Utilisation Of Risk-Adjusted German Inpatient Administrative Data For The Evaluation Of In-Patient Mortality

Based on a risk-adjusted multivariable model, we investigated the feasibility of inpatient mortality quality analyses using German administrative. Administrative data (required by law, (SIC) 21 Kran-kenhausentgeltgesetz) of six German tertiary hospitals were used to include 557 989 stationary patients (>= 18 years of age, length of stay > 24 hours, valid DRG-coding). Probabilities were calculated for each hospitalized case by a logistic regression model mortality. Then they were compared with the observed mortality rates. The reproducibility of the regression model on specific subgroups was tested using Standardized Mortality Ratios (SMRs). External validation was performed against the 2010 data set of the QKK project (N = 390 489). (QKK: Qualitatsindikatoren fur Kirchliche Krankenhauser, quality indicators for denominational hospitals)Valid predictors for inpatient mortality were: age, number of concurrent diseases, number of coded procedures, duration of ventilation, DRG-partition, DRG-value (cost weight), gender, emergency admission status, patient clinical complexity level (PCCL) and length of stay. The regression model showed a Nagelkerke-Pseudo-R2 of 0.36, equivalent to an AUC of 0.921 [CI95 (0.919; 0.923)]. Differences between unadjusted and adjusted mortality rates were quantified between hospitals and the subgroups examined. The validation against the external data set showed a good transferability with limited stability.The regression model showed a very good prediction for inpatient mortality. A risk-adjusted comparison of mortality rates indicated potential for misinterpretation in quality assessments of hospitals and subgroups using unadjusted mortality rates alone.Further analyses with larger datasets are required to ensure model stability.