Propensity Score Estimation Research Articles

A semiparametric multiply robust multiple imputation method for causal inference.

Evaluating the impact of non-randomized treatment on various health outcomes is difficult in observational studies because of the presence of covariates that may affect both the treatment or exposure received and the outcome of interest. In the present study, we develop a semiparametric multiply robust multiple imputation method for estimating average treatment effects in such studies. Our method combines information from multiple propensity score models and outcome regression models, and is multiply robust in that it produces consistent estimators for the average causal effects if at least one of the models is correctly specified. Our proposed estimators show promising performances even with incorrect models. Compared with existing fully parametric approaches, our proposed method is more robust against model misspecifications. Compared with fully non-parametric approaches, our proposed method does not have the problem of curse of dimensionality and achieves dimension reduction by combining information from multiple models. In addition, it is less sensitive to the extreme propensity score estimates compared with inverse propensity score weighted estimators and augmented estimators. The asymptotic properties of our method are developed and the simulation study shows the advantages of our proposed method compared with some existing methods in terms of balancing efficiency, bias, and coverage probability. Rubin's variance estimation formula can be used for estimating the variance of our proposed estimators. Finally, we apply our method to 2009-2010 National Health Nutrition and Examination Survey (NHANES) to examine the effect of exposure to perfluoroalkyl acids (PFAs) on kidney function.

SCMRSA: a New Approach for Identifying and Analyzing Anti-MRSA Peptides Using Estimated Propensity Scores of Dipeptides.

Staphylococcus aureus is deemed to be one of the major causes of hospital and community-acquired infections, especially in methicillin-resistant S. aureus (MRSA) strains. Because antimicrobial peptides have captured attention as novel drug candidates due to their rapid and broad-spectrum antimicrobial activity, anti-MRSA peptides have emerged as potential therapeutics for the treatment of bacterial infections. Although experimental approaches can precisely identify anti-MRSA peptides, they are usually cost-ineffective and labor-intensive. Therefore, computational approaches that are able to identify and characterize anti-MRSA peptides by using sequence information are highly desirable. In this study, we present the first computational approach (termed SCMRSA) for identifying and characterizing anti-MRSA peptides by using sequence information without the use of 3D structural information. In SCMRSA, we employed an interpretable scoring card method (SCM) coupled with the estimated propensity scores of 400 dipeptides. Comparative experiments indicated that SCMRSA was more effective and could outperform several machine learning-based classifiers with an accuracy of 0.960 and Matthews correlation coefficient of 0.848 on the independent test data set. In addition, we employed the SCMRSA-derived propensity scores to provide a more in-depth explanation regarding the functional mechanisms of anti-MRSA peptides. Finally, in order to serve community-wide use of the proposed SCMRSA, we established a user-friendly webserver which can be accessed online at http://pmlabstack.pythonanywhere.com/SCMRSA. SCMRSA is anticipated to be an open-source and useful tool for screening and identifying novel anti-MRSA peptides for follow-up experimental studies.

Impact of Soil and Water Conservation on Household Income in West Arsi Zone of Oromia

Land degradation due to soil erosion and nutrient depletion is one of the main problems constraining the development of the agricultural sector in Ethiopia. As part of intervention activities number of soil and water conservation (SWC) practices have been promoted to smallholder farmers living in highly degraded and drought prone areas of the country. This study was conducted to assess the impact of SWC intervention on the livelihood of smallholder farm households in terms household income and productivity. To meet this objective primary data was gathered from 150 sample responds (67 SWC program participants and 83 non-participants). Descriptive and inferential statistics and propensity score matching (PSM) models were used to address the stated objectives. Results of the descriptive statistics showed that before matching there was statistically significant difference between program participants and their counterfactual households in terms of sex, education and farm experience generally in favor of non-program participants whereas access to market information and amount of land allocated for production in favor of program participant in the zone. The analysis of mean difference in outcome variables before matching result indicated that the mean total crop yield for SWC practiced respondent households is higher (22.102quintal per hectare per household) than SWC non-practiced respondent households (17.26quintal per hectare per household) with mean total crop yield difference equivalent to 4.842 quintal per hectare per household in the zone. In the mean-time, even though, the results of the PSM model revealed that SWC intervention did not result in significant difference between program participant and non-participant households in terms of maize productivity and gross crop income total annual crop income of households who participated in SWC program was 10,838.13 birr whereas its 6075.48 birr for non-participant farmers accounting 4762.65birr difference suggesting that on average participant households earned 43.9 percent more crop value per hectare than the control group. Logistic regression model was employed to estimate propensity scores for matching SWC program households with their counterfactuals. Except sex of the household, Market information, amount of land cultivated, Education and farm experience influenced the probability of HH participation in SWC positively and significantly at 10% expect land cultivated at 5% level. The implication could be that farm household participation was more guided by demographic than economics factors (defined by farm size and herd size). Nominal results of analysis of treatment effects indicate that there was a sign of positive impact on both of the variables considered due to SWC program. However, the changes in crop productivity and gross household income could not be statistically justified as there was no statistically significant difference between the two groups in terms of these variables. The possible reason could be that SWC programs are not short-term nature and impacts are to be realized gradually with increased adoption and intensification of activities. Finally it was to be noted that there were positive trends which all together should guide SWC policy makers to identify important factors influencing the contribution of such a program and reconsider the design and implementation of the interventions. Keywords: Impacts, Income, participant, non- participant, household, Propensity Score Matching. DOI: 10.7176/JBAH/12-18-01 Publication date: September 30 th 2022

Nonparametric tests for treatment effect heterogeneity in observational studies

en

Blood culture positivity in patients with acute appendicitis: A propensity score-matched prospective cohort study.

The prevalence of bacteremia in acute appendicitis is unknown. We aimed to assess prevalence and predictive factors of bacteremia in adult patients with appendicitis. In this prospective propensity score-matched cohort study, patients were recruited as part of one single-center prospective observational study assessing appendicitis microbiology in concurrence with two randomized controlled trials on non-operative treatment of uncomplicated acute appendicitis. All patients evaluated for enrollment in these three trials between April 2017 and December 2018 with both a confirmed diagnosis of appendicitis and available blood culture on admission were included in this study. Potential predictive factors of bacteremia (age, sex, body mass index (BMI), body temperature, C-reactive protein (CRP), leukocyte count, comorbidities, symptom duration, and appendicitis severity) were assessed. Prevalence of bacteremia was determined by all available blood cultures followed by propensity score matching using sex, age, BMI, CRP, leukocyte count, and body temperature of the patients without available blood culture. Out of the 815 patients with appendicitis, 271 patients had available blood culture and the prevalence of bacteremia was 12% (n = 33). Based on propensity score estimation, the prevalence of bacteremia in the whole prospective appendicitis cohort was 11.1%. Bacteremia was significantly more frequent in complicated acute appendicitis (15%; 29/189) compared with uncomplicated acute appendicitis (5%; 4/82) (p = 0.015). Male sex (p = 0.024) and higher body temperature (p = 0.0044) were associated with bacteremia. Estimated prevalence of bacteremia in patients with acute appendicitis was 11.1%. Complicated appendicitis, male sex, and higher body temperature were associated with bacteremia in acute appendicitis.

Flexible propensity score estimation strategies for clustered data in observational studies.

Existing studies have suggested superior performance of nonparametric machine learning over logistic regression for propensity score estimation. However, it is unclear whether the advantages of nonparametric propensity score modeling are carried to settings where there is clustering of individuals, especially when there is unmeasured cluster-level confounding. In this work we examined the performance of logistic regression (all main effects), Bayesian additive regression trees and generalized boosted modeling for propensity score weighting in clustered settings, with the clustering being accounted for by including either cluster indicators or random intercepts. We simulated data for three hypothetical observational studies of varying sample and cluster sizes. Confounders were generated at both levels, including a cluster-level confounder that is unobserved in the analyses. A binary treatment and a continuous outcome were generated based on seven scenarios with varying relationships between the treatment and confounders (linear and additive, nonlinear/nonadditive, nonadditive with the unobserved cluster-level confounder). Results suggest that when the sample and cluster sizes are large, nonparametric propensity score estimation may provide better covariate balance, bias reduction, and 95% confidence interval coverage, regardless of the degree of nonlinearity or nonadditivity in the true propensity score model. When the sample or cluster sizes are small, however, nonparametric approaches may become more vulnerable to unmeasured cluster-level confounding and thus may not be a better alternative to multilevel logistic regression. We applied the methods to the National Longitudinal Study of Adolescent to Adult Health data, estimating the effect of team sports participation during adolescence on adulthood depressive symptoms.

Testing for the presence of treatment effect under selection on observables

The evaluation of the possible effects of a treatment on an outcome plays a central role in both theoretical and applied statistical and econometrical literature. This paper focuses on nonparametric tests for possible difference in the distribution of potential outcomes due to receiving or not receiving a treatment. The approach is based on weighting observed data on the basis on the estimated propensity score. Kolmogorov–Smirnov type and Wilcoxon–Mann–Whitney type tests are constructed, and their limiting distributions are studied. Rejection regions are obtained by inverting confidence intervals. This involves the study of appropriate estimators of the limiting variance of test statistics. Approximations of quantiles via subsampling are also considered. The merits of the different tests are studied by Monte Carlo simulation. An application to the construction of tests for stochastic dominance is provided.

Estimation for volunteer web survey samples using a model-averaging approach

Propensity score approach is a popular technique for estimating the population based on volunteer web survey samples. Various models have been used to estimate propensity scores and produce different population estimates. To obtain more accurate population estimators, we propose a model-averaging estimation approach based on propensity score estimates from a parametric logistic regression model and a nonparametric generalized boosted model. Consistency and asymptotic normality of the proposed estimators are established. A computation algorithm is also developed to implement the proposed method. Simulation studies are conducted to compare the performance of the proposed method with the other methods. A survey data from the Netizen Social Awareness Survey (NSAS) is used to illustrate the proposed methodology.

Propensity score methodology for confounding control in health care utilization databases

Propensity score (PS) methodology is a common approach to control for confounding in nonexperimental studies of treatment effects using health care utilization databases. This methodology offers researchers many advantages compared with conventional multivariate models: it directly focuses on the determinants of treatment choice, facilitating the understanding of the clinical decision-making process by the researcher; it allows for graphical comparisons of the distribution of propensity scores and truncation of subjects without overlapping PS indicating a lack of equipoise; it allows transparent assessment of the confounder balance achieved by the PS at baseline; and it offers a straightforward approach to reduce the dimensionality of sometimes large arrays of potential confounders in utilization databases, directly addressing the “curse of dimensionality” in the context of rare events. This article provides an overview of the use of propensity score methodology for pharmacoepidemiologic research with large health care utilization databases, covering recent discussions on covariate selection, the role of automated techniques for addressing unmeasurable confounding via proxies, strategies to maximize clinical equipoise at baseline, and the potential of machine-learning algorithms for optimized propensity score estimation. The appendix discusses the available software packages for PS methodology. Propensity scores are a frequently used and versatile tool for transparent and comprehensive adjustment of confounding in pharmacoepidemiology with large health care databases.

Covariate distribution balance via propensity scores

SummaryThis paper proposes new estimators for the propensity score that aim to maximize the covariate distribution balance among different treatment groups. Heuristically, our proposed procedure attempts to estimate a propensity score model by making the underlying covariate distribution of different treatment groups as close to each other as possible. Our estimators are data‐driven and can be used to estimate different treatment effect parameters under different identifying assumptions, including unconfoundedness and local treatment effects. We derive the asymptotic properties of inverse probability weighted estimators for the average, distributional, and quantile treatment effects based on the proposed propensity score estimator and illustrate their finite sample performance via Monte Carlo simulations and an empirical application.

BART 기반 기계학습 인과효과 추정 사례 연구

The study aims to introduce a causal inference method using machine learning to general education researchers, and in particular, focus on the theory and practice of Bayesian Additive Regression Trees algorithm. To analyze the empirical data, public data from the Korean Children and Youth Panel Survey 2018 were used. For an illustrative purpose, this study estimated the causal effect of participation in activities related to self (personality) development on students’ life satisfaction and self-esteem and discussed the feasibility of the BART method in educational impact studies. The applicability of the BART-based machine learning causal inference technique in the field of education was discussed in comparison with model-based propensity score and causal effect estimation. Finally future research topics and limitations of the study were addressed.

MSR65 Comparison of Machine Learning Approach with Traditional Method to Predict Risk of Serious Events Associated with Individual Cholinesterase Inhibitors Use in Older Adults with Dementia

Machine learning techniques have been recommended as promising alternatives to logistic regression for the estimation of propensity scores. This study compared the performance of the machine learning approach Inverse Probability Weighting- Gradient Boosted Modeling (IPTW-GBM) with traditional Propensity Scores (PS) method in predicting risk of serious events across individual Cholinesterase inhibitors (ChEIs) use in older adults with dementia.

Community-Based Advanced Case Management for Patients with Complex Multimorbidity and High Medical Dependence: A Longitudinal Study.

This longitudinal study aimed to evaluate a community-based and nurse-led advanced case management model centered on disease management. Participants were chronically ill patients aged 20 years and older who were highly dependent on medical care. The case management group (CMG) received nurse-led advanced case management, and the comparison group (CG) was selected by matching estimated propensity scores with the CMG. We compared the changes in medico-economic indicators between the two groups and analyzed the physical and psychological indicators of the CMG over time. The CMG comprised 51 participants, of which eight dropped out by 12 months after registration. After 1:1 propensity score matching, there were 40 participants in the CMG and CG, respectively. At 12 months after the registration, there was no significant difference between the two groups and no change in the CMG. At 24 months after the registration, the CMG’s medical and long-term care costs decreased significantly, while the CG’s costs increased. Moreover, there was a significant reduction in the number of hospital days and hospital admissions in the CMG. Our findings revealed that nurse-led advanced case management could be useful for patients with complex needs to avoid hospitalization due to exacerbations.

Influence of contrast enhancement at the contrast injection location for the arm or leg in neonatal and infant patients during cardiac computed tomography

Introduction and ObjectivesObtaining CCTA images with optimal injection location such as the arm or leg is important to avoid the artifacts caused by the CM. This study compares the computed tomography (CT) numbers and visualization scores of the three-dimensional (3D) images of the lumens of the blood vessels in the arm or leg during cardiac computed tomography angiography (CCTA) in neonatal and infant patients. Patients or Materials and MethodsBetween January 2017 and January 2020, 253 consecutive patients were considered for inclusion. We used the estimated propensity scores as a function of the demographic data, including age, body weight, and injection location (right or left side) in the arm (n = 58) and leg (n = 58) of neonatal and infant patients. We compared the mean CT numbers of the pulmonary artery, ascending aorta, and left superior vena cava; contrast–noise ratios (CNR); and visualization scores between the arm and leg as the injection locations. ResultsThe mean CT numbers during CCTA for the arm and leg were 479.4 and 461.3 HU in the ascending aorta, 464.2 and 448.1 HU in the pulmonary artery, and 232.8 and 220.1 HU in the left superior vena cava, respectively. The mean image noise (SD) and CNR values, respectively, were 38.9 HU and 12.1 for the arm as the injection location and 39.1 HU and 12.3 for the leg as the injection location. The median visualization scores of volume rendering of the 3D images were 3.0 and 3.0 for the arm and leg injection sites, respectively. There were no significant differences in the mean CT numbers of the ascending aorta, pulmonary artery, and left superior vena cava; SD value; CNR; and visualization scores between the arm and leg injection locations. ConclusionsThe CT numbers of the lumen of the blood vessel and visualization scores of the 3D images of the arm and leg injection locations are equal during CCTA in neonatal and infant patients with congenital heart disease.

Single-staged laparotomy versus multiple-staged laparotomy for traumatic massive hemoperitoneum with hemodynamic instability: a single-center, propensity score-matched analysis

BackgroundCurrently, damage control surgery (DCS) employing multiple-staged laparotomy (MSL) is a standard hemostatic approach for treating trauma patients with unstable hemodynamics attributable to massive hemoperitoneum. Based on these findings, we had frequently employed MSL as a part of our hemostatic strategy for the patients, but with unsatisfactory outcomes. On the other hand, with the establishment of damage control resuscitation (DCR), it has become possible to avoid trauma-induced coagulopathy and to achieve adequate hemostasis with single-staged laparotomy (SSL). Consequently, our institutional strategy for surgical hemostasis of the patients has gradually shifted from MSL to SSL with implementation of DCR. The purpose of the study is to evaluate the impact of this shift in the strategy by comparing outcomes of the patients between those underwent MSL and those underwent SSL employing propensity score matching.MethodsThis retrospective, single-center, observational study evaluated outcomes of hemodynamically unstable patients with traumatic massive hemoperitoneum requiring surgical intervention between 2005 and 2020. The patient population was divided into two groups: a SSL group and a MSL group. Propensity score matching was used to adjust for differences in baseline characteristics in the two groups, a one-to-one matched analysis using nearest-neighbor matching was performed based on the estimated propensity score of each group. The primary outcome was in-hospital mortality, and secondary outcomes were 48-h mortality and 28-day mortality.ResultsA total of 170 patients met the inclusion criteria; 141 patients underwent SSL, and 29 underwent MSL. In the propensity-matched analysis with 27 pairs, the SSL group had significantly lower in-hospital mortality (odds ratio [OR] 0.154; 95% confidence interval (CI) 0.035 to 0.682) and 28-day mortality (OR 0.200; 95% CI 0.044 to 0.913) than the MSL group, but the 48-h mortality did not differ significantly between the two groups (25.9% vs. 44.4%; OR 0.375; 95% CI 0.099–1.414).ConclusionsSingle-staged laparotomy may be an effective surgical treatment for the traumatic massive hemoperitoneum cases with hemodynamic instability, if conducted following sufficient damage control resuscitation and performed by an experienced surgeon.

Reader reaction to "Outcome-adaptive lasso: Variable selection for causal inference" by Shortreed and Ertefaie (2017).

Shortreed and Ertefaie introduced a clever propensity score variable selection approach for estimating average causal effects, namely, the outcome adaptive lasso (OAL). OAL aims to select desirable covariates, confounders, and predictors of outcome, to build an unbiased and statistically efficient propensity score estimator. Due to its design, a potential limitation of OAL is how it handles the collinearity problem, which is often encountered in high-dimensional data. As seen in Shortreed and Ertefaie, OAL's performance degraded with increased correlation between covariates. In this note, we propose the generalized OAL (GOAL) that combines the strengths of the adaptively weighted L1 penalty and the elastic net to better handle the selection of correlated covariates. Two different versions of GOAL, which differ in their procedure (algorithm), are proposed. We compared OAL and GOAL in simulation scenarios that mimic those examined by Shortreed and Ertefaie. Although all approaches performed equivalently with independent covariates, we found that both GOAL versions were more performant than OAL in low and high dimensions with correlatedcovariates.

Long-term effectiveness of a disease management program to prevent diabetic nephropathy: a propensity score matching analysis using administrative data in Japan

BackgroundExisting reviews indicated that disease management for patients with diabetes may be effective in achieving better health outcomes with less resource utilization in the short term. However, the long-term results were inconsistent because of the heterogeneous nature of the study designs. In the present study, we evaluated the 5-year follow-up results of a local disease management program focused on diabetic nephropathy prevention under the universal public health insurance scheme in Japan.MethodsPatients diagnosed with type 2 diabetes who had stage 3 or 4 diabetic kidney disease and were aged between 20 and 75 years were invited to join a disease management program to support self-management and receive a recommended treatment protocol between 2011 and 2013. Follow-up data were collected from an electronic claims database for the public insurance scheme. Considering the non-random selection process, we prepared two control groups matched by estimated propensity scores to compare the incidence of diabetes-related complications, death, and resource utilization.ResultsThe treatment group was more likely to receive clinical management in accordance with the guideline-recommended medication. After propensity score matching, the treatment group had lower incidence of diabetic nephropathy and emergency care use than the control group selected from a beneficiary pool mainly under primary care. Comparisons between the treatment group and the control group with more selected clinical conditions did not show differences in the incidence rate and resource utilization.ConclusionsThe present results demonstrated limited effectiveness of the program for reducing complication incidence and resource utilization during the 5-year follow-up. Further research on the long-term effectiveness of co-management by primary care physicians, subspecialists in endocrinology and nephrology, and nurse educators is required for effective management of diabetes-related nephropathy.

Angiotensin-converting enzyme inhibitors and risk of age-related macular degeneration in individuals with hypertension.

AimsSeveral observational studies have examined the potential protective effect of angiotensin‐converting enzyme inhibitor (ACE‐I) use on the risk of age‐related macular degeneration (AMD) and have reported contradictory results owing to confounding and time‐related biases. We aimed to assess the risk of AMD in a base cohort of patients aged 40 years and above with hypertension among new users of ACE‐I compared to an active comparator cohort of new users of calcium channel blockers (CCB) using data obtained from IQVIA Medical Research Data, a primary care database in the UK.MethodsIn this study, 53 832 and 43 106 new users of ACE‐I and CCB were included between 1995 and 2019, respectively. In an on‐treatment analysis, patients were followed up from the time of index drug initiation to the date of AMD diagnosis, loss to follow‐up, discontinuation or switch to the comparator drug. A comprehensive range of covariates were used to estimate propensity scores to weight and match new users of ACE‐I and CCB. Standardized mortality ratio weighted Cox proportional hazards model was used to estimate hazard ratios of developing AMD.ResultsDuring a median follow‐up of 2 years (interquartile range 1–5 years), the incidence rate of AMD was 2.4 (95% confidence interval 2.2–2.6) and 2.2 (2.0–2.4) per 1000 person‐years among the weighted new users of ACE‐I and CCB, respectively. There was no association of ACE‐I use on the risk of AMD compared to CCB use in either the propensity score weighted or matched, on‐treatment analysis (adjusted hazard ratio: 1.07 [95% confidence interval 0.90–1.27] and 0.87 [0.71–1.07], respectively).ConclusionWe found no evidence that the use of ACE‐I is associated with risk of AMD in patients with hypertension.

Utilizing stratified generalized propensity score matching to approximate blocked randomized designs with multiple treatment levels

ABSTRACT Conducting causal inference in settings with more than one treatment level can be challenging. Classical methods, such as propensity score matching (PSM), are restricted to only a binary treatment. To extend propensity score methods beyond a binary treatment, generalized propensity score methods have been proposed, with generalized propensity score matching (GPSM) standing as the multi-level treatment analog to PSM. One drawback of GPSM is it is only capable of emulating a completely randomized trial (CRT) design and not the more efficient blocked randomized trial design. Motivated by the desire to emulate the more efficient design, we expand on GPSM estimating literature and develop a new estimator incorporating relevant stratifying variables into the GPSM framework. We examine the variance estimation methods available for GPSM and demonstrate how to extend the estimator to one where stratifying variables are included. While it would be straightforward to include relevant stratifying variables as covariates in the propensity score estimation, our method provides for researchers to conduct retrospective analyses more consistently with the prospective experiment they would have designed if permitted. Namely, our method permits researchers to approximate a stratified randomized trial as opposed to the CRT otherwise obtainable by GPSM. We apply our proposed method to an analysis of how the number of children in a household affects systolic blood pressure in adults. We conduct a simulation study assessing how the relationship between response, treatment, and strata affect the performance of our method and compare the results to non-stratified GPSM.

Handling high-dimensional data with missing values by modern machine learning techniques

High-dimensional data have been regarded as one of the most important types of big data in practice. It happens frequently in practice including genetic study, financial study, and geographical study. Missing data in high dimensional data analysis should be handled properly to reduce nonresponse bias. We discuss some modern machine learning techniques including penalized regression approaches, tree-based approaches, and deep learning (DL) for handling missing data with high dimensionality. Specifically, our proposed methods can be used for estimating general parameters of interest including population means and percentiles with imputation-based estimators, propensity score estimators, and doubly robust estimators. We compare those methods through some limited simulation studies and a real application. Both simulation studies and real application show the benefits of DL and XGboost approaches compared with other methods in terms of balancing bias and variance.