Modelling hospital outcome: problems with endogeneity

Abstract

Background Mortality modelling in the critical care paradigm traditionally uses logistic regression, despite the availability of estimators commonly used in alternate disciplines. Little attention has been paid to covariate endogeneity and the status of non-randomized treatment assignment. Using a large registry database, various binary outcome modelling strategies and methods to account for covariate endogeneity were explored. Methods Patient mortality data was sourced from the Australian & New Zealand Intensive Society Adult Patient Database for 2016. Hospital mortality was modelled using logistic, probit and linear probability (LPM) models with intensive care (ICU) providers as fixed (FE) and random (RE) effects. Model comparison entailed indices of discrimination and calibration, information criteria (AIC and BIC) and binned residual analysis. Suspect covariate and ventilation treatment assignment endogeneity was identified by correlation between predictor variable and hospital mortality error terms, using the Stata™ “eprobit” estimator. Marginal effects were used to demonstrate effect estimate differences between probit and “eprobit” models. Results The cohort comprised 92,693 patients from 124 intensive care units (ICU) in calendar year 2016. Patients mean age was 61.8 (SD 17.5) years, 41.6% were female and APACHE III severity of illness score 54.5(25.6); 43.7% were ventilated. Of the models considered in predicting hospital mortality, logistic regression (with or without ICU FE) and RE logistic regression dominated, more so the latter using information criteria indices. The LPM suffered from many predictions outside the unit [0,1] interval and both poor discrimination and calibration. Error terms of hospital length of stay, an independent risk of death score and ventilation status were correlated with the mortality error term. Marked differences in the ventilation mortality marginal effect was demonstrated between the probit and the "eprobit" models which were scenario dependent. Endogeneity was not demonstrated for the APACHE III score. Conclusions Logistic regression accounting for provider effects was the preferred estimator for hospital mortality modelling. Endogeneity of covariates and treatment variables may be identified using appropriate modelling, but failure to do so yields problematic effect estimates.

The benefit of dexamethasone in patients with COVID‐19 infection is preserved in patients with diabetes

“…One of the measures that were chosen a priori was FiO 2 , that is, the maximum oxygen requirement during admission—a direct measure of disease severity. Rather than arbitrarily remove this factor from this analysis, we elected to control for the potential endogeneity 7 of FiO 2 (which would otherwise invalidate the statistical analysis), we ran a first stage regression of FiO 2 on a set of exogenous regressors (clinical frailty score, admission temperature, respiratory rate, heart rate and systolic blood pressure), which showed a good fit ( R 2 of 30%). Thereafter the logit specifications were run in STATA, adding the residuals from the first stage regressions.…”

Unlocking Sub-Saharan Africa's Export Potential: A Reassessment of the Influence of Manufacturing Firms Access to finance

“…The result from Table 6 further reveals whether access to bank nance is rightly an endogenous variable. Speci cally, the error correlation is an estimate of the correlation between the error from the endogenous covariate equation and the error term from the outcome equation (He and Jiang, 2019; Moran et al, 2021;StataCorp, 2021). The rejection of the null hypothesis of no endogeneity in access to bank nance is indicated by the signi cance of the estimated correlation between the errors of access to bank nance and rm export performance at the 1 per cent level.…”

Reducing Crowding in Emergency Departments With Early Prediction of Hospital Admission of Adult Patients Using Biomarkers Collected at Triage: Retrospective Cohort Study

“…To assess potential overoptimism, we also calculated a 10-fold cross-validated AUROC. Calibration, the extent to which the model-predicted probabilities agree with observed binary outcomes [ 90 ], is a more appropriate gauge of model performance [ 91 ] and was measured by Hosmer-Lemeshow goodness of fit and evaluated graphically using a “calibration belt” [ 92 ] for internal validation. The calibration belt methodology formulated the relationship between the predictions and the true probabilities of admission with a second logit regression model based on a polynomial transformation of the predictions.…”

“…The calibration belt methodology formulated the relationship between the predictions and the true probabilities of admission with a second logit regression model based on a polynomial transformation of the predictions. The degree of the polynomial was forwardly selected, beginning with the second order on the basis of a sequence of likelihood-ratio tests [ 91 ].…”