| LOS Efficiency Index |
|
|
In order to obtain homogeneity of LOS in admissions for statistical analysis purposes, all admissions were grouped by the All Patient Refined Diagnosis Related Groups (APR-DRG**) and severity. (All copyrights in and to APR-DRGs** are owned by 3M. All rights reserved.) The APR-DRGs** have four levels of severity for each APR-DRG** – (1) minor, (2) moderate, (3) severe, and (4) extreme. However, within these groupings significant variance in the LOS still exists.
Certain admissions are eliminated from the analysis. These include admissions transferred out to another acute facility, deaths, and extreme high and low outliers. After the statistical models are developed, some of the transfers and deaths are added back. In effect, after the add backs, only the very short LOS transfers and deaths are excluded (generally less than 0 to 3 days but varies by APR-DRG**/severity and can be as high as 10 days).
Excluded long LOS outlier thresholds are generally set at 15 to 150 days but vary by APR-DRG**/severity and can exceed 300 days. Long Term Care admissions (acute, not SNF) and Rehab are treated separately. Also, the statistically estimated additional LOS from surgical mishaps and hospital-acquired infections are excluded for the purpose of applying the models. In effect, extra days resulting from these quality problems in excess of those implicitly found in the benchmark hospitals are considered avoidable. Step-wise linear regression models within the APR-DRG**/severity groups were built using additional statistically significant variables with sufficient volume. These modeled variables include diagnosis and procedure codes, source of admission, discharge disposition, Medicaid status (state data only), and broad age categories (<18, 18-64, 65+). Severities may be combined for low volume APR-DRGs**, but in these admissions the severity is added to the model as a variable. In some instances, groups of similar ICD-9 codes are combined. Interaction terms are added for variables with high correlation. Included in each regression model is a variable to determine each hospital’s efficiency for that APR-DRG**/severity grouping. The R2 for the resulting models are generally in the 90% range or above. Very few, usually low frequency groupings, are below 80%. Once the models are developed, benchmark analysis is conducted on the Efficiency Index variable using a t-test and Duncan’s Test. In applying the models to determine benchmarks and avoidable days, only the discharge disposition with the lowest coefficient (subject to significant volume) is considered. Statistical comparisons of potential benchmark groupings are run on hospitals, metropolitan statistical areas (MSAs) and states. These are all analyzed so that if low admission volume would not produce sufficient hospitals meeting minimum criteria (which include at least 30 admissions per entity for each APR-DRG**/severity grouping and a minimum number of hospitals/metropolitan areas and a minimum number of total admissions) the MSA or state results are used. In most benchmarks, the minimums are far exceeded. If the eligible benchmarks are clear outliers compared to other eligible benchmarks, they are excluded from consideration as the benchmark. Once the benchmarks are developed, the LOS models are all re-indexed to the benchmark levels, and a benchmark LOS is developed for each admission based on its characteristics. The actual LOS of each admission is compared to its unique benchmark LOS to develop avoidable days. The individual admission results are summarized by APR-DRG**/severity, specialty and higher levels of aggregation. Individual admission results are not meaningful per se, but only in the context of the aggregations of admissions through the application of statistical principles. |
** 3M™ All Patient Refined Diagnostic Related Groups. All copyrights in and to APR-DRGs are owned by 3M. All rights reserved.