Dataset Information

Evaluating large-scale propensity score performance through real-world and synthetic data experiments.

ABSTRACT: Background:Propensity score adjustment is a popular approach for confounding control in observational studies. Reliable frameworks are needed to determine relative propensity score performance in large-scale studies, and to establish optimal propensity score model selection methods. Methods:We detail a propensity score evaluation framework that includes synthetic and real-world data experiments. Our synthetic experimental design extends the 'plasmode' framework and simulates survival data under known effect sizes, and our real-world experiments use a set of negative control outcomes with presumed null effect sizes. In reproductions of two published cohort studies, we compare two propensity score estimation methods that contrast in their model selection approach: L1-regularized regression that conducts a penalized likelihood regression, and the 'high-dimensional propensity score' (hdPS) that employs a univariate covariate screen. We evaluate methods on a range of outcome-dependent and outcome-independent metrics. Results:L1-regularization propensity score methods achieve superior model fit, covariate balance and negative control bias reduction compared with the hdPS. Simulation results are mixed and fluctuate with simulation parameters, revealing a limitation of simulation under the proportional hazards framework. Including regularization with the hdPS reduces commonly reported non-convergence issues but has little effect on propensity score performance. Conclusions:L1-regularization incorporates all covariates simultaneously into the propensity score model and offers propensity score performance superior to the hdPS marginal screen.

SUBMITTER: Tian Y

PROVIDER: S-EPMC6280944 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:BackgroundCommunity-acquired pneumonia (CAP) is a leading cause of morbidity and mortality worldwide despite correct antibiotic use. Corticosteroids have long been evaluated as a treatment option, but heterogeneous effects on survival have precluded their widespread implementation. We aimed to evaluate whether corticosteroids might improve clinical outcomes in patients with severe CAP and high inflammatory responses.Study design and methodsWe analyzed two prospective observational cohorts of patients with CAP in Barcelona and Rome who were admitted to intensive care with a high inflammatory response. Propensity score (PS) matching was used to obtain balance among the baseline variables in both groups, and we excluded patients with viral pneumonia or who received hydrocortisone.ResultsOf the 610 patients admitted with severe CAP, 198 (32%) received corticosteroids and 387 had major criteria for severe CAP. All patients had a baseline serum C-reactive protein above 15 mg/dL. Patients who received corticosteroids were more commonly male, had more comorbidities (e.g., cancer or chronic obstructive pulmonary disease), and presented with significantly higher sequential organ failure assessment scores. Eighty-nine patients met major severity criteria (invasive mechanical ventilation and/or septic shock) and were matched per group. Twenty-eight-day mortality was lower among patients receiving corticosteroids (16 patients, 18%) than among those not receiving them (28 patients, 31%; p = 0.037). After PS matching, corticosteroid therapy reduced the 28-day mortality risk in patients who met major severity criteria (hazard ratio (HR) 0.53, 95% confidence interval (CI) 0.29-0.98) (p = 0.043). In patients who did not meet major severity criteria, no benefits were observed with corticosteroid use (HR 0.88 (95%CI 0.32-2.36).ConclusionsCorticosteroid treatment may be of benefit for patients with CAP who have septic shock and/or a high inflammatory response and requirement for invasive mechanical ventilation. Corticosteroids appear to have no impact on mortality when these features are not present.

Project description:Real-world data are needed to establish SARS-CoV-2 rapid antigen testing (RAT) as an effective and reliable approach for SARS-CoV-2 screening. This study included 1,952,931 individuals who provided upper respiratory specimens during SARS-CoV-2 screening at CityMD urgent care locations in the New York metropolitan area from October 2020 to March 2021. Positive and negative results, as determined by the BD Veritor™ System for Rapid Detection of SARS-CoV-2 antigen (Veritor), were obtained for all individuals, with reflex reverse transcriptase-polymerase chain reaction (RT-PCR) testing performed on a case-by-case basis, per standard of care. Using verification bias adjustment, two alternative model assumptions were utilized for RAT results with missing reflex RT-PCR results. The worst antigen diagnostic performance estimates asserted that missing RT-PCR results would show a distribution similar to those RT-PCR results actually obtained, based on symptom category. The best antigen diagnostic performance estimates asserted that individuals without RT-PCR results had a clinical presentation consistent with RAT results, and, therefore, missing RT-PCR results would agree with RAT results. For patients with symptoms or high-risk exposure, 25.3% (n = 86,811/343,253) of RAT results were positive; vs. 3.4% (n = 53,046/1,559,733) positive for asymptomatic individuals without high-risk exposure. Reflex RT-PCR results were obtained from 46.3% (n = 158,836/343,253) and 13.8% (n = 215,708/1,559,733) of symptomatic and asymptomatic individuals, respectively. RT-PCR confirmed 94.4% (4,265/4,518) of positive and 90.6% (139,759/154,318) of negative RAT results in symptomatic individuals; and confirmed 83.4% (6,693/8,024) of positive and 95.3% (197,955/207,684) of negative RAT results in asymptomatic individuals. Applied assumptions for missing reflex RT-PCR results led to worst performance sensitivity estimates of 77.2 and 38.5% in the symptomatic and asymptomatic populations, respectively; assumptions for best performance estimates led to sensitivity values of 85.6 and 84.2%, respectively. Specificity values, regardless of assumptions or symptom category, ranged from 97.9-99.9%. At 10% SARS-CoV-2 prevalence, RAT positive predictive value was 86.9 and 99.0% for worst and best performance estimates across the total population, respectively; negative predictive values were >95% regardless of the applied assumption. Veritor test performance was consistent with that listed in the manufacturer instructions for use for symptomatic individuals. Real-world evidence should be gathered on RATs to support their efficacy as SARS-CoV-2 persists.

Project description:BackgroundThe safety, feasibility, and prognosis of sleeve lobectomy by minimally invasive surgery (MIS) remain to be validated. The purpose of this study was to investigate outcomes in real-world patients receiving minimally invasive sleeve lobectomy in a balanced large cohort.MethodsBetween January 2013 and December 2018, 578 consecutive patients undergoing sleeve resection at a high-volume center were retrospectively analyzed. Surgical and oncologic outcomes were compared between MIS and thoracotomy patients after propensity-score matching (PSM).ResultsMIS sleeve lobectomy was increasingly used as a time-trend in real-world. Before PSM, the MIS group had smaller tumor size, more T2-stage cases, and more right upper lobe sleeve lobectomies compared to the Open group. After 1:4 PSM by patient demographics and tumoral characteristics, 100 cases of MIS and 338 cases of Open sleeve lobectomy were further analyzed. Although median operation time was longer in the MIS group than in the Open group (170.5 minutes vs.149.5 minutes, P < 0.001), patients in MIS group had significantly less estimated intraoperative blood loss (100 ml vs. 200 ml, P = 0.003), shorter drainage duration (5 days vs. 6 days, P = 0.027) and less amount of drainage (1280 ml vs. 1640 ml, P < 0.001) after surgery. Complete resection rate, combined angioplasty, number of dissected lymph nodes, post-operative length of stay, postoperative morbidity and mortality rate, and application of adjuvant therapy were similar between the two matched groups. Conversion to open thoracotomy was necessary in 13.6% patients, but with similar perioperative outcomes compared to Open cases except for longer operation time. More lower lobe sleeve lobectomies were accomplished via robot-assisted thoracoscopic surgery than via video-assisted thoracoscopic surgery (40.0% vs. 12.0%, P = 0.017) in MIS patients. Five-year overall survivals (MIS vs. Open: 72.7% vs. 64.4%, P = 0.156) and five-year progression-free survivals (MIS vs. Open: 49.2% vs. 50.5%, P = 0.605) were similar between the two matched groups.ConclusionsMIS sleeve lobectomy is associated with similar or even better perioperative results and oncologic outcomes to open thoracotomy. Conversion to thoracotomy does not compromise perioperative outcomes. Robot surgery may be preferable for more complex sleeve resections.

Project description:BackgroundLenvatinib is approved for patients with advanced hepatocellular carcinoma (HCC) due to its non-inferiority to sorafenib of overall survival (OR) in clinical trials. This study was to compare the effectiveness and safety of lenvatinib and sorafenib in the real world.MethodsWe retrospectively evaluated 338 patients with unresectable HCC who had undergone lenvatinib or sorafenib treatment between January 2018 and August 2020. Propensity-score matching analysis was performed with a 1:2 ratio to reduce the real-life baseline difference between the two groups.ResultsA total of 210 patients (Male/Female: 150/60, mean age: 65.8 years) were recruited including 70 patients in the Lenvatinib group and 140 patients in the Sorafenib group. Compared with sorafenib, lenvatinib had significantly longer progression-free survival (PFS) (5.2 vs 3.3 months, p=0.019) but similar OR (13.3 vs 11.8 months, p=0.714). Additionally, lenvatinib had better disease control rates (62.3 vs 48.6%, p=0.029) and equivalent incidences of treatment-related adverse events over sorafenib. In multivariate analysis, lenvatinib was associated with better PFS over sorafenib (hazard ratio: 0.49, 95% confidence interval: 0.3-0.79, p=0.004) after adjustments of albumin-bilirubin grade and alpha-fetoprotein level; however, different agents using lenvatinib or sorafenib did not contribute to OS, whether in univariate or multivariate analysis. Patients who failed lenvatinib had a lower proportion of having sequential systemic therapies compared with the Sorafenib group (36.2 vs 47.8%, p=0.02). The most frequently used sequential therapy following lenvatinib and sorafenib was chemotherapy (n=9, 42.8%) and regorafenib (n=33, 50.8%), respectively.ConclusionsIn clinical real-life practice, lenvatinib illustrated promising survival benefits and acceptable safety for patients with unresectable HCC, while reducing the risk of progression disease compared with sorafenib. Additionally, lack of approved post-lenvatinib systemic therapies is a serious issue in the real world.

Dataset Information

Evaluating large-scale propensity score performance through real-world and synthetic data experiments.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets