Electronic data modeling to predict 30-day hospital readmission for older adults
Recommended Citation
Khan A, Hook ML, Pagel P, Vollbrecht MA, Singh M. Electronic data modeling to predict 30-day hospital readmission for older adults. J Patient-Centered Res Rev. 2014;1:142-143.
Presentation Notes
Presented at 2014 Aurora Scientific Day, Milwaukee, WI
Abstract
Background/Significance: Approximately 20% of Medicare beneficiaries are readmitted within 30 days, costing $17.4 billion annually. Research predicting readmission (readmit) has focused on administrative and diagnosis data.
Purpose: The aim of this study was to identify electronic health record (EHR)-based clinical factors to predict readmit for older adults.
Methods: This retrospective cohort study used demographic, diagnoses, and clinical EHR data to identify readmit predictors at a large quaternary medical center. The population was limited to adults > 65 years, index length of stay < 30 days and those not discharged to an acute care facility or inpatient rehabilitation. Logistic regression modeling evaluated clinical predictors with diagnoses from two sources: medical history and postdischarge ICD9 coding. Univariate analysis was done for categorical and continuous variables. For multivariate logistic regression, the population was divided into derivation (70%) and validation (30%) cohorts.
Results: The sample (N=4,503; mean age ± standard deviation (SD): 77 ± 8 years; female: 54%) included patients hospitalized between July 2012 and Dec. 2012. Index length of stay ± SD was 4.9 ± 4; disposition to home was 65%, to home care was 18% and to skilled nursing was 18%; readmit rate was 12.3%. Readmit predictors were: age, heart failure, COPD, depression, anxiety, gastrointestinal disease, malnutrition, chronic pain, Medicaid insurance, length of stay, smoking, respiratory symptoms, social work consult, hypertension and acute respiratory failure (ICD9 only), pneumonia and kidney disease (medical history only). The receiver-operating characteristic (ROC) C-statistic using ICD9 diagnoses was 0.64 (95% confidence interval [CI]: 0.61-0.67) for derivation and 0.63 (95% CI: 0.58- 0.67) for validation cohorts, respectively, with significant predictors being age 75-84 (odds ratio [OR]: 1.33; 95% CI: 1.04-1.71), depression (OR: 1.42; 95% CI: 1.01-2.01), hypertension (OR: 0.76; 95% CI: 0.61-0.95); smoking past (OR: 0.69; 95% CI: 0.53-0.88), length of stay = 0.95 (95% CI: 0.93-0.98). The model using medical history data produced similar findings (ROC: 0.64, 95% CI: 0.61-0.67) with somewhat different predicators. Limitations included single site and missing clinical values.
Conclusion: EHR-based clinical factors were found to predict readmission. Medical history produced similar results to ICD9 coding, suggesting that risk can be predicted using clinical data available during patient care. More work is needed to isolate clinical predictors for use in creating real time scoring mechanisms.
Document Type
Abstract
Affiliations
Department of Geriatrics, Aurora Research Institute, Senior Services