Background: Older patients with multiple chronic conditions are major users of health care, as they often see multiple specialists, frequently visit the emergency department (ED), and have multiple hospital admissions per year. One model to improve care for these high-risk patients is case management, with a nurse or social worker (SW) taking responsibility for coordinating and implementing a patient's care plan. Case managers are now frequently used at health systems participating in the Medicare Accountable Care Organization (ACO) program. However, case management is resource- and time-intensive; programs vary widely, with mixed effectiveness, and there is little guidance for health systems to identify patients who would benefit the most.
Objectives: The specific aims of the project were to (1) create a classification scheme to characterize existing case management programs (CMPs) and support future comparative effectiveness studies; (2) examine different approaches to assessing the relationship of 3 selected CMPs in preventing the composite outcome of hospital events (inpatient admissions and ED visits) for Medicare ACO patients; (3) evaluate alternative approaches to CMP staffing and caseloads within a single site; and (4) develop a strategy to identify patients who are likely to benefit from case management in 1 health system and test this strategy in a second health system.
Methods: For aim 1, we characterized CMPs across 20 health systems in 2 National Patient-Centered Clinical Research Network (PCORnet) clinical data research networks (CDRNs) using a mixed-methods design that included semistructured interviews of patients, health system leaders, and program directors at 2 sites, followed by a comprehensive survey of program directors that cataloged the elements of each program.
For aim 2, we examined the effectiveness of 3 existing CMPs compared with usual care in preventing hospital events for high-risk Medicare ACO patients in 3 academic health systems in the Midwest and Northeast; the 2 Midwest sites included both urban and rural patient populations. We used a retrospective cohort study design with multiple before-and-after measures of the outcome for each case management patient and their matched comparisons. The final sample size of case management patients meeting the inclusion criteria was 1935, which comprised site A (n = 139), site B (n = 718), and site C (n = 1078). We matched each case management patient to ∼2 of the closest eligible comparison patients using a propensity score constructed from baseline measures. To assess the importance of health services use outside the ACO in the matching process, we matched to comparison patients using 2 different approaches based on either electronic health record (EHR) data alone or EHR data combined with Medicare claims data. The final sample size of comparison patients meeting these criteria in the EHR-matched cohort was 3833, which comprised site A (n = 275), site B (n = 1417), and site C (n = 2141). The final sample size of comparison patients meeting these criteria in the EHR/claims-matched cohort was 1336, which comprised site A (n = 226) and site B (n = 1140). Our outcome measure was a count of event-days from the Medicare claims, defined as the number of days spent in the hospital or at an ED visit.
For aim 3, an in-depth mixed-methods evaluation of a single site was used to understand details of the variations in staffing and caseloads, using surveys of patients and providers, and conducting semistructured interviews of nurse and social work case managers to assess perceptions of the program as well as variations.
For aim 4, because targeting of patients is a critical component of successful case management interventions, we tested a strategy to identify patients who might benefit from enrollment in a CMP through the use of a benefit score. Specifically, we examined whether a score developed at 1 site (site X) to identify patients for enrollment in case management could be applied to a second site (site Y). We linked Medicare claims data to participation in case management for the 2 sites. We used the same inclusion criteria, matching the approach, outcome, and analytic approach for aim 2 of the study. We identified case management patients meeting these criteria at site X (n = 469) and site Y (n = 1483) and comparison patients meeting these criteria at site X (n = 1578) and site Y (n = 4788).
Results: For aim 1, of the 20 eligible sites contacted, 18 responded; of these, 3 sites did not have CMPs. Most programs provided a range of services. All had a registered nurse (RN) or advanced practice RN on staff as case managers, and most had an SW. High variability was seen in the number of cases served by the program and the caseloads per case manager. Most programs used cost and use data to identify patients and track outcomes. The widely varied case or care management program processes and structures made classification challenging.
For aim 2, across our 3 sites with linked data, case management patients tended to be female, be White, and have slightly different mean ages (69 years at site A, 71 years at site B, and 74 years at site C). Without matching on claims-based health care use, the CMP at 2 of 3 health systems was associated with fewer hospital admissions and ED visits over the subsequent 12 months. After matching on claims-based health care use, case management was no longer associated with the admissions and ED visit volumes at those 2 programs.
For aim 3, from our in-depth review of a single site, patients with SW team leads had more contacts with the program and were nearly twice as likely to be asked about social needs as were patients in practices where case management was nurse led only (59% vs 35%, respectively). Panel sizes differed substantially; panels where the patients had an RN team lead only had an average panel size of 126.5 patients, while panels where any patient had an social work team lead had an average panel size of 11.8 patients. Physicians saw a positive impact when an SW was part of the team, although not all team members were aware of case management being done by SWs.
For aim 4, our identification strategy was successful in identifying patients who would benefit from case management to prevent high-cost events when applied to the site where the strategy was developed (site X, 443 event-days prevented; 95% CI, 283-553 event-days) but was not successful in identifying patients who would benefit at a second site. The patients at the development site were sicker, were more likely to be enrolled in Medicaid, and had higher rates of chronic conditions than those at the second site, and the program at the development site had a more intensive inclusion of social work.
Conclusions: We found wide variation in CMP characteristics, with differing levels of panel size, duration, intensity, and integration, as well as a wide range of services. We also determined that the rapid evaluation of programs for high-need, high-cost patients required the use of claims data to match cases to comparison patients. Social work was identified as a valuable, but underused, resource from which patients might benefit, but our strategy to identify specific patients who might benefit from enrollment in case management did not identify such patients at a second site.
Limitations: We cannot be certain about the generalizability of our survey findings outside the 2 participating CDRNs. We only followed outcomes for 1 year after enrollment in case management. Our sample of CMPs was limited to large health systems affiliated with academic medical centers. The moderate number of SWs in our samples may limit comparisons. Both self-report and administrative health care data have limitations. There may still be unobserved patient and program characteristics that explain our findings.
Copyright © 2021. University of Wisconsin-Madison. All Rights Reserved.