Abstract
Introduction
Identifying social risk factors during the perinatal period may improve pregnancy-related outcomes. In 2018, the Massachusetts Medicaid program implemented an accountable care organization (ACO)—which required social risk factor screening—under 2 ACO model types: Model A (health system/Medicaid managed care plan partnership model) and Model B (primary care practice model).
Methods
Using the 2016-2020 Massachusetts All Payer Claims Database and a difference-in-differences (DiD) design, we compared changes in any social risk factor documentation (ie, ICD-10 Z-codes) for Medicaid-enrolled deliveries in non-ACOs vs ACO Model A vs Model B, measured separately for the prenatal period, 60 days postpartum, 12 months postpartum, and the perinatal period. Secondary outcomes included documentation of specific Z-codes related to homelessness and food insecurity.
Results
Both Model A (DiD = 1.03 percentage points [PP], P < .001) and Model B (DiD = 0.65 PP, P < .001) were associated with significant increases in Z-code documentation of food insecurity in the perinatal period. Model B was associated with a significant increase in Z-code documentation of homelessness in the perinatal period (DiD = 1.10 PP, P < .001) but not Model A.
Conclusion
Massachusetts' Medicaid ACO was associated with modest increases in Z-code documentation during the perinatal period, with some variation by ACO model type.
Keywords: Medicaid, social risk factors, social determinants of health, accountable care organizations, maternal health
Key points.
Compared to non-accountable care organizations (ACOs), both A (Medicaid managed care plan/health system partnership) and Model B (primary care practice [PCP]-led) experienced increases in Z-code documentation in the perinatal period, including significant increases in documentation of food insecurity.
Increases in Z-code documentation in Model B were largely concentrated in the prenatal period, while increases in Model A were largely concentrated in the 12-month postpartum period.
Compared to non-ACOs, there was a significant increase in Z-code documentation of homelessness for Model B in the perinatal period, but not Model A.
Background
Medicaid—the federal-state public health insurance program for low-income people1—disproportionately serves patients who report social risk factors, such as food insecurity.2,3 Some state Medicaid programs have implemented efforts to directly address enrollees' social needs through Medicaid Accountable Care Organizations (ACOs).4 Accountable care organizations aim to improve quality of care and lower costs for an attributed patient population through enhanced care coordination, increased care engagement, and financial incentives.5,6 Medicaid ACOs may take different approaches to addressing or intervening on social needs, such as referrals to social service programs or community-based supports.7,8 Given that over 40% of deliveries nationwide are financed by Medicaid—and that pregnant patients often report unmet social needs1,9—Medicaid ACOs may disproportionately improve care for pregnant and postpartum patients.
International Classification of Diseases (ICD)-10 diagnosis codes for social risk factors, referred to as Z-codes, were introduced in 2016 to begin systematically capturing patients' social contexts in claims data; however, their uptake has been slow in comparison to documentation in electronic health records.10-15 Understanding patients' unmet social needs (ie, social risk factors for which they would like assistance) in the perinatal period may provide crucial information for clinicians serving pregnant patients.9 However, evidence on the impacts of Medicaid ACOs on Z-code documentation of social risk factors is still emerging.16-18
Despite heterogeneity in how Medicaid ACOs are structured,19 few studies have compared outcomes based on model type. States have taken different approaches to designing their Medicaid ACO models: for example, in some states, Medicaid managed care organizations (MCOs) are responsible for financial risk but partner with a provider network. Other states have adopted provider-led ACOs where a provider organization (eg, hospital system) establishes a collaborative network of providers to deliver care for an attributed patient population.20 Accountable care organization model designs can have implications for care delivery and, specifically, screening and documentation of social risk factors.
Beginning in 2018, the Massachusetts Medicaid program (MassHealth) implemented an ACO model with the goals of integrating physical, behavioral, and social needs of enrollees.21 Sixteen ACOs were formed, with approximately 70% of members under age 65 years enrolled.21 Unique to Massachusetts was the development of 2 distinct Medicaid ACO models: the first (Model A) was a partnership between a health system and Medicaid MCO and the second (Model B) was a primary care provider (PCP) model.21 Studies have begun documenting the early impacts of the MassHealth ACO model on health care utilization, quality, and documentation of social needs among pregnant people.5,16,17,22 Although overall MassHealth ACO implementation was associated with significant increases in Z-code documentation among pregnant and postpartum patients,16 it is unknown whether Z-code documentation varied by ACO model design.
Using data on all live deliveries to Medicaid enrollees in Massachusetts between 2016 and 2020, we evaluated whether changes in social risk factor (ie, Z-code) documentation were differential based on enrollment in a health system-managed care plan partnership model or PCP-led model.
Methods
Study context
Of the 16 ACOs formed in Massachusetts, 13 developed health system-Medicaid MCO partnerships (“Model A”), wherein a single Medicaid MCO partnered with a health system (inclusive of hospital[s], specialists, and PCPs) to deliver services for ACO members within a network. The remaining 3 were implemented under the primary care provider (PCP) model (“Model B”), wherein groups of primary care practices formed an ACO, primary care services were delivered by the ACO, and other services were delivered on a fee-for-service basis through the state's Medicaid network. Each model type covered approximately 50% of MassHealth ACO members. Attribution to an ACO was based on whether a patient's PCP was part of an ACO, creating a novel nested natural experiment.
Accountable care organization contracts required screening and documentation of social risk factors, and included risk adjustment for housing instability (based on Z-code documentation for “homelessness” or 3 or more addresses in a calendar year).21 Accountable care organizations were required to screen all members for 4 domains (housing, food security, transportation, and utility needs) at least once annually, and there was a financial performance measure related to social risk factor screening.8 MassHealth provided ACOs some flexibility in screening, including selection of one supplemental screening domain and selection of their screening instrument.8,17
All MassHealth ACOs were required to link members to social supports and community-based organizations when appropriate. Accountable care organizations also implemented a “Flex Services Program,” where qualified individuals with food insecurity or housing insecurity were provided with nutritional supports (eg, meal boxes) or housing supports (eg, transitional assistance such as first month's rent or moving expenses). Accountable care organizations had discretion in how they integrated screenings and referrals into their workflows and in the types of social support services that they provided, including the specific supports offered through Flex Services: for example, resource staff at one Medicaid ACO in Massachusetts reported addressing food insecurity by assisting with applications for the Supplemental Nutrition Assistance Program, providing information on community food sources (eg, food banks), and offering one-time gift cards for emergency food assistance,23 while other ACOs may have implemented different types of social supports and workflows. Partnering with an MCO and/or the inclusion of specialists within the ACO network may have differentially affected these workflow and service provision decisions. All social risk factor provisions in the MassHealth ACO applied to both Model A and Model B; however, model designs may differentially impact Z-code documentation. Medicaid-enrolled pregnant and postpartum people receiving care from non-ACO PCPs and hospitals could also have social risk factors documented via Z-codes. For example, it is possible that some health systems not participating in the ACO had social risk screening processes in place; however, non-ACO providers did not have a financially incentivized screening requirement, nor did they have access to Flex Services.
Data and study population
The unit of analysis was the delivery-quarter. We used 2014-2020 Massachusetts All-Payer Claims Database to identify all Medicaid-enrolled live deliveries occurring between 2016 and 2020 among members ages 18-44 years with at least 60 days of prenatal and postpartum Medicaid enrollment. We required a minimum of 60 days postpartum, rather than the full postpartum year, because many postpartum people lose Medicaid coverage after 60 days postpartum; however, as of April 2022, MassHealth provides coverage through 12 months postpartum. We excluded deliveries that were dually enrolled in other types of insurance, deliveries occurring in the first 2 quarters of 2018 (which were considered ACO transition quarters), or deliveries among people with emergency-only Medicaid. Data were linked to the 2018-2020 Massachusetts Registration of Provider Organization files, which linked primary care provider national provider identifiers to ACOs; MassHealth provider directories; and the American Community Survey, which was used to characterize sociodemographic characteristics of member ZIP codes.
Main measures
Our primary outcome was documentation of any social risk factor-related Z-code in claims (yes/no), measured separately for the prenatal period (between 4 weeks gestation and delivery date), 60 days postpartum, 12 months postpartum, and the full perinatal period (ie, prenatal period through 12 months postpartum). Z-code documentation was flagged if a Z-code was billed on at least one claim for any visit type (eg, outpatient, inpatient) in a given time period. Secondary outcomes included documentation of specific Z-codes related to (1) homelessness, (2) food insecurity, (3) other problems related to housing or economic circumstances (excluding codes related to homelessness and food insecurity), and (4) all other Z-codes (Appendix Table S1).
Our primary exposure was ACO model type (non-ACO, Model A, or Model B). In the MassHealth ACO program, PCP attribution determined ACO attribution. We identified the primary care practice that provided the plurality of primary care services to attribute a member to a PCP and, consequently, an ACO, as described in previous work.5,24 Deliveries were then categorically defined as those attributed to Model A, deliveries attributed to Model B, and Medicaid deliveries that were not part of an ACO.
Statistical analysis
We first described member and delivery characteristics by ACO model type, using Pearson's chi-square tests and t-tests to compare the distributions of categorical and continuous variables, respectively. We then described overall Z-code documentation rates by ACO model type. We used a difference-in-differences (DiD) design with linear probability models to compare Z-code documentation before (2016-2017) vs after (2018-2020) ACO implementation for deliveries attributed to Model A (exposure group 1), attributed to Model B (exposure group 2), and non-ACO deliveries (comparison group). We excluded the first 2 quarters of 2018 as this was a transitional implementation period.5,16,24 Regression models adjusted for age, delivery type, multiple gestation, documentation of 3 or more clinical comorbidities (including body mass index [BMI] 25-40 kg/m2, BMI > 40 kg/m2, diabetes, hypertension, hyperlipidemia, cardiovascular disease, asthma, major depression, other depression, and anxiety), and ZIP-code-level rurality and sociodemographic characteristics (eg, racial composition, median income), with delivery hospital-level fixed effects and standard errors clustered at the individual level. Clinical covariates (multiple gestation, comorbidities) were derived using ICD-10 diagnosis codes, which have been described in our previous work.5
Sensitivity analyses and robustness checks
We conducted multiple sensitivity analyses. First, the underlying assumption of the DiD framework is that trends in the comparison group reflect what would have occurred in the absence of the intervention (eg, the Medicaid ACO). We visually inspected the pre-intervention trends and tested the interaction between an indicator for each intervention group and a linear time trend in the pre-intervention period (Appendix Table S2). In the Appendix, we also present event study plots. Second, we re-ran all analyses by excluding deliveries in 2020, which coincided with the early COVID-19 pandemic and subsequent changes in utilization of health services and increases in unmet social needs.25,26 Third, because we assessed differences in 20 outcomes with 2 intervention groups, we accounted for multiple comparisons by using a Bonferroni correction, wherein the threshold for statistical significance was P < .00125.
Results
Our study sample included 116 169 deliveries, of which 31 320 (27.0%) were in a non-ACO; 49 123 (30.8%) were in Model A, the health system/MCO partnership model; and the remaining 35 726 (42.3%) were in Model B, the PCP-led model (Table 1). There were some statistically significant baseline differences in sociodemographic characteristics and clinical comorbidities based on model type.
Table 1.
Characteristics of Medicaid-insured deliveries by MassHealth Accountable Care Organization model type, 2016-2020.
| Non-ACO (n = 31 320) | Model A: health system/MCO partnership (n = 49 123) | Model B: primary care practice-led (n = 35 726) | Total (n = 116 169) | |
|---|---|---|---|---|
| Age at delivery, mean (SD) | 28.2 (5.7) | 27.9 (5.7) | 28.2 (5.7) | 28.0 (5.7) |
| Delivery type | ||||
| Vaginal | 19 388 (65.0%) | 30 081 (65.7%) | 22 440 (66.5%) | 71 909 (65.7%) |
| C-section | 10 432 (35.0%) | 15 726 (34.3%) | 11 306 (33.5%) | 37 464 (34.3%) |
| Multiple gestation | 575 (1.8%) | 868 (1.8%) | 601 (1.7%) | 2 044 (1.8%) |
| Three or more comorbid conditions, % | 3 271 (10.4%) | 8 270 (16.8%) | 5 561 (15.6%) | 17 102 (14.7%) |
| Rural ZIP code | 2 265 (7.2%) | 1 971 (4.0%) | 2 106 (5.9%) | 6 342 (5.5%) |
| Select diagnoses (ever) | ||||
| Mental health | 9 781 (31.2%) | 20 377 (41.5%) | 14 977 (41.9%) | 45 135 (38.9%) |
| Diabetes | 893 (2.9%) | 1 816 (3.7%) | 1 309 (3.7%) | 4 018 (3.5%) |
| Asthma | 3 587 (11.5%) | 7 616 (15.5%) | 5 474 (15.3%) | 16 677 (14.4%) |
| Hypertension | 2 009 (6.4%) | 4 169 (8.5%) | 3 041 (8.5%) | 9 219 (7.9%) |
| Overweight or obesity | 6 460 (20.6%) | 13 663 (27.8%) | 10 142 (28.4%) | 30 265 (26.1%) |
| ZIP code-level characteristics, mean (SD) | ||||
| % Non-Hispanic Black | 11.2 (14.7) | 13.2 (16.3) | 13.3 (16.9) | 12.7 (16.1) |
| % Non-Hispanic White | 62.0 (24.7) | 55.9 (26.2) | 57.2 (26.1) | 57.9 (25.9) |
| % Hispanic | 18.6 (17.7) | 21.9 (21.9) | 21.8 (19.9) | 21.0 (20.3) |
| Median household income, $ | 80 843 (29 636) | 75 580 (30 378) | 77 690 (28 903) | 77 648 (29 807) |
The unit of analysis was delivery. Differences by enrollment in MassHealth ACO were measured using Pearson's chi-square tests for categorical variables and t-tests for age and zip code-level continuous variables. P < .001 for all comparisons; multiple gestation (P = .316).
Unadjusted rates of Z-code documentation in the prenatal period, 60 days postpartum, 12 months postpartum, and the entire perinatal period were lower among non-ACO deliveries when compared to Model A and Model B deliveries (Figure 1; Appendix Table S3). For example, in the post-period (ie, 2018-2020), rates of any Z-code documentation in the perinatal period were 4.64% in non-ACOs, 8.36% in Model A, and 8.27% in Model B.
Figure 1.
Changes in documentation of any social risk factor by MassHealth Accountable Care Organization (ACO) model type. The figure presents quarterly proportions of deliveries with Z-code documentation by Medicaid ACO model type: non-ACO, model A (health system/managed care organization [MCO] partnership), and model B (primary care practice [PCP]-led). (A) Presents changes in the prenatal period. (B) Presents changes 60 days postpartum. (C) Presents changes 12 months postpartum. (D) Presents changes in the entire perinatal period.
Prenatal period
Compared to concurrent trends in non-ACOs, in the prenatal period, Model A was associated with significant increases in Z-code documentation of food insecurity (DiD = 0.58 percentage points [PP]; 95% CI, 0.34-0.82; P < .001), but not with any Z-code documentation or other indicators (Table 2). Model B was associated with significant increases in any Z-code documentation (DiD = 1.57 PP; 95% CI, 0.64-2.50; P = .001), including Z-code documentation of homelessness (DiD = 0.94 PP; 95% CI, 0.37-1.61; P = .001), food insecurity (DiD = 0.52 PP; 95% CI, 0.29-0.75; P < .001), and other problems related to housing and economic circumstances (DiD = 0.79; 95% CI, 0.03-1.54; P = .04) in the prenatal period.
Table 2.
Changes in documentation of social risk factors associated with MassHealth Accountable Care Organization, 2016-2020.
| Non-ACO, % | Model A, % | Model B, % | Difference (non-ACO and Model A) | Difference (non-ACO and Model B) | ||||
|---|---|---|---|---|---|---|---|---|
| Pre (2016-2017) | Post (2018-2020) | Pre (2016-2017) | Post (2018-2020) | Pre (2016-2017) | Post (2018-2020) | |||
| Prenatal period | ||||||||
| Any social risk factor | 2.17 | 3.54 | 3.46 | 5.90 | 3.14 | 5.87 | 0.62 (−0.22, 1.47) | 1.57 (0.64, 2.50)** |
| Homelessness | 0.61 | 0.80 | 0.74 | 1.41 | 0.89 | 1.90 | 0.34 (−0.18, 0.86) | 0.94 (0.37, 1.51)** |
| Food insecurity | 0.08 | 0.35 | 0.04 | 0.91 | 0.07 | 0.81 | 0.58 (0.34 0.82)*** | 0.52 (0.29, 0.75)*** |
| Other problems related to housing and economic circumstances | 0.23 | 0.94 | 0.48 | 1.60 | 0.48 | 1.53 | 0.30 (−0.02, 0.80) | 0.36 (−0.02, 0.74) |
| Other social risk factor | 1.47 | 2.08 | 2.43 | 3.55 | 1.90 | 3.25 | 0.27 (−0.32, 0.85) | 0.79 (0.03, 1.54)* |
| 60 days postpartum | ||||||||
| Any social risk Factor | 0.50 | 1.01 | 0.67 | 1.69 | 0.69 | 1.38 | 0.40 (−0.15, 0.94) | 0.08 (−0.31, 0.48) |
| Homelessness | 0.09 | 0.14 | 0.15 | 0.33 | 0.12 | 0.50 | 0.08 (−0.08, 0.25) | 0.27 (0.08, 0.47)** |
| Food insecurity | 0.02 | 0.15 | 0.03 | 0.23 | 0.05 | 0.09 | 0.06 (−0.10, 0.23) | −0.10 (−0.21, 0.01) |
| Other problems related to housing and economic circumstances | 0.09 | 0.21 | 0.10 | 0.54 | 0.14 | 0.37 | 0.24 (−0.07, 0.55) | 0.05 (−0.10, 0.20) |
| Other social risk factor | 0.32 | 0.58 | 0.47 | 0.95 | 0.42 | 0.54 | 0.18 (−0.30, 0.66) | −0.15 (−0.45, 0.16) |
| 12 months postpartum | ||||||||
| Any social risk factor | 1.57 | 2.13 | 2.79 | 4.62 | 3.06 | 4.57 | 0.92 (0.18, 1.66)* | 0.86 (0.06, 1.66)* |
| Homelessness | 0.39 | 0.53 | 1.05 | 1.06 | 0.87 | 1.61 | −0.29 (−0.69, 0.11) | 0.61 (0.10, 1.12)* |
| Food insecurity | 0.09 | 0.25 | 0.19 | 0.64 | 0.13 | 0.45 | 0.26 (−0.02, 0.55) | 0.21 (−0.02, 0.44) |
| Other problems related to housing and economic circumstances | 0.20 | 0.44 | 0.43 | 1.16 | 0.64 | 1.16 | 0.40 (0.04, 0.76)* | 0.26 (−0.17, 0.70) |
| Other social risk factor | 1.02 | 1.19 | 1.54 | 2.79 | 1.69 | 2.51 | 0.94 (0.36, 1.53)** | 0.58 (0.03, 1.11)* |
| Entire perinatal period | ||||||||
| Any social risk factor | 3.50 | 4.64 | 5.48 | 8.36 | 5.51 | 8.27 | 1.62 (0.81, 2.43)*** | 1.62 (0.76, 2.48)*** |
| Homelessness | 0.97 | 1.20 | 1.60 | 2.11 | 1.59 | 2.72 | 0.29 (−0.15, 0.73) | 1.10 (0.63, 1.58)*** |
| Food insecurity | 0.16 | 0.59 | 0.22 | 1.64 | 0.21 | 1.20 | 1.03 (0.75, 1.31)*** | 0.65 (0.35, 0.95)*** |
| Other problems related to housing and economic circumstances | 0.50 | 1.30 | 0.92 | 2.23 | 1.08 | 2.30 | 0.51 (0.10, 0.91)* | 0.42 (−0.00, 0.85) |
| Other social risk factor | 2.33 | 2.80 | 3.71 | 5.16 | 3.38 | 4.95 | 0.88 (0.02, 1.54)** | 0.92 (0.22, 1.63)* |
*P < .05, **P < .01, ***P < .001. “Entire perinatal period” refers to prenatal through 12 months postpartum. In each “Pre” and “Post” column, unadjusted percentages of deliveries with Z-code documentation are presented. In the “Difference (non-ACO and Model A)” and “Difference (non-ACO and Model B)” columns, adjusted difference-in-differences coefficients are presented. Linear probability models were used, including indicators for ACO model type (reference group: non-ACO), pre- vs post-period, and their interaction (ACO model type × post), which were the coefficients of interest. Covariates included age at delivery, delivery type, multiple gestations, documentation of 3 or more clinical comorbidities, and residence in a rural zip code. Models also adjusted for zip-code-level sociodemographic characteristics, included hospital-level fixed effects, and clustered standard errors at the individual level.
Postpartum period
In the 60-day postpartum period, there were no statistically significant changes in outcomes for Model A, though Model B was associated with significant increases in Z-code documentation of homelessness (DiD = 0.27 PP; 95% CI, 0.08-0.47; P = .007). In the 12-months postpartum period, Model A was associated with significant increases in Z-code documentation (DiD = 0.92 PP; 95% CI, 0.18-1.66; P = .03), including other problems related to housing and economic circumstances (DiD = 0.40 PP; 95% CI, 0.04-0.76; P = .03), and other social risk factors (DiD = 0.94; 95% CI, 0.36-1.53; P = .001). Model B was associated with a significant 0.86 PP increase (95% CI, 0.06-1.66; P = .034) in any Z-code documentation, including significant increases in Z-code documentation of homelessness (DiD = 0.61 PP; 95% CI, 0.10-1.12; P = .018), and other social risk factors (DiD = 0.58 PP; 95% CI, 0.02-1.11; P = .037) in the 12-month postpartum period.
Full perinatal period
Over the entire perinatal period, Model A was associated with a significant increase in documentation of any Z-code (DiD = 1.62 PP; 95% CI, 0.81-2.43; P < .001), including Z-code documentation of food insecurity (DiD = 1.03 PP; 95% CI, 0.75-1.31; P < .001), other problems related to housing and economic circumstances (DiD = 0.51 PP; 95% CI, 0.10-0.91; P = .014), and other social risk factors (DiD = 0.88 PP; 95% CI, 0.02-1.54; P = .009). Model B was associated with statistically significant increases in documentation of any Z-code (DiD = 1.62 PP; 95% CI, 0.76-2.48; P < .001), including Z-code documentation of homelessness (DiD = 1.10 PP; 95% CI, 0.63-1.58; P < .001), food insecurity (DiD = 0.65 PP; 95% CI, 0.35-0.95; P < .001), and other social risk factors (DiD = 0.92 PP; 95% CI, 0.22-1.63; P = .01) in the perinatal period.
Sensitivity analyses
Visual inspection and statistical assessments indicated that pre-intervention trends between non-ACOs, Model A ACOs, and Model B ACOs were parallel (Appendix Figures S1-S8). Estimates excluding 2020 data were broadly consistent with our main model (Appendix Table S4). Following correction for multiple comparisons, fewer differences were statistically significant. However, several of the main findings remained: in the prenatal period, Model A was associated with increased Z-code documentation of food insecurity and Model B was associated with increased documentation of any Z-code and of homelessness (P < .001 for all). In the full perinatal period, Model A was associated with significant increases in documentation of any Z-code and Z-code documentation of food insecurity, while Model B was associated with significant increases in documentation of any Z-code and Z-code documentation of homelessness (P < .001 for all).
Discussion
Using data from 2016-2020 Medicaid-enrolled live deliveries in Massachusetts, we found that implementation of 2 distinct Medicaid ACO models was associated with significant, albeit modest, increases in Z-code documentation in the perinatal period, including increases in Z-code documentation of food insecurity for both models. The increase in Z-code documentation among Medicaid-enrolled live deliveries in Massachusetts was modest in absolute terms but large in relative terms, given low baseline rates. Our post-period estimates in both model types were larger than (1) Z-code documentation in states with Medicaid ACOs in 2018 (2.12%) and (2) one national estimate of Z-code documentation in Medicaid in 2020 and 2021 (1.6%), potentially suggesting responsiveness to financial incentives to screen for and document social risk through the ACO.18,26 While heterogeneity between the 2 model types was limited, in the entire perinatal period, Model B was associated with increased probability of Z-code documentation of homelessness, unlike Model A. Accountable care organization-related increases in Z-code documentation were largest in the 12-month postpartum for Model A and in the prenatal period for Model B, though overall documentation rates were low (5% or less).
To our knowledge, one other study to date has documented the impact of Medicaid ACOs on social risk documentation, finding that ACO implementation was associated with modest increases in Z-code documentation of social risk factors among pregnant and postpartum Medicaid enrollees, and that increases largely occurred in the prenatal period.16 Another recent study examined the impact of the MassHealth ACO model design on process and outcome measures among pregnant and postpartum Medicaid enrollees, and found heterogeneous effects: Model A (health system/MCO partnership) was associated with timely postpartum care while Model B (PCP-led model) was associated with increased office visits during the prenatal and postpartum periods.24 Our study fills a crucial gap—whether and how Z-code documentation varies by ACO model type—which is important for potentially informing the design of care delivery models that may better increase documentation of patient social risk factors. Most previous work assessing the impacts of different Medicaid ACO model designs have compared outcomes in different states,19 and cannot account for different state environments or details of their Medicaid programs. We add to the literature by documenting some heterogeneity by model design within the same state Medicaid program.
Changes in Z-code documentation may reflect how ACO model designs differentially change care processes. The heterogeneous effects identified in our study may be explained by several concurrent mechanisms: we observed that both Medicaid ACO model types were associated with significant increases in any Z-code documentation throughout the perinatal period, including significant increases in documentation of food insecurity. These results may reflect the design of the ACO contract, which requires all ACOs to screen for social risk factors and develop infrastructure for referring patients with unmet social needs to social services, where requirements were the same for both model types. The ACO contract requirement to screen for social risk factors included Z-code documentation as one approach but ultimately granted flexibility with how social risk factor data were assessed and documented.
Significant increases in documentation of homelessness were observed in Model B, but not Model A, particularly in the prenatal period. The 3 ACOs that adopted the PCP-led model may have used Z-code documentation of homelessness more than the health system/MCO partnership model, given the state's risk adjustment payment model: Medicaid payments were risk-adjusted for housing instability, which was measured through Z-code documentation or changes of address 3 times in a year. The PCP-led model may have also had increased opportunities for engagement in the prenatal period, as supported by our earlier work on variation by Medicaid ACO model design,24 resulting in a higher likelihood of being screened for and having documentation of Z-codes. Model A ACOs may have developed shorter screening tools, limiting opportunities for identification and documentation of less common social risk factors, or perhaps used other forms of social risk factor screening or documentation (ie, not Z-codes) to align or share information with integrated MCOs.27
In previous work, increases in Z-code documentation following MassHealth's ACO implementation were largely concentrated in the prenatal period.16 Our study findings indicate that Model B was associated with significant increases in documentation of any Z-code in the prenatal period, but not Model A. The frequency of visits during the prenatal period may provide an opportunity for providers to screen and refer pregnant patients with unmet social needs to necessary resources. This aligns with findings from previous work,24 which found that Model B was associated with increases in care engagement during the prenatal period, which likely led to more opportunities to screen for social needs.28-30 Nonetheless, overall low rates of social needs documentation compared to other survey-based estimates may reflect an ongoing need to better screen for social risk factors in the prenatal period, as recommended in 2025 ACOG guidance on addressing unmet social needs as part of prenatal care delivery.9,31,32
As more states consider adopting Medicaid ACOs—and whether to incentivize the assessment and documentation of social risk factors—their designs are crucial. State Medicaid programs may need to provide reporting requirements or incentives (eg, reimbursable codes) to promote wider adoption of Z-codes. Because claims data are often one of the few standardized data sources that allow for comparison of needs across Medicaid enrollees, increasing adoption of Z-code documentation is an opportunity for states to better understand the social needs of their Medicaid population. While our study suggests modest improvements in Z-code documentation when social risk screenings are required, in March 2025, Centers for Medicare & Medicaid Services rescinded guidance to state Medicaid programs on covering health-related social needs services via Section 1115 waivers, which may lead to decreased social risk factor screening and documentation for perinatal enrollees and Medicaid enrollees.
Limitations
Our study has several limitations. First, we were unable to identify mechanisms to explain why changes in Z-code documentation might differ between Model A and Model B. Our results could be explained by different levels of health-related social needs, social risk factor screening patterns, or changes in visit patterns related to ACO adoption. Second, we lacked race and ethnicity data and were therefore unable to account for or stratify outcomes for people of color, who experience inequitably worse health outcomes because of structural racism. Third, the COVID-19 pandemic may have differentially affected ACO vs non-ACO provider groups, though sensitivity analysis findings excluding 2020 were robust. Fourth, we could not account for variation in social risk screening practices across ACOs or population prevalence of social risk factors. It is also possible that non-ACO providers or health systems had social risk screening practices in place. Fifth, Z-code documentation alone may not necessarily reflect social risk factor screening; rather, it is a proxy. Medicaid ACOs may screen for (eg, via paper social risk factor screening tools) and document (eg, in electronic health records) social risk factors, but not use Z-codes, and this could differ by Medicaid ACO. While our study period precedes coding for social risk factor screening (such as Healthcare Common Procedure Coding System code G0136, “administration of a standardized, evidence-based social determinants of health risk assessment”), which were introduced in 2024, future research should assess impacts on screening. Sixth, while our estimates are lower than some prevalences of social risk factors among Medicaid-insured people and Medicaid-insured pregnant people,2,9 they are consistent with estimates of Z-code documentation, therefore likely reflecting low Z-codes uptake.10,11,15,18,33 Documentation may also not indicate that a patient's unmet health-related social need is being met, and some providers may not screen for social risk factors if they lack resources to address them. Seventh, we were unable to examine changes for other specific social risk factor domains because of smaller sample sizes or non-parallel pre-intervention trends.
Conclusion
In Massachusetts, Medicaid ACO implementation was associated with modest, significant increases in Z-code documentation in the prenatal period—particularly for food insecurity—for both PCP-led model and health system/MCO partnership ACO model types, with some variation by model type. Requirements of social risk factor screening and documentation as part of a Medicaid ACO contract may increase adoption of Z-codes among perinatal populations, regardless of model design. As more states consider adoption or refinement of their Medicaid ACOs to potentially account for unmet social needs, multiple strategies may be needed to improve adoption of Z-code documentation.
Supplementary Material
Acknowledgments
None.
Contributor Information
Kevin H Nguyen, Department of Health Law, Policy, and Management, Boston University School of Public Health, Boston, MA 02118, United States.
Kenneth Lim, Department of Health Law, Policy, and Management, Boston University School of Public Health, Boston, MA 02118, United States.
Kathryn Thompson, Department of Health Law, Policy, and Management, Boston University School of Public Health, Boston, MA 02118, United States; Department of Community Health Sciences, Boston University School of Public Health, Boston, MA 02118, United States.
Sarah H Gordon, Department of Health Law, Policy, and Management, Boston University School of Public Health, Boston, MA 02118, United States.
Collette N Ncube, Department of Epidemiology, Boston University School of Public Health, Boston, MA 02118, United States.
Lois McCloskey, Department of Community Health Sciences, Boston University School of Public Health, Boston, MA 02118, United States.
Megan B Cole, Department of Health Law, Policy, and Management, Boston University School of Public Health, Boston, MA 02118, United States; Department of Population Medicine, Harvard Medical School & Harvard Pilgrim Health Care Institute, Boston, MA 02215, United States.
Supplementary material
Supplementary material is available at Health Affairs Scholar online.
Funding
Research reported in this publication was supported by the National Institute on Minority Health and Health Disparities of the National Institutes of Health under Award Number 5R01MD017703. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Notes
- 1. Medicaid and CHIP Payment and Access Commission . Report to Congress on Medicaid and CHIP. Chapter 5, Medicaid's role in maternal health. MACPAC; 2020. https://www.macpac.gov/wp-content/uploads/2020/06/June-2020-Report-to-Congress-on-Medicaid-and-CHIP.pdf
- 2. Nguyen KH, Cole MB. Social risk factors, health insurance coverage, and inequities in access to care. Am J Prev Med. 2025;68(1):145–153. 10.1016/j.amepre.2024.09.005 [DOI] [PubMed] [Google Scholar]
- 3. Alderwick H, Gottlieb LM. Meanings and misunderstandings: a social determinants of health lexicon for health care systems. Milbank Q. 2019;97(2):407–419. 10.1111/1468-0009.12390 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Kaiser Family Foundation . States that reported Accountable Care Organizations in place. 2023. Accessed May 2, 2025. https://www.kff.org/medicaid/state-indicator/states-that-reported-accountable-care-organizations-in-place/?currentTimeframe=0&sortModel=%7%22colId%22%22Location%22%22sort%22%22asc%22%7D
- 5. Cole MB, Kim J, Gordon SH, et al. Massachusetts Medicaid ACO program may have improved care use and quality for pregnant and postpartum enrollees. Health Aff (Millwood). 2024;43(9):1209–1218. 10.1377/hlthaff.2024.00230 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Rosenthal MB, Alidina S, Ding H, Kumar A. Realizing the potential of accountable care in Medicaid. New York: The Commonwealth Fund. Updated 2023. Accessed August 5, 2026. https://www.commonwealthfund.org/publications/issue-briefs/2023/apr/realizing-potential-accountable-care-medicaid
- 7. Sabatino MJ, Sullivan K, Alcusky MJ, Nicholson J. Identifying and addressing health-related social needs: a Medicaid member perspective. BMC Health Serv Res. 2024;24(1):1203. 10.1186/s12913-024-11605-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Tobin-Tyler E, Ahmad B. Marrying value-based payment and the social determinants of health through Medicaid ACO's. Milbank Memorial Fund. Updated May 2020. Accessed August 5, 2026. https://www.milbank.org/wp-content/uploads/2020/05/Medicaid-AC0s-and-SD0H.ver5_.pdf [Google Scholar]
- 9. Daw JR, Underhill K, Liu C, Allen HL. The health and social needs of Medicaid beneficiaries in the postpartum year: evidence from a multistate survey. Health Aff (Millwood). 2023;42(11):1575–1585. 10.1377/hlthaff.2023.00541 [DOI] [PubMed] [Google Scholar]
- 10. Truong HP, Luke AA, Hammond G, Wadhera RK, Reidhead M, Maddox KEJ. Utilization of social determinants of health ICD-10 Z-codes among hospitalized patients in the United States, 2016-2017. Med Care. 2020;58(12):1037–1043. 10.1097/MLR.0000000000001418 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Torres JM, Lawlor J, Colvin JD, et al. ICD social codes. Med Care. 2017;55(9):810–816. 10.1097/MLR.0000000000000764 [DOI] [PubMed] [Google Scholar]
- 12. Cole MB, Nguyen KH, Byhoff E, Murray GF. Screening for social risk at federally qualified health centers: a national study. Am J Prev Med. 2022;62(5):670–678. 10.1016/j.amepre.2021.11.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Nguyen KH, Giron NC, Cole MB. National social risk factor screening rates among federally qualified health center patients. J Gen Intern Med. 2024;39(13):2621–2624. 10.1007/s11606-024-08879-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Weeks WB, Cao SY, Lester CM, Weinstein JN, Morden NE. Use of Z-codes to record social determinants of health among fee-for-service Medicare beneficiaries in 2017. J Gen Intern Med. 2020;35(3):952–955. 10.1007/s11606-019-05199-w [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Aswani MS, Do LA, Shafer PR. Use of social determinants of health Z codes was sparse, 2016-22. Health Aff (Millwood). 2025;44(5):631–635. 10.1377/hlthaff.2024.01033 [DOI] [PubMed] [Google Scholar]
- 16. Nguyen KH, Gordon SH, Lim K, Thompson KD, Ncube CN, Cole MB. Medicaid Accountable Care Organization implementation and perinatal claims documentation of social risk factors. JAMA Netw Open. 2025;8(4):e255999. 10.1001/jamanetworkopen.2025.5999 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Schiavoni KH, Helscel K, Vogeli C, et al. Prevalence of social risk factors and social needs in a Medicaid Accountable Care Organization (ACO). BMC Health Serv Res. 2022;22(1):1375. 10.1186/s12913-022-08721-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Ubri PS, Bailey L, Melendez M, Sawyer J, Stead M. The role of state policy in use of Z Codes to document social need in Medicaid data. Updated March 2022. Accessed August 5, 2026. https://www.norc.org/content/dam/norc-org/pdfs/The%20Role%20of%20State%20Medicaid%20Policy%20in%20Documentation%20of%20SDOH%20in%20Medicaid%20Data_032422.pdf
- 19. Rutledge RI, Romaire MA, Hersey CL, Parish WJ, Kissam SM, Lloyd JT. Medicaid Accountable Care Organizations in four states: implementation and early impacts. Milbank Q. 2019;97(2):583–619. 10.1111/1468-0009.12386 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Matulis R, Lloyd J. The history, evolution, and future of Medicaid accountable care organizations. Center for Health Care Strategies. Updated February 2018. Accessed May 2, 2025. https://www.chcs.org/resource/history-evolution-future-medicaid-accountable-care-organizations/
- 21. Blue Cross Blue Shield Foundation of Massachusetts . What to know about MassHealth ACOs. 2023. Accessed August 5, 2026. https://www.bluecrossmafoundation.org/sites/g/files/csphws2101/files/2023-10/ACO%20Primer_2023_FINAL.pdf
- 22. Ranchoff BL, Geissler KH, Attanasio LB, Jeung C. Association of Medicaid Accountable Care Organizations and postpartum mental health care utilization. Health Serv Res. 2025;60:e14421. 10.1111/1475-6773.14421 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Browne J, Mccurley JL, Fung V, Levy DE, Clark CR, Thorndike AN. Addressing social determinants of health identified by systematic screening in a Medicaid Accountable Care Organization: a qualitative study. J Prim Care Community Health. 2021;12:2150132721993651. 10.1177/2150132721993651 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Cole MB, Lim K, Nguyen KH, et al. Medicaid accountable care model designs and maternal health measures. JAMA Netw Open. 2025;8(10):e2536565. 10.1001/jamanetworkopen.2025.36565 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Anderson KE, McGinty EE, Presskreischer R, Barry CL. Reports of forgone medical care among US adults during the initial phase of the COVID-19 pandemic. JAMA Netw Open. 2021;4(1):e2034882. 10.1001/jamanetworkopen.2020.34882 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Bertoldo J, Wolfson JA, Sundermeir SM, et al. Food insecurity and delayed or forgone medical care during the COVID-19 pandemic. Am J Public Health. 2022;112(5):776–785. 10.2105/AJPH.2022.306724 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Houston R, Lloyd J, Crumley D, Matulis R, Keehn A. The MassHealth Accountable Care Organization program: uncovering opportunities to drive future success. Blue Cross Blue Shield of Massachusetts Foundation, May. Updated 2021. Accessed May 19, 2025. https://www.bluecrossmafoundation.org/sites/g/files/csphws2506/files/2021-05/ACO_Qual-Assess_FullReport_Final_0.pdf
- 28. Nianogo RA, Wang MC, Basurto-Davila R, et al. Economic evaluation of California prenatal participation in the Special Supplemental Nutrition Program for Women, Infants and Children (WIC) to prevent preterm birth. Prev Med. 2019;124:42–49. 10.1016/j.ypmed.2019.04.011 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Reyes AM, Akanyirige PW, Wishart D, et al. Interventions addressing social needs in perinatal care: a systematic review. Health Equity. 2021;5(1):100–118. 10.1089/heq.2020.0051 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Muoto I, Luck J, Yoon J, Bernell S, Snowden JM. Oregon's coordinated care organizations increased timely prenatal care initiation and decreased disparities. Health Aff (Millwood). 2016;35(9):1625–1632. 10.1377/hlthaff.2016.0396 [DOI] [PubMed] [Google Scholar]
- 31. Stanhope KK, Goebel A, Simmonds M, et al. The impact of screening for social risks on OBGYN patients and providers: a systematic review of current evidence and key gaps. J Natl Med Assoc. 2023;115(4):405–420. 10.1016/j.jnma.2023.06.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Tailored prenatal care delivery for pregnant individuals: ACOG Clinical Consensus No. 8. Obstet Gynecol. 2025;145(5):565–577. 10.1097/AOG.0000000000005889 [DOI] [PubMed] [Google Scholar]
- 33. Gibbons JB, Cram P, Meiselbach MK, Anderson GF, Bai G. Comparison of social determinants of health in Medicaid vs commercial health plans. Health Aff Sch. 2023;1(6):qxad074. 10.1093/haschl/qxad074 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.