BACKGROUND
Unplanned hospital readmission is frequent and costly. Existing readmission reduction solutions focus on complementing inpatient care with enhanced care transition and post-discharge interventions, which are initiated near or after discharge when clinicians’ impact on inpatient care is ending. Preventive intervention during hospitalization is an under-explored area and holds potential for reducing readmission risk. However, it is challenging for clinicians to identify high-risk patients early during hospitalization.
OBJECTIVE
The objective was to build a predictive model for early detection of hospital-wide all-cause 30-day unplanned hospital readmission. We were also interested at identifying novel readmission predictors.
METHODS
We extracted index admissions and previous encounters up to one year from the Cerner Health Facts® database. The model was only built with data of previous encounters and index admission data that can be available within 24 hours. Candidate models were evaluated in terms of performance, interpretability, timeliness, and generalizability. Multivariate analysis was used to identify readmission predictors.
RESULTS
Based on 96,550 patients’ data, we developed a readmission predictive model with AUC of 0.754. By multivariate analysis, we identified 16 novel readmission predictors, including patients with 1 maintenance chemotherapy last year (OR 1.476, 95% CI 1.218-1.790), the number of lymphocyte count test with abnormal result last year was 1 (OR 1.247, 95% CI 1.144-1.359) or ≥ 2 (OR 1.257, 95% CI 1.091-1.447), the number of monocyte count test with abnormal result last year was 1 (OR 1.199, 95% CI 1.056-1.362), the number of monocyte percent test with abnormal result last year was ≥ 2 (OR 1.371, 95% CI 1.178-1.596), the number of serum calcium quantitative test with abnormal result last year was 1 (OR 1.254 95% CI 1.107-1.420) or ≥ 2 (OR 1.345, 95% CI 1.122-1.612), the number of prescriptions of albuterol ipratropium last year was 1 (OR 1.073, 95% CI 1.010-1.141) or ≥ 2 (OR 1.157, 95% CI 1.052-1.272), the number of prescriptions of cefazolin last year was 1 (OR 0.884, 95% CI 0.822-0.950), the index admission hospital was in the Northeast census region (OR 1.441, 95% CI 1.345-1.543), prescribed gabapentin in index admission (OR 1.176, 95% CI 1.113-1.243), prescribed ondansetron in index admission (OR 1.111, 95% CI 1.057-1.168), prescribed polyethylene glycol 3350 in index admission (OR 1.076, 95% CI 1.017-1.139), prescribed cefazolin in index admission (OR 0.863, 95% CI 0.798-0.934), the number of lab tests with abnormal results in index admission was ≥ 16 (OR 1.151, 95% CI 1.043-1.269).
CONCLUSIONS
The performance of our model is better than the most widely used models in the US health care settings. By multivariate analysis, we identified 16 novel readmission predictors. This model can help clinicians to identify readmission risk early during hospitalization so that clinicians can pay extra attention to high-risk patient’s discharge process.
A Predictive Model for Early Detection of Hospital-Wide All-Cause 30-Day Hospital Readmission (Preprint)
