Case-mix adjustment using objective measures of severity: the case for laboratory data

Prior research has shown diagnostic-related group (DRG) systems are widely used to categorize inpatient stays for billing and resource planning. However, these systems often fail to capture variations in patient severity within broad diagnostic categories. It was already known that DRG-based classifications may not fully explain differences in length of stay (LOS) across similar diagnoses. No prior work had resolved how to improve LOS prediction without increasing classification complexity. This gap motivated investigation into alternative data sources for subclassification. Researchers have explored clinical variables, but few have focused on routinely collected laboratory data. The uncertainty around the predictive value of lab data in this context drove the need for new analysis. This study aimed to assess whether lab test results could enhance LOS prediction within major diagnostic categories.

The aim of this study was to evaluate the potential of laboratory data to improve the accuracy of LOS prediction within surgical and nonsurgical major diagnostic categories. The specific problem addressed is the limited explanatory power of DRG classifications for inpatient resource utilization. The motivation stems from the need for more precise case-mix adjustment in hospital planning. Researchers sought to determine if lab data could provide a more homogeneous grouping of patients than DRG-based systems. They focused on seven routine lab tests as potential predictors. The study population was derived from a merged database of two medical centers. The goal was to compare predictive accuracy across three classification schemes. The study aimed to validate whether lab data could enhance LOS prediction without overfitting.

The study used a cross-sectional, retrospective design with data from the Combined Patient Experience (COPE) database. The database merged records from two medical centers, yielding 73,117 admissions. The unit of analysis was an individual admission. Researchers selected 13 sub-MDCs representing 45% of inpatient stays. Nine predictor variables were derived from seven lab tests (WBC, Na, K, CO2, BUN, ALB, HCT). Minimum and maximum values were recorded for each test during the hospital stay. Patients were randomly assigned to two datasets in a 2:1 ratio. One dataset was used to create models, the other to validate them. Three classification schemes were compared: DRG classes, lab data-based classes, and a combination of both.

The study compared predictive accuracy across three classification schemes for eight of the largest sub-MDCs. DRG classes explained 23% of the variance in LOS within these sub-MDCs. Laboratory data classes explained 31% of the variance. The combined approach explained 37% of the variance. These results are weighted average R2 values. The number of LOS classes used in partitioning was 20 for DRGs, 10 for lab data, and 10 for the combined approach. In six of the eight sub-MDCs, lab data alone explained more variance than DRG classes. The improvement in prediction was not due to overfitting. The number of classes used was fewer than the number of DRGs. These findings suggest lab data can enhance LOS prediction accuracy.

The authors propose that laboratory data improve the accuracy of LOS prediction compared to DRG classifications. They emphasize that the improvement is not due to overfitting the data. The number of classes used in prediction is fewer than the number of DRGs. This suggests lab data can provide a more efficient classification scheme. The findings indicate that lab data alone may better capture patient severity than DRG-based systems. The study supports the use of lab data for case-mix adjustment in hospital planning. The authors suggest that integrating lab data into existing classification systems could enhance resource allocation. They conclude that routine lab test data offer a promising alternative for improving LOS prediction.