Multipayer Primary Care Transformation: Impact for Medicaid Managed Care Beneficiaries

The Michigan Primary Care Transformation project generated cost savings among adults in Medicaid managed care, particularly high-risk adults, while largely maintaining quality of care.

ABSTRACT

Objectives: To evaluate the effects of Michigan Primary Care Transformation (MiPCT), a statewide multipayer patient-centered medical home (PCMH) demonstration in 2012-2015, on cost, utilization, and quality among Medicaid managed care beneficiaries.

Study Design: Observational longitudinal study with comparison groups.

Methods: Difference-in-differences (DID) analyses compared changes in outcomes among beneficiaries whose primary care providers participated in MiPCT, non-MiPCT PCMH, and non-PCMH practices. Net cost savings were derived.

Results: The study included 173,179 MiPCT, 209,181 non-MiPCT PCMH, and 148,657 non-PCMH beneficiaries. Against 1 or both comparison groups relative to 2011, MiPCT adults had significant reductions in cost, emergency department (ED) visits, and hospitalization risk in 2015. Against both comparison groups, MiPCT high-risk adults showed significant cost reduction in 2014-2015, ED reduction in 2015, and reduced hospitalization risk in 2013-2015. For children, no significant relative change in cost occurred, but both ED and hospitalization risk were reduced in 2015. In 2013-2015, cumulative net cost savings were $15,569,526 (95% CI, $3,416,832-$27,722,219) (return on investment [ROI], $3.60) for adults and $23,998,180 (95% CI, $11,782,031-$36,214,347) (ROI, $10.69) for high-risk adults, and a cost increase of $16,517,948 (95% CI, $7,712,286-$25,323,609) (ROI, —$1.30) for children. Quality metrics were significantly higher in MiPCT in most years, although most DID estimates were not significant.

Conclusions: Evidence of cost savings exists among MiPCT Medicaid managed care adults; it was driven by high-risk adults, who also had reduced hospitalization risk. For children, no cost reductions occurred, but hospital and ED utilization were reduced in 2015. MiPCT maintained equal or higher quality of care but did not show consistent improvement.

Am J Manag Care. 2019;25(11):e349-e357Takeaway Points

Compared with similar nonparticipating groups and relative to baseline, Medicaid managed care beneficiaries assigned to physicians in practices participating in a large statewide multipayer patient-centered medical home demonstration exhibited:

  • Significant cost savings among adults, driven by savings among high-risk adults
  • Significantly reduced risk of hospitalization among high-risk adults
  • No cost savings and utilization reductions among children until the project’s fourth year
  • Better or equal quality of care, but no improvement over time

The patient-centered medical home (PCMH) has become the standard for primary care practices in the United States. Medicaid provides health insurance for populations who may benefit from PCMH.1 More than half of US states have implemented PCMH within Medicaid.2 Findings from one study indicated that 64% of Medicaid beneficiaries report having a usual source of care with some attributes of a PCMH.3 Results have been mixed.4-7 Iowa experienced reduced healthcare spending primarily due to reductions in emergency department (ED) utilization, whereas Louisiana and Alabama experienced no reduction in overall costs.4-6 Mixed results are similarly reflected in PCMH initiatives within commercial populations.8-10 Additionally, many Medicaid PCMH evaluations have identified greater utilization of services addressing unmet socioeconomic needs of beneficiaries, often through efforts of care managers within the practices.2,9

Medicaid PCMH initiatives have often been introduced through multipayer demonstrations, such as the Multi-Payer Advanced Primary Care Program (MAPCP), Comprehensive Primary Care initiatives (CPC and CPC+), and State Innovation Model initiatives.11 Few of these have been comprehensively evaluated for Medicaid outcomes.12-15

Several payers and health systems in Michigan were early adopters in the transformation of primary care. The Blue Cross Blue Shield of Michigan (BCBSM) Physician Group Incentive Program (PGIP)’s PCMH model began in 2009 and the Priority Health PCMH pilot began in 2008.16-18

The Michigan Primary Care Transformation (MiPCT) project, the first multipayer PCMH program in Michigan, was the largest participant in MAPCP. MiPCT operated from January 2012 to December 2016.19 Michigan Medicaid, Medicare, and 3 commercial payers (BCBSM, Blue Care Network, and Priority Health) participated with up to $9.50 per member per month (PMPM) contributed from Medicare and $7.50 PMPM from the other payers, plus $0.26 PMPM for administration. Medicaid accounted for 26% of the total MiPCT population (more than 1.1 million as of 2015).

This study focuses on the Medicaid managed care population, as Medicaid fee-for-service beneficiaries were not part of MiPCT. By 2011, 88% of Michigan Medicaid beneficiaries were enrolled in managed care.20 Medicaid health plans are required to maintain certification from a national certifying body (eg, the National Committee for Quality Assurance) and provide telephonic case management. Managed care beneficiaries either self-select their primary care providers (PCPs) or accept health plan—assigned PCPs.

A key feature of MiPCT was the provision of embedded care management services, with at least 2 trained care managers per 5000 patients. Care managers, often nurses or social workers embedded within primary care practice teams, are becoming increasingly utilized in new models of primary care, as evidenced in several national and regional PCMH and primary care demonstrations.21-25 These embedded care managers have been found to be acceptable to primary care teams26 and more effective in patient engagement than traditional case management by disease management companies.27,28

MiPCT care managers were physically located within the practice, documented patients’ visits in their electronic health record, communicated directly with physicians and other care team members electronically and in person, and were provided lists of high-risk beneficiaries and encouraged to work with providers to target those who could most benefit. MiPCT provided training for care managers and facilitated learning across practices. Other MiPCT requirements were to have an all-patient registry to address gaps in care and to provide advanced access (open access scheduling and options for care outside of business hours).

Other practices in Michigan continued to build PCMH capacity throughout the study period, albeit with fewer resources and without multipayer alignment. This study compared MiPCT beneficiaries with beneficiaries served in both other PCMH and non-PCMH practices. Given the ongoing PCMH programming of key commercial payers, and mature managed care in Medicaid, a key question is whether the multipayer approach to transformation had benefits over and above existing efforts. Because care management was targeted to high-risk beneficiaries, improvements were expected to be largest for this population.

METHODS

Study Design

This difference-in-differences (DID) study analyzed 2011-2015 Medicaid encounter data collected from health plans by the state of Michigan. The project was determined to be non—human subjects research by Michigan Public Health Institute’s institutional review board.

Intervention Group

Practices eligible to participate in MiPCT had to maintain PCMH29 designation from 2010 throughout the entire project period. During the project period, practices could be dropped from MiPCT if they did not meet program requirements. At the end of 2012, when comparison groups were being selected, 395 of the 473 originally eligible practices were participating in MiPCT (15 lost PCMH designation in 2012, 48 declined, and 15 dropped out). Medicaid managed care beneficiaries whose PCPs were in MiPCT practices were eligible for the intervention group.

Comparison Groups

The practices in the 2012 PGIP provider list served as the sampling frame to create 2 comparison groups: PCMH and non-PCMH. BCBSM requires all PGIP providers to designate their primary practice location and assigns a unique practice identifier for each provider. All 539 PCMH practices that were ineligible for MiPCT (namely, that were designated a PCMH in 2011 or 2012, not in 2010) served as the PCMH comparison practices. Because there was a large pool of non-PCMH practices (n = 1436), propensity score nearest-neighbor matching without replacement was performed to select 395 non-PCMH comparison practices. The 395 non-PCMH comparison practices were statistically equivalent, on average, to the MiPCT practices along the following observable practice-level variables: number of Medicaid beneficiaries per PCP, ratio of PCPs to physicians, average risk score of BCBSM beneficiaries, average patient age, federally qualified health center (FQHC) indicator, and Dartmouth hospital service area (HSA). The only variable not matched was practice size, measured by number of PCPs. The mean MiPCT practice size was significantly larger than the mean non-PCMH comparison practice size (4.1 vs 2.4 PCPs, respectively) (eAppendix Table 1 [eAppendix available at ajmc.com]). Medicaid managed care beneficiaries whose PCPs were in the PCMH and non-PCMH comparison practices were eligible for the PCMH and non-PCMH comparison groups, respectively.

Exclusions

Practices were excluded if they dropped out of MiPCT (beneficiary data were no longer available for the dropouts), joined MiPCT after 2013, became a PCMH after 2012, or lost PCMH designation. Throughout the project period, 38 practices dropped out, commonly due to being unwilling to implement the program.

Beneficiaries had to have a visit with their PCP’s practice in the measurement year or previous year, have at least 6 months enrollment, and not be pregnant in the measurement year. Also, Children’s Special Health Care Services and Medicaid expansion populations were excluded due to lack of baseline. Beneficiaries in the top 0.01% of costs were excluded from the cost analysis.

Weighting

Beneficiary-level data were not available at the time of selecting comparison groups. To compensate for not being able to match beneficiaries, an entropy balancing weighting procedure30 was used. Weights were computed and trimmed for each comparison beneficiary for each year. These positive weights minimized a distance function to satisfy balancing constraints such that the weighted means of selected observable variables over the comparison group equaled the means of those variables over MiPCT. The variables included age, gender, concurrent risk score, practice-level indicators for primary care focus and FQHC, median household income, and Michigan’s prosperity regions31 of practice location. Enrollment weights were calculated as fractions of enrolled months in the measurement year to down-weight beneficiaries with less exposure. The product of the entropy balancing and enrollment weights was used as the regression weight.

Data Sources and Variables

Data sources. All the outcomes were claims based, calculated from the administrative data from the Michigan Department of Health and Human Services Data Warehouse. Other data sources included PGIP’s physician lists and other practice-level data provided by BCBSM; FQHC and rural health center lists from the state of Michigan; HSA codes; and Rural-Urban Commuting Area (RUCA) codes.

Outcome variables. ED rate and hospital inpatient (IP) rate were measured as the number of visits/admissions per 1000 beneficiaries per year. Quality outcomes were indicators for diabetes care (glycated hemoglobin testing, eye exam, attention to nephropathy, and all 3); breast cancer, cervical cancer, and chlamydia screening; childhood and adolescent immunizations; and well-child/well-care visits at 15 months, 3 to 6 years, and adolescence. All calculations were based on Healthcare Effectiveness Data and Information Set measures. MiPCT stakeholders selected quality metrics in 2012.

PMPM cost was the total paid amount divided by the number of enrollment months in the measurement year. For reasons related to data completeness, comparability across payers, and confidentiality, PMPM cost excluded pharmacy, dental, vision, Prepaid Inpatient Health Plan services, chiropractic, nonemergency transportation, and nontobacco substance abuse claims.

Control variables. Beneficiary-level control variables included concurrent risk score and eligibility category (Aged, Blind, Disabled, or Temporary Assistance for Needy Families). Practice characteristics included size (solo, 2-3, 4-6, >6 physicians), FQHC indicator, pediatric (pediatric analyses), hospital employment (85% PCPs in the practice were hospital employed), location (based on RUCA 3.0: urban core, other urban, rural), and Michigan’s prosperity regions.

The concurrent risk score was calculated from the claims data by Truven Health Analytics, using the DCG models from Verisk Health (now Verscend Technologies), based on age, gender, health conditions, and condition interactions. The scores were categorized into 5 levels—very low (0-20), low (21-75), medium (76-225), high (226-700), and very high (701-9999)—based on the percentiles of cost in Truven’s Commercial MarketScan population.

Statistical Methods

Analyses were performed on the following subpopulations separately: adult (≥18 years), high-risk adult (in the high or very high risk levels), and pediatric (<18 years).

Mixed and generalized mixed models (IBM SPSS version 23 [IBM; Armonk, New York]) were used, with repeated measures to account for correlations among yearly measurements, random intercept to account for practice-level clustering, and robust parameter estimate covariance to handle violations of model assumptions. ED visits were estimated with a negative binomial distribution and log-link function to account for overdispersion. IP admissions were estimated using 2-part models. First, the dependent variable was an indicator of admission estimated with a binomial distribution and logit link function. Second, nonzero admissions were estimated with a negative binomial distribution and log-link function. The 2 parts were combined to estimate admissions.

An assumption of DID is that intervention and comparison groups would have parallel trends in outcome absent the intervention. Due to incompleteness of the Medicaid data prior to 2011, this assumption was tested on quarterly PMPM cost, ED visits, and hospital admissions in 2011. Differences in slopes for each outcome between MiPCT and the comparison groups from quarters 1 to 2, 2 to 3, and 3 to 4 were estimated. The P value for each slope difference was compared with a rejection criterion α, adjusted for multiple comparisons using Benjamini-Hochberg’s “step-down” method.

Results include the model-adjusted means from 2011 to 2015 and DIDs from baseline 2011 between MiPCT and the comparison groups for 2013-2015. For cost, the DIDs are changes in PMPM cost from baseline for MiPCT, minus changes in PMPM cost from baseline for a comparison group. Negative DIDs demonstrate relative cost reductions for MiPCT. For indicator and utilization outcomes, the DIDs are multiplicative: ratios of odds ratios for indicator outcomes, and ratios of incidence rate ratios for ED rate. DIDs less than 1 show relative reduced ED from baseline for MiPCT. For example, a DID of 0.81 for ED in 2015 means that the ratio of incidence rate in 2015 to baseline for MiPCT is 81% of the ratio for comparison group, implying that if both groups had same ED rate at baseline, and the comparison ED rate changed to 1000 in 2015, then the ED rate for MiPCT in 2015 would be 810. For IP, DIDs less than 1 show relative reduced risk of admission from baseline for MiPCT. For quality outcomes, DIDs greater than 1 show relative improved rates for MiPCT from baseline. For example, a DID of 1.4 for breast cancer screening in 2015 means that the ratio of odds of having breast cancer screening in 2015 for MiPCT, compared with baseline, is 40% higher than the ratio for the comparison group. All CIs are 95%, P <.05 was considered statistically significant, and Sidak corrections were applied when comparing multiple groups at each time point.

RESULTS

Included in the analyses were 173,179 (49,214 adults and 123,965 children) MiPCT beneficiaries, 209,181 (75,583 adults and 133,598 children) PCMH comparison beneficiaries, and 148,657 (45,235 adults and 103,422 children) non-PCMH beneficiaries. Of the adults, 24,056 (49%) MiPCT beneficiaries, 36,554 (48%) PCMH comparison beneficiaries, and 21,942 (49%) non-PCMH beneficiaries were high-risk adults. There were 326 MiPCT, 479 PCMH comparison, and 329 non-PCMH practices included. Beneficiary characteristics at baseline, entropy-weighted and unweighted, are in eAppendix Table 2. The trimmed entropy weight ranged from 0.05 to 13.1 (mean [SD] = 0.72 [0.93]) in the PCMH comparison group and from 0.02 to 19.70 (1.05 [1.53]) in the non-PCMH group.

Parallel Trends Assumption

No statistically significant preintervention quarterly slope difference existed between MiPCT and its comparison groups for all the subpopulations in cost and utilization outcomes (eAppendix Table 3; eAppendix Figure).

Cost

Figure 1 (also eAppendix Tables 4 and 5) shows model-adjusted mean PMPM costs and estimated total net cost savings for 3 postdemonstration years (2013-2015). Table 1 presents the DIDs with CIs for 2013-2015 and baseline model-adjusted means.

For adults, MiPCT showed significant cost reduction relative to both comparison groups in 2015 compared with the baseline PMPM cost of $255 (DID, —$90). In 2013 and 2014, the relative changes in cost were not significant (DIDs ranging from –$12 to $21).

For high-risk adults, MiPCT showed significant cost reduction relative to both comparison groups in 2014 and 2015 compared with the baseline PMPM cost of $464 (DID, —$53 in 2014 and –$187 in 2015 against PCMH; DID, –$72 in 2014 and –$212 in 2015 against non-PCMH). The DIDs in 2013 were not significant ($11 and –$17, respectively).

For children, the DIDs compared with the baseline PMPM cost of $103 ranged from —$3 to $7 and were not significant.

The net cost savings were derived from multiplying the DIDs against PCMH comparison by member-months, discounting the investment in these beneficiaries ($7.76 PMPM). In 2013-2015, the estimated net cost savings for high-risk adults was $23,178,313 (95% CI, $11,782,031-$36,214,347), 16% of their total cost; for adults, it was $15,569,526 (95% CI, $3,416,832-$27,722,219), 9% of their total cost. The pediatric population had a significant net cost increase of $16,517,948 (95% CI, $7,712,286-$25,323,609), 14% of its total cost. Overall, MiPCT beneficiaries (adult plus pediatric) had a nonsignificant net cost increase of $948,422 (95% CI, —$14,059,164 to $15,956,007), 0.3% of their total cost.

Utilization

Figure 2 (also eAppendix Table 4) presents model-adjusted means. Table 1 presents multiplicative DIDs with CIs and baseline model-adjusted means. In 2015, all MiPCT populations demonstrated significant reduction in ED use (adults: baseline ED rate, 883; DID, 0.93 against non-PCMH; high-risk adults: baseline ED rate, 1553; DID, 0.81 and 0.77 against PCMH and non-PCMH; pediatrics: baseline ED rate, 692; DID, 0.91 and 0.88 against PCMH and non-PCMH). These translated to significant ED rate reduction by 71 for adults against non-PCMH; by 362 and 444 for high-risk adults against PCMH and non-PCMH, respectively; and by 54 and 81 for pediatrics against PCMH and non-PCMH, respectively. However, MiPCT adults experienced increased ED rate in 2013-2014 relative to both groups (by 65 to 95), and pediatrics experienced increased ED rate in 2013 relative to both groups and in 2014 relative to PCMH (by 61 to 68).

All MiPCT populations had significantly reduced relative risk of inpatient admission in 2015 (adults: baseline IP rate, 143; DID, 0.72 and 0.42 against PCMH and non-PCMH; high-risk adults: baseline IP rate, 292; DID, 0.51 and 0.54 against PCMH and non-PCMH; pediatrics: baseline IP rate, 55; DID, 0.86 and 0.75 against PCMH and non-PCMH). The high-risk adults demonstrated consistently significant reduced risk of inpatient admission in 2013-2015 against PCMH (DID, 0.77, 0.71, and 0.51, respectively) and against non-PCMH (DID, 0.83, 0.83, and 0.54, respectively), except for 2013 when DID was marginally significant. In magnitude, IP rate significantly decreased for adults by 72 against non-PCMH in 2015; for high-risk adults by 25, 36, and 86 against PCMH in 2013-2015 and by 96 against non-PCMH in 2015; and for pediatrics by 10, 2, and 5 against non-PCMH in 2013-2015 and by 4 against PCMH in 2013.

Quality

MiPCT’s cross-sectional adult quality rates were either significantly higher than or equal to comparison groups (Figure 3; eAppendix Table 6). However, the only significant DID was a 10 percentage-point increase in the chlamydia screening rate in 2014 relative to the non-PCMH group (Table 2; eAppendix Table 6).

For pediatric quality outcomes, MiPCT maintained the same or higher rates than comparison groups, except in 2012 when non-PCMH had a significantly higher rate (by 4 percentage points) in well-child visits for those aged 3 to 6 years (Figure 3; eAppendix Table 6). The only significant DIDs included a 5 percentage-point increase in childhood immunization in 2013 relative to non-PCMH; 2 and 4 percentage-point increases in well-child visits for those aged 3 to 6 years in 2013 and 2014, respectively, relative to PCMH; 3 and 4 percentage-point increases in well-child visits for those aged 3 to 6 years in 2013 and 2014, respectively, relative to non-PCMH; 6 and 5 percentage-point increases in adolescent well-care visits in 2014 relative to PCMH and non-PCMH, respectively; and a 6 percentage-point decease in adolescent immunization in 2014 relative to non-PCMH (Table 2; eAppendix Table 6). By 2015, no difference existed between the groups.

DISCUSSION

This study is important in that evaluations of multipayer PCMH initiatives conducted by the Center for Medicare and Medicaid Innovation to date have been unable to include Medicaid beneficiaries in their analysis due to delays in receipt of data.32

The study results among high-risk adults suggest that MiPCT, including embedded care management targeting high-risk beneficiaries, was effective at holding down costs and inpatient hospitalizations while maintaining quality. The MiPCT Medicaid general adult population started to see relative cost and utilization reductions in 2015. However, additional analysis on the non—high-risk adults (not presented) did not show cost and utilization reductions, suggesting that continuing the MiPCT project may be holding cost and utilization trends in check for high-risk adults.

Among pediatric beneficiaries, MiPCT did not show a reduction in cost or utilization in 2013 or 2014. In 2015, comparison groups experienced significant increases relative to MiPCT in both utilization measures. MiPCT was equivalent or higher in most pediatric quality rates when compared in any given year, yet quality was not improving and the gaps between MiPCT and comparisons were narrowing over time.

In retrospect, metrics chosen by MiPCT could have been better targeted to measure intervention effects for pediatric populations. Those assessed are typical measures of good primary care but are not necessarily expected to respond to care management. It therefore makes sense that as other practices added PCMH capacity, they were catching up to MiPCT. A 2013 query of MiPCT pediatric care managers on their preferences for training topics is a window into the patient needs they were addressing: autism spectrum, depression, anxiety, and oppositional defiant disorders; developmental and mental health screening; working with schools and parents; pediatric-specific documentation tools; and legal issues (guardianship, foster care, divorce situations). Future evaluations of Michigan pediatric care management will address these domains. Additionally, the literature acknowledges that focusing on pediatric interventions may not result in immediate and direct healthcare savings but rather in long-term savings and positive effects in other systems such as schools and juvenile justice.33-35 In this study, the average model-adjusted PMPM cost for children was $152 compared with $278 for adults; thus, the potential cost savings for children are limited.

Finally, when looking at utilization trends, context is important. Michigan expanded Medicaid in mid-2014. By 2015, 30% of Medicaid managed care beneficiaries were new expansion members, who—according to Michigan’s policy—were supposed to schedule a primary care visit within 60 days of enrollment. Although this population was excluded from our analysis, ramifications could have ensued for existing patients. Perhaps MiPCT practices were better able to handle the increased volume without sacrificing access for all patients.

Limitations

As with similar quasi-experimental designs, causal inference may be impaired by self-selection bias. It is empirically demonstrated that MiPCT practices, as early PCMH adopters, differed from later PCMH adopters and non-PCMH practices. These differences included practice size, hospital employment, and region. Additionally, they were higher performers on most quality measures at baseline. This problem, common to PCMH evaluations, was addressed by using 2 comparison groups as reference points, DID modeling, and controlling for confounding variables. Propensity score matching was successful in matching practices along most variables, but it did not yield a matched comparison group of beneficiaries. This has occurred in other PCMH evaluations.36 Entropy balancing weights were used to make the comparison and MiPCT beneficiaries more comparable.

The lack of multiple years of baseline data made testing of preintervention parallel yearly trends impossible. This was somewhat remedied by testing the quarterly trends of the available baseline year.

Other factors may also be impacting trends in Michigan. PCMH was expanding and accountable care organizations were being created. These entities were building on the lessons of MiPCT and expanding care management into non-MiPCT practices. Thus, MiPCT was having spillover effects—potentially into comparison practices. This works against being able to affirm the study hypothesis.

CONCLUSIONS

This study provides evidence of success of a multipayer PCMH initiative, with an emphasis on practice-based care management, among the high-risk adult Medicaid managed care population in Michigan when considered within the framework of the 3-part aim. Results show significant cost reduction and consistent significantly reduced risk of hospitalization among high-risk adults. A cumulative cost increase occurred among the pediatric beneficiaries. However, significant reductions in cost, ED use, and risk of hospitalization started in the fourth year into the project. Meanwhile, MiPCT generally maintained equal or better quality of care.

Acknowledgments

The authors wish to acknowledge the codirectors of MiPCT (in addition to Dr Jean M. Malouin and Kathy Stiffler), Susan Moran, MPH; Theresa Landfair; and Carol Callaghan, MPH; and the MiPCT project manager, Diane Marriott, DrPH. We also thank the Michigan Data Collaborative and Chris Wojcik, MPH, at Michigan Public Health Institute for data management.

Coauthor Jean M. Malouin, MD, MPH, died March 10, 2018.Author Affiliations: Michigan Public Health Institute (SZ, CLT), Okemos, MI; Michigan State University (RAM), East Lansing, MI; Department of Family Medicine, University of Michigan Medical School (JMM), Ann Arbor, MI; Medical Services Administration, Michigan Department of Health and Human Services (KS), Lansing, MI.

Source of Funding: The evaluation was funded by the participating payers in the demonstration through an administrative fee charged on a per-member per-month basis.

Author Disclosures: Dr Rebecca Malouin is a paid consultant to the Michigan Public Health Institute assisting with evaluation of the Michigan Primary Care Transformation program. Dr Tanner is an employee of the Michigan Public Health Institute, which was contracted to evaluate the program. 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 (SZ, RAM, JMM, KS, CLT); acquisition of data (SZ); analysis and interpretation of data (SZ, RAM, CLT); drafting of the manuscript (SZ, RAM, JMM, KS); critical revision of the manuscript for important intellectual content (SZ, RAM, JMM, KS, CLT); statistical analysis (SZ); obtaining funding (CLT); administrative, technical, or logistic support (KS, CLT); and supervision (CLT).

Address Correspondence to: Clare L. Tanner, PhD, Michigan Public Health Institute, 2501 Jolly Rd, Ste 180, Okemos, MI 48864. Email: ctanner@mphi.org.REFERENCES

1. Christensen AL, Zickafoose JS, Natzke B, McMorrow S, Ireys HT. Associations between practice-reported medical homeness and health care utilization among publicly insured children. Acad Pediatr. 2015;15(3):267-274. doi: 10.1016/j.acap.2014.12.001.

2. Takach M. About half of the states are implementing patient-centered medical homes for their Medicaid populations. Health Aff (Millwood). 2012;31(11):2432-2440. doi: 10.1377/hlthaff.2012.0447.

3. Cunningham PJ. Many Medicaid beneficiaries receive care consistent with attributes of patient-centered medical homes. Health Aff (Millwood). 2015;34(7):1105-1112. doi: 10.1377/hlthaff.2015.0141.

4. Shane DM, Nguyen-Hoang P, Bentler SE, Damiano PC, Momany ET. Medicaid Health Home reducing costs and reliance on emergency department: evidence from Iowa. Med Care. 2016;54(8):752-757. doi: 10.1097/MLR.0000000000000555.

5. Cole ES, Campbell C, Diana ML, Webber L, Culbertson R. Patient-centered medical homes in Louisiana had minimal impact on Medicaid population’s use of acute care and costs. Health Aff (Millwood). 2015;34(1):87-94. doi: 10.1377/hlthaff.2014.0582.

6. Bronstein JM, Morrisey MA, Sen B, Engler S, Smith WK. Initial impacts of the Patient Care Networks of Alabama initiative. Health Serv Res. 2016;51(1):146-166. doi: 10.1111/1475-6773.12319.

7. Harman JS, Lemak CH, Al-Amin M, Hall AG, Duncan RP. Changes in per member per month expenditures after implementation of Florida’s Medicaid reform demonstration. Health Serv Res. 2011;46(3):787-804. doi: 10.1111/j.1475-6773.2010.01226.x.

8. Jabbarpour Y, DeMarchis E, Bazemore A, Grundy P. The impact of primary care practice transformation on cost, quality, and utilization: a systematic review of research published in 2016. Patient-Centered Primary Care Collaborative website. pcpcc.org/sites/default/files/resources/pcmh_evidence_report_08-1-17%20FINAL.pdf. Published July 2017. Accessed October 11, 2017.

9. Jones C, Finison K, McGraves-Lloyd K, et al. Vermont’s community-oriented all-payer medical home model reduces expenditures and utilization while delivering high-quality care. Popul Health Manag. 2016;19(3):196-205. doi: 10.1089/pop.2015.0055.

10. Sinaiko AD, Landrum MB, Meyers DJ, et al. Synthesis of research on patient-centered medical homes brings systematic differences into relief. Health Aff (Millwood). 2017;36(3):500-508. doi: 10.1377/hlthaff.2016.1235.

11. Comprehensive Primary Care initiative. CMS website. innovation.cms.gov/initiatives/comprehensive-primary-care-initiative. Accessed July 10, 2017.

12. Dale SB, Ghosh A, Peikes DN, et al. Two-year costs and quality in the Comprehensive Primary Care initiative. N Engl J Med. 2016;374(24):2345-2356. doi: 10.1056/NEJMsa1414953.

13. Rosenthal MB, Alidina S, Friedberg MW, et al. A difference-in-difference analysis of changes in quality, utilization and cost following the Colorado multi-payer patient-centered medical home pilot. J Gen Intern Med. 2016;31(3):289-296. doi: 10.1007/s11606-015-3521-1.

14. Kern LM, Edwards A, Kaushal R. The patient-centered medical home and associations with health care quality and utilization: a 5-year cohort study. Ann Intern Med. 2016;164(6):395-405. doi: 10.7326/M14-2633.

15. Peikes D, Anglin G, Dale S, et al. Evaluation of the Comprehensive Primary Care initiative: fourth annual report. Mathematica Policy Research website. mathematica-mpr.com/our-publications-and-findings/publications/evaluation-of-the-comprehensive-primary-care-initiative-fourth-annual-report. Published May 23, 2018. Accessed December 5, 2018.

16. Hearld LR, Weech-Maldonado R, Asagbra OE. Variations in patient-centered medical home capacity: a linear growth curve analysis. Med Care Res Rev. 2013;70(6):597-620. doi: 10.1177/1077558713498117.

17. Lemak CH, Nahra TA, Cohen GR, et al. Michigan’s fee-for-value physician incentive program reduces spending and improves quality in primary care. Health Aff (Millwood). 2015;34(4):645-652. doi: 10.1377/hlthaff.2014.0426.

18. Cross DA, Cohen GR, Harris Lemak C, Adler-Milstein J. Sustained participation in a pay-for-value program: impact on high-need patients. Am J Manag Care. 2017;23(2):e33-e40.

19. Malouin JM, Malouin RA, Sarinopoulos I, et al. Physician and staff acceptance of care managers in primary care offices. J Am Board Fam Med. 2017;30(2):140-149. doi: 10.3122/jabfm.2017.02.160246.

20. Medicaid managed care enrollment report. Medicaid website. medicaid.gov/medicaid-chip-program-information/by-topics/data-and-systems/downloads/2011-medicaid-mc-enrollment-report.pdf. Published July 1, 2011. Accessed October 4, 2019.

21. Friedberg MW, Rosenthal MB, Werner RM, Volpp KG, Schneider EC. Effects of a medical home and shared savings intervention on quality and utilization of care [erratum in JAMA Intern Med. 2015;175(8):1426. doi: 10.1001/jamainternmed.2015.3461]. JAMA Intern Med. 2015;175(8):1362-1368. doi: 10.1001/jamainternmed.2015.2047.

22. Hudon C, Chouinard MC, Diadiou F, Lambert M, Bouliane D. Case management in primary care for frequent users of health care services with chronic diseases: a qualitative study of patient and family experience. Ann Fam Med. 2015;13(6):523-528. doi: 10.1370/afm.1867.

23. Bogner HR, Joo JH, Hwang S, et al. Does a depression management program decrease mortality in older adults with specific medical conditions in primary care? an exploratory analysis. J Am Geriatr Soc. 2016;64(1):126-131. doi: 10.1111/jgs.13711.

24. Gallo JJ, Hwang S, Joo JH, et al. Multimorbidity, depression, and mortality in primary care: randomized clinical trial of an evidence-based depression care management program on mortality risk. J Gen Intern Med. 2016;31(4):380-386. doi: 10.1007/s11606-015-3524-y.

25. Peikes DN, Reid RJ, Day TJ, et al. Staffing patterns of primary care practices in the Comprehensive Primary Care initiative. Ann Fam Med. 2014;12(2):142-149. doi: 10.1370/afm.1626.

26. Howard HA, Malouin R, Callow-Rucker M. Care managers and knowledge shift in primary care patient-centered medical home transformation. Hum Organ. 2016;75(1):10-20. doi: 10.17730/0018-7259-75.1.10.

27. Luo Z, Chen Q, Annis AM, et al. A comparison of health plan— and provider-delivered chronic care management models on patient clinical outcomes. J Gen Intern Med. 2016;31(7):762-770. doi: 10.1007/s11606-016-3617-2.

28. Annis AM, Holtrop JS, Tao M, Chang HC, Luo Z. Comparison of provider and plan-based targeting strategies for disease management. Am J Manag Care. 2015;21(5):344-351.

29. Patient-centered medical home and patient-centered medical home—neighbor: interpretive guidelines. Blue Cross Blue Shield of Michigan website. bcbsm.com/content/dam/public/Providers/Documents/value/pcmh-and-pcmh-n-interpretive-guidelines.pdf. Accessed March 22, 2019.

30. Hainmueller J. Entropy balancing for causal effects: a multivariate reweighting method to produce balanced samples in observational studies. Polit Anal. 2012;20(1):25-46. doi: 10.1093/pan/mpr025.

31. State of Michigan prosperity regions. State of Michigan website. michigan.gov/documents/dmb/Prosperity_Map1_430346_7.pdf. Accessed September 5, 2018.

32. Cohen M, Russo A, Kennell D, et al; Kennel and Associates, Inc; RTI International. Systematic review of CMMI primary care initiatives: final report. CMS website. innovation.cms.gov/Files/reports/primarycare-finalevalrpt.pdf. Published February 2018. Accessed September 2018.

33. Pelham WE, Foster EM, Robb JA. The economic impact of attention-deficit/hyperactivity disorder in children and adolescents. J Pediatr Psychol. 2007;32(6):711-727. doi: 10.1093/jpepsy/jsm022.

34. Jaycox LH, Stein BD, Paddock S, et al. Impact of teen depression on academic, social, and physical functioning. Pediatrics. 2009;124(4):e596-e605. doi: 10.1542/peds.2008-3348.

35. Delaney L, Smith JP. Childhood health: trends and consequences over the life course. Future Child. 2012;22(1):43-63.

36. Nichols D, Haber S, Romaire M, et al; RTI International; The Urban Institute; National Academy for State Health Policy. Evaluation of the Multi-Payer Advanced Primary Care Practice (MAPCP) demonstration: final report. CMS website. downloads.cms.gov/files/cmmi/mapcp-finalevalrpt.pdf. Published June 2017. Accessed October 11, 2017.