Cloudy with a Chance of Discharge: An Evaluation of General Medicine Discharge Prediction Forecasts

1Duke University Medical Center, Durham, NC
2Durham VA Medical Center, Durham, NC

Meeting: Hospital Medicine 2014, March 24-27, Las Vegas, Nev.

Abstract number: 153

Background:

Hospital throughput and discharge coordination are strategic priorities for the hospital to ensure patients are safely transitioned across episodes of care and to predict bed availability for patients being admitted. Coordinating bed flow and discharge requires anticipation of when discharges are likely to occur. Physicians make these determinations to help care teams plan for anticipated discharges, however it is unclear if these predictions are accurate. To our knowledge, there has not been an evaluation of provider discharge prediction accuracy. This quality improvement effort was designed to evaluate discharge prediction accuracy of general medicine providers.

Methods:

General medicine providers were surveyed at 7‐8AM (morning handoff) and 12‐2PM (post rounds) to determine discharge predictions. Providers were asked to indicate which of their current patients had an >/= 80% likelihood of discharge during the same day or the following day. Predictions were tracked to evaluate for sensitivity of predictions as well as positive and negative predictive values. The time of the prediction and its effect on prediction accuracy was also assessed.

Results:

Over 2 weeks, there were 1741 observations. The baseline rate of same‐day discharge was 20%. In patients in whom same‐day discharge was predicted, the overall positive predictive value was 69% (CI 65%‐72%) with a sensitivity of 70%. When predictions were made prior to morning rounds, PPV was 62% (CI 57%‐67%) compared with 79% (CI 74%‐82%) when predictions were made after morning rounds. The PPV was also influenced by the training level of the predicting provider. When residents predicted same‐day discharge, PPV was 63% (CI 56%‐68%); when attendings, however, predicted same‐day discharge, PPV was 72% (CI 67%‐76%). When patients were predicted to be discharged on the day following prediction, the prediction was less accurate. Overall, 40% (CI 35%‐45%) of patients in whom next‐day discharge was predicted were discharged the next day with a sensitivity of 32%. Similar to same‐day discharges predictions, next day predictions were more accurate when made after morning rounds and when made by attending physicians (data not shown).

Conclusions:

Predicting discharges has important implications on coordination of care and hospital throughput. Physician discharge prediction accuracy is limited and is influenced by level of training, time before discharge, and time of day. Additional study is needed to determine what level of precision in predictions is required to begin standardized discharged processes and to make triggered interventions cost and time effective. Furthermore, additional study may help define objective measures to improve physician prediction of discharge readiness.

To cite this abstract:

Bae J, Ming D, Choi J, Clark A, O’brien C, Gentry J, Schulteis R. Cloudy with a Chance of Discharge: An Evaluation of General Medicine Discharge Prediction Forecasts. Abstract published at Hospital Medicine 2014, March 24-27, Las Vegas, Nev. Abstract 153. Journal of Hospital Medicine. 2014; 9 (suppl 2). https://www.shmabstracts.com/abstract/cloudy-with-a-chance-of-discharge-an-evaluation-of-general-medicine-discharge-prediction-forecasts/. Accessed September 19, 2019.

« Back to Hospital Medicine 2014, March 24-27, Las Vegas, Nev.