Multiple imputation for propensity score analysis with covariates missing at random: some clarity on within and across methods.

In epidemiology and social sciences, propensity score methods are popular for estimating treatment effects using observational data, and multiple imputation is popular for handling covariate missingness. However, how to appropriately use multiple imputation for propensity score analysis is not completely clear. This paper aims to bring clarity on the consistency (or lack thereof) of methods that have been proposed, focusing on the within approach (where the effect is estimated separately in each imputed dataset and then the multiple estimates are combined) and the across approach (where typically propensity scores are averaged across imputed datasets before being used for effect estimation). We show that the within method is valid and can be used with any causal effect estimator that is consistent in the full-data setting. Existing across methods are inconsistent, but a different across method that averages the inverse probability weights across imputed datasets is consistent for propensity score weighting. We also comment on methods that rely on imputing a function of the missing covariate rather than the covariate itself, including imputation of the propensity score and of the probability weight. Based on consistency results and practical flexibility, we recommend generally using the standard within method. Throughout, we provide intuition to make the results meaningful to the broad audience of applied researchers.

Survival disparities in non-Hispanic Black and White cervical cancer patients vary by histology and are largely explained by modifiable factors.

PURPOSE: We investigated racial disparities in survival by histology in cervical cancer and examined the factors contributing to these disparities. METHODS: Non-Hispanic Black and non-Hispanic White (hereafter known as Black and White) patients with stage I-IV cervical carcinoma diagnosed between 2004 and 2017 in the National Cancer Database were studied. Survival differences were compared using Cox modeling to estimate hazard ratio (HR) or adjusted HR (AHR) and 95% confidence interval (CI). The contribution of demographic, socioeconomic and clinical factors to the Black vs White differences in survival was estimated after applying propensity score weighting in patients with squamous cell carcinoma (SCC) or adenocarcinoma (AC). RESULTS: This study included 10,111 Black and 43,252 White patients with cervical cancer. Black patients had worse survival than White cervical cancer patients (HR = 1.40, 95% CI = 1.35-1.45). Survival disparities between Black and White patients varied significantly by histology (HR = 1.20, 95% CI = 1.15-1.24 for SCC; HR = 2.32, 95% CI = 2.12-2.54 for AC, interaction p < 0.0001). After balancing the selected demographic, socioeconomic and clinical factors, survival in Black vs. White patients was no longer different in those with SCC (AHR = 1.01, 95% CI 0.97-1.06) or AC (AHR = 1.09, 95% CI = 0.96-1.24). In SCC, the largest contributors to survival disparities were neighborhood income and insurance. In AC, age was the most significant contributor followed by neighborhood income, insurance, and stage. Diagnosis of AC (but not SCC) at ≥65 years old was more common in Black vs. White patients (26% vs. 13%, respectively). CONCLUSIONS: Histology matters in survival disparities and diagnosis at ≥65 years old between Black and White cervical cancer patients. These disparities were largely explained by modifiable factors.

Benefits of Living Over Deceased Donor Kidney Transplantation in Elderly Recipients. A Propensity Score Matched Analysis of a Large European Registry Cohort.

Although kidney transplantation from living donors (LD) offers better long-term results than from deceased donors (DD), elderly recipients are less likely to receive LD transplants than younger ones. We analyzed renal transplant outcomes from LD versus DD in elderly recipients with a propensity-matched score. This retrospective, observational study included the first single kidney transplants in recipients aged ≥65 years from two European registry cohorts (2013-2020, n = 4,257). Recipients of LD (n = 408), brain death donors (BDD, n = 3,072), and controlled cardiocirculatory death donors (cDCD, n = 777) were matched for donor and recipient age, sex, dialysis time and recipient diabetes. Major graft and patient outcomes were investigated. Unmatched analyses showed that LD recipients were more likely to be transplanted preemptively and had shorter dialysis times than any DD type. The propensity score matched Cox's regression analysis between LD and BDD (387-pairs) and LD and cDCD (259-pairs) revealing a higher hazard ratio for graft failure with BDD (2.19 [95% CI: 1.16-4.15], p = 0.016) and cDCD (3.38 [95% CI: 1.79-6.39], p < 0.001). One-year eGFR was higher in LD transplants than in BDD and cDCD recipients. In elderly recipients, LD transplantation offers superior graft survival and renal function compared to BDD or cDCD. This strategy should be further promoted to improve transplant outcomes.

Propensity score analysis of low-dose aspirin and bleeding complications in pregnancy.

OBJECTIVE: Low-dose aspirin (LDA) has been shown to reduce the risk of preterm pre-eclampsia and it has been suggested that it should be recommended for all pregnancies. However, some studies have reported an association between LDA and an increased risk of bleeding complications in pregnancy. Our aim was to evaluate the risk of placental abruption and postpartum hemorrhage (PPH) in patients for whom their healthcare provider had recommended prophylactic aspirin. METHODS: This multicenter cohort study included 72 598 singleton births at 19 hospitals in the USA, between January 2019 and December 2021. Pregnancies complicated by placenta previa/accreta, birth occurring at less than 24 weeks' gestation, multiple pregnancy or those with data missing for aspirin recommendation were excluded. Propensity scores were calculated using 20 features spanning sociodemographic factors, medical history, year and hospital providing care. The association between LDA recommendation and placental abruption or PPH was estimated by inverse-probability treatment weighting using the propensity scores. RESULTS: We included 71 627 pregnancies in the final analysis. Aspirin was recommended to 6677 (9.3%) and was more likely to be recommended for pregnant individuals who were 35 years or older (P < 0.001), had a body mass index of 30 kg/m2 or higher (P < 0.001), had prepregnancy hypertension (P < 0.001) and who had a Cesarean delivery (P < 0.001). Overall, 1.7% of the study cohort (1205 pregnancies) developed preterm pre-eclampsia: 1.3% in the no-aspirin and 5.8% in the aspirin group. After inverse-probability weighting with propensity scores, aspirin was associated with increased risk of placental abruption (adjusted odds ratio (aOR), 1.44 (95% CI, 1.04-2.00)) and PPH (aOR, 1.21 (95% CI, 1.05-1.39)). The aOR translated to a number needed to harm with LDA of 79 (95% CI, 43-330) for PPH and 287 (95% CI, 127-3151) for placental abruption. CONCLUSIONS: LDA recommendation in pregnancy was associated with increased risk for placental abruption and for PPH. Our results support the need for more research into aspirin use and bleeding complications in pregnancy before recommending it beyond the highest-risk pregnancies. © 2023 The Authors. Ultrasound in Obstetrics & Gynecology published by John Wiley & Sons Ltd on behalf of International Society of Ultrasound in Obstetrics and Gynecology.

Low-Volume Acute Normovolemic Hemodilution Does Not Reduce Allogeneic Red Blood Cell Transfusion in Cardiac Surgery in the Modern Era of Patient Blood Management: A Propensity Score-Matched Cohort Study.

OBJECTIVES: Patients undergoing cardiac surgery often require blood transfusions, which are associated with increased morbidity and mortality. Patient blood management (PBM) strategies, including acute normovolemic hemodilution (ANH), have been implemented to minimize allogeneic transfusion requirements. Older studies suggested that ANH is associated with reduced transfusions; however, its effectiveness in the modern era of PBM remains unclear. DESIGN: This was a retrospective cohort study. SETTING: The study was held at a single university hospital. PARTICIPANTS: 542 patients who underwent elective cardiac surgery with cardiopulmonary bypass (CPB) using low-priming-volume circuits between January 2017 and March 2022. INTERVENTIONS: Patients who received ANH were matched with those who did not receive ANH, using propensity scores. MEASUREMENTS AND MAIN RESULTS: The primary outcome was the proportion of patients who received perioperative red blood cell (RBC) transfusion. Of the 542 eligible patients, 49 ANH cases were propensity-score matched to 97 controls. The median ANH volume was 450 mL (IQR, 400-800 mL). There was no significant difference in perioperative RBC transfusion rates between the 2 groups (24.5% in the ANH group vs 30.9% in the control group, p = 0.42). The odds ratio for perioperative RBC transfusion in the ANH group versus the control group was 0.72 (95% CI, 0.32-1.55, p = 0.42). CONCLUSIONS: Low-volume ANH was not associated with a significant reduction in perioperative allogeneic RBC transfusion during cardiac surgery with CPB using low-priming-volume circuits. The benefits of low-volume ANH in reducing the requirement for RBC transfusion in the modern era of PBM may be smaller than reported previously.

Real-world efficacy of adjuvant therapy for totally resected stage I lung adenocarcinoma patients with pathological high-risk factors: propensity score analysis.

BACKGROUND: We investigated the real-world efficacy of adjuvant therapy for stage I lung adenocarcinoma patients with pathological high-risk factors. METHODS: Study participants were enrolled from November 1, 2016 and December 31, 2020. Clinical bias was balanced by propensity score matching. Disease-free survival (DFS) outcomes were compared by Kaplan-Meier analysis. The Cox proportional hazards regression was used to identify survival-associated factors. p ≤ 0.05 was the threshold for statistical significance. RESULTS: A total of 454 patients, among whom 134 (29.5%) underwent adjuvant therapy, were enrolled in this study. One hundred and eighteen of the patients who underwent adjuvant therapy were well matched with non-treatment patients. Prognostic outcomes of the treatment group were significantly better than those of the non-treatment group, as revealed by Kaplan-Meier analysis after PSM. Differences in prevention of recurrence or metastasis between the targeted therapy and chemotherapy groups were insignificant. Adjuvant therapy was found to be positive prognostic factors, tumor size and solid growth patterns were negative. CONCLUSIONS: Adjuvant therapy significantly improved the DFS for stage I lung adenocarcinoma patients with high-risk factors. Larger prospective clinical trials should be performed to verify our findings.

The relationship between serum uric acid and gastrointestinal bleeding in peritoneal dialysis patients: a propensity score analysis.

PURPOSE: Gastrointestinal bleeding is an important gastrointestinal complication among peritoneal dialysis patients and correlated with a higher risk of mortality. Increased uric acid levels are a significant complication for peritoneal dialysis patients and have been associated with an increased risk of hemorrhagic stroke. The objective of the present study was to investigate the relationship between serum uric acid levels and gastrointestinal bleeding in peritoneal dialysis patients. METHODS: A total of 2498 peritoneal dialysis patients were recruited. Based on the optimal uric acid cutoff value, two groups of patients were divided. We constructed a propensity-score-matched population of 1762 patients by matching sex, age, and body mass index. Survival outcomes between the two groups were compared using adjusted Kaplan-Meier curves. We constructed the restricted cubic splines regression to assess the correlation between levels of uric acid and gastrointestinal bleeding. A multivariate Cox proportional hazards regression was performed to test whether higher levels of uric acid are an independent risk factor for gastrointestinal bleeding. We performed a forest plot to show interaction effects in different subgroups. RESULTS: According to restricted cubic splines regression, uric acid levels were positively correlated with the risk of gastrointestinal bleeding events. After adjusted different confounding factors, patients with high levels of uric acid were prone to experience gastrointestinal bleeding (HR 1.868, 95%CI 1.001-3.486). In subgroups, the interaction between higher levels of uric acid and utilizing proton pump inhibitors was significant (P for interaction = 0.034). Further research found that taking proton pump inhibitors could decrease the risk of gastrointestinal bleeding in peritoneal dialysis patients accompanied high levels of uric acid. CONCLUSION: The baseline high levels of uric acid are an independent risk factor for gastrointestinal bleeding in patients undergoing peritoneal dialysis.

Adjuvant chemotherapy for stage II and III gastric adenocarcinoma: A retrospective analysis with long-term follow-up.

BACKGROUND: Whether adjuvant chemotherapy should be different for patients with stage II and III gastric cancer is unknown. METHODS: We retrospectively analyzed the effects of adjuvant chemotherapy on the outcomes of 140 and 256 patients with stage II and III gastric cancer, respectively, between January 2008 and December 2018. Chemotherapies were stratified as fluoropyrimidine plus platinum versus fluoropyrimidine alone, tegafur/gimeracil/octeracil (S-1)-containing versus non-S-1-containing regimens, and S-1 plus cisplatin versus S-1 alone. RESULTS: The median age of patients was 67.0 (range 24.6-98.8) years. With a median follow-up of 105 months, recurrence occurred in 32 (22.9%) and 130 (50.8%) patients with stage II and III disease, respectively. Adjuvant chemotherapy was administered as fluoropyrimidine monotherapy to 68 (48.6%) and 73 (28.5%) patients, fluoropyrimidine plus platinum to 9 (6.4%) and 104 (40.6%) patients, and none to 63 (45.0%) and 79 (30.9%) patients with stage II and III gastric cancer, respectively. Doublet chemotherapy was associated with longer disease-free survival (DFS) (26.5 vs. 15.2 months, P = 0.001) and overall survival (OS) (41.2 vs. 22.0 months, P < 0.001) than fluoropyrimidine monotherapy for stage IIIB-IIIC disease. Furthermore, S-1-containing regimens prolonged DFS (57.4 vs. 21.9 months, P = 0.044) and OS (81.4 vs. 28.6 months, P = 0.023) compared with non-S-1-containing chemotherapy in stage III disease. CONCLUSION: Although fluoropyrimidine monotherapy is feasible for stage II-IIIA disease, doublet chemotherapy is significantly associated with longer survival than monotherapy for stage IIIB-IIIC disease. S-1-containing regimens might lead to longer survival than non-S-1-containing chemotherapy in stage III gastric cancer.

A Generalized Bootstrap Procedure of the Standard Error and Confidence Interval Estimation for Inverse Probability of Treatment Weighting.

The inverse probability of treatment weighting (IPTW) approach is commonly used in propensity score analysis to infer causal effects in regression models. Due to oversized IPTW weights and errors associated with propensity score estimation, the IPTW approach can underestimate the standard error of causal effect. To remediate this, bootstrap standard errors have been recommended to replace the IPTW standard error, but the ordinary bootstrap (OB) procedure might still result in underestimation of the standard error because of its inefficient resampling scheme and untreated oversized weights. In this paper, we develop a generalized bootstrap (GB) procedure for estimating the standard error and confidence intervals of the IPTW approach. Compared with the OB procedure and other three procedures in comparison, the GB procedure has the highest precision and yields conservative standard error estimates. As a result, the GB procedure produces short confidence intervals with highest coverage rates. We demonstrate the effectiveness of the GB procedure via two simulation studies and a dataset from the National Educational Longitudinal Study-1988 (NELS-88).

Unreliable Continuous Treatment Indicators in Propensity Score Analysis.

Propensity score analyses (PSA) of continuous treatments often operationalize the treatment as a multi-indicator composite, and its composite reliability is unreported. Latent variables or factor scores accounting for this unreliability are seldom used as alternatives to composites. This study examines the effects of the unreliability of indicators of a latent treatment in PSA using the generalized propensity score (GPS). A Monte Carlo simulation study was conducted varying composite reliability, continuous treatment representation, variability of factor loadings, sample size, and number of treatment indicators to assess whether Average Treatment Effect (ATE) estimates differed in their relative bias, Root Mean Squared Error, and coverage rates. Results indicate that low composite reliability leads to underestimation of the ATE of latent continuous treatments, while the number of treatment indicators and variability of factor loadings show little effect on ATE estimates, after controlling for overall composite reliability. The results also show that, in correctly specified GPS models, the effects of low composite reliability can be somewhat ameliorated by using factor scores that were estimated including covariates. An illustrative example is provided using survey data to estimate the effect of teacher adoption of a workbook related to a virtual learning environment in the classroom.