Using structured pathology data to predict hospital-wide mortality at admission

Abstract

Early prediction of in-hospital mortality can improve patient outcome. Current prediction models for in-hospital mortality focus mainly on specific pathologies. Structured pathology data is hospital-wide readily available and is primarily used for e.g. financing purposes. We aim to build a predictive model at admission using the International Classification of Diseases (ICD) codes as predictors and investigate the effect of the self-evident DNR ("Do Not Resuscitate") diagnosis codes and palliative care codes. We compare the models using ICD-10-CM codes with Risk of Mortality (RoM) and Charlson Comorbidity Index (CCI) as predictors using the Random Forests modeling approach. We use the Present on Admission flag to distinguish which diagnoses are present on admission. The study is performed in a single center (Ghent University Hospital) with the inclusion of 36 368 patients, all discharged in 2017. Our model at admission using ICD-10-CM codes (AUCROC = 0.9477) outperforms the model using RoM (AUCROC = 0.8797 and CCI (AUCROC = 0.7435). We confirmed that DNR and palliative care codes have a strong impact on the model resulting in a decrease of 7% for the ICD model (AUCROC = 0.8791) at admission. We therefore conclude that a model with a sufficient predictive performance can be derived from structured pathology data, and if real-time available, can serve as a prerequisite to develop a practical clinical decision support system for physicians.

Fig 1. ICD-10-CM diagnosis code hierarchy and with example: S52, fracture of the forearm.

The ICD-10-CM code consists of a chapter (S), category (S52) and full code (S52.521A).

Fig 2

A (Upper panel). ROC curves showing the results using three different sets of predictors in the Random Forests model. The figure on the left shows the ROC curve with all diagnoses known at admission, while the figure on the right shows all diagnoses known at discharge. The ROC curve in the middle is for the models using only the diagnoses known at admission, excluding all admissions with DNR and palliative care codes at admission. B (Lower panel). Precision-Recall plot showing the results using three different sets of predictors in the Random Forests model. The figure on the left shows the PR curve with all diagnoses known at admission and on the right all diagnoses known at discharge. The PR curve in the middle are the models using only the diagnoses known at admission, excluding all admissions with codes DNR or palliative care codes at admission. Both approaches show low performance using CCI as a predictor for in-hospital mortality, while the models using ICD as predictors perform best overall. Legend: CCI = Charlson Comorbidity Index; RoM = Risk of Mortality; ICD = International Classification of Diseases.

Fig 3. Variable importance plot for the models using ICD-10-CM diagnosis codes as predictors at admission and discharge, either using all diagnosis codes, without the Do Not Resuscitate or palliative care codes at admission, or without these codes and without the principal diagnosis.

Whereas ‘Encounter for palliative care’, ‘Do not resuscitate’ and ‘Encounter for other aftercare and medical care’ are the three most important variables in the models with all diagnosis, ‘Cardiac arrest’ is the most important variable in the model with the excluded patients.