A community-based care management program in Rhode Island reduced hospitalizations and inpatient costs. There were stark differences across subgroups based on intensity of care.
Objectives: A variety of care coordination and delivery models have been used to address the social and medical needs of high-need, high-cost patient populations. However, the evidence on the effectiveness of such models is far from clear. The purpose of this study is to determine whether the Community Health Team (CHT) program, a community-based care management program in Rhode Island, had impacts on health care utilization and cost.
Study Design: We used data from 2014 to 2018 to evaluate the effects of the CHT program on health care utilization and cost. Our analytical sample consisted of a total of 12,830 patients, with 2282 in the intervention group and 10,548 in the matched comparison group.
Methods: We used a combination of propensity score–matched difference-in-differences framework and generalized linear models.
Results: The program led to an overall decrease in hospitalizations (incidence rate ratio [IRR], 0.89; P = .028) and inpatient costs (IRR, 0.79; P = .024). This translates into a reduction of 7 hospitalizations per 1000 people per month and a reduction of inpatient cost of $289 per person per month. Impacts varied considerably across subgroups. For patients with 1 to 2 encounters with the program, there was a significant decrease in emergency department visits, hospitalizations, inpatient cost, outpatient cost, professional cost, and total cost. Although no significant impacts were observed for patients with 3 to 5 encounters with the program, patients with more than 6 encounters with the program saw an increase in pharmacy cost and total cost.
Conclusions: There is a need for a tailored approach to addressing patients’ needs in primary care.
Am J Manag Care. 2022;28(4):187-191. https://doi.org/10.37765/ajmc.2022.88862
A community-based care management program targeting high-risk patients in Rhode Island reduced hospitalizations and inpatient costs.
The top 5% of utilizers account for approximately 50% of overall health care spending in the United States.1 High utilizers often have unaddressed, complex health needs that result in frequent hospital use. In attempts to improve care quality and reduce costs for high utilizers, payers across all markets, including Medicare, Medicaid, and commercial insurers, have been experimenting with various care coordination models.2 One particular model, advocated by the Camden Coalition of Healthcare Providers, caught national attention.3 In recognizing not only the medical needs but also the social needs of high utilizers, the Camden model brings together an interdisciplinary team, often a combination of nurses, community health workers, and behavioral specialists. Many care organizations across the country have adopted variants of the Camden model.
Despite the popularity of these care coordination models that address both medical and social needs of high utilizers, the evidence on their effectiveness is far from clear. Although some evaluations of these models reported significant improvement in quality of care and reduction in utilization and spending among participants, others were inconclusive.2,4-7 Such evaluations are often limited by small sample sizes or lack of an appropriate control group. Without a comparable control group, important factors that could confound the study results are unaddressed because the participants in these models are often selected based on their recent health services utilization. Even in the absence of intervention, some reduction in utilization is to be expected as severity of their condition wanes. Additionally, many of these programs focused on Medicare beneficiaries, who may have different medical and social needs from those not covered by Medicare.6
In this study, we evaluated the impacts of the Community Health Team (CHT) program, which is a community-based care management program implemented at the Thundermist Health Center, a federally qualified health center in Rhode Island. Our study has several advantages over previous studies. First, the CHT program did not restrict patients to a specific population, such as those with a specific coverage type or chronic condition. This allowed us to examine outcomes in a more general population whereas previous studies could not.
Second, we used the state’s all-payers claims database (APCD) for our analysis, which contains deidentified enrollment files and health care claims from Medicare, Medicaid, and the 9 largest commercial health insurers in the state, enabling us to examine utilization and cost patterns even when patients switched between insurers. Finally, our study sample included 2282 program patients, providing us with a much larger sample size than similar evaluations.
Using a propensity score–matched difference-in-differences (DID) framework,8 we evaluated the impact of the CHT program on health care utilization and cost across 7 outcome measures.
The Thundermist CHT Program
The Thundermist Health Center provides care to more than 51,000 patients annually.9 The Thundermist CHT program consists of a multidisciplinary team of nurses, community health workers, behavioral health care managers, and administrative and management support staff. The program serves as an extension of primary care at the center, offering social and behavioral health support, care coordination with health care providers, chronic disease management, and follow-up care after emergency department (ED) visits, hospital admissions, or transitions from skilled nursing facilities. It also provides nonmedical support, such as applications to social services programs (eg, housing, transportation, financial resources) and assistance with entitlement programs such as Medicaid and Social Security disability benefits.10 Details on the CHT program’s patient identification process and the types of services offered are provided in part H of the eAppendix (available at ajmc.com).
Between August 2014 and December 2018, the CHT enrolled 3393 program patients and made 8.8 face-to-face encounters with each patient on average. Patients remained in the program ideally until their health and social conditions stabilized, at which point the CHT could graduate them from the program/intervention.
Data and Study Population
We used Rhode Island’s APCD data from 2014 to 2018. The APCD contains deidentified data for nearly 85% of the state population.11 Our analysis included 2282 of 3393 CHT patients for whom we were able to obtain enrollment and claims records in the APCD.
To identify potential comparison individuals from the APCD population, we used a propensity score matching (PSM) technique, which estimated a propensity score based on a bivariate logit model. The propensity score was then used to create a matched comparison group with observable baseline characteristics similar to those of the intervention group. Matching was done at the individual level using one-to-many matching (1:5) without replacement. Matching variables included a combination of demographic characteristics, CHT’s selection criteria, and comorbidities. Additional details on PSM are provided in part B of the eAppendix.
We selected a total of 10,548 patients in the comparison group for the overall study sample. For each comparison group patient, a pseudoenrollment date was created to indicate the date that the patient would have enrolled in the CHT program had they been in the intervention group.
A propensity score–matched DID framework12 was used to estimate the difference in outcomes between CHT patients and comparison group patients, 12 months before and 12 months after the program start date. Two sets of outcomes were analyzed: utilization and cost. Utilization outcomes were ED visit rate and hospitalization, and the cost outcomes were inpatient cost, outpatient cost, professional cost, pharmacy cost, and total cost. The methodology used to categorize health services into these outcomes was based on the 2017 Health Care Cost and Utilization Report, published by the Health Care Cost Institute.13 Although both sets of outcomes were modeled using generalized linear models, utilization outcomes were modeled using negative binomial distribution with log link,14-16 whereas cost outcomes were modeled using gamma distribution with log link.15-17 Estimates were presented as relative effects in the form of incidence rate ratios (IRRs). An IRR less than 1 indicated that the CHT program was associated with a decrease in the outcome, whereas an IRR greater than 1 indicated that CHT was associated with an increase in the outcome. All matched models fulfilled the parallel trends assumption. The unit of analysis was the person-month, and covariates included demographic characteristics and comorbidities, number of face-to-face encounters between patients and the CHT, and month fixed effects. Robust standard errors were used to correct for heteroscedasticity at the person level.18 See eAppendix part A for details on parallel trends and eAppendix part D for details on the regression specification.
Because we observed large variation in the number of face-to-face encounters between the patients and the CHT program, we divided the patients into 3 subgroups: low, medium, and high. The low subgroup included patients with 1 or 2 encounters (0-50th percentile in terms of the number of encounters), the medium subgroup included patients with 3 to 6 encounters (50th-75th percentile), and the high subgroup included patients with more than 6 encounters (> 75th percentile). We did subgroup-level matching and reanalyzed the same model for each subgroup.
To test whether the program had differential impact among patients who participated in the program in 2017 or 2018 (late participants) vs those who participated in the program in 2014, 2015, or 2016 (early participants), we looked at the 3-way interaction among the intervention, the post period, and the late indicators.
Table 1 displays the characteristics of the CHT patients (ie, the intervention group) and the patients in the matched comparison group. The study sample consisted of 2282 CHT patients in the intervention group and 10,548 patients in the matched comparison group. As shown in Table 1, the intervention group and the matched comparison group had similar baseline characteristics. Subgroup-level baseline characteristics are shown in eAppendix Table 2 in part C of the eAppendix.
Table 2 displays the adjusted IRR estimates for the program impacts. In the overall sample, the program led to significant reductions in hospitalizations (IRR, 0.89; P = .028) and inpatient cost (IRR, 0.79; P = .024). This translates into a reduction of 7 hospitalizations per 1000 people per month and a reduction of inpatient cost amounting to $289 per person per month. Impacts varied considerably across patient subgroups. For patients in the low subgroup, there was a significant decrease in ED visits, hospitalizations, inpatient cost, outpatient cost, professional cost, and total cost. Although no statistically significant impacts were observed for patients in the medium subgroup, patients in the high subgroup saw an increase in pharmacy cost and total cost. See eAppendix Table 4 in part F of the eAppendix for details on marginal effects.
Additionally, we saw significant differential impacts between late participants and early participants. In the overall sample, late participants saw an increase in outpatient cost but a decrease in pharmacy cost. See eAppendix Table 10 in part G of the eAppendix.
Using APCD data, this study examined the impacts of the CHT program, a community-based care management program in Rhode Island, on health care utilization and cost. Overall, the program reduced hospitalizations and hospitalization-related inpatient costs. Reduction in hospitalizations could have occurred, for example, because the program helped patients with medication adherence by helping them devise and stick to a medication regimen. Similarly, by ensuring timely referrals to counselors and providers, including specialists, the program may have averted unnecessary hospitalizations. Reduction in hospitalizations likely translated into reduction in inpatient costs.
Patients in the low subgroup (ie, those with 1 or 2 encounters with the CHT program) saw a decrease in all outcomes analyzed, except pharmacy cost, which remained unchanged. Findings for this subgroup could suggest that health care cost and utilization associated with this subgroup were less complex and more quickly manageable and may be addressed effectively by an intervention, such as CHT, that caters to both medical and nonmedical needs. This is supported by eAppendix Table 2, which shows that participants in the low subgroup were less likely to have multiple chronic conditions or dual-eligible status.
The increase in pharmacy costs among patients with more than 6 encounters suggests that patients need longer-term monitoring and management of conditions. This subgroup may also be more likely to experience unstable housing, which can reduce access to prescription drugs and could worsen health outcomes.19 The increases in pharmacy costs may indicate that the patients in this subgroup were receiving the care they needed. Finally, the program’s differential impacts on outpatient costs and pharmacy costs for late participants suggest the varying effects of the program as it evolves and matures (see eAppendix Table 10).
Our study has limitations. First, because the intervention was tailored based on each patient’s needs, we could not identify the specific program attributes that resulted in the changes in our outcomes. Second, we did not know the patients’ final status in the program (ie, whether they had graduated or were still engaged by the end of our study period). Nevertheless, our data confirm that nearly 95% of the study participants were enrolled for at least 12 months. Third, the PSM technique is based on observable factors and does not account for unobserved confounders. Finally, the lack of data on health care utilization in the APCD database does not mean that uninsured people did not get medical care. These people are likely to receive uncompensated/charity care, and such care does not show up in the claims data. Additionally, evidence suggests that uninsured people are more likely to have poorer health outcomes.20 This means that the absence of uninsured people in our analysis would have likely biased the results toward the null.
Our study examined the impacts of a community-based care management program among high utilizers of health care. The study contributes to the growing evidence base on the effectiveness of care management programs to improve patient outcomes and reduce cost. The study showed that even within the group of high-risk (ie, high-need, high-cost) patients, there may still be large heterogeneities in the type and amount of care needed. This highlights the need for health care professionals to tailor their interventions to suit patient needs. Further, the study also highlights that health care may be an important consideration but not the only consideration in addressing health needs. Interventions such as CHT that go beyond health care in addressing health needs may be more suited for improving some patient outcomes.
The authors would like to thank Deepak Adhikari, MS, for his help and insights in preparing the analytical file. The authors also would like to thank the team of Care Transformation Collaborative of Rhode Island, in particular Debra Hurwitz, Pano Yeracaris, Linda Cabral, and Jazmine Mercado, for help and cooperation during this evaluation.
Author Affiliations: Department of Health Services, Policy, and Practice, School of Public Health, Brown University (BBT, XL, OG), Providence, RI; RTI International (XL), Washington, DC.
Source of Funding: Care Transformation Collaborative of Rhode Island.
Author Disclosures: The 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 (BBT, XL, OG); acquisition of data (XL, OG); analysis and interpretation of data (BBT, XL, OG); drafting of the manuscript (BBT, XL); critical revision of the manuscript for important intellectual content (BBT, XL, OG); statistical analysis (BBT, XL); provision of patients or study materials (OG); obtaining funding (OG); administrative, technical, or logistic support (OG); and supervision (OG).
Address Correspondence to: Omar Galárraga, PhD, Department of Health Services, Policy, and Practice, Brown University School of Public Health, 121 S Main St, Providence, RI 02912. Email: email@example.com.
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 No. 455. October 2014. Accessed June 10, 2020. http://www.meps.ahrq.gov/mepsweb/data_files/publications/st455/stat455.pdf
2. McCarthy D, Ryan J, Klein S. Models of care for high-need, high-cost patients: an evidence synthesis. Issue Brief (Commonw Fund). 2015;31:1-19.
3. Gawande A. The hot spotters. New Yorker. January 16, 2011. Accessed June 10, 2020. https://www.newyorker.com/magazine/2011/01/24/the-hot-spotters
4. Williams BC. Limited effects of care management for high utilizers on total healthcare costs. Am J Manag Care. 2015;21(4):e244-e246.
5. Hong CS, Siegel AL, Ferris TG. Caring for high-need, high-cost patients: what makes for a successful care management program? Issue Brief (Commonw Fund). 2014;19:1-19.
6. Bodenheimer T. Strategies to reduce costs and improve care for high-utilizing Medicaid patients: reflections on pioneering programs. Center for Health Care Strategies. October 2013. Accessed August 9, 2020. https://www.chcs.org/resource/strategies-to-reduce-costs-and-improve-care-for-high-utilizing-medicaid-patients-reflections-on-pioneering-programs/
7. Peterson GG, Zurovac J, Brown RS, et al. Testing the replicability of a successful care management program: results from a randomized trial and likely explanations for why impacts did not replicate. Health Serv Res. 2016;51(6):2115-2139. doi:10.1111/1475-6773.12595
8. Wing C, Simon K, Bello-Gomez RA. Designing difference in difference studies: best practices for public health policy research. Annu Rev Public Health. 2018;39:453-469. doi:10.1146/annurev-publhealth-040617-013507
9. 2019 annual report. Thundermist Health Center. August 2020. Accessed October 24, 2020. https://www.thundermisthealth.org/wp-content/uploads/2020/08/2019_THCAnnualReport.pdf
10. SIM community health team final evaluation report. Rhode Island Executive Office of Health and Human Services. August 2019. Accessed May 25, 2021. https://eohhs.ri.gov/sites/g/files/xkgbur226/files/Portals/0/Uploads/Documents/SIM/CommunityHealthTeamStateEvaluation-Final.pdf
11. HealthFacts RI Database. State of Rhode Island Department of Health. Accessed May 25, 2021. https://health.ri.gov/data/healthfactsri/
12. Stuart EA, Huskamp HA, Duckworth K, et al. Using propensity scores in difference-in-differences models to estimate the effects of a policy change. Heal Serv Outcomes Res Methodol. 2014;14(4):166-182. doi:10.1007/s10742-014-0123-z
13. 2017 health care cost and utilization report. Health Care Cost Institute. 2019. Accessed October 28, 2020. https://healthcostinstitute.org/annual-reports/2017-health-care-cost-and-utilization-report
14. Cole MB, Wright B, Wilson IB, Galárraga O, Trivedi AN. Medicaid expansion and community health centers: care quality and service use increased for rural patients. Health Aff (Millwood). 2018;37(6):900-907. doi:10.1377/hlthaff.2017.1542
15. Deb P, Norton EC. Modeling health care expenditures and use. Annu Rev Public Health. 2018;39:489-505. doi:10.1146/annurev-publhealth-040617-013517
16. Buntin MB, Zaslavsky AM. Too much ado about two-part models and transformation? comparing methods of modeling Medicare expenditures. J Health Econ. 2004;23(3):525-542. doi:10.1016/j.jhealeco.2003.10.005
17. Deb P, Norton EC, Manning WG. Health Econometrics Using Stata. Stata Press; 2017.
18. Daw JR, Hatfield LA. Matching and regression to the mean in difference-in-differences analysis. Health Serv Res. 2018;53(6):4138-4156. doi:10.1111/1475-6773.12993
19. Galárraga O, Rana A, Rahman M, et al. The effect of unstable housing on HIV treatment biomarkers: an instrumental variables approach. Soc Sci Med. 2018;214:70-82. doi:10.1016/j.socscimed.2018.07.051
20. McWilliams JM. Health consequences of uninsurance among adults in the United States: recent evidence and implications. Milbank Q. 2009;87(2):443-494. doi:10.1111/j.1468-0009.2009.00564.x