A Cross-Sectional Study to Predict Mortality for Medicare Patients Based on the Combined Use of HCUP Tools

Prediction of inpatient mortality is not an easy problem since multiple comorbidities and other factors in synergy have a variable effect on inpatient death risk. This research combined Healthcare Cost and Utilization Project (HCUP) tools (clinical classification software, CCS; Chronic Condition Indicator, CCI) to recommend a critical set of CCS diagnosis and procedure predictors for mortality. The study motivation is to provide the research community an up-to-date critical set of inhospital mortality predictors. The study follows a cross-sectional design. An inpatient CMS claims file (N = 418,529) was combined with the HCUP grouper to transform the ICD-10-CM and CPT codes to CCS categories and to enhance the data with the acuity and the diagnosis presence/non-presence on admission. Five logistic regressions were conducted to progressively enhance the feature set with the aforementioned dimensions. The Sensitivity_death and positive predictive value (PPV_death) were estimated for each consecutive step to examine the attributable predictive power of each dimension. When all information were inserted, the PPV_death was 65.5%, a 10% increase over a single representation of secondary diagnoses. A critical collection of significant CCS diagnoses and procedures were extracted as predictors of inpatient mortality. The chronicity and POA status of a diagnosis improve the prediction of inpatient mortality. Furthermore, the combined use of these dimensions provides better predictions against the Elixhauser Comorbidity Index. The combined use of HCUP tools provides a reasonable estimate of inpatient mortality. This is the first study that uses the updated HCUP groupers for ICD-10-CM to provide insights about drivers of inpatient mortality.