Risk adjustment models play a critical role in directing medical care, adjusting case mix in research, and supporting health care planning and financing, including adjusting payments based on patient complexity. However, these models, such as the Elixhauser Comorbidity Index, often focus only on comorbidities and fail to capture the medical complexity of the patient. This is especially true in Israel, where hospitals are compensated with a fixed payment per day of hospitalization, regardless of the complexity of the medical case. Few studies to date have extended the Elixhauser model to include broader clinical information to improve treatment outcome predictions.
Researchers from the Hebrew University were able to externally improve the Elixhauser model to improve its accuracy. The study, led by Prof. Adam Rose from the Faculty of Medicine at the Hebrew University, was published in the journal BMC Health Services Research, incorporated additional clinical and demographic data into the model. By combining these data, the study improved the accuracy of predictions of key outcomes such as length of hospitalization, mortality during hospitalization, readmission to hospitalization within 30 days, and increased care, such as intensive care units or similar types of care.
Assistance in making clinical decisions
The improved model not only assists healthcare professionals in making more informed clinical decisions, but also enables a more efficient allocation of resources in healthcare institutions. The updated model improves the overall quality of patient care, may lead to cost savings and contributes to progress in health research.
Using Israel's unique central health database, the study performed a retrospective analysis of 55,946 hospitalizations in the internal medicine department of Shaare Zedek Medical Center in Jerusalem. By including additional variables such as laboratory test results, vital signs and demographic information, which are not included in Elixhauser's basic model, the new model achieved significant improvements in predictive accuracy compared to the basic model that had been in use for decades.
The study shows that the upgraded model was more accurate in predicting certain health outcomes. For example, in predicting length of stay, accuracy improved from an R2 of 10.1% to 28.1%. In addition, in the forecast regarding the risk of mortality during hospitalization, the accuracy index (c-statistic) increased from 71.1% to 87.9%. These improvements emphasize that the improved model is better in estimating the length of hospitalization and predicting the risk of mortality compared to the standard model.
"Our improved model fills a critical gap in Elixhauser's original model by providing a more comprehensive assessment of patient complexity," said Professor Adam Rose. "The model has broad applications in other medical contexts, both within the field of internal medicine and beyond, and it can support decisions regarding hospitalization and treatment, adjustment of home care and payments based on the complexity of the medical case.
Improving a disease prediction model. The image was prepared using DALEE and is not a scientific image.
