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

Area of Science:

• Medical Informatics
• Clinical Prediction Models
• Health Services Research

Background:

• Early prediction of in-hospital mortality is vital for improving patient outcomes.
• Existing models often focus on specific diseases, limiting broad applicability.
• Hospital-wide structured pathology data, like ICD codes, offers a readily available resource.

Purpose of the Study:

• To develop and evaluate a predictive model for in-hospital mortality using International Classification of Diseases (ICD) codes at admission.
• To compare the performance of ICD-based models against established predictors like Risk of Mortality (RoM) and Charlson Comorbidity Index (CCI).
• To investigate the impact of "Do Not Resuscitate" (DNR) and palliative care codes on predictive accuracy.

Main Methods:

• Utilized Random Forests modeling approach for prediction.
• Employed ICD-10-CM codes as primary predictors, distinguishing diagnoses present on admission using the Present on Admission (PoA) flag.
• Compared model performance using Area Under the Receiver Operating Characteristic curve (AUCROC).

Main Results:

• The ICD-10-CM model achieved a high AUCROC of 0.9477, significantly outperforming RoM (AUCROC = 0.8797) and CCI (AUCROC = 0.7435).
• Inclusion of DNR and palliative care codes notably impacted the model, reducing AUCROC to 0.8791, indicating their strong association with mortality.
• The study included 36,368 patients discharged in 2017 from Ghent University Hospital.

Conclusions:

• Structured pathology data, specifically ICD codes, can form the basis for a highly predictive in-hospital mortality model.
• Real-time availability of such data could facilitate the development of practical clinical decision support systems for physicians.
• The model's performance highlights the potential of leveraging routinely collected hospital data for enhanced clinical decision-making.