Automatic variable selection for exposure-driven propensity score matching with unmeasured confounders

Predictors and outcomes of acute pulmonary embolism in COVID-19; insights from US National COVID cohort collaborative

OBJECTIVES

To investigate whether COVID-19 patients with pulmonary embolism had higher mortality and assess the utility of D-dimer in predicting acute pulmonary embolism.

PATIENTS AND METHODS

Using the National Collaborative COVID-19 retrospective cohort, a cohort of hospitalized COVID-19 patients was studied to compare 90-day mortality and intubation outcomes in patients with and without pulmonary embolism in a multivariable cox regression analysis. The secondary measured outcomes in 1:4 propensity score-matched analysis included length of stay, chest pain incidence, heart rate, history of pulmonary embolism or DVT, and admission laboratory parameters.

RESULTS

Among 31,500 hospitalized COVID-19 patients, 1117 (3.5%) patients were diagnosed with acute pulmonary embolism. Patients with acute pulmonary embolism were noted to have higher mortality (23.6% vs.12.8%; adjusted Hazard Ratio (aHR) = 1.36, 95% CI [1.20-1.55]), and intubation rates (17.6% vs. 9.3%, aHR = 1.38[1.18-1.61]). Pulmonary embolism patients had higher admission D-dimer FEU (Odds Ratio(OR) = 1.13; 95%CI [1.1-1.15]). As the D-dimer value increased, the specificity, positive predictive value, and accuracy of the test increased; however, sensitivity decreased (AUC 0.70). At cut-off D-dimer FEU 1.8 mcg/ml, the test had clinical utility (accuracy 70%) in predicting pulmonary embolism. Patients with acute pulmonary embolism had a higher incidence of chest pain and history of pulmonary embolism or deep vein thrombosis.

CONCLUSIONS

Acute pulmonary embolism is associated with worse mortality and morbidity outcomes in COVID-19. We present D-dimer as a predictive risk tool in the form of a clinical calculator for the diagnosis of acute pulmonary embolism in COVID-19.

Solutions to problems of nonexistence of parameter estimates and sparse data bias in Poisson regression

Poisson regression can be challenging with sparse data, in particular with certain data constellations where maximum likelihood estimates of regression coefficients do not exist. This paper provides a comprehensive evaluation of methods that give finite regression coefficients when maximum likelihood estimates do not exist, including Firth’s general approach to bias reduction, exact conditional Poisson regression, and a Bayesian estimator using weakly informative priors that can be obtained via data augmentation. Furthermore, we include in our evaluation a new proposal for a modification of Firth’s approach, improving its performance for predictions without compromising its attractive bias-correcting properties for regression coefficients. We illustrate the issue of the nonexistence of maximum likelihood estimates with a dataset arising from the recent outbreak of COVID-19 and an example from implant dentistry. All methods are evaluated in a comprehensive simulation study under a variety of realistic scenarios, evaluating their performance for prediction and estimation. To conclude, while exact conditional Poisson regression may be confined to small data sets only, both the modification of Firth’s approach and the Bayesian estimator are universally applicable solutions with attractive properties for prediction and estimation. While the Bayesian method needs specification of prior variances for the regression coefficients, the modified Firth approach does not require any user input.