Longitudinal, Relationship-Based Case Management: A Prospective Cohort Trial

ABSTRACT

Objectives: This study addresses the challenge of improving outcomes for the 5% of individuals with complex chronic diseases who utilize 50% of health care resources. Previous interventions targeting this population have shown limited impact, often due to transactional and time-limited approaches. This study proposes a longitudinal, relationship-based case management framework as an alternative solution.

Study Design: A nonrandomized, prospective cohort study was conducted among Medicaid enrollees with complex medical and social needs.

Methods: The intervention involved case managers building strong interpersonal relationships over a minimum of 1 year, addressing barriers to care and facilitating solutions. Primary outcomes were total health care expenditures and patient-rated quality of life.

Results: The intervention group exhibited a significant reduction in total health care costs over 1 year ($8568 per patient), with greater savings observed for patients with higher preintervention costs. Additionally, an estimated annual savings net of program costs of $248,121 was observed. Patient-rated quality of life showed substantial improvement, evident at both 6 months and 1 year post enrollment.

Conclusions: This study demonstrates the effectiveness of a longitudinal, relationship-based case management approach in improving outcomes for individuals with complex medical, social, and behavioral needs. Unlike transactional interventions, this approach emphasizes partnership and customization, yielding substantial cost reductions and enhanced quality of life. Although limitations exist, including nonrandomization and staff diversity, this study provides a foundation for future research and scalability of similar interventions.

Am J Manag Care. 2025;31(5):In Press

Takeaway Points

Implementing a relationship-based case management approach for patients with complex health needs can lead to significant cost savings and improved quality of life.

• Establishing long-term relationships with case managers can reduce health care costs, especially for patients with high preintervention expenses.
• Patients experience better health and satisfaction when personalized care addresses their medical and social needs.
• This approach promotes lasting improvements, unlike short-term, transactional interventions.
• Adopting such a framework can enhance patient well-being and optimize health care spending, making it a strategic investment for organizations.

To improve value, significant energy has been focused on the subset of individuals with complex chronic disease who have the worst outcomes and utilize the most care—the 5% of the population who utilize 50% of resources.1 This population, which disproportionately struggles to overcome social influencers that both negatively impact health and interfere with engagement,2-4 has proven difficult to impact.Most notably, a randomized controlled trial of Medicaid enrollees with complex lives and high costs assessed the effectiveness of the Camden Coalition case management intervention, which sought to solve barriers to care but failed to show any significant difference in cost between the control and intervention groups.5 This and similar efforts have led many to write this population off as “unimpactable” and suggest focusing resources on individuals with fewer barriers to care instead.6-8 We hypothesize that efforts to improve outcomes for individuals with the most complex lives failed to demonstrate an impact because they were time limited and transactional rather than longitudinal and relationship based. This study contrasts transactional care (task oriented, focuses on resolving a specific need) with relationship-based care (focuses first on developing trust, respect, understanding, and other interpersonal dynamics before focusing on solving specific problems).

We have previously published a proposed framework for longitudinal, relationship-based case management that prioritizes the development of relationships as the initial, primary intervention for individuals with complex lives because trust and partnership are prerequisites to solving complex problems.9 To test our hypothesis, we developed and implemented a mixed-methods study, utilizing a nonrandomized, prospective cohort protocol to assess the intervention impact on expenditures and a quasi-experimental preest/posttest methodology to assess the intervention impact on quality of life for high-cost individuals enrolled in Medicaid.

METHODS

Trial Design and Context

The trial was approved by the University Hospitals Cleveland Medical Center’s Human Subjects Institutional Review Board and funded by the Robert Wood Johnson Foundation Achieving Health Equity for Medicaid Individuals grant. The COVID-19 pandemic presented significant challenges, causing delays and necessitating a rolling enrollment study design. Efforts to recruit a racially diverse team of case managers were only partially successful, potentially impacting the intervention’s effectiveness for racial minority groups.

Ethical Declaration

The study protocol was approved by the institutional review board, ensuring ethical standards and participant protection. Informed consent was obtained, and confidentiality was maintained.

Eligibility

Enrollment was limited to adults 18 years and older in greater Cleveland, Ohio, who were enrolled in Medicaid with a behavioral health diagnosis, were in the top quartile of medical expenditures, and had at least 2 annual unplanned care instances within 1 year prior to identification. Exclusions included cognitive disorders, active cancer driving utilization, or use of existing case management services.

Recruitment for Intervention and Comparison Group

Eligible patients were identified through 2 primary methods:

1. Internal screening and outreach: Internal claims data and diagnostic information were used to identify patients who met the eligibility criteria. These prescreened patients were then contacted via phone or through their primary care providers for potential enrollment.
2. Referrals: Referrals were received from health care providers, Medicaid payer partners, and the psychiatry intake team. Referred patients were subsequently screened to ensure they met the same eligibility criteria as those identified through internal screening.

Enrollment and follow-up data collection occurred from July 2021 to February 2023.

Controls were selected from patients with qualifying primary care or emergency department (ED) visits during the same enrollment period as the intervention group and were followed for 1 year, ensuring a consistent time frame for outcome comparisons. Control data were retrospectively collected, and participants were risk matched to the intervention group using University Hospitals’ (UH) Population Health Risk Score, a proprietary model developed to predict both clinical and financial risk. This model, tailored for the UH Accountable Care Organization (ACO), integrates a comprehensive comorbidity index (based on CMS’ Chronic Conditions Data Warehouse and an expanded Charlson Comorbidity Index) and risk-adjusted expenditures from the previous 12 months. Patients are categorized into 6 risk levels, enabling targeted resource allocation for high-intensity care among the top-risk tiers, which represent a disproportionate share of health care spending. The model’s development and application are detailed in a recent study10; however, specific performance metrics and external validation were not available.

Intervention

The primary goal of the case managers was to improve adherence to planned treatments (eg, scheduled primary care visits, specialist appointments, or elective procedures) and reduce unplanned care (eg, ED visits and unscheduled admissions). This involved identifying and addressing both internal barriers (eg, low motivation, treatment-interfering behaviors) and external barriers (eg, social impediments to care, poor care pathway design) and then collaborating with patients to create solutions. Case managers employed strategies including building therapeutic engagement (motivational interviewing11 and acceptance and commitment therapy12), facilitating referrals to community resources, managing appointments, attending appointments or admissions, facilitating patient-provider communication, arranging transportation, and serving as advocates for the patients’ needs at all points of the patient journey.

Case managers met with patients weekly to build trust and rapport. Each manager was assigned a maximum caseload of 30 patients to ensure personalized attention. Meetings were flexible, taking place in patients’ homes, medical facilities, public areas, or via phone. The team included 2 nurses and 3 social workers.

Personnel

This program was staffed by a team of 5 clinical case managers, comprising 2 nurses and 3 licensed social workers. The majority of patients were assigned to a single case manager, whereas a select few with more complex needs were supported by 2 case managers. In addition to the case managers, a dedicated community health worker was also part of the program. Their role centered on facilitating connections between patients and vital community resources, thereby enhancing the program’s outreach and support capabilities. Lastly, a program manager provided oversight of administrative operations and ensured smooth implementation of the program.

Outcomes

The 2 primary outcomes were total health care expenditure (all billed services only) and patient-rated quality of life. For total health care expenditure, a baseline monthly medical expenditure was determined for each patient in both the control and intervention groups by calculating the mean monthly spend over the previous 2-year period. After 1 year of participation in the case management program, the postenrollment monthly medical expenditure was calculated using all medical claims within the 1-year period. All patients enrolled in this study were members of our ACO, and as such, all Medicaid billing was captured across all systems.

Quality of life was assessed using the Recovering Quality of Life 20 questionnaire (ReQoL-20). ReQoL-20 is a validated, self-reported measure designed to evaluate quality of life specifically for individuals with mental health conditions.13,14 The questionnaire consists of items covering various domains, including meaningful activity, belonging, autonomy, hope, self-perception, well-being, and physical health. Research assistants administered the questionnaire either in person or by phone at 3 points: at enrollment, 6 months after enrollment, and 1 year after enrollment. The maximum score on the ReQoL-20 is 84, representing the highest level of quality of life. To facilitate visual interpretation, domain scores were standardized into percentages. The ReQoL-20 was not administered to the control group.

Secondary outcomes included both planned and unplanned monthly health care expenditures. Unplanned expenditures included claims associated with ED visits and inpatient hospital admissions, and planned spend comprised the remaining claims data.

Statistical Analysis

The adjusted mean difference in monthly health care expenditures (post intervention) across the treated and control participants was estimated by way of ordinary least squares regression using Stata 16.1 (StataCorp LLC). SEs were corrected for clustering within pairs as the econometrics literature recommends.15 Our preferred specification controlled for treatment status, mean monthly spending prior to intervention, age and age squared, an indicator for female sex, and an indicator for non-White race/ethnicity (Black or Hispanic). Auxiliary models tested for differential effects by preintervention spending levels and differentiated between planned and unplanned spending.

RESULTS

Sample Characteristics

The treatment population was younger (mean age, 42.6 vs 45.8 years; P = .126), more likely to be female (71.3% vs 63.2%; P = .261), and more likely to be Black or Hispanic (41.4% vs 21.8%; P = .005) (Table, panel A). Mean monthly expenditures prior to the intervention were similar in the treated and control samples ($3162 vs $3046, respectively; P = .477) (Table, panel B). In the prior 2 years, planned expenditures were slightly higher in the treatment population and unplanned expenditures were slightly lower, but neither difference was statistically significant. Figure 1 depicts the difference in mean monthly spending partitioned into quarters. Our analysis reveals a nonsignificant linear trend over the preintervention period, with an ordinary least squares coefficient of 16.1 (P = .68). This detailed examination demonstrates that preintervention spending trends do not significantly differ from zero, suggesting stability before the intervention.

Program Implementation

On average, patients had regular contact with their assigned case manager approximately 4.2 days per month (SD, 1.6), totaling a mean (SD) of 2.6 (1.4) hours per month, or 50.4 days and 31.2 hours a year of direct contact between each patient and their case manager. This time estimate does not include time case managers spent outside of direct contact that still benefits the patient’s care (eg, transportation time, coordination with specialists, contacting insurance representatives).

Program Implementation and Costs

The program was scaled for a capacity of 150 patients, with estimated annual operational costs of $427,786. These costs consisted of the salary and benefits for 3 licensed social workers, 2 nurse case managers, a community health worker, and a program manager; a patient health transportation service through Roundtrip; separately contracted case management supervision; acceptance and commitment therapy training; and all overhead costs.

The program was initiated in July 2021, with enrollments starting that month. Maximum capacity (150 patients) was achieved in July 2022. Total enrollments climbed steadily over the ramp-up period, averaging 12.43 per month. The first 87 enrolled patients comprised our sample of treated patients.

Monthly Spending Post Intervention

Panel C of the Table reports summary statistics on the spending outcome: mean monthly spending post intervention. Mean monthly postintervention spending was significantly higher in the control group vs the treatment group ($3382 vs $2710; P < .001). The spending difference primarily reflects higher unplanned spending in the control group ($1671 vs $1125; P < .001). The control group also had higher planned spending in the postintervention period, but this difference was smaller and not statistically significant ($1711 vs $1585; P = .273). The difference in mean monthly spending grew increasingly negative over the 4 postintervention quarters (Figure 1).

eAppendix Figure 1 (eAppendix available at ajmc.com) displays the preintervention and postintervention monthly spending levels of each participant, differentiated by treatment status. The distribution of prior spending levels for the matched samples largely overlap. Fitted linear regression lines show a positive association between preintervention and postintervention spending levels. This association reveals a steeper slope in the control group.

Effects of the Intervention

eAppendix Table 1 presents estimated effects of the intervention on monthly postintervention spending. Controlling only for preintervention spending, treatment was predictive of a $733 reduction in monthly expenditures (column 1), an estimate 9% larger than the unadjusted difference ($672) (see Table, panel C). Age, female sex, and non-White racial identification were jointly significant predictors of postintervention spending levels (P = .067); however, their inclusion as controls had little effect on the estimated reduction in monthly expenditures ($733 to $714 in column 2 of eAppendix Table 1). In column 3, we find evidence of significant increases in the treatment effect for participants with higher preintervention spending levels. A $100 increase in preintervention spending was associated with a $65 increase in postintervention spending in the control group but only a $38 increase in the treated group.

Columns 4 and 5 of eAppendix Table 1 indicate these estimates are robust to specifications controlling for the distinct types of preintervention spending (planned and unplanned). Planned preintervention spending was a stronger predictor of postintervention spending than unplanned preintervention spending and also appeared to more strongly mediate the treatment effect, but none of the pertinent differences were statistically significant.

eAppendix Table 2 reports estimates of the treatment effect on planned and unplanned postintervention spending, treated as distinct outcomes. Results in columns 1 and 3 indicate treatment had a stronger effect on unplanned spending ($502) than planned spending ($224). These results also suggest that the predictiveness of preintervention spending was specific to the type of spending. Planned preintervention expenditures strongly predicted planned postintervention expenditures but were not predictive of unplanned postintervention expenditures (and vice versa). Planned (but not unplanned) preintervention spending also appeared to mediate the effect on planned spending (column 2). In contrast, the treatment effect on unplanned spending appeared to be similarly mediated by both types of preintervention spending.

In eAppendix Table 3, we replicate the results of eAppendix Table 1 using log-transformed values of all spending variables. The implied magnitudes and significance of key results are essentially unchanged. However, the fit is somewhat poorer.

Figure 2 summarizes the results of the ReQoL-20. The quality of life of the treatment group was assessed at 3 distinct time points to evaluate the effectiveness of the intervention. We employed a repeated measures analysis of variance to identify significant changes over time. The results revealed a significant difference in ReQoL-20 scores across all 3 time points (F86,2 = 104.73; P < .001). Subsequent post hoc comparisons, utilizing the t test with Bonferroni correction, demonstrated that mean (SD) ReQoL-20 scores at baseline (34.63 [176.88]) were significantly lower than scores at 6 months (52.89 [165.74]). Moreover, ReQoL-20 scores at 1-year follow-up (57.21 [94.58]) were significantly higher than both baseline and 6-month scores. Notably, although the most substantial improvement in ReQoL-20 scores was observed between baseline and 6 months, the improvement continued to persist beyond 6 months to the 1-year mark.

Accumulation of Net Program Savings Over Time

eAppendix Figure 2 simulates the accumulation of program costs and aggregate savings for a hypothetical program patterned after ours. Program costs were assumed to accrue at a constant rate of $35,650 per month. Patient enrollments accrued at a rate of 12.43 per month until a capacity of 150 was reached. Monthly savings were based on the estimated intervention effect, multiplied by the count of enrolled patients in that month. In version A, we assumed that a constant savings of $703.7 accrued per patient-month (see eAppendix Table 1, column 2). In version B, we employed estimates from an alternative specification allowing the savings effect to evolve by quarter of patient participation time (see eAppendix Figure 2 notes for details). All monthly cost and savings values were discounted assuming a monthly discount rate of 0.247% (3% annual).

During the initial months of operation, program costs accrued at a faster pace than predicted savings, but this reversed as enrollments increased. Cumulative savings overtook cumulative program costs in month 8 when we employed the homogeneous treatment effect estimate (version A) but did not do so until month 12 when quarterly effect estimates were used (version B). By month 18, the 2 methods produced similar estimates of cumulative savings ($1.28 million vs $1.23 million, respectively), approximately twice the accumulated program costs ($628,400). Additional simulation results can be found in eAppendix Tables 4 and 5.

DISCUSSION

In this nonrandomized trial of 174 participants, longitudinal case management significantly reduced health care costs over 1 year, saving $8568 per patient (program cost: $2852). The cost reduction was proportional to patients’ prior health care spending, indicating greater savings for higher-cost individuals. Although the program initially incurred a deficit, it ultimately achieved a net savings of $248,121, suggesting long-term financial benefits. Quality of life also improved significantly in the intervention group, although no comparison was made with the control group.

Like researchers of other studies, including the Camden Coalition study, we targeted individuals on Medicaid with complex medical and social needs. Our longitudinal approach built relationships over time, contrasting with the Camden Coalition study’s transactional focus on immediate postdischarge needs. Transactional programs often target easily measurable metrics such as rapidity of engagement and time to first physician appointment after discharge. This is for good reason: Ample data suggest that these interventions do impact outcomes across large populations.16-19 Yet for the individuals with complex medical, social, and behavioral needs, these metrics may be unrealistic to achieve and irrelevant to long-term outcomes because the reason for admission itself is not easily reducible to a few distinctive problems. In particular, our median total invested time of 31.2 hours and 50 unique days over a minimum of 1 full year of enrollment contrasts with the Camden Coalition study, which had a median of 7.6 home visits and 8.8 calls across a median intervention time of 91.5 days.

Limitations

Several limitations exist. Unlike the Camden Coalition study, our nonrandomized design introduces potential bias, although the large effect size makes it unlikely that bias alone accounts for the results. We did not collect quality of life data from the control group, limiting our ability to assess the full impact of the intervention on this outcome. Additionally, less than half (47%) of referred patients were successfully engaged, which may have introduced selection bias because those who were engaged may have some underlying characteristic that motivated their outcomes differently from those who were not engaged. Future studies should match groups more comprehensively across demographic and socioeconomic variables to strengthen internal validity.

Recruitment through referrals may have further biased outcomes, as referred patients might differ from randomly selected controls in ways we could not fully measure. Although we adjusted for baseline characteristics, unmeasured differences may remain. Future studies could address this limitation through randomized designs or more rigorous matching.

Ethical concerns also influenced our choice to forgo a randomized controlled trial. Building trust with participants was crucial, and assigning them to a control group could have undermined that trust. Despite this limitation, our outreach led to a 100% enrollment rate among those we engaged, highlighting the success of our approach in reaching a traditionally hard-to-engage population.

CONCLUSIONS

Despite these limitations, we see several exciting avenues for future research, including enrolling and studying more patients, at larger scale, and utilizing a randomized trial design that randomizes control participants to another standard of care (perhaps traditional case management). Additionally, because this intervention is not scalable with licensed social workers and nurses alone, expanding research to study the impact of this work utilizing case managers who have less intensive training—such as peers or community health workers—will be important. Furthermore, the impact of case manager–patient racial concordance on outcomes may also be of future interest.

This study provides strong evidence that longitudinal, relationship-based case management can benefit individuals with high medical, social, and behavioral complexities and serves as a template for future interventions. 

Author Affiliations: University Hospitals Health System (PR, RM, PJP, JP), Cleveland, OH; Case Western Reserve University School of Medicine (PR, PJP, AA), Cleveland, OH; Case Western Reserve University Weatherhead School of Management (MV), Cleveland, OH; Case Western Reserve University Frances Payne Bolton School of Nursing (JP), Cleveland, OH.

Source of Funding: Robert Wood Johnson Foundation Clinical Transformation Grant Award for Achieving Health Equity for Medicaid Individuals.

Author Disclosures: Dr Pronovost reports receipt of a Robert Wood Johnson Foundation grant to support the work in this paper. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design (PR, RM, PJP, JP); acquisition of data (PR, RM); analysis and interpretation of data (PR, RM, MV, PJP, AA); drafting of the manuscript (PR, RM, MV, PJP, AA); critical revision of the manuscript for important intellectual content (PR, RM, MV, PJP, AA); statistical analysis (RM, MV, PJP); provision of patients or study materials (RM); obtaining funding (PR); administrative, technical, or logistic support (RM, JP); and supervision (PR, PJP, JP).

Address Correspondence to: Ryan Muskin, MA, University Hospitals Health System, 11100 Euclid Ave, Cleveland, OH 44106. Email: ryan.muskin@uhhospitals.org.

REFERENCES

1. Cohen SB. The concentration of health care expenditures and related expenses for costly medical conditions, 2012. Agency for Healthcare Research and Quality statistical brief 455. October 2014. Accessed July 17, 2023. https://www.meps.ahrq.gov/data_files/publications/st455/stat455.pdf

2. The World Health Organization on health inequality, inequity, and social determinants of health. Popul Dev Rev. 2007;33(4):839-843.

3. Wilkinson R, Marmot M, eds. Social Determinants of Health: The Solid Facts. 2nd ed. World Health Organization; 2003.

4. Marmot M. Social determinants of health inequalities. Lancet. 2005;365(9464):1099-1104. doi:10.1016/S0140-6736(05)71146-6

5. Finkelstein A, Zhou A, Taubman S, Doyle J. Health care hotspotting — a randomized, controlled trial. N Engl J Med. 2020;382(2):152-162. doi:10.1056/NEJMsa1906848

6. Runnels P, Pronovost PJ. Labeling complex and costly patients as “unimpactable”: a morally questionable practice likely to worsen outcomes. Popul Health Manag. 2022;25(4):571-572. doi:10.1089/pop.2021.0208

7. DuBard CA, Jackson CT. Active redesign of a Medicaid care management strategy for greater return on investment: predicting impactability. Popul Health Manag. 2018;21(2):102-109. doi:10.1089/pop.2017.0122

8. Baah FO, Teitelman AM, Riegel B. Marginalization: conceptualizing patient vulnerabilities in the framework of social determinants of health—an integrative review. Nurs Inq. 2019;26(1):e12268. doi:10.1111/nin.12268

9. Runnels P, Muskin R, Ansah A, Penman J, Pronovost P. A longitudinal, relationship-based model for managing complex chronic disease in the Medicaid population. Popul Health Manag. 2022;25(4):535-541. doi:10.1089/pop.2022.0013

10. Coran JJ, Schario ME, Pronovost PJ. Stratifying for value: an updated population health risk stratification approach. Popul Health Manag. 2022;25(1):91-99. doi:10.1089/pop.2021.0096

11. Baker A, Hambridge J. Motivational interviewing: enhancing engagement in treatment for mental health problems. Behav Change. 2002;19(3):138-145. doi:10.1375/bech.19.3.138

12. Hayes SC, Luoma JB, Bond FW, Masuda A, Lillis J. Acceptance and commitment therapy: model, processes and outcomes. Behav Res Ther. 2006;44(1):1-25. doi:10.1016/j.brat.2005.06.006

13. Keetharuth AD, Brazier J, Connell J, et al. Recovering Quality of Life (ReQoL): a new generic self-reported outcome measure for use with people experiencing mental health difficulties. Br J Psychiatry. 2018;212(1):42-49. doi:10.1192/bjp.2017.10

14. Keetharuth AD, Bjorner JB, Barkham M, Browne J, Croudace T, Brazier J. An item response theory analysis of an item pool for the recovering quality of life (ReQoL) measure. Qual Life Res. 2021;30(1):267-276. doi:10.1007/s11136-020-02622-2

15. de Chaisemartin C, Ramirez-Cuellar J. At what level should one cluster standard errors in paired experiments, and in stratified experiments with small strata? Am Econ J Appl Econ. 2024;16(1):193-212. doi:10.1257/app.20210252

16. DuBard CA, Jackson CT. Active redesign of a Medicaid care management strategy for greater return on investment: predicting impactability. Popul Health Manag. 2018;21(2):102-109. doi:10.1089/pop.2017.0122

17. Stamp KD, Machado MA, Allen NA. Transitional care programs improve outcomes for heart failure patients: an integrative review. J Cardiovasc Nurs. 2014;29(2):140-154. doi:10.1097/JCN.0b013e31827db560

18. Wee SL, Loke CK, Liang C, Ganesan G, Wong LM, Cheah J. Effectiveness of a national transitional care program in reducing acute care use. J Am Geriatr Soc. 2014;62(4):747-753. doi:10.1111/jgs.12750

19. Hirschman KB, Shaid E, McCauley K, Pauly MV, Naylor MD. Continuity of care: the transitional care model. Online J Issues Nurs. 2015;20(3):1.