched Case Control Studies

Background: The matched case-control study design is frequently used in cardiovascular disease research and can result in significant gains in efficiency; however, prediction in this setting is not straightforward. We propose an inverse-probability weighting approach to estimate the predictive ability associated with a set of covariates. Methods: We consider AUC estimation in 1:m matched case control studies and compare its bias with logistic regression in simulations. Let X denote the vector of exposures, Z the vector of matching factors and Y the binary outcome. We assume the population logistic model: logit [Pr(Y = 1

On weighting approaches for missing data

We review the class of inverse probability weighting (IPW) approaches for the analysis of missing data under various missing data patterns and mechanisms. The IPW methods rely on the intuitive idea of creating a pseudo-population of weighted copies of the complete cases to remove selection bias introduced by the missing data. However, different weighting approaches are required depending on the missing data pattern and mechanism. We begin with a uniform missing data pattern (i.e. a scalar missing indicator indicating whether or not the full data is observed) to motivate the approach. We then generalise to more complex settings. Our goal is to provide a conceptual overview of existing IPW approaches and illustrate the connections and differences among these approaches.

Observation plans in longitudinal studies with time-varying treatments

In longitudinal studies individuals are followed over time and their time-varying information on treatment, outcomes, and covariates is collected at certain intervals. Ideally, one would want these intervals as short as possible to consider data collection continuous for all practical purposes. Unfortunately, continuous recording, though sometimes approximately possible for certain outcomes (e.g., death status), is infeasible for most study variables. Rather, the time-varying information is typically collected only during the times when subjects are observed. These times of observation are usually pre-specified in randomized experiments and in certain observational studies in which the subjects' data are recorded at regular intervals. For example, participants in the Multicenter AIDS Cohort Study (MACS) are scheduled to show up at the research center every 6 months,1 and participants in the Nurses' Health Study are asked to return a self-reported questionnaire every two years.2 On the other hand, in many observational studies the observation regime or plan varies across individuals. For example, some observational studies are based on information that is collected for clinical purposes (e.g., databases from health care organizations) and thus the recording of information occurs at irregular and possibly subject-specific intervals. Because the intervals may depend of an individual's clinical history (treatment, symptoms, etc.), we refer to these as dynamic observation plans. Some studies combine the features of dynamic and pre-specified (static) observation plans. For example, the French Hospital Database on HIV (FHDH) collects information whenever an individual comes to an outpatient or inpatient hospital visit for clinical reasons (dynamic observation plan) but also attempts to collect information at least every six months (static observation plan) by reviewing medical records.3 Even studies with a static observation plan like the MACS may have a dynamic component if individuals who do not show up for a scheduled visit do so because of reasons related to their clinical history. In epidemiology, cohort studies with a predominantly static observation plan are often referred to as interval cohorts because they have regular intervals between visits, whereas cohort studies with a dynamic observation plan are often referred to as clinical cohorts because the interval between visits depends on the clinical evolution of the patient. In this paper we discuss the biases that may arise in interval cohorts with static observation plans, and in clinical cohorts when the dynamic observation plans are not explicitly incorporated in the analysis. We also discuss some analytic approaches that, under certain conditions, allow to adjust for the bias caused by dynamic observation plans. We highlight issues that are implicit in previous work by Robins and colleagues on marginal structural models4,5 and missing data.6 This paper is structured as follows. Section 2 describes inverse probability weighting (IPW) techniques to estimate treatment effects in longitudinal studies with a single static plan (e.g., an interval cohort study). To fix ideas, we summarize the methods previously used to estimate the effect of anti-retroviral treatment on mean CD4 cell count in the MACS/Women's Interagency HIV Study (WIHS) cohorts.7,8 We describe the key assumption of coincidence between observation times and times of potential treatment change. Section 3 extends the discussion to studies in which subjects follow one of several static regimes. Section 4 considers studies with dynamic observation plans (e.g., clinical cohorts), describes two potential biases that may arise in these studies, and proposes IPW-based approaches to adjust for them. Section 5 presents an application of IPW to estimate the effect of antiretroviral treatment on the evolution of the mean CD4 cell count in the FHDH, and Section 6 presents some brief conclusions. Throughout the paper we use causal diagrams to describe the causal structure and some of the key assumptions underlying the IPW approach.

Direct estimation of the area under the receiver operating characteristic curve in the presence of verification bias

The area under a receiver operating characteristic (ROC) curve (AUC) is a commonly used index for summarizing the ability of a continuous diagnostic test to discriminate between healthy and diseased subjects. If all subjects have their true disease status verified, one can directly estimate the AUC nonparametrically using the Wilcoxon statistic. In some studies, verification of the true disease status is performed only for a subset of subjects, possibly depending on the result of the diagnostic test and other characteristics of the subjects. Because estimators of the AUC based only on verified subjects are typically biased, it is common to estimate the AUC from a bias-corrected ROC curve. The variance of the estimator, however, does not have a closed-form expression and thus resampling techniques are used to obtain an estimate. In this paper, we develop a new method for directly estimating the AUC in the setting of verification bias based on U-statistics and inverse probability weighting (IPW). Closed-form expressions for the estimator and its variance are derived. We also show that the new estimator is equivalent to the empirical AUC derived from the bias-corrected ROC curve arising from the IPW approach.