Abstract
Aim:
Evaluate forecasting models applied to smaller geographic locations within the hospital.
Materials & methods:
Damped trend models were applied to blood glucose measurements of progressively smaller inpatient geographic subpopulations. Mean absolute percentage error (MAPE) and 95% prediction intervals (PIs) assessed validity of the models to forecasts 48 weeks into the future.
Results:
MAPE values increased, and 95% PIs widened, when data from progressively smaller geographic areas were analyzed. MAPE values were highest and 95% PIs were broadest with the smallest geographic areas. In contrast, observations missed at larger geographical locations were more evident with smaller subpopulations.
Conclusion:
The utility of damped trend models to forecast inpatient glucose control diminished when applied to smaller geographic areas within the hospital.
Hyperglycemia (high blood sugar) is associated with worse outcomes in hospitalized patients. Predicting future glucose control could identify problems sooner and lead to earlier quality improvement changes. This study examined whether a forecasting tool used across an entire hospital might be useful in smaller geographic areas of the facility. The models became less reliable when tested with progressively smaller geographic subpopulations.
Collapse