Association between physician practice Medicaid acceptance and employing nurse practitioners and physician assistants: A longitudinal analysis

Abstract

Background:

Access to care is often a challenge for Medicaid beneficiaries due to low practice participation. As demand increases, practices will likely look for ways to see Medicaid patients while keeping costs low. Employing nurse practitioners (NPs) and physician assistants (PAs) is one low-cost and effective means to achieve this. However, there are no longitudinal studies examining the relationship between practice Medicaid acceptance and NP/PA employment.

Purpose:

The purpose of this study was to examine the association of practice Medicaid acceptance with NP/PA employment over time.

Methods:

Using SK&A data (2009–2015), we constructed a panel of 102,453 unique physician practices to assess for changes in Medicaid acceptance after newly employing NPs and PAs. We employed practice-level fixed effects linear regressions.

Results:

Our results showed that, among practices employing both NPs and PAs, there was a roughly 2% increase in the likelihood of Medicaid participation over time. When stratifying our sample by practice size and specialty, the positive correlation localized to small primary care and medical practices. When both NPs and PAs were present, small primary care practices had a 3.3% increase and small medical practices had a 6.9% increase in the likelihood of accepting Medicaid.

Conclusion:

NP and PA employment was positively associated with increases in Medicaid participation.

Practice Implications:

As more individuals gain coverage under Medicaid, organizations will need to decide how to adapt to greater patient demand. Our results suggest that hiring NPs and PAs may be a potential lower cost strategy to accommodate new Medicaid patients.

Access to care for Medicaid beneficiaries has lagged behind individuals covered by Medicare and private insurance (Hing et al., 2015; Polsky et al., 2017). Research finds that Medicaid patients constitute less than 10% of physicians’ patient panels, and many physicians report not accepting new Medicaid patients (Hing et al., 2015; Neprash et al., 2018). However, this persistent access gap may be a growing issue, given the increasing number of individuals covered under Medicaid since the Affordable Care Act expansions. There are almost 71 million Medicaid beneficiaries (Medicaid.gov, 2020), but greater access to Medicaid coverage may not always translate to access to care.

Workforce challenges, such as provider shortages and geographic maldistribution, have increased interest in relying more on nurse practitioners (NPs) and physician assistants (PAs) to deliver ambulatory care, especially for vulnerable populations (Everett et al., 2009; Maier et al., 2017). Currently, there are over 290,000 NPs and 131,000 PAs in the United States (American Academy of Physician Assistants, 2019; American Association of Nurse Practitioners, 2020), with both workforces continuing to expand (Salsberg, 2018). NP and PA employment in physician practices is growing (Barnes et al., 2018; Martsolf et al., 2018), and a body of evidence shows that these clinicians offer safe, high-quality care, which can be equivalent to comparable physician-delivered services (Kurtzman & Barnow, 2017; Stanik-Hutt et al., 2013; Xue et al., 2016). Moreover, interdisciplinary models of care that include NPs and PAs have the potential to alleviate workforce challenges and enhance care delivery (Everett et al., 2013; Richards & Polsky, 2016). Yet, the research and policy communities have an incomplete understanding of if and how these evolutions in provider configurations map to access to care improvements for Medicaid beneficiaries.

Prior research has found NPs and PAs are more likely than physicians to care for patients covered under public insurance (Benitez et al., 2015; Buerhaus et al., 2015; Everett et al., 2009). Likewise, other studies have found evidence consistent with improved Medicaid access among physician practices that employ NPs and PAs (Barnes et al., 2017; Richards & Polsky, 2016). However, most of this work is cross-sectional in design (Barnes et al., 2017; Buerhaus et al., 2015; Richards & Polsky, 2016) and limited to primary care (Buerhaus et al., 2015; Everett et al., 2009). Benitez et al.’ (2015) study is an exception in that it incorporated multiple years of the National Hospital Ambulatory Medical Care Survey, but the data lack a practice-level longitudinal component and are from a previous decade. For these reasons, the research and policy-making communities are generally reliant on constrained snapshots of the relationship between Medicaid participation and practice provider mix. We consequently can only draw cautious inferences due to the known limitations of cross-sectional analyses, and importantly, we have no empirical understanding of whether these key practice characteristics (i.e., Medicaid acceptance and provider configuration) evolve together over time.

Using a national, balanced panel of physician practices from 2009 to 2015, we were able to examine the associations between within-practice changes in Medicaid participation and newly employing NPs and PAs. Also, we provide new evidence of heterogeneity across practice size and specialty (i.e., primary care, single medical or surgical subspecialty, and multispecialty). As the presence of NPs and PAs increases in both primary and nonprimary care settings (Barnes et al., 2018; Martsolf et al., 2018), our practice-level longitudinal analyses represent a novel contribution to the existing literature, and the findings have relevance to contemporary health care delivery reform and policy debates.

Theoretical Basis for the Study

The underlying conceptual basis for this study considers that, in response to external market changes, such as growing patient demand and evolving payer mix, diversifying a practice’s provider configurations (i.e., hiring NPs and PAs) is one way of potentially lowering a practice’s marginal costs of delivering care while expanding its capacity to start seeing new patients. However, variations in reimbursement rates between private and public insurers may create challenges for physician practices when making management decisions surrounding insurance program participation (Richards, Nikpay, & Graves, 2016; Richards & Polsky, 2016). Specifically, the same patient visit will often be reimbursed at a higher rate if the visit is covered under private insurance versus Medicaid, which is well known for comparatively low payments for physician services. At the same time, 20% of the U.S. population is now covered by the Medicaid program (Kaiser Family Foundation, 2020). Such growth in recent years may be the impetus for many practices to begin identifying ways to accept Medicaid patients and/or planning to do so in the near future.

As noted above, one plausible strategy to increasing a practice’s capacity to serve Medicaid patients is to begin employing lower cost providers. The labor costs associated with hiring NPs and PAs (i.e., salaries), for example, are below those of physicians (Bureau of Labor Statistics, 2020), and quality of care has not been shown to be compromised when relying more on high-skill, nonphysician clinicians (Kurtzman & Barnow, 2017; Stanik-Hutt et al., 2013; Xue et al., 2016). Adopting a lower cost approach to care delivery (i.e., reducing the marginal costs of supplying patient visits) via employing NPs and PAs may consequently make it financially feasible for practices to grow their patient panel and/or accommodate a different mix of payers and thereby incorporate Medicaid patients into their revenue stream for the first time.

Both of these practice-level changes (i.e., introducing a new clinical staffing structure and newly participating in the Medicaid program) could be conceptualized as markers of practice transformation in relation to management and business strategies, and it is consequently our intent to empirically determine if they systematically happen in tandem among physician practices. A positive association between the two markers of significant practice change would then be consistent with an underlying practice management decision that is common to both of them.

Methods

This longitudinal study was a secondary analysis of a balanced panel (i.e., same exact practices are present in all years of data) of 102,453 physician practices from 2009 to 2015.

Study Data and Measures

The primary data sources were the SK&A physician and nurse files. SK&A (now OneKey; IQVIA, 2020) is a commercial market research firm that maintains data sets of office-based health care providers in all 50 states and Washington, DC; these data are updated twice yearly via telephone verification. Available practice-level variables include practice specialty, number and type of providers within the practice, practice integration status (i.e., horizontal or vertical), practice location (address, zip code), average daily practice volume, and practice Medicare and Medicaid acceptance. These data have been found to be comprehensive and representative of greater than 90% of ambulatory practices in the United States (Rhodes et al., 2014).

We combined separate SK&A files for odd-numbered years (i.e., 2009, 2011, 2013, 2015) to create one longitudinal set of data of physician practices, including the providers employed by the practice. Identification of individuals (i.e., physicians, NPs, PAs) within a given practice allowed for flexibility when examining provider configuration in a given year and over time. These data have been used in prior research examining access to care for Medicaid beneficiaries (Bond et al., 2017; Richards & Polsky, 2016; Richards, Saloner et al., 2016). Using common county-level FIPS codes, we merged the SK&A file with the Area Health Resource File to obtain county-level market characteristics for each practice.

Our main outcome variable was practice-reported Medicaid acceptance, which was a binary variable indicating whether a practice accepted Medicaid or not. Our main predictor variables were NP presence and PA presence in a practice. These were binary variables labeled as “1” if a practice employed at least one NP or one PA and zero otherwise. These were not mutually exclusive as some practices in our sample employed both types of providers, which we address in our analyses.

Data Analysis

We calculated summary statistics for all study variables. To examine our main association of interest, we employed a practice-level fixed effects (FE) linear regression model. The practice FEs allowed us to estimate the association between changes in NP or PA presence and changes in Medicaid acceptance within the same practice over time. In all models, our time-varying control variables included the number of physicians in a practice, integration status, rural location, percentage of the population living in poverty, percentage of non-White population, percentage of adults who are older than 65 years, percentage of adults 18–64 years old that are uninsured, and unemployment rate (Table 1). We controlled separately for the number of physicians in a practice in order to separate the outcome’s relationship with changes in NP or PA presence from changes in the number of physicians. We included separate time-varying dummy variables in our models, indicating full NP practice authority (Kuo et al., 2013; Phillips, 2016) and state adoption of Medicaid expansion (Kaiser Family Foundation, 2019). We included year FEs, and robust standard errors were clustered at the practice level. We ran additional models controlling for other county-level, time-varying variables (e.g., physician, NP, and PA supply; physician-to-population ratios; and practice volume), but the addition of these variables did not change the results and were not included in the reported results (available upon request). This study received institutional review board exempt status for human subjects research. All analyses were conducted using StataMP-14.

TABLE 1:

Practice and county characteristics: 2009–2015

	2009	2011	2013	2015
Primary care (%)	34.4	33.9	33.3	32.0
Medical specialty (%)	26.8	26.7	26.5	26.2
Surgical specialty (%)	26.0	26.4	26.0	25.1
Multispecialty (%)	12.8	13.0	14.2	16.7
Horizontal (%)a	25.0	23.3	21.7	21.3
Vertical (%)a	6.7	11.2	14.0	20.7
Full NP practice authority (%)b	7.9	9.7	13.2	16.6
Medicaid expansionc	0.0	20.0	22.0	61.2
Rural (%)d	11.5	11.5	11.5	11.5
High poverty (%)e	10.6	15.9	16.0	15.5
No. of physicians (M)	3.0	3.2	3.1	3.0
% Non-White population (M)	21.8	23.0	23.4	24.3
% Population 65+ (M)	13.0	13.4	14.2	14.5
% Adults 18–64 uninsured (M)	20.8	21.2	20.5	16.5
Unemployment rate (persons 16+)	9.7	9.0	7.5	5.3

Note. Calculations based on a panel of 102,453 individual physician practices present in all 4 years of data (i.e., 2009, 2011, 2013, 2015).

NP = nurse practitioner; M = mean.

^a“Horizontal” practices are those that are part of a larger physician group. “Vertical” practices are those that report being owned by an integrated health/hospital system.

^bPractice is located in a state that allows for full NP practice authority.

^cPractice is located in state that expanded Medicaid.

^dRural location established using core-based statistical area.

^eHigh poverty is >20% of population in poverty.

Results

Medicaid Acceptance in Physician Practices With and Without NPs and PAs

Using our balanced panel of physician practices (N = 102,453), we first pooled all observations to describe Medicaid acceptance rate across all possible provider configurations: no NPs or PAs (physician-only), only NPs, only PAs, and both NPs and PAs. We also stratified our sample by practice size: “small” (one to three physicians) and “not-small” (four or more physicians). For both size categories, the lowest percentage of Medicaid acceptance was among physician-only practices with 58.9% of small and 73.4% of not-small practices accepting Medicaid (Figure 1). Medicaid acceptance was meaningfully greater in practices that employed only NPs or employed both NPs and PAs. Among small practices, there was roughly a 10-percentage point difference in Medicaid acceptance between these practices and physician-only practices. Larger practices demonstrated a gap of approximately 4 and 8 percentage points, respectively. Interestingly, practices that only employed PAs had Medicaid acceptance rates that were quite close to physician-only practices; in the case of not-small practices, they are virtually indistinguishable. Practices relying on all three provider types had the highest Medicaid acceptance rates.

Figure 1.

Medicaid acceptance in physician practices with and without nurse practitioners (NPs) and physician assistants (PAs) by practice size. Calculations based on pooled observations of Medicaid acceptance in 102,453 physician practices present in all 4 years of data: 2009, 2011, 2013, 2015. Small practice = 1–3 physicians; not-small practices = 4+ physicians

Additional results when we stratified the sample by the number of physicians in the practices at baseline (i.e., 2009) were consistent with our main analyses and can be seen in Supplemental Digital Content 1 (see http://links.lww.com/HCMR/A73).

Association Between Physician Practice Medicaid Acceptance and Newly Employing NPs and PAs

To move beyond a pooled cross-sectional view, we leveraged our longitudinal analyses using practice-level FEs (“within estimator”) and time-varying control variables listed above.

The first two columns of Table 2 are the results from our regression models for all practices. The strongest, positive result was among practices that employed both NPs and PAs (Column 1). Employing both types of providers was associated with a 1.4-percentage point (p < .001) increase in the probability of a practice now accepting Medicaid—a roughly 2% increase relative to the sample mean. This relationship held when we included our full set of covariates (Column 2). There was a small, but positive and statistically significant, association between newly employing just NPs and Medicaid acceptance in both models. The results were consistently nonsignificant among PA-only practices.

TABLE 2:

Association between physician practice Medicaid acceptance and employing NPs/PAs using fixed effects linear regression: 2009–2015

	All practices		1–3 physicians		4+ physicians
	(1) Model 1	(2) Model 2	(3) Model 1	(4) Model 2	(5) Model 1	(6) Model 2
	β (SE)	β (SE)	β (SE)	β (SE)	β (SE)	β (SE)
NPs onlya	0.006 (0.003)*	0.005 (0.003)*	0.007 (0.003)*	0.006 (0.003)	0.003 (0.004)	0.003 (0.004)
PAs onlya	0.006 (0.003)	0.006 (0.003)	0.010 (0.004)*	0.008 (0.004)*	−0.001 (0.005)	−0.0004 (0.0053)
Both NPs/PAsa	0.014 (0.004)***	0.013 (0.004)**	0.024 (0.005)***	0.021 (0.005)***	0.000 (0.006)	−0.000 (0.006)
Year FEs	Yes	Yes	Yes	Yes	Yes	Yes
Practice FEs	Yes	Yes	Yes	Yes	Yes	Yes
Covariates	No	Yes	No	Yes	No	Yes
Observations	409,812	409,800	307,400	307,394	102,412	102,406
Unique practices	102,453	102,453	76,850	76,850	25,603	25,603
Sample mean	0.64	0.64	0.61	0.61	0.75	0.75

Note. Practice size was established based on the number of physicians employed in the practice in 2009. Time-varying covariates include the number of physicians in the practice, dummy variables (Medicaid expansion, full NP practice authority, and practice horizontal and vertical integration status), and county-level demographics. Standard errors clustered at the practice level. Clustering at the state level produced either the same standard error estimates or estimates that were less conservative than clustering at the practice level. NP = nurse practitioner; PA = physician assistant; SE = standard errors; FE = fixed effects.

^aReference category is physician-only practices.

^*p < .05.

^**p < .01.

^***p < .001.

When stratifying by the number of physicians employed in the practices in 2009 (i.e., baseline year), we see that the previous associations are driven by small physician practices. Employing both NPs and PAs was associated with a 2.4-percentage point (p < .001) increase in the probability that a small practice accepted Medicaid (a 3.9% increase relative to the sample mean). In addition, among smaller practices, newly introducing PA-only employment was associated with a 1.0-percentage point (p < .05) increase in the probability of a small practice accepting Medicaid (a 1.6% relative improvement). The association between Medicaid acceptance and NP-only employment was again nonsignificant in the adjusted model. The coefficients are uniformly small and statistically nonsignificant for larger practices (Columns 5 and 6 in Table 2).

To further examine if there was any influence of the number of NPs and PAs in a practice (i.e., intensive margin) on a practice’s decision to accept Medicaid, we conducted analyses using continuous measures of the number of NPs, PAs, and both clinicians in a practice. All associations were nonsignificant for all practices and when we stratified the sample by practice size (results available upon request). These results suggest that the extensive margins for both practice staffing (i.e., employing NPs and PAs for the first time) and practice Medicaid participation represent a shift in the practice’s business strategy within the cost and revenue domains compared to adding more NPs and PAs to those already working in the practice.

We then stratified our sample by practice specialty (i.e., primary care, single medical or surgical subspecialty, and multispecialty) as well as practice size. As evident in Table 3 (adjusted models only), the results mirror our previous analyses with the strongest, positive associations being between newly employing both NPs and PAs and Medicaid acceptance in small primary care practices (β = 0.020, p < .01) and small single medical subspecialty practices (β = 0.041, p < .001). Small primary care practices had a 2.0-percentage point increase in the probability of accepting Medicaid when both NPs and PAs were present (a 3.3% relative change), and similarly, small medical practices demonstrated a 4.1-percentage point increase after incorporating both provider types (a 6.9% relative change). PA-only employment was associated with greater Medicaid acceptance among small primary care practices (β = 0.012, p < .05), and a similar relationship was seen for NP-only employment in small medical practices (β = 0.012, p < .05). There were nonsignificant relationships for surgical and multispecialty practices, as well as not-small physician practices, irrespective of specialty classification.

TABLE 3:

Association between physician practice Medicaid acceptance and employing NPs/PAs using fixed effects linear regression: 2009–2015 (by practice specialty)

	Primary care		Medical		Surgical		Multispecialty
	1–3 Physicians	4+ Physicians	1–3 Physicians	4+ Physicians	1–3 Physicians	4+ Physicians	1–3 Physicians	4+ Physicians
	B (SE)	β (SE)	β (SE)	β (SE)	β (SE)	β (SE)	β (SE)	β (SE)
NPs onlya	0.008 (0.005)	−0.000 (0.010)	0.012 (0.006)*	0.005 (0.007)	−0.002 (0.008)	0.012 (0.011)	−0.009 (0.010)	−0.000 (0.008)
PAs onlya	0.012 (0.006)*	−0.007 (0.014)	0.014 (0.007)	0.004 (0.009)	−0.006 (0.009)	0.006 (0.012)	0.001 (0.012)	−0.004 (0.009)
Both NPs/PAsa	0.020 (0.008)**	0.015 (0.015)	0.041 (0.011)***	−0.007 (0.011)	0.030 (0.016)	−0.009 (0.015)	−0.015 (0.015)	−0.002 (0.010)
Year FEs	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Practice FEs	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Covariates	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	118,766	22,370	82,004	27,608	83,248	23,264	23, 376	29,164
Unique practices	29,693	5,594	20,501	6,902	20,812	5,816	5,844	7,291
Sample mean	0.60	0.64	0.59	0.81	0.63	0.74	0.67	0.78

Note. Practice size was established based on the number of physicians employed in the practice in 2009. Time-varying covariates include the number of physicians in the practice, dummy variables (Medicaid expansion, full NP practice authority, and practice horizontal and vertical integration status), and county-level demographics. Standard errors clustered atthe practice level. Clustering atthe state level produced eitherthe same standard error estimates or estimates that were less conservative than clustering at the practice level. NP = nurse practitioner. PA = physician assistant; SE = standard errors; FE = fixed effects.

^aReference category is physician-only practices.

^*p < .05.

^**p < .01.

^***p < .001.

Discussion

This was the first study to use a national, balanced panel of physician practices to examine the association of newly employing NPs and PAs and changes in practice Medicaid acceptance over time. We found positive and statistically significant associations between NP and PA presence and physician practice Medicaid participation. The strongest relationships were among practices that employed both NPs and PAs. Among all practices, there was a 2% increase in the likelihood that a practice accepted Medicaid when both NPs and PAs had been recently incorporated into the practice. Furthermore, the finding was driven by small practices, particularly primary care and single medical subspecialty practices.

Although largely descriptive, our results expand on earlier research through a longitudinal design and examination of heterogeneity across practice size and specialty. First, our pooled cross-sectional results (Figure 1) suggested that NP presence may predominantly correspond to greater Medicaid acceptance, but the longitudinal regression results revealed that the addition of both provider types positively correlated with Medicaid acceptance. The stronger relationships we found among smaller practices are also noteworthy. Lower reimbursement rates paid to physicians for Medicaid services compared to private insurance have been cited as a reason for limited physician acceptance of Medicaid (Decker, 2012). In addition, physicians working in smaller practices (less than three physicians) have been shown to be more than 20 percentage points less likely to accept new Medicaid patients than physicians in larger practices (Decker, 2012).

Second, one cross-sectional analysis found primary care practices with NPs had higher rates of Medicaid acceptance (Barnes et al., 2017). A second study demonstrated greater appointment availability for new Medicaid beneficiaries among primary care practices with NPs and/or PAs, though the data could not disentangle the precise provider configuration within a practice and the positive associations were confined to states with favorable legal scope of practice environments for NPs (Richards & Polsky, 2016). Our results show, however, a 3.3% and 6.9% increase in the likelihood that a practice accepts Medicaid among small primary care and small single medical subspecialty practices, respectively, after newly employing NPs and PAs together. These findings are at least consistent with the proposition that some physician practices may begin hiring NPs and PAs as a means to offer lower cost services in order to accommodate Medicaid patients for the first time.

Our study is the first to provide evidence of the temporal relationship between changes in the extensive margins for staffing (i.e., employing NPs and PAs for the first time) and practice Medicaid participation. These co-occurring changes on their respective extensive margins plausibly reflect a significant shift in the practice’s business strategy on the revenue (lower payments from Medicaid) and cost (lower labor cost associated with NP and PA employment) sides. Put differently, as a physician practice adopts a new provider mix, it may make it more financially viable to create access for Medicaid patients that did not previously exist.

Limitations

Our results should be interpreted within the context of the study’s limitations. First, NPs and PAs working in practices that are not physician-led are not included in these data. We also do not observe how practices are utilizing NPs and PAs within practices. We do not have information on clinicians’ full-time or part-time status, or how new and existing patients are assigned to and managed by clinicians employed in practices. In addition, we do not have information on practices’ rationale or motivation underlying the decision to newly employ NPs or PAs or start accepting Medicaid, which may be driven by a variety of cost, practice capacity, and other business and environmental factors. Because the SK&A data are self-reported, our outcome measure (Medicaid acceptance) maybe an overestimation by some practices (Coffman et al., 2016).

Relatedly, our longitudinal analysis design is a clear improvement on the existing literature, but we still cannot claim direct causality or establish the precise timing of events. For example, we acknowledge the relationship could exist in the opposite direction (i.e., a practice first decides to participate in the Medicaid program and then expands its provider capacity via NPs and PAs). However, we maintain that regardless of the exact sequence of these decisions, our study enhances our understanding of the association between different provider configurations and changes in Medicaid participation. That is, if greater provider mixing leads to increased Medicaid acceptance or if greater Medicaid exposure encourages more blended provider staffing, the downstream impact for improving disadvantage patient access is largely equivalent.

Finally, we acknowledge that the decision to participate in Medicaid for the first time is likely a complex decision for many practices and influenced by state- and plan-level Medicaid policies (American Academy of Physician Assistants, 2020; Buppert, 2016; Chapman et al., 2010). For example, Medicaid fee-for-service reimbursement rates for NP and PA services vary across states and is often lower than the already low physician rates. Also, individual Medicaid Managed Care plans follow different rules for NP and PA reimbursement and are often decided at the plan level. Thus, varying local policies and the complexity of the local Medicaid program may restrain participation among physician practices irrespective of use of more blended, interdisciplinary provider configurations.

Practice Implications

Researchers and policy makers recommend interdisciplinary models of care for the delivery of high-quality care and improved patient outcomes (Bodenheimer et al., 2014; Castellucci, 2017). Although our study did not aim to examine these specific downstream consequences, our results reveal positive changes in practice Medicaid acceptance after adding NPs and PAs to a practice’s provider configuration. At the same time, the observed greater reliance on NPs and PAs by these physician practices may align with other, complementary policy and care delivery objectives.

The benefits to practices that employ NPs and PAs may also not be confined to seeing Medicaid patients. Hiring NPs and PAs may have broader implications by improving the financial benefits across payers (e.g., providing more care and/or providing care at lower costs to other patient–payer combinations). In this way, our results should encourage future research to gain a deeper understanding of how increased provider diversification influences overall practice capacity as well as practice financial performance and stability. That said, our focus on a practice’s decision to newly participate in Medicaid is not inconsequential. Policy makers and plan administrators have long grappled with how to increase physician participation in state Medicaid programs (Decker, 2012; Neprash et al., 2018). Our results suggest that rethinking how the practice is structured and adopting new approaches to care delivery can be key ingredients.