A flexible approach for variable selection in large-scale healthcare database studies with missing covariate and outcome data.

BACKGROUND: Prior work has shown that combining bootstrap imputation with tree-based machine learning variable selection methods can provide good performances achievable on fully observed data when covariate and outcome data are missing at random (MAR). This approach however is computationally expensive, especially on large-scale datasets. METHODS: We propose an inference-based method, called RR-BART, which leverages the likelihood-based Bayesian machine learning technique, Bayesian additive regression trees, and uses Rubin's rule to combine the estimates and variances of the variable importance measures on multiply imputed datasets for variable selection in the presence of MAR data. We conduct a representative simulation study to investigate the practical operating characteristics of RR-BART, and compare it with the bootstrap imputation based methods. We further demonstrate the methods via a case study of risk factors for 3-year incidence of metabolic syndrome among middle-aged women using data from the Study of Women's Health Across the Nation (SWAN). RESULTS: The simulation study suggests that even in complex conditions of nonlinearity and nonadditivity with a large percentage of missingness, RR-BART can reasonably recover both prediction and variable selection performances, achievable on the fully observed data. RR-BART provides the best performance that the bootstrap imputation based methods can achieve with the optimal selection threshold value. In addition, RR-BART demonstrates a substantially stronger ability of detecting discrete predictors. Furthermore, RR-BART offers substantial computational savings. When implemented on the SWAN data, RR-BART adds to the literature by selecting a set of predictors that had been less commonly identified as risk factors but had substantial biological justifications. CONCLUSION: The proposed variable selection method for MAR data, RR-BART, offers both computational efficiency and good operating characteristics and is utilitarian in large-scale healthcare database studies.

Post-Anticoagulant D-dimer as a Highly Prognostic Biomarker of COVID-19 Mortality

ImportanceClinical biomarkers that accurately predict mortality are needed for the effective management of patients with severe COVID-19 illness. ObjectiveTo determine whether D-dimer levels after anticoagulation treatment is predictive of in-hospital mortality. DesignRetrospective study using electronic health record data. SettingA large New York City hospital network serving a diverse, urban patient population. ParticipantsAdult patients hospitalized for severe COVID-19 infection who received therapeutic anticoagulation for thromboprophylaxis between February 25, 2020 and May 31, 2020. ExposuresMean and trend of D-dimer levels in the 3 days following the first therapeutic dose of anticoagulation. Main OutcomesIn-hospital mortality versus discharge. Results1835 adult patients (median age, 67 years [interquartile range, 57-78]; 58% male) with PCR-confirmed COVID-19 who received therapeutic anticoagulation during hospitalization were included. 74% (1365) of patients were discharged and 26% (430) died in hospital. The study cohort was divided into four groups based on the mean D-dimer levels and its trend following anticoagulation initiation, with significantly different in-hospital mortality rates (p<0.001): 49% for the high mean-increase trend (HI) group; 27% for the high-decrease (HD) group; 21% for the low-increase (LI) group; and 9% for the low-decrease (LD) group. Using penalized logistic regression models to simultaneously analyze 67 variables (baseline demographics, comorbidities, vital signs, laboratory values, D-dimer levels), post-anticoagulant D-dimer groups had the highest adjusted odds ratios (ORadj) for predicting in-hospital mortality. The ORadj of in-hospital death among patients from the HI group was 6.58 folds (95% CI 3.81-11.16) higher compared to the LD group. The LI (ORadj: 4.06, 95% CI 2.23-7.38) and HD (ORadj: 2.37; 95% CI 1.37-4.09) groups were also associated with higher mortality compared to the LD group. Conclusions and RelevanceD-dimer levels and its trend following the initiation of anticoagulation have high and independent predictive value for in-hospital mortality. This novel prognostic biomarker should be incorporated into management protocols to guide resource allocation and prospective studies for emerging treatments in hospitalized COVID-19 patients. Key PointsO_ST_ABSQuestionC_ST_ABSAre D-dimer levels following therapeutic anticoagulation predictive of mortality in hospitalized COVID-19 patients? FindingIn a retrospective study of 1835 adult COVID-19 patients who received therapeutic anticoagulation for thromboprophylaxis during hospitalization, 1365 (74%) patients were discharged and 470 (26%) died. Post-anticoagulant D-dimer levels and trends were significantly and independently predictive of mortality. MeaningActive monitoring of post-anticoagulant D-dimer levels in hospitalized COVID-19 patients is a novel strategy for stratifying individual risk of in-hospital mortality that can help guide resource allocation and prospective studies for emerging treatments for severe COVID-19 illness.