Prioritized Post-Discharge Telephonic Outreach Reduces Hospital Readmissions for Select High-Risk Patients | Page 2

Stratified randomization design created comparable prioritized and unprioritized groups at baseline. A daily list of eligible patients discharged from the previous day was produced by utilization managers and called a Census Discharge File. The Census Discharge File contained discharge-related information (ie, dates, diagnosis, procedure codes, etc) describing the inpatient episode and we confirmed that Census Discharge File information was consistent with data in paid claims 90 days later. Our stratified randomization logic assigned study subjects into 1 of several strata defined by a combination of the following variables: gender, ERG score, major ICD-9-CM diagnostic group, count of prior 12-month admissions, health insurance plan type, and hospital facility. Within each stratum, patients were randomly assigned to either the prioritized or unprioritized group. Under this randomization protocol we prospectively matched patients one-to-one on gender, risk score, major diagnostic group, recent admissions history, plan type, and discharge hospital. The matching variables were selected because previous studies showed they were strong predictors of readmissions.^5,11-13

Outcomes Studied

The primary outcome was the percent of unique urgent emergent readmissions at 30 days and 60 days. The secondary outcome was readmission rates per 1000. Urgent emergent readmissions were defined as all-cause unscheduled admissions following the initial discharge. We explored measuring clinically related readmissions, but decided against it because secondary claims did not provide the level of clinical detail needed for appropriate measurement, such as chart abstraction information. All outcomes are derived from insurance claims data and CM utilization data including facility, professional, pharmaceutical, and CM call activity.

Statistical Analyses

A group proportions power calculation determined needed sample size. We hypothesized that the intervention would yield a 2% to 3% difference in readmissions between the comparison groups based on the results of a prior internal retrospective analysis.¹⁰ Using the power calculation inputs of power 0.80 and a 2-sided P value of .05, the estimated sample size was 3988. We aimed for a cohort of 4786 to account for the 15% to 17% of patients likely to be excluded because of having no comparable match, inactive coverage, or mortality. Baseline demographics and readmission outcomes were analyzed by χ² tests for discrete variables, 1-way analysis of variance for normally distributed continuous variables, and the Wilcoxon and Mann-Whitney test for skewed distribution of continuous variables. For all statistical analyses, alpha was set to 0.05 and SAS software version 9.1 was used (SAS Institute Inc, Cary, North Carolina). All reported readmission results used intent-to-treat estimation.

Post Hoc Assessments

Three retrospective post hoc assessments further assessed the relationship between the intervention and outcomes. Each post hoc assessment used one-to-one retrospective matched case control without replacement to control for bias due to non-random assignment of the newly created post hoc comparison groups formed from within the original data set. The groups were retrospectively matched on key demographics and were statistically indistinguishable (P >.05) at baseline by age, gender, ERG score, major diagnosis at initial discharge, 12-month admission history, and plan type.

The first post hoc assessment explored the readmission trend of the second control group. The second control group consisted of patients eligible for the pilot that were not randomly assigned to the prioritized or unprioritized group because they had no comparable match when first entering the pilot (N = 521). The second control group was outreached to the same as the first control group. We postulated that the readmission trend of the second control group should be similar to the unprioritized group since both groups did not receive the prioritization outreach intervention.

The second assessment examined the rates of outpatient physician visits and prescription drug fills within 30 days post discharge between the comparison groups. The hypothesis was that patients in the prioritized group, who experienced a lower rate of readmissions, would have a higher rate of prescriptions filled as well as outpatient visits, prompted by more expedient follow-up calls.

The third assessment explored whether the readmission results impacted total medical cost (TMC). TMC was defined as the sum of eligible charge amounts for inpatient, outpatient, professional, and facility claims. To measure the change in TMC consistent with the 60-day readmission results, we captured TMC 60 days prior to, and 60 days after, the initial discharge. To minimize the effect outliers we capped the 4% of participants with TMC greater than $100,000.00 at $100,000.00. We used a difference-in-differences regression model to test the impact on the change in TMC between the 2 comparisons groups pre- and post-intervention.

RESULTS

Baseline Characteristics and Post-Discharge Intervention

We identified 4807 patients eligible for the study. Of those identified, 298 were excluded because of inactive healthcare coverage and 521 were excluded after randomization because they did not have a comparable match. The Table shows the baseline characteristics of the 3988 randomly matched patients in the study (1994 prioritized and 1994 unprioritized). After random assignment, the comparison groups had similar distributions by gender, age, inpatient history, discharge diagnosis, chronic conditions, and health plan type. No single hospital made up a disproportionate number of discharges. The ERG scores were used to obtain an aggregate measure of patients’ health risk and the risk distribution was at baseline. All statistical tests confirmed no statistically significant differences between the 2 groups following randomization.