Background:

Preventing unplanned readmissions within 30 days of hospital discharge is a major focus of quality improvement and payment reform. Identification of patients at high risk for 30–day readmissions is an important step towards improving care and reducing readmissions. The growing adoption of electronic health records (EHR) may prove to be an important component of strategies designed to risk stratify patients and introduce targeted interventions.

Purpose:

To develop and implement an automated predictive model integrated into our hospital’s EHR that identifies on admission patients at high risk for readmission within 30 days of discharge.

Description:

We first performed a systematic literature review to identify risk factors for 30–day readmissions. We then examined the data available from our hospital EHR at admission for those factors identified in our review, identified three variables that were consistently accurate, and then developed and tested 30 candidate prediction rules using a combination of these variables, including >0 and >1 prior admissions in the last 6 and 12 months, >0 and >1 prior ED visits in the last 6 and 12 months, and prior 30 day readmissions in the last 6 and 12 months. Rules were tested on 24 months of historic data with a baseline 30–day readmission rate of 5%. A single risk factor, >1 inpatient admissions in the past 12 months, was found to have the best balance of sensitivity (25%), positive predictive value (20%), and proportion of patients flagged (7%). An automated readmission risk flag was then integrated into the hospital’s EHR, such that patients with this risk factor are flagged in the EHR on admission (Fig. 1). The flag can be double–clicked to display information relevant to discharge planning including inpatient and ED visits in the prior 12 months, as well as information about the primary team, length of stay, and admitting problem for those admissions. In the 30–day period after risk flag implementation, 13% of inpatients were flagged, and of those 21% were readmitted to our institution. The distribution of flags across hospital units corresponds closely with the distribution of readmissions as expected.

Conclusions:

An automated prediction rule was easily and effectively integrated into an existing EHR and identified patients on admission who are at risk for readmission within 30 days of discharge. Future work will further prospectively evaluate the flags performance, gather qualitative data on how providers and care teams use the information provided by the flag, and examine the impact of the flag on readmission rates.