Characteristics of Specialty Mental Health Provider Networks in Oregon Medicaid

Abstract

Objective:

Provider networks are narrower for mental health than for other specialties. Their influence on access to care is potentially greater in Medicaid because enrollees are generally limited to contracted providers, without “out-of-network” options for non-emergent mental health care. This study describes variation in specialty mental health provider networks using claims-based metrics.

Methods:

In a cross-sectional analysis of 2018 Oregon Medicaid claims data, adults aged 18–64 with a mental health diagnosis were identified. “In-network” providers were identified as those associated with any medical claims filed for at least 5 unique Medicaid beneficiaries enrolled in a health plan (Coordinated Care Organization, CCO) during the study period. Specialty mental health prescribers were categorized as prescribers (psychiatrists and mental health nurse practitioners) and non-prescribers (therapists, counselors, clinical nurse specialists, psychologists, and social workers). Measures of network composition, provider to population ratio, continuity, and concentration of care were calculated at the CCO level, and the correlation between these measures was estimated to describe the degree to which they capture unique dimensions of provider networks.

Results:

Across all 15 CCOs, the number of prescribing providers per 1000 patients was relatively stable. CCOs that expanded their networks did this by increasing provider-to-population ratios for non-prescribing providers. There were moderately negative correlations between non-prescriber provider to population ratio and proportions of visits with prescribers, as well as with usual provider continuity.

Conclusions:

This analysis advances future research and policy applications by offering a more nuanced view of provider network measurement and describing empirical variation across networks.

A provider network is a group of all health care providers and facilities that a health plan has contracted with to provide medical care to its members. The numbers and types of “in-network” mental health providers with which a health plan contracts may vary, though evidence suggests that provider networks tend to be narrower for mental health than for other specialties,¹ in part because psychiatrist participation in health plans is the lowest among specialists and continues to decline.^2,3 A chief concern among policymakers is that health plans with only a limited set of in-network providers may impact access to care and treatment outcomes.^4,5,6 In particular, those who need broader or more integrated networks for care (for example, more complex individuals with serious mental illness) may be adversely affected when provider networks are too restrictive.^7,8 Conversely, there may be efficiencies from certain network designs, as health plans may contract with a set of high-quality, low-cost providers, utilizing behavioral specialists, therapists, psychologists, advanced nurse practitioners, and social workers rather than psychiatrists.⁹

Addressing access to mental health services through the design of adequate provider networks is particularly important in Medicaid. Medicaid is currently the single largest payer for mental health care services in the United States, covering at least one-fifth of all Americans with mental health disorders and continuing to increase under Medicaid expansion. The challenges in accessing mental health services are magnified in the Medicaid program for a number of reasons, including lower willingness among psychiatrists to accept Medicaid than other forms of insurance.^10,11 Geographic access to Medicaid mental health providers is even more restricted, with over one-third of counties –the majority of them rural– lacking outpatient mental health facilities that accept Medicaid.¹² Moreover, social and medical complexities facing the patient population contribute to greater care needs and barriers in accessing care. Finally, the influence of provider networks is potentially greater in the Medicaid program as cash pay is often unaffordable, and because enrollees are generally limited to contracted providers and do not typically have cost-sharing options for going “out-of-network” for non-emergent care.^13,14

Thus, understanding which provider network strategies lead to increased efficiencies versus suboptimal care is a critical question for Medicaid enrollees, particularly as it relates to mental health care. A number of recent policies have been implemented to regulate and monitor provider networks. Medicaid managed care programs, for example, are required to use any of a range of quantitative standards to monitor whether their networks include sufficient numbers and types of mental health providers.¹⁵ However, network adequacy remains a vaguely defined concept, and existing standards do not always translate to acceptable levels of access.¹⁶ One persistent challenge is a relative dearth of evidence on appropriate metrics to describe network characteristics beyond network breadth,^1,17,18 which measures theoretical access (proportion of covered providers within a given geographical area) but may not reflect other important measures of network adequacy or quality. Additionally, regulators often employ only a single metric, like provider-to-enrollee ratio, that may not take into account additional dimensions of provider networks.^15,19

Studies focused on hospital-based and medical specialty provider networks have found that claims-based measures of provider networks represent provider-level coordination and care delivery patterns well.^20,21,22 Only one prior study used claims-based measures to identify connections between providers caring for patients with severe mental illness in Colorado, finding variation across networks that is not captured by traditional measures of care.²³ In this setting, we use 2018 Medicaid claims data in Oregon to test several alternative network measures, and to describe empirical variation in mental health specialty networks. We describe mental health network variation along a number of claims-based dimensions, including composition of provider types,²⁴ provider to population ratio, continuity of care (as operationalized by the Usual Provider Continuity index),²⁵ and concentration of care (via the Herfindahl-Hirschman Index). We then estimate the correlation between these measures to describe the degree to which they capture unique aspects of provider networks. Taken together, these analyses can advance future research and policy applications by offering a more nuanced view of measuring provider networks and the potential tradeoffs associated with different network structures and characteristics.

Methods

Setting

Oregon’s Medicaid managed care program reflects the state of health care delivery in many states. Oregon’s Medicaid program is managed through fifteen Coordinated Care Organizations (CCOs), with substantial variability in provider networks and no variability in patient cost-sharing. Implemented in 2012, CCOs cover distinct geographic regions and combine elements of Medicaid managed care organizations and accountable care organizations in how they accept financial risk and pay for care.^26–28 Our focus on a single state allows us to study these provider networks within a relatively homogenous administrative setting.

Study population

We identified adults aged 18–64 enrolled in Oregon’s Medicaid program for 9 months or longer between January 1, 2018, and December 31, 2018, regardless of whether the enrollment was continuous. We excluded all dual-eligible and fee-for-service beneficiaries, due to data availability limitations and differences in population, health care needs, service utilization, claims, and reimbursement (Appendix A). We also excluded the 4% of enrollees that switched CCOs during the study period due to a move to a different coverage area. We then restricted to patients with a mental health diagnosis for any encounter during the study year, including those who had primary psychiatric diagnoses listed in Clinical Classification System Refined categories MBD001-MBD013, MBD027, or EXT021.

Provider network construction

We constructed a sample of in-network providers empirically using medical claims data. Providers were considered to be “in-network” for a given CCO if they were associated with any medical claims filed for at least 5 unique Medicaid beneficiaries enrolled in that CCO during the study period..²⁹ Using information from the National Plan and Provider Enumeration System (NPPES), we identified the following types of individual specialty mental health providers: mental health prescribers (psychiatrists and mental health nurse practitioners); and non-prescribing mental health specialists (therapists and counselors, clinical nurse specialists, psychologists, and social workers) (Appendix B).

Other variables

We identified unique outpatient mental health visits based on member, service date, and performing provider, and included Evaluation & Management and Psychotherapy visits only.

Patient demographics included age, sex, self-reported race and ethnicity, and rural or urban residence as defined by the Oregon Office of Rural Health. We also measured total months enrolled in Medicaid during the study year and total number of condition categories from the Chronic Illness and Disability Payment System (CDPS).³⁰ The Chronic Illness and Disability Payment System (CDPS) a claims-based model based on 20 diagnosis codes, widely used by Medicaid programs for risk adjustment.

Analyses

We calculated summary measures of mental health visits, as well as measures of network composition, provider to population ratio, continuity, and concentration of care (Table 1 provides overview and detailed definitions for these measures).

Table 1.

Overview of selected network measures

Network Measure	Definition	Justification	Advantages	Limitations	Interpretation
Composition	Total number of in-network mental health providers with a specific specialty, divided by the total number of all in-network mental health providers We also stratify network composition via measures that relate to prescribing vs non-prescribing providers	Describes an element of network structure, or the resources available to provide access to care	Easy to apply and interpret	Does not take into account services provided or density of care
Percent of prescribing providers	Total number of in-network mental health providers with a specialty that allows them to prescribe medications, divided by the total number of all in-network mental health providers
Proportion of mental health visits with prescribers	Number of outpatient mental health visits with an in-network provider who can prescribe medications, divided by the total number of outpatient mental health visits with any in-network provider
Provider to population ratio	Total number of in-network providers divided by the total number of patients with a mental health diagnosis	Proxy measure for network capacity –whether there are sufficient numbers of providers for patient population	Used in real-world policy and regulatory settings to assess network adequacy	Does not take into account services provided or density of care	Higher provider to population ratio suggests greater capacity for mental health care
Usual Provider Continuity (UPC)	Total number of outpatient mental health visits with a patient’s usual in-network provider, divided by the total number of outpatient mental health visits with any in-network provider.	Continuity is associated with higher quality of care, including for those with serious mental illness	Validated measure; reflects the “density” of a network	Lower continuity may not directly correlate with poorer quality in team-based mental health deliver; may also reflect a network outcome	Ranges from 0 (no visits with the same provider) to 1 (all visits with same provider).
Herfindahl-Hirschman index (HHI)	The share of all outpatient mental health visits for a given CCO, for each in-network provider, then squared and summed across providers.	Common measure of market concentration	Validated measure; has been applied to assess concentration of care	Offers limited information on aspects of care coordination; may also reflect a network outcome	Ranges from zero to 10,000. Lower HHI values indicate that outpatient mental health visits are spread relatively evenly across a large number of providers. Higher HHI values suggest that visits are concentrated among a small set of providers.

Notes: All measures are analyzed at the plan (CCO) level.

First, we measured network composition – the inclusion of different types and specialties of providers.^31,32 Specifically, we computed the percentage of all mental health specialists of different types (prescribers including psychiatrists and mental health nurse practitioners, and non-prescribing clinicians including psychologists, counselors and therapists, clinical nurse specialists, and social workers). For example, a provider network could technically be large but offer a lower supply of prescribers vs. non-prescribing clinicians. To this end, we also calculated the percentage of mental health visits with prescribers.

Second, provider to population ratio measures how many providers are available in a service area and is commonly used in the literature,³³ as well as by state and federal regulators,³⁴ to assess the sufficiency of a provider network. We defined this measure as the total number of in-network providers per 1000 enrollees with any mental health condition, as derived from Medicaid claims.

Third, we evaluated continuity of care at the network-level via the Usual Provider Continuity (UPC) index, a frequently used continuous measure representing the proportion of all visits that occur with a patient’s usual provider. Greater continuity of care has been associated with better patient outcomes,³⁵ including for those with severe mental illness.^36,37 Using claims data, we identified the ‘usual in-network provider’ for each patient as the provider that the patient saw the most frequently during the study period. In the case of a tie, we randomly assigned the usual provider. A UPC index score, ranging from 0 (no visits with the same provider) to 1 (all visits with same provider), was calculated for each patient and the median score calculated at the CCO level.

To describe the concentration of mental health outpatient visits across providers, we computed the share of total mental health outpatient visits within a CCO that was provided by each in-network mental health specialist. A higher concentration of care has been associated with lower costs and higher quality,²² and prior work has shown that mental health provision in Medicaid may be concentrated among specific subsets of providers.³⁸ We thus computed the Herfindahl-Hirschman index (HHI), a commonly accepted measure of market concentration that has been applied to assess concentration of care and degree of care coordination.^21,39,40 HHI was calculated by squaring the share of visits to each in-network clinician and summing across all in-network clinicians that saw patients within a given CCO. Values range from 0 to 10,000, with larger values representing a more concentrated care delivery pattern, i.e., a smaller set of in-network clinicians comprising a higher share of mental health visits.

In order to assess the level of association between these provider network dimensions, we computed Pearson correlations among the various measures. Finally, in sensitivity analyses, we evaluated additional definitions of care continuity (Appendix D), and other measures including proportion of patients with 1 or more prescriber visit, and UPC calculated separately for prescribers and non-prescribers (Appendix E).

Descriptive analyses were conducted using R, version 4.0.3 from December 2020 through August 2021. The study protocol was deemed exempt by the Oregon Health & Science University’s Institutional Review Board (#1017760).

Results

Our final cohort included 100,515 Medicaid enrollees who had at least one mental health condition. The number of study cohort members enrolled in each CCO ranged from 933 to 35,994 (Appendix C). Across the 15 CCOs, the average age of enrollees ranged from 37.5 to 40.3, and a majority were female (64.0% - 68.7%), white (49.1% - 62.3%). The proportion of patients living in urban areas varied substantially, from 0% to 94.8%. The average number of health conditions ranged from 2.3 to 2.7, and the proportion of those with mental health diagnoses that had a serious mental illness ranged from 9.3% to 14.8%. Across all CCOs, we identified a total of 292 psychiatrists, 259 nurse practitioners specializing in mental health, 285 psychologists, 3,379 counselors and therapists, 8 clinical nurse specialists, and 797 social workers as in-network in 2018.

Across CCOs, the majority of specialty mental health networks was comprised of non-prescribing mental health specialists, such as counselors, therapists, and social workers (Figure 1). Counselors and therapists comprised the largest proportion of in-network mental health specialty providers, ranging from 53.9–81.7% (mean, 69.2%), followed by social workers (8.6–26.1% of network providers, mean 15.3%), and psychologists (1.6–11.9%, mean 4.9%). Prescribers comprised a consistently small proportion of mental health specialty networks across CCOs, with psychiatrists representing 1.9–13.0% (mean, 5.4%) and psychiatric advanced practice providers representing 3.5–6.8% (mean, 5.1%) of providers. Figure 2 shows that as networks increase in size, they generally do so by incorporating more non-prescribing providers. Also, higher plan enrollment of individuals with mental health conditions was not associated with higher provider to population ratios. Provider to population ratios ranged from 27.8 to 92.2 for non-prescribing mental health providers (mean, 52.8 providers per 1000 mental health enrollees) and from 3.3 to 8.9 for prescribers (mean, 6.0 providers per 1000 mental health enrollees) (Table 2).

Figure 1. Composition of specialty mental health provider networks, by CCO health plan.

Note: Figure 1 shows the composition of specialty mental health provider networks across health plans in Oregon’s Medicaid program. Providers with specialties that allow them to prescribe medications are shown in shades of brown. Non-prescribing provider specialties are shown in shades of blue. The majority of plans’ mental health workforce is comprised of behavioral health specialists, such as counselors and therapists, ranging from 54–81% of in-network specialty mental health providers, while psychiatrists represented between 2%−13% of in-network mental health providers.

Figure 2. Provider to population ratio of prescribing vs. non-prescribing mental health networks, by CCO health plan.

Note: Figure 2 displays provider to population ratios (for prescribing clinicians, Y-axis; and non-prescribing clinicians, X-axis) at the CCO health plan level. The size of each bubble denotes the size of the Medicaid enrollee population for that given CCO.

Table 2.

Summary metrics for mental health specialty networks, by CCO health plan

CCO	Total mental health visits (N)	Total prescriber visits (N)	Total non-prescriber visits (N)	Average prescriber visits among mental health patients	Average non-prescriber visits among mental health patients	Prescribing providers (per 1000 mental health patients)	Non-prescribing providers (per 1000 mental health patients)	Prescribing providers (%)	Proportion of mental health visits with prescribers (%)	UPC Index (median)	HHI
A	128,031	16,107	111,924	1.2	8.4	6.1	54.6	10.0	12.6	0.98	23
B	11,957	1,964	9,993	0.7	3.7	8.9	63.8	12.2	16.4	0.94	192
C	23,533	2,910	20,623	0.6	4.4	5.5	47.0	10.5	12.4	0.93	131
D	15,354	1,954	13,400	0.8	5.6	7.1	71.8	8.9	12.7	0.77	245
E	28,090	3,194	24,896	0.6	4.9	5.3	62.8	7.8	11.4	0.86	126
F	74,300	12,091	62,209	1.2	6.2	4.9	40.4	10.9	16.3	1.00	70
G	224,818	34,662	190,156	1.0	5.3	7.7	46.2	14.2	15.4	0.95	4
H	29,787	5,174	24,613	0.8	3.9	7.8	38.1	17.1	17.4	1.00	126
I	23,749	2,248	21,501	0.6	5.9	6.8	67.6	9.2	9.5	1.00	121
J	9,107	2,124	6,983	1.0	3.3	3.3	47.3	6.6	23.3	1.00	366
K	6,092	1,147	4,945	1.0	4.2	5.9	44.6	11.7	18.8	1.00	435
L	4,171	454	3,717	0.5	4.0	7.5	92.2	7.5	10.9	0.90	377
M	19,516	4,602	14,914	1.3	4.1	4.2	27.8	13.0	23.6	1.00	250
N	41,625	3,951	37,674	0.7	6.2	4.4	48.7	8.4	9.5	1.00	103
O	8,995	2,239	6,756	1.0	2.9	4.3	39.8	9.8	24.9	1.00	375

^*CCO = Coordinated Care Organization (Oregon’s Medicaid managed care organizations). UPC index = Usual Provider Continuity Index, a continuous measure representing the proportion of all visits that occur with a patient’s usual provider. HHI = Herfindahl-Hirschman index (HHI), a commonly accepted measure of market concentration, with larger numbers represented more concentrated care of mental health patients among a set of providers.

Table 2 dispays summary measures of mental health visits across CCOs, as well as network metrics including variation in the proportion of visits with mental health prescribers, median Usual Provider Continuity index, and concentration of care as operationalized by HHI. The proportion of visits with mental health prescribers ranged from 9.5% to 24.9% (mean, 15.7%), suggesting that across CCOs, most visits were conducted with non-prescribing mental health clinicians. The UPC ranged from 0.77 to 1.00, suggesting that majority of outpatient mental health visits were made to a regular provider. While outpatient mental health care appeared to be highly unconcentrated among providers, some CCO networks concentrated care among a smaller group of providers than did others (HHI range, 435 vs. 4).

We found moderately negative correlations between provider to population ratio for non-prescribers and prescriber composition (−0.5) and proportion of visits with prescribers (−0.6) (Figure 3). In other words, networks with fewer prescribing clinicians were characterized by utilization patterns whereby mental health services were more frequently managed by non-prescribing clinicians. We also found a moderately negative correlation between non-prescribing provider to population ratio and usual provider continuity (−0.6). Our results were robust to sensitivity analyses (Appendices D and E).

Figure 3. Correlation matrix for provider network-level metrics.

Note: Figure 3 displays correlation coefficients between metrics (with higher numbers indicating a stronger association in either the negative direction (−1.0) or positive direction (+1.0)). Increasing size of non-prescribing specialty mental health clinicians (number of providers per 1000 mental health enrollees) is negatively correlated with prescriber composition, visits with prescribers, and usual provider continuity. That is, with a greater proportion of non-prescribing clinicians, there appears to be relatively lower continuity of care. Asterisks denote a statistically significant correlation.

Discussion

In this cross-sectional study, we used 2018 Oregon Medicaid claims data to describe variation in the characteristics of specialty mental health provider networks via a number of network-based metrics, including provider to population ratio, network composition, continuity, and concentration of care. We found that across CCOs, the majority of the mental health workforce that was actively seeing patients consisted of non-prescribing mental health providers, and there was considerable variation in the size of these non-prescriber networks. However, the size of networks for prescribing mental health providers remained relatively stable across CCOs.

Provider networks are ultimately limited by workforce supply constraints and provider participation, which may be driving our finding of relatively stable proportions of in-network mental health prescribers across CCOs. A broad body of literature suggests that low psychiatrist participation within Medicaid is attributable to a number of factors,^,41 including relatively low reimbursement rates; administrative burdens and delays; clinical case complexity within Medicaid, which disproportionately serves those with severe mental illness; as well as an aging workforce and overall shortage in psychiatrist supply.⁴² We find evidence that CCOs, to some extent, may make up for prescriber shortages by contracting with variable numbers and types of non-prescribing mental health providers, such as psychologists, nurses, social workers, counselors, and therapists. While these workforce types may be active complements for prescribers, there may also be tradeoffs, including, as we found, potentially lower continuity of care particularly if patients rely on non-prescribing clinicians who may disproportionately experience burnout, fragmented care, and high turnover.⁴³

There are additional reasons that may explain the significant correlations we observed between some measures, including between non-prescribing provider to population ratio and usual provider continuity. For example, as CCOs increase network size primarily by increasing proportions of non-prescribing clinicians, it is possible that lower continuity of care may be due to greater provider choice or increased access to care due to larger provider networks. We are unable to distinguish these causal mechanisms in this analysis.

It is yet unclear the extent to which different mental health provider network characteristics may be associated with quality of care as well as downstream patient outcomes. For instance, a highly concentrated network structure, in which a small set of in-network providers is seeing a large share of mental health visits, may reflect greater specialization of care and be associated with higher quality. Alternatively, a loosely concentrated network structure, in which many in-network providers are seeing small shares of mental health visits, may be associated with lower provider burnout and greater continuity. The extent to which there are tradeoffs inherent in these network structures is yet unknown. For example, provider network breadth reflects known tradeoffs between access and cost containment; Medicaid managed care plans may choose to offer a more selective network in order to steer enrollees to lower cost or higher quality providers, by providing enrollees less freedom of choice. Similarly, networks with greater access to mental health prescribers vs. non-prescribers could reflect care delivery specialization and population health needs. Care delivery models for those with serious mental illness may rely on different network structures than for those with other mental health conditions.^23,44 Additional research is therefore needed to understand these tradeoffs, and what might constitute an evidence-based benchmark for high-quality networks.

One chief advantage of this work is the use of administrative claims data to describe realized access to provider networks. While federal and state regulators lean on provider directories to monitor the adequacy of provider networks, prior studies have shown that provider directories are highly inaccurate,⁴⁵ often listing incorrect contact information or including health care professionals who are inactive, do not accept certain insurance types, or have closed panels. Given that provider directories are challenging to validate and maintain in real-time, claims data may help policymakers measure more directly how provider networks and access to care intersect.

This study has important limitations. Most notably, we purposefully excluded primary care physicians (PCPs) from our analysis to focus on measures of specialty mental health networks, in line with many states that separately monitor provider networks for primary care and mental health.¹⁵ It is possible that including PCPs may have altered our network measures. Second, we do not incorporate geographic measures related to provider access, such as travel time and distance, nor were we able to incorporate non-claims based measures like appointment wait times, all of which could reflect additional considerations that distinguish mental health provider networks. Finally, within the scope of this paper, we did not assess the relationship between provider network characteristics and quality of care or patient outcomes. However, our findings lay a solid foundation on which to conduct these future analyses.

Conclusion

Using 2018 Oregon Medicaid claims data, we described variation in the characteristics of specialty mental health provider networks via a number of network-based metrics, including provider to population ratio, network composition, continuity, and concentration of care. Across 15 health plans, the number of prescribing providers per 1000 patients was relatively stable. As CCOs that expanded their networks did this almost exclusively by increasing the number of non-prescribing providers per 1000 patients. We found moderately negative correlations between non-prescriber provider to population ratio and proportions of visits with prescribers, as well as with usual provider continuity. Taken together, our analyses advance future research and policy applications by offering a more nuanced view of provider network measurement and empiric variation in network structures and characteristics.

Highlights:

• Provider networks are narrower for mental health than for other specialties

• Across 15 health plans in Oregon’s Medicaid program, networks were stable in numbers of mental health prescribers per 1000 mental health enrollees.

• As networks increased in size, they generally did so by expanding the provider to population ratio for non-prescribing providers.