The Impact of Nurse Practitioner-Led Primary Care on Quality and Cost for Medicaid-Enrolled Patients in States With Pay Parity

Abstract

Studies have established that nurse practitioners (NPs) deliver primary care comparable to physicians in quality and cost, but most focus on Medicare, a program that reimburses NPs less than physicians. In this retrospective cohort study, we evaluated the quality and cost implications of receiving primary care from NPs compared to physicians in 14 states that reimburse NPs at the Medicaid fee-for-service (FFS) physician rate (i.e., pay parity). We linked national provider and practice data with Medicaid data for adults with diabetes and children with asthma (2012-2013). We attributed patients to primary care NPs and physicians based on 2012 evaluation & management claims. Using 2013 data, we constructed claims-based primary care quality measures and condition-specific costs of care for FFS enrollees. We estimated the effect of NP-led care on quality and costs using: (1) weighting to balance observable confounders and (2) an instrumental variable (IV) analysis using differential distance from patients’ residences to primary care practices. Adults with diabetes received comparable quality of care from NPs and physicians at similar cost. Weighted results showed no differences between NP- and physician-attributed patients in receipt of recommended care or diabetes-related hospitalizations. For children with asthma, costs of NP-led care were lower but quality findings were mixed: NP-led care was associated with lower use of appropriate medications and higher rates of asthma-related emergency department visits but similar rates of asthma-related hospitalization. IV analyses revealed no evidence of differences in quality between NP- and physician-led care. Our findings suggest that in states with Medicaid pay parity, NP-led care is comparable to physician-led care for adults with diabetes, while associations between NP-led care and quality were mixed for children with asthma. Increased use of NP-led primary care may be cost-neutral or cost-saving, even under pay parity.

• What do we already know about this topic?

• Studies have established that nurse practitioners (NPs) deliver primary care comparable to that of physicians in quality and cost, but most focus on Medicare, a program that reimburses NPs less than physicians.

• How does your research contribute to the field?

• In this retrospective cohort study, we evaluated the quality and cost implications of receiving primary care from NPs compared to physicians in 14 states that reimburse NPs at the Medicaid fee-for-service physician rate (i.e., Medicaid pay parity).

• What are your research’s implications toward theory, practice, or policy?

• Our findings suggest NP-led primary care in Medicaid could help address primary care shortages without sacrificing quality or raising costs for adults, even under pay parity, but further research is needed to understand the outcomes of NP-led care for children.

Introduction

High-quality primary care is important to prevent and manage chronic disease, avoid acute illness, and control health care costs, yet many Americans with chronic conditions such as diabetes and asthma lack access to a primary care provider.^1,2 The Association of American Medical Colleges has projected a shortage of 17,800 to 48,000 primary care physicians by 2034. ³ Growing primary care shortages in many areas contribute to challenges with access to care, particularly for underserved populations such as Medicaid enrollees. Medicaid-enrolled patients have historically faced lower appointment availability and longer wait times, largely due to lower Medicaid reimbursement rates relative to other payers. ⁴ Ensuring access to care for Medicaid enrollees is an ongoing challenge for states, and Medicaid expansion in many states has led to increased demand. ⁵ As the supply of primary care physicians continues to decline, nurse practitioners (NPs) play a growing role in meeting demands for care. ⁶

NPs are advanced practice registered nurses trained at the masters or doctoral level and licensed by states to assess patients, order and interpret diagnostic tests, make diagnoses, initiate and manage treatment plans, and prescribe medications.^7,8 NPs practice independently or as part of interprofessional teams with physicians. The NP workforce has grown rapidly in recent years, more than doubling between 2010 and 2017. ⁹ Among the 355,000 licensed NPs in the US, 89% are credentialed to deliver primary care and 70% practice in primary care. ¹⁰ Previous studies suggest primary care NPs are more likely than primary care physicians to practice in rural, underserved, and socioeconomically disadvantaged areas and accept Medicaid-enrolled patients.^11-13

Numerous studies have established that NPs deliver high-quality primary care comparable to physicians,^14-20 and previous studies have found similar or lower costs of primary care delivered by NPs versus physicians.^14-18 However, these previous studies have focused primarily on the Medicare population, and there is little evidence on the impact of NP-led care among younger populations. Similarly, there is little evidence on the effects of NP-led care on health care costs when NPs and physicians are reimbursed at the same rate (i.e., pay parity). Some state Medicaid programs reimburse NPs at the Medicaid fee-for-service (FFS) physician rate, but in many states Medicaid reimbursement for NPs is up to 25% less than reimbursement for physicians. ¹⁹ Total patient costs may be similar for NPs and physicians under pay parity if the services received are similar. Under pay parity, impacts of NP-led care on costs should be driven by impacts on service use: total patient costs may be lower for NPs than physicians if NP-led care lowers overall service use, or vice versa.

Scope of practice laws may also have implications for the quality of care provided by NPs and total costs of care. In many states scope of practice laws limit NPs’ abilities to practice without physician oversight or collaboration. ²⁰ Studies suggest that quality of NP-led care in states with full practice authority for NPs (FPA states) is similar or better compared to non-FPA states.^21-24 Findings regarding FPA and health care spending have been mixed. Some studies suggest FPA is associated with cost savings due to greater efficiency of care delivery,^21,25-27 while others suggest expanded access to care under FPA may generate higher costs due to increased service use.^28,29 Thus it remains unknown how patient costs may vary for NP-led care in states with Medicaid pay parity with and without FPA.

To address these gaps in the literature, our objective was to evaluate the quality and cost implications of receiving primary care from NPs compared to physicians in 14 states that reimburse NPs at the Medicaid fee-for-service physician rate (i.e., Medicaid pay parity), including five FPA states and nine non-FPA states. We conducted a retrospective cohort study focused on Medicaid-enrolled patients from 2012 to 2013 with two of the most prevalent chronic conditions: adults with diabetes and children with asthma. We used a validated method to attribute primary care patients to NPs and physicians and then used two complementary empirical strategies to estimate the effect of NP-led care on quality and costs: (1) entropy balancing to derive weights to adjust for observed differences in patients attributed to NPs and physicians, and (2) an instrumental variables analysis using differential distance from patients’ residences to primary care practices with NPs on staff to adjust for unobserved differences in patients attributed to NPs and physicians.

Methods

This study was approved with a waiver of consent by the lead author’s Human Subjects Protection Committee.

Data Sources

We obtained data on health care service use and costs from the 2012 and 2013 Medicaid Analytic Extract (MAX) files. Developed by the Centers for Medicare & Medicaid Services (CMS) to support research and policy analysis for Medicaid populations, MAX files include person-level information on Medicaid eligibility, service utilization, and payments. ³⁰ MAX files include National Provider Identifiers (NPI) but do not include Tax Identification Numbers (TIN) or clinician addresses or ZIP codes. Though more recent data is available from the T-MSIS Analytic File (TAF), the new generation of Medicaid data available for all states from 2016 onward, further work is needed to understand usability of the TAF data for policy analyses, including the extent of quality issues common in Medicaid claims such as missingness and variability across states. ³¹

We sampled primary care practices and providers using national health care provider and practice data from the 2012 and 2013 SK&A files (now called OneKey by IQVIA). ³² The SK&A files contain data on office-based physician practices across the U.S., including practice specialty, number and type of providers within the practice, provider specialties, NPIs, and practice location (ZIP code and latitude/longitude). This information allowed us to identify NPs working in primary care, an advantage over other data sources. ³³

Sample Construction

To select the sample of states, we reviewed 2012 and 2013 MAX data validation reports for the 27 states with Medicaid pay parity in 2012 (AppendixTable A1) to identify those with the most complete NPI data. We also considered usability of MAX encounter data for managed care enrollees. ³⁴ Based on these criteria, we included 5 FPA states (Iowa, Maine, Oregon, Washington, Wyoming) and 10 non-FPA states (California, Delaware, Michigan, Missouri, ⁱ New Jersey, New York, Ohio, Tennessee, Utah, and Virginia).

We used 2012 to 2013 SK&A data to identify primary care practices in the states of interest. We defined the following practice specialties as primary care: adolescent medicine, family practice, general practice, general preventative medicine, geriatric medicine, internal medicine, internal medicine/pediatrics, multi-specialty, and pediatrics. ³⁵ We characterized the provider mix in each practice as the percentage of providers (including physicians, NPs, and physician assistants) that were NPs. To ensure adequate representation of practices with a large share of NPs, we included in our sample all primary care practices in which at least 80% of providers were NPs. From the remaining pool of primary care practices, we sampled 100% of practices in small states and 33 or 67% of practices in mid- and large-sized states (AppendixTable A1). This was to ensure the size of the patient cohort was not intractably large. We generated a deduplicated list of NPIs for all providers in the sampled primary care practices in 2012 or 2013, resulting in a sample of approximately 60,000 NPIs. Any patient with one or more outpatient claims from any of the NPIs on our list was included in the final cohort of approximately 12 million Medicaid enrollees.

Next, we identified adults with diabetes and children with persistent asthma who were continuously enrolled in Medicaid during 2012 to 2013 (inclusive of both enrollees in FFS and comprehensive managed care) and had one or more primary care visits in 2012. We excluded individuals dually enrolled in Medicaid and Medicare because we did not have information on care paid by Medicare. The diabetes sample included adults ages 18 to 64 with diagnoses of diabetes reported in the Medicaid Enrollee Supplemental File (MESF) chronic conditions file. The pediatric asthma sample included children ages 5 to 17 who met claims-based criteria for persistent asthma. ³⁶

Patient Attribution

We used a validated method to attribute patients to providers. ³⁷ To ensure an established relationship with the attributed primary care provider, patient attribution was based on 2012 evaluation and management (E&M) services from the Berenson-Eggers Type of Services Codes for new (M1A: 99201-99205) and established office visits (M1B: 99211-99215). Patients were attributed to the provider who delivered the plurality of E&M visits paid by Medicaid (at least 30% of visits). We excluded E&M services from inpatient, emergency, or specialty care settings so that attribution would be based on primary care. After applying these criteria, no patients in Missouri were attributed to NPs, therefore we excluded Missouri from our analytic sample. We caution that, under typical coding practices, billing data cannot identify care delivered by NPs under indirect billing (when an NP performs services but bills under a physician’s name). Some care delivered by NPs may consequently be inaccurately attributed to physicians (see Discussion for details). The final patient sample included 49,907 adults with diabetes and 39,765 children with asthma (Appendix Figures A1 and A2).

Control Variables

Our analyses adjusted for patient demographic and clinical characteristics including age, sex, race/ethnicity, county rural/urban residence (metropolitan, non-metropolitan—adjacent to metropolitan areas, non-metropolitan—not adjacent to metropolitan areas), ³⁸ state of residence, Medicaid enrollment type (FFS or comprehensive managed care), Medicaid eligibility category, ⁱⁱ and chronic condition count.

Dependent Variables

Quality measures

We constructed claims-based quality measures using 2013 data based on specifications from the Healthcare Effectiveness Data and Information Set (HEDIS) and the Agency for Healthcare Research and Quality (AHRQ). For adults with diabetes, we constructed the following indicators of comprehensive diabetes care (CDC): receipt of hemoglobin A1c testing, annual low density lipoprotein (LDL) screening, and medical attention for nephropathy. ³⁶ We also constructed indicators of having all three CDC measures and having zero CDC measures. We did not examine receipt of a retinal eye exam, another indicator of comprehensive diabetes care, because we could not reliably identify provider specialty in the MAX data. We also identified adults having any preventable hospitalizations for complications of diabetes. ³⁹ For children with persistent asthma, we constructed a measure of receipt of appropriate medications, ³⁶ an indicator of having one or more emergency department (ED) visits for asthma, ⁴⁰ and an indicator of having one or more hospital admissions for asthma. ³⁹

Costs

Our cost analyses were limited to FFS enrollees (<10% of our patient sample) in a subset of states with available data. ⁱⁱⁱ Among the patients in our sample with FFS enrollment, we evaluated diabetes- and asthma-related Medicaid costs in 2013. We identified FFS claims with a principal diagnosis or second listed diagnosis for diabetes or asthma, respectively, and summed total annualized costs for inpatient visits, outpatient visits, and prescription drugs. Costs in each category were winsorized at the 99th percentile for those with non-zero spending. We calculated total costs per patient as the sum of inpatient, outpatient, and prescription costs. Due to documented issues with incomplete and inaccurate cost data for Medicaid managed care payments,^41-43 including missing encounters, inappropriate zeros, and negative values, we were unable to measure costs for managed care enrollees.

Statistical Analysis

We used two approaches to address observable and unobservable differences between patients receiving primary care from NPs and those receiving primary care from physicians. Analyses were estimated using data from all states and then for groups of states stratified by FPA versus non-FPA status. Analyses were performed in Stata version 16 and SAS version 9.4. For all analyses we used two-sided tests with alpha level of 0.05.

Weighting estimates of the effects of NP-led care

We used entropy balancing ⁴⁴ to derive weights that balance differences in observable patient characteristics between those attributed to NPs and those attributed to physicians, including age, sex, race/ethnicity, rural/urban residence, state of residence, Medicaid enrollment type, and Medicaid eligibility category. Entropy balancing was chosen over propensity score methods for ease of use and computational speed. We evaluated outcomes before and after weighting to examine whether differences in quality and cost outcomes associated with NP attribution are driven by observable differences between patient groups.

Instrumental variables (IV) estimates of the effects of NP-led care

To avoid confounding due to unobservable differences between NP- and physician-attributed patients (i.e., due to self-selection into NP-led care based on unobservable factors affecting quality or cost of care), we also conducted an IV analysis.

In addition to examining provider-level attribution to an NP (which is the explanatory variable of interest in our weighting estimates), we constructed a practice-level attribution measure intended to capture whether patients were attributed to providers at a practice with one or more NPs on staff. Practice-level exposure to NPs is of interest because the presence of NPs in a practice may affect quality of care even if the patient’s attributed provider is a physician. Practice-level attribution may also help address mismeasurement of NP-led care due to indirect billing.

We followed previous studies in using differential distance as an instrument,^45-48 where differential distance is defined as the additional distance beyond the nearest primary care practice that a patient needs to travel to reach a practice that has one or more NPs on staff. We counted a practice as having NPs on staff if one or more NPs were linked to that practice in the SK&A data. Our differential distance calculations included the universe of primary care practices in the states of interest in SK&A. We used Google Maps Application Programming Interfaces to calculate the driving distance between patient ZIP code centroids and practice coordinates (latitude and longitude) obtained from SK&A (Appendix B-1).

We report estimates from linear regression models of quality measures on an indicator for NP attribution, patient demographic and clinical characteristics, and state fixed effects. Heteroskedasticity-robust variance estimates were used for statistical inference. ⁴⁹ Linear regression models that do not use the instrument were estimated using ordinary least squares (OLS), while the IV models were estimated using two-stage least squares (2SLS). We also estimated nonlinear IV models using two-stage residual inclusion (2SRI). ⁵⁰ Nonlinear models (logistic regression for binary patient-level quality measures, and Poisson regression for a count outcome) may be more efficient for the outcomes considered here, but in practice both sets of IV models were imprecisely estimated and yield similar findings: we therefore do not report the 2SRI estimates in this paper.

Differential distance to an NP practice is a strong predictor of NP attribution. The first-stage F-statistics on the differential distance instruments are 116 for adults with diabetes and 50 for children with asthma, both well above the threshold of 10 customarily viewed as a warning sign of weak instruments. First-stage F-statistics for practice-level attribution were even higher: 289 for children with asthma and 540 for adults with diabetes. Reduced-form and first stage estimates are reported in the Appendix (Tables A6 and A7), where we also discuss why additional assumptions required for IV to identify causal effects are likely to be met in this setting (Appendix B-2).

We also used our IV specification to estimate effects of provider-level and practice-level attribution on costs for the subsample of FFS patients for whom costs could be measured reliably. However, these IV estimates were too imprecise to be informative, and we do not report them.

Results

Rates and Correlates of NP-Led Care

Patient attribution to an NP was somewhat rare; 3.51% of children with asthma and 4.55% of adults with diabetes were attributed to an NP as their primary care provider, with the remainder attributed to a physician. Among adults with diabetes (AppendixTable A2), those attributed to NPs were slightly younger, more likely to be female, non-Hispanic White, and live in rural areas, and had differences in Medicaid eligibility category: more likely to be blind/disabled, more likely to qualify through Section 1931, less likely to qualify through Section 1115, less likely to be medically needy. Among children with asthma (AppendixTable A3), those attributed to NPs were slightly older, more likely to be non-Hispanic White and live in rural areas, and had differences in Medicaid eligibility category: less likely to be blind/disabled, more likely to qualify through Section 1115, less likely to be medically needy. Entropy balancing weights achieved balance on all characteristics (AppendixTables A2 and A3).

Estimates of the Effects of NP-Led Care

Quality outcomes: Adults with diabetes

After weighting, adult patients with diabetes attributed to NPs had no significant differences in performance on hemoglobin A1c testing, LDL cholesterol screening, or medical attention for nephropathy compared to those attributed to physicians (Table 1; Figure 1A). Though patients attributed to NPs had significantly lower unadjusted performance on these outcomes, there were no significant differences after weighting for patient characteristics. The proportion of patients with diabetes-related hospital admissions did not differ significantly between patients attributed to NPs and physicians before or after weighting (Table 1; Figure 1B). Findings were similar in FPA and non-FPA states; there were no significant differences in weighted outcomes between NPs and physicians. Performance on diabetes quality measures was slightly better overall in FPA states (AppendixTable A4).

Table 1.

Standardized Differences in Quality Outcomes for Patients Attributed to Nurse Practitioners and Physicians, Overall and by FPA and Non-FPA States.

	Overall		Non-FPA states		FPA states
	Standardized difference a		Standardized difference a		Standardized difference a
	Unweighted	Weighted	Unweighted	Weighted	Unweighted	Weighted
Adults with diabetes
Hemoglobin A1c testing performed	−0.04*	0.02	−0.07**	0.02	−0.06	0.01
LDL-C screening	−0.11***	0.02	−0.12***	0.03	−0.09	−0.01
Medical attention for nephropathy	−0.05*	0	−0.05*	−0.01	−0.03	0
All 3 performed	−0.12***	0	−0.13***	0.02	−0.13*	−0.06
None performed	0	−0.03	0.01	−0.02	−0.04	−0.1
Has one or more claims for diabetes-related admissions	−0.01	0.01	0	0.02	−0.05	−0.06
Children with asthma
Use of appropriate medications for children with asthma	−0.14***	−0.10**	−0.17***	−0.11**	−0.02	−0.03
Has one or more claims for asthma-related emergency department visits	0.02	0.06*	0.02	0.06	0.06	0.06
Has one or more claims for asthma-related admissions	−0.04	0	−0.05	0	0	0.01

^aStandardized difference calculated as average rate of outcome for NP-attributed patients minus average rate of outcome for physician-attributed patients.

^*P < .05. **P < .01. ***P < .001.

Figure 1.

Quality outcomes for adults with diabetes and children with asthma. A. Percent of Adults with Diabetes Receiving Recommended Care; B. Diabetes-Related Hospital Admissions; C. Percent of Children with Asthma Receiving Recommended Care; D. Asthma-Related Emergency Department Visits; E. Asthma-Related Hospital Admissions.

Note. P-values represent standardized differences in quality measure performance between (1) NP versus Physician and (2) NP versus Physician, Weighted. NP = nurse practitioner.

*P < .05. **P < .01. ***P < .001.

All regression models (OLS and IV) indicated that receiving NP-led primary care (measured via provider-level attribution) was not associated with any differences in quality of care for adults with diabetes (Table 2). While the IV point estimates for the effects of NP attribution were large, they were imprecisely estimated, having large standard errors, and none are statistically significant at 5%. Receiving care from a practice that had one or more NPs on staff was also not associated with any differences in quality of care across all IV models. Instrumental variables estimates for the effects of NP care in FPA versus non-FPA states were statistically insignificant at both the provider level and the practice level in both groups of states (AppendixTable A8).

Table 2.

Linear Regression and Instrumental Variables Analysis of Quality for Adults With Diabetes.

Outcomes	Hemoglobin A1c testing performed	LDL-C screening	Medical attention for nephropathy	All 3 procedures performed	None performed	Has one or more claims for diabetes-related admissions in 2013
Explanatory variable: Provider-level attribution to NP
OLS estimate (standard error [SE])	0.012 (0.010)	0.009 (0.011)	−0.003 (0.008)	0.003 (0.011)	−0.005 (0.005)	0.001 (0.003)
2SLS (IV) estimate (SE)	−0.393 (0.249)	−0.420 (0.265)	−0.335 (0.193)	−0.454 (0.285)	0.209 (0.131)	−0.117 (0.089)
Mean of outcome	0.759	0.709	0.873	0.615	0.055	0.023
N	49 855	49 855	49 855	49 855	49 855	49 855
Explanatory variable: Practice-level attribution to practice with NP(s) on staff
OLS estimate (SE)	0.025*** (0.005)	−0.002 (0.005)	0.008* (0.004)	0.005 (0.005)	−0.009*** (0.003)	−0.001 (0.002)
2SLS (IV) estimate (SE)	−0.077 (0.048)	−0.082 (0.051)	−0.065 (0.037)	−0.089 (0.055)	0.041 (0.025)	−0.023 (0.017)
Mean of outcome	0.759	0.709	0.873	0.615	0.055	0.023
N	49 855	49 855	49 855	49 855	49 855	49 855

Note. Standard errors are heteroskedasticity-robust. Instrument is differential distance from nearest primary care practice to nearest primary care practice with one or more NPs on staff. First-stage F-statistic for individual-level NP attribution was 116.1. First-stage F-statistic for practice-level NP attribution was 540.3. All models control for state fixed effects and all covariates listed in AppendixTable A2.

OLS estimate = coefficient on indicator for patient attribution to NP from OLS regression of quality measure on attribution indicator and covariates; 2SLS (IV) estimate = coefficient on indicator for patient attribution to NP from OLS regression of quality measure on attribution indicator and covariates; N = sample size for 2SLS model.

^*P < .05. ***P < .001.

Quality outcomes: Children with asthma

Children with asthma who were attributed to NPs had lower unadjusted performance on use of appropriate medications for people with asthma (59.0 vs 65.7%, P < .001) (Table 1; Figure 1C). After weighting, this difference was reduced, but not eliminated (59.0 vs 63.8%, P < .001). NP-physician differences in performance on the asthma medication measure were driven by non-FPA states; we did not observe a significant difference in FPA states (AppendixTable A5). Children with asthma attributed to NPs had a similar unadjusted rate of asthma-related ED visits compared to those attributed to physicians (10.7 vs 10.3%, P = .567) (Table 1; Figure 1D). However, after weighting, children with asthma attributed to NPs had a higher rate of asthma-related ED visits compared to those attributed to physicians (10.7 vs 8.9%, P = .033), with similar findings in FPA and non-FPA states. The proportion of patients with asthma-related hospital admissions did not differ significantly between patients attributed to NPs and physicians before or after weighting (Table 1; Figure 1E), and findings were similar in FPA and non-FPA states.

In IV models, NP-led primary care (measured via provider-level attribution) was not associated with any differences in use of appropriate medications, ED visits, or hospital admissions for children with asthma (Table 3). Results were similar in the IV models examining receipt of care from a practice that had one or more NPs on staff. IV estimates for the effects of NP-led care in FPA versus non-FPA states were also statistically insignificant at both the provider level and the practice level in both groups of states (AppendixTable A9).

Table 3.

Linear Regression and Instrumental Variables Analysis of Quality for Children With Asthma.

Outcomes	Use of appropriate medications for children with asthma	Has one or more claims for asthma-related emergency department visits in 2013	Has one or more claims for asthma-related admissions in 2013
Explanatory variable: Provider-level attribution to NP
OLS estimate (SE)	–0.047*** (0.013)	0.0180* (0.00853)	–0.000115 (0.003)
2SLS (IV) estimate (SE)	0.012 (0.460)	0.247 (0.298)	0.002 (0.122)
Mean of outcome	0.655	0.103	0.016
N	39 744	39 744	39 744
Explanatory variable: Practice-level attribution to practice with NP(s) on staff
OLS estimate (SE)	–0.012* (0.006)	0.00151 (0.00376)	–0.003 (0.001)
2SLS (IV) estimate (SE)	0.002 (0.072)	0.0386 (0.0464)	0.000 (0.019)
Mean of outcome	0.655	0.103	0.016
N	39 744	39 744	39 744

Note. Standard errors are heteroskedasticity-robust. Instrument is differential distance from nearest primary care practice to nearest primary care practice with one or more NPs on staff. First-stage F-statistic for individual-level NP attribution was 49.95. First-stage F-statistic for practice-level NP attribution was 289.11. All models control for state fixed effects and all covariates listed in AppendixTable A3.

OLS estimate = coefficient on indicator for patient attribution to NP from OLS regression of quality measure on attribution indicator and covariates; 2SLS (IV) estimate = coefficient on indicator for patient attribution to NP from OLS regression of quality measure on attribution indicator and covariates; N = sample size for 2SLS model.

^*P < .05.***P < .001.

Cost outcomes: Adults with diabetes

Among adults with diabetes, there were no statistically significant differences in FFS costs for NP- and physician-attributed patients (Figure 2A). Although unadjusted annual mean costs were lower for NP-attributed patients ($5525) than for physician-attributed patients ($8123, P = .065), the difference shrunk after weighting to achieve balance on observable characteristics (weighted annual mean cost for physician-attributed patients = $5815, P = .74). Findings with regard to NP-physician cost differences were similar in FPA and non-FPA states (AppendixTable A10).

Figure 2.

Fee-for-service cost outcomes for adults with diabetes and children with asthma. A. Mean Annualized Cost for Adults with Diabetes; B. Mean Annualized Costs for Children with Asthma.

Note. P-values represent standardized differences in mean annualized costs between (1) NP versus Physician and (2) NP versus Physician, Weighted. Total costs are the sum of inpatient, outpatient, and prescription costs.

NP = nurse practitioner.

*P < .05. **P < .01. ***P < .001.

Cost outcomes: Children with asthma

Among children with asthma, those attributed to NPs had lower total FFS costs compared to those attributed to physicians (Figure 2B). Unadjusted annual mean costs were lower for NP-attributed patients ($588) than physician-attributed patients ($1586, P = .02). After weighting, costs remained significantly lower among children attributed to NPs (weighted annual mean cost for physician-attributed patients = $1306, P < .01). Findings with regard to NP-physician cost differences were similar in FPA and non-FPA states (AppendixTable A11).

Discussion

To our knowledge, no previous studies have evaluated the cost and quality outcomes of NP-led primary care in a context with pay parity. Examining states with Medicaid pay parity, we found that NP-attributed adults with diabetes received comparable quality of care at similar cost to physician-attributed patients. Quality findings were mixed for children with asthma: NP-led care was associated with lower use of appropriate medications and a higher rate of asthma-related ED visits. However, rates of asthma-related hospitalization were not associated with NP-led care, and children with asthma had lower FFS costs under NP-led care. The association between NP-led care and health care quality or cost did not differ substantially in FPA states versus non-FPA states.

We observed notable demographic, geographic, and eligibility category differences among the population of Medicaid enrollees served by primary care NPs compared to those served by primary care physicians. These characteristics may be associated with quality and cost outcomes. For example, rural patients may face barriers with access to care, which could lead to lower rates of hemoglobin A1c testing and LDL cholesterol screening. In both populations examined here (adults with diabetes and children with asthma), patients attributed to NPs were more likely to live in rural areas. We also found that adults with diabetes who were attributed to NPs had differences in Medicaid eligibility category (i.e., more likely to qualify through Section 1931) that may suggest greater poverty compared to patients attributed to physicians. These findings are consistent with previous studies that suggest NPs are more likely than physicians to deliver primary care in rural and socioeconomically disadvantaged areas.^11-13

Our findings for adults with diabetes suggest that if NPs and physicians treated the same patients, then quality of care would be similar. Without adjusting for patient characteristics, adult patients with diabetes attributed to NPs were less likely to receive hemoglobin A1c testing, LDL cholesterol screening, and medical attention for nephropathy. The fact that weighting to balance these characteristics between NP- and physician-attributed patients eliminated most quality differences suggests that quality differences in the unadjusted data were driven by differences in patient population rather than differences in provider behavior. However, for children with asthma, weighting for observable patient characteristics did not eliminate NP-physician differences in quality, suggesting a need for further research to better understand outcomes of NP-led care for children with asthma and other pediatric populations.

A limitation of many previous studies comparing quality and cost for NP- and physician-attributed patients is potential confounding due to unobserved differences in patient characteristics that may influence quality and cost outcomes. In our main analysis, in which we used weights based on observable patient characteristics, some NP-physician differences in quality remained (i.e., lower performance on use of appropriate medications and a higher rate of ED visits for children with asthma attributed to NPs). In IV analyses, we found no evidence of a causal relationship between NP attribution and quality of care, or between presence of NPs at the practice level and quality of care. We caution that, as is typical in IV analyses, the IV estimates were less precise than other estimates in our analysis, and so our results do not rule out clinically meaningful quality differences. However, the findings do not provide any evidence that such quality differences exist.

We also note that the focus of our study (the effects of receiving primary care from an NP, which we measure via patient attribution) is a patient-level variable. Potential concerns about the misuse of patient-level IVs to estimate the effects of practice- or facility-level characteristics therefore should not apply in this context. ⁵¹

Few studies comparing costs for NP- and physician-attributed patients have accounted for lower reimbursement of NPs. In one study, Medicare beneficiaries attributed to NPs had 18% lower costs than those attributed to physicians even after accounting for case mix and differences in reimbursement. ¹⁴ Another study found that lower costs for NP-attributed Medicare beneficiaries were driven by lower service use by NPs, and NP-physician cost differences were greatest for low acuity patients. ⁵² We found that Medicaid FFS costs were similar for NP- and physician-attributed adults with diabetes, but children with asthma who received care from NPs versus physicians had significantly lower FFS costs across all spending categories. Though we did not have detailed clinical data to examine NP-physician cost differences by patient acuity, children with asthma had fewer chronic conditions, on average, than adults with diabetes (1.0 vs 4.5). Further research is needed to understand the potential drivers of lower FFS cost for children with asthma attributed to NPs versus physicians, including potential differences in patient acuity and differences in the volume or type of services provided.

We interpret these results in the context of several limitations. First, our comparisons of costs are limited to Medicaid FFS enrollees and are imprecise, due in part to small sample sizes. Our findings related to cost may not be generalizable to managed care enrollees, as we excluded managed care enrollees from cost analyses due to concerns about inaccurate encounter payments.^42,43 In addition, the completeness of provider NPI data in Medicaid claims varies widely across states, ⁵³ limiting the number of patients we could attribute to providers. Newer Medicaid data in the T-MSIS Analytic File (TAF) may have more complete information on NPI, so this problem may be mitigated in future work using more recent data. Our patient sample was further restricted by limiting to patients attributed to providers in primary care specialties, as designated in SK&A. Further, it is unclear to what extent Medicaid claims data accurately distinguish services provided by NPs versus physicians. Current coding practices do not identify care delivered by NPs under indirect billing when an NP performs services but bills under a physician’s name.^54,55 Studies that rely on billing provider NPIs may inaccurately attribute care delivered by NPs to physicians, in which case comparisons of NP- and physician-led care may be biased toward the null. Finally, recognizing that the best Medicaid data available for this study were several years old, and predated state Medicaid expansions, future research should explore if findings have changed as the Medicaid-insured population has grown.

Conclusion

In the context of shifts in the composition of the primary care workforce and growing demand for care, NPs play an important role in delivery of high-quality primary care to Medicaid-enrolled patients, particularly in rural areas. Our findings suggest that in states with Medicaid pay parity, NP-led care is comparable to physician-led care for adults with diabetes, while associations between NP-led care and quality were mixed for children with asthma. The associations estimated here also suggest that increased use of NP-led primary care could be cost-neutral or cost-saving. However, further work is needed to obtain more precise estimates of the causal effect of NP-led care on cost and quality, both in Medicaid and in other settings beyond Medicare.

Appendix A

Table A1.

State Sample Selection.

States with Medicaid pay parity for nurse practitioners in 2012	Nurse practitioner scope of practice in 2012	Included in study sample?	Percent of primary care practices sampled from SK&A (%)
California	Non-FPA	Yes	33
Delaware	Non-FPA	Yes	100
District of Columbia	FPA
Illinois	Non-FPA
Iowa	FPA	Yes	100
Louisiana	Non-FPA
Maine	FPA	Yes	100
Maryland	Non-FPA
Michigan	Non-FPA	Yes	67
Missouri	Non-FPA	Yes	67
Nebraska	Non-FPA
New Hampshire	FPA
New Jersey	Non-FPA	Yes	67
New York	Non-FPA	Yes	33
North Carolina	Non-FPA
Ohio	Non-FPA	Yes	67
Oklahoma	Non-FPA
Oregon	FPA	Yes	100
Pennsylvania	Non-FPA
Rhode Island	FPA
Tennessee	Non-FPA	Yes	67
Utah	Non-FPA	Yes	100
Virginia	Non-FPA	Yes	67
Washington	FPA	Yes	100
West Virginia	Non-FPA
Wisconsin	Non-FPA
Wyoming	FPA	Yes	100

Note. FPA = full practice authority.

Table A2.

Patient Characteristics: Adults With Diabetes.

	Unweighted							Weighted
	Total		NP-attributed		MD-attributed		Standardized difference +	Standardized difference +
	N or mean	% or SD	N or mean	% or SD	N or mean	% or SD
Number of patients	49 907	100.00	2272	4.55	47 635	95.45
Demographics
Age (mean, SD)	48.19	0.05	46.14	0.23	48.29	0.05	−0.20***	0.00
Female, n (%)	31 618	63.35	1621	71.35	29 997	62.97	0.18***	0.00
Race/ethnicity, n (%)
Non-Hispanic White	22 258	44.60	1486	65.40	20 772	43.61	0.45***	0.00
Non-Hispanic Black	12 082	24.21	392	17.25	11 690	24.54	−0.18***	0.00
Hispanic	8429	16.89	209	9.20	8220	17.26	−0.24***	0.00
Other race/ethnicity	7188	14.40	185	8.14	7003	14.70	−0.21***	0.00
Medicaid variables
Eligibility category, n (%)
Blind/disabled	28 772	57.65	1429	62.90	27 343	57.40	0.11***	0.00
Section 1931	8472	16.98	486	21.39	7986	16.76	0.12***	0.00
Other poverty	2091	4.19	87	3.83	2004	4.21	−0.02	0.00
Section 1115	7605	15.24	116	5.11	7489	15.72	−0.35***	0.00
Medically needy	1203	2.41	++	++	++	++	−0.13***	0.00
Foster care	10	0.02	++	++	++	++	0.01	0.00
Other	1754	3.51	134	5.90	1620	3.40	0.12***	0.00
MMC 2012, n (%)
CMC	42 951	86.06	2 018	88.82	40 933	85.93	0.09***	0.00
FFS	3317	6.65	181	7.97	3136	6.58	0.05*	0.00
Other	3639	7.29	73	3.21	3566	7.49	−0.19***	0.00
Health status
Number of comorbidities (mean, SD)	4.53	0.01	4.30	0.06	4.54	0.01	−0.08***	0.00
Geographic variables
Urban/rural setting, n (%)
Metro	42 703	85.63	1339	58.96	41 364	86.90	−0.66***	0.00
Non-metro, adjacent	5553	11.13	709	31.22	4844	10.18	0.54***	0.00
Non-metro, not adjacent	1615	3.24	223	9.82	1392	2.92	0.29***	0.00
State
CA	7011	14.05	29	1.28	6982	14.66	−0.51***	0.00
DE	628	1.26	20	0.88	608	1.28	−0.04	0.00
IA	852	1.71	66	2.90	786	1.65	0.08***	0.00
ME	521	1.04	73	3.21	448	0.94	0.16***	0.00
MI	3989	7.99	44	1.94	3945	8.28	−0.29***	0.00
NJ	3597	7.21	78	3.43	3519	7.39	−0.18***	0.00
NY	15 673	31.40	357	15.71	15 316	32.15	−0.39***	0.00
OH	7387	14.80	263	11.58	7124	14.96	−0.10***	0.00
OR	1348	2.70	157	6.91	1191	2.50	0.21***	0.00
TN	4991	10.00	1015	44.67	3976	8.35	0.90***	0.00
UT	269	0.54	11	0.48	258	0.54	−0.01	0.00
VA	3082	6.18	78	3.43	3004	6.31	−0.13***	0.00
WA	454	0.91	69	3.04	385	0.81	0.16***	0.00
WY	105	0.21	12	0.53	93	0.20	0.06**	0.00

⁺Standardized difference was calculated as the average rate of outcome for NP-attributed patients minus average rate of outcome for physician-attributed patients. Medicaid eligibility categories are defined based on uniform eligibility codes: Blind/disabled = 14, 32, 42; Section 1931 = 14, 15, 16, 17; Other poverty = 31, 34, 35; Section 1115 = 51, 52, 54, 55; Medically needy = 21, 22, 24, 25; Foster care = 48; Other = all else.

⁺⁺Data have been censored to mask cells with fewer than 10 observations.

^*P < .05. **P < .01. ***P < .001.

Table A3.

Patient Characteristics: Children With Asthma.

	Unweighted							Weighted
	Total		NP-attributed		MD-attributed		Standardized difference +	Standardized difference +
	N or mean	% or SD	N or mean	% or SD	N or mean	% or SD
Number of patients	39 765	100.00	1396	3.51	38 369	96.49
Demographics
Age, mean (SD)	8.91	0.02	9.37	0.09	8.90	0.02	0.14***	0.00
Female, n (%)	16 536	41.58	605	43.34	15931	41.52	0.04	0.00
Race/ethnicity, n (%)
Non-Hispanic White	14 834	37.30	765	54.80	14 069	36.67	0.37***	0.00
Non-Hispanic Black	11 470	28.84	322	23.07	11 148	29.05	–0.14***	0.00
Hispanic	10 079	25.35	184	13.18	9895	25.79	–0.32***	0.00
Other race/ethnicity	3568	8.97	129	9.24	3439	8.96	0.01	0.00
Medicaid variables
Eligibility category, n (%)
Blind/disabled	3836	9.65	103	7.38	3733	9.73	–0.08***	0.00
Section 1931	16 197	40.73	579	41.48	15 618	40.70	0.02	0.00
Other poverty	16 138	40.58	580	41.55	15 558	40.55	0.02	0.00
Section 1115	141	0.35	++	++	126	0.33	0.09***	0.00
Medically needy	903	2.27	++	++	897	2.34	–0.16***	0.00
Foster care	970	2.44	31	2.22	939	2.45	–0.02	0.00
Other	1580	3.97	82	5.87	1498	3.90	0.09***	0.00
MMC 2012, n (%)
CMC	35 080	88.22	1218	87.25	33 862	88.25	–0.03	0.00
FFS	1818	4.57	96	6.88	1722	4.49	0.10***	0.00
Other	2867	7.21	82	5.87	2785	7.26	–0.06*	0.00
Health status
Number of comorbidities (mean, SD)	1.02, 0.01		0.94, 0.03		1.03, 0.01		–0.07**	0.00
Geographic variables
Urban/rural setting, n (%)
Metro	35 350	88.90	950	68.05	34 400	89.66	–0.55***	0.00
Non-metro, adjacent	3495	8.79	321	22.99	3174	8.27	0.41***	0.00
Non-metro, not adjacent	918	2.31	125	8.95	793	2.07	0.31***	0.00
State, n (%)
CA	3793	9.54	37	2.65	3756	9.79	–0.30***	0.00
DE	697	1.75	21	1.50	676	1.76	–0.02	0.00
IA	855	2.15	68	4.87	787	2.05	0.15***	0.00
ME	263	0.66	23	1.65	240	0.63	0.10***	0.00
MI	2742	6.90	++	++	2737	7.13	–0.36***	0.00
NJ	7140	17.96	77	5.52	7063	18.41	–0.41***	0.00
NY	9952	25.03	182	13.04	9770	25.46	–0.32***	0.00
OH	4273	10.75	195	13.97	4078	10.63	0.10***	0.00
OR	862	2.17	69	4.94	793	2.07	0.16***	0.00
TN	4050	10.18	488	34.96	3562	9.28	0.65***	0.00
UT	271	0.68	15	1.07	256	0.67	0.04	0.00
VA	3782	9.51	79	5.66	3703	9.65	–0.15***	0.00
WA	984	2.47	132	9.46	852	2.22	0.31***	0.00
WY	101	0.25	++	++	96	0.25	0.02	0.00

Note. SD = standard deviation.

⁺Standardized difference was calculated as the average rate of outcome for NP-attributed patients minus average rate of outcome for physician-attributed patients. Medicaid eligibility categories are defined based on uniform eligibility codes: Blind/disabled = 14, 32, 42; Section 1931 = 14, 15, 16, 17; Other poverty = 31, 34, 35; Section 1115 = 51, 52, 54, 55; Medically needy = 21, 22, 24, 25; Foster care = 48; Other = all else.

⁺⁺Data have been censored to mask cells with fewer than 10 observations.

^*P < .05. **P < .01. ***P < .001.

Table A4.

Quality Outcomes: Adults With Diabetes.

	Unweighted							Weighted
	Total		NP–attributed		MD–attributed		Standardized difference +	Standardized difference +
	N	%	N	%	N	%
Number of observations	49 907	100	2272	4.552	47 635	95.45
Outcomes
Hemoglobin A1c (HbA1c) testing performed
Overall	37 890	75.92	1685	74.16	36 205	76.01	−0.04*	0.02
Non-FPA states	35 071	75.22	1368	72.19	33 703	75.34	−0.07**	0.02
FPA states	2819	85.95	317	84.08	2502	86.19	−0.06	0.01
LDL-C screening
Overall	35 398	70.93	1499	65.98	33 899	71.16	−0.11***	0.02
Non-FPA states	33 040	70.86	1241	65.49	31 799	71.09	−0.12***	0.03
FPA states	2358	71.89	258	68.44	2100	72.34	−0.09	−0.01
Medical attention for nephropathy
Overall	43 542	87.25	1944	85.56	41 598	87.33	−0.05*	0.00
Non-FPA states	40 729	87.35	1624	85.70	39 105	87.42	−0.05*	−0.01
FPA states	2813	85.76	320	84.88	2493	85.88	−0.03	0.00
All 3 performed
Overall	30 666	61.45	1266	55.72	29 400	61.72	−0.12***	0.00
Non-FPA states	28 582	61.30	1047	55.25	27 535	61.56	−0.13***	0.02
FPA states	2084	63.54	219	58.09	1 865	64.24	−0.13*	−0.06
None performed
Overall	2754	5.52	123	5.41	2631	5.52	0.00	−0.03
Non-FPA states	2612	5.60	109	5.75	2503	5.60	0.01	−0.02
FPA states	142	4.33	14	3.71	128	4.41	−0.04	−0.10
Has one or more claims for diabetes-related admissions in 2013
Overall	1122	2.25	48	2.11	1074	2.25	−0.01	0.01
Non-FPA states	1060	2.27	43	2.27	1017	2.27	0.00	0.02
FPA states	62	1.89	++	++	57	1.96	−0.05	−0.06

⁺Standardized difference was calculated as the average rate of outcome for NP-attributed patients minus average rate of outcome for physician-attributed patients.

⁺⁺Data have been censored to mask cells with fewer than 10 observations.

^*P < .05. **P < .01. ***P < .001.

Table A5.

Quality Outcomes: Children With Asthma.

	Unweighted							Weighted
	Total		NP-attributed		MD-attributed		Standardized difference +	Standardized difference +
	N or mean	% or SD	N or mean	% or SD	N or mean	% or SD
Number of observations	39 765	100.00	1396	3.51	38 369	96.49
Outcomes
Use of appropriate medications for children with asthma, n (%)
Overall	26 042	65.49	824	59.03	25 218	65.72	–0.14***	–0.10**
Non-FPA states	23 987	65.36	628	57.14	23 359	65.61	–0.17***	–0.11**
FPA states	2055	67.05	196	65.99	1859	67.16	–0.02	–0.03
Has one or more claims for asthma-related emergency department visits in 2013, n (%)
Overall	4091	10.29	150	10.74	3941	10.27	0.02	0.06*
Non-FPA states	3826	10.43	120	10.92	3706	10.41	0.02	0.06
FPA states	265	8.65	30	10.10	235	8.49	0.06	0.06
Has one or more claims for asthma-related admissions in 2013, n (%)
Overall	632	1.59	16	1.15	616	1.61	–0.04	0.00
Non-FPA states	580	1.58	11	1.00	569	1.60	–0.05	0.00
FPA states	52	1.70	++	++	47	1.70	0.00	0.01

⁺Standardized difference was calculated as the average rate of outcome for NP-attributed patients minus average rate of outcome for physician-attributed patients.

⁺⁺Data have been censored to mask cells with fewer than 10 observations.

^*P < .05. **P < .01. ***P < .001.

Table A6.

Reduced-Form Estimates of Association Between Differential Distance and Quality for Adults With Diabetes.

Outcomes	Hemoglobin A1c testing performed	LDL-C screening	Medical attention for nephropathy	All 3 procedures performed	None performed	Has one or more claims for diabetes-related admissions in 2013
Explanatory variable: Differential distance to NP practice
OLS estimate (standard error (SE))	0.00258 (0.00162)	0.00276 (0.00172)	0.00220 (0.00126)	0.00297 (0.00185)	−0.00137 (0.000852)	0.000770 (0.000580)
Mean of outcome	0.759	0.709	0.873	0.615	0.055	0.023
N	49 855	49 855	49 855	49 855	49 855	49 855

Note. Differential distance = ln(dist_NP)—ln(dist), where dist_NP is the distance to the nearest primary care practice with one or more NPs, and dist is the distance to the nearest primary care practice. Table reports coefficients (standard errors) on differential distance in reduced-form models corresponding to 2SLS estimates reported in Table 2 of manuscript. Regressions control for all covariates used in regression models. All reduced-form estimates were statistically insignificant at the 5% level.

Table A7.

Reduced-Form Estimates of Association Between Differential Distance and Quality for Children With Asthma.

Outcomes	Use of appropriate medications for children with asthma	Has one or more claims for asthma-related emergency department visits in 2013	Has one or more claims for asthma-related admissions in 2013
Explanatory variable: Differential distance to NP practice
OLS estimate (SE)	−0.00005 (0.00201)	−0.00108 (0.00129)	−0.00001 (0.00053)
Mean of outcome	0.655	0.103	0.016
N	39 744	39 744	39 744

Note. Differential distance = ln(dist_NP)—ln(dist), where dist_NP is the distance to the nearest primary care practice with 1 or more NPs, and dist is the distance to the nearest primary care practice. Table reports coefficients (standard errors) on differential distance in reduced-form models corresponding to 2SLS estimates reported in Table 3 of manuscript. Regressions control for all covariates used in regression models. All reduced-form estimates were statistically insignificant at the 5% level.

Table A8.

Linear Regression and Instrumental Variables Analysis of Quality for Adults With Diabetes, Stratified by FPA and Non-FPA States.

Outcomes	Hemoglobin A1c testing performed	LDL-C screening	Medical attention for nephropathy	All 3 procedures performed	None performed	Has one or more claims for diabetes-related admissions in 2013
Explanatory variable: Provider-level attribution to NP
Non-FPA states
OLS estimate (SE)	0.0158 (0.0109)	0.0145 (0.0116)	−0.00350 (0.00850)	0.0124 (0.0120)	−0.00324 (0.00571)	0.0158 (0.0109)
2SLS (IV) estimate (SE)	−0.460 (0.294)	−0.531 (0.309)	−0.392 (0.225)	−0.572 (0.333)	0.242 (0.155)	−0.460 (0.294)
Mean of outcome	0.7521	0.7086	0.8736	0.6131	0.0560	0.7521
N	46 579	46 579	46 579	46 579	46 579	46 579
FPA states
OLS estimate (SE)	−0.00290 (0.0211)	−0.00312 (0.0258)	−0.00219 (0.0202)	−0.0331 (0.0276)	−0.0183 (0.0112)	−0.00290 (0.0211)
2SLS (IV) estimate (SE)	−0.186 (0.304)	−0.0637 (0.401)	0.00914 (0.296)	−0.00572 (0.425)	0.0723 (0.166)	−0.186 (0.304)
Mean of outcome	0.8599	0.7192	0.8574	0.6355	0.0433	0.8599
N	3276	3276	3276	3276	3276	3276
Explanatory variable: Practice-level attribution to practice with NP(s) on staff
Non-FPA states
OLS estimate (SE)	0.0254** (0.00506)	−0.00174 (0.00544)	0.00808* (0.00388)	0.00566 (0.00571)	−0.00786** (0.00267)	0.0254** (0.00506)
2SLS (IV) estimate (SE)	−0.0859 (0.0546)	−0.0992 (0.0571)	−0.0732 (0.0416)	−0.107 (0.0616)	0.0452 (0.0287)	−0.0859 (0.0546)
Mean of outcome	0.7521	0.7086	0.8736	0.6131	0.0560	0.7521
N	46 579	46 579	46 579	46 579	46 579	46 579
FPA states
OLS estimate (SE)	0.0264** (0.0125)	−0.000973 (0.0160)	0.00485 (0.0125)	−0.000245 (0.0172)	−0.0171** (0.00737)	0.0264** (0.0125)
2SLS (IV) estimate (SE)	−0.0493 (0.0802)	−0.0169 (0.107)	0.00243 (0.0786)	−0.00152 (0.113)	0.0192 (0.0439)	−0.0493 (0.0802)
Mean of outcome	0.8599	0.7192	0.8574	0.6355	0.0433	0.8599
N	3276	3276	3276	3276	3276	3276

Note. Differential distance = ln(dist_NP)—ln(dist), where dist_NP is the distance to the nearest primary care practice with 1 or more NPs, and dist is the distance to the nearest primary care practice. Table reports coefficients (standard errors) on differential distance in first-stage models corresponding to 2SLS estimates reported in Table 2 of manuscript (First stage is identical for all outcomes). Regressions control for all covariates used in regression models.

^*P < .05. **P < .01.

Table A9.

Linear Regression and Instrumental Variables Analysis of Quality for Children With Asthma, Stratified by FPA and Non-FPA States.

Outcomes	Use of appropriate medications for children with asthma	Has one or more claims for asthma-related emergency department visits in 2013	Has one or more claims for asthma-related admissions in 2013
Explanatory variable: Provider-level attribution to NP
Non-FPA states
OLS estimate (SE)	–0.0532*** (0.0151)	0.0169 (0.00964)	–0.000683 (0.00313)
2SLS (IV) estimate (SE)	0.0912 (0.528)	0.324 (0.342)	–0.0663 (0.141)
Mean of outcome	0.6535	0.1042	0.0158
N	36 682	36 682	36 682
FPA states
OLS estimate (SE)	–0.0168 (0.0286)	0.0184 (0.0183)	0.000410 (0.00794)
2SLS (IV) estimate (SE)	–0.329 (0.710)	–0.223 (0.462)	0.285 (0.199)
Mean of outcome	0.6708	0.08654	0.0170
N	3062	3062	3062
Explanatory variable: Practice-level attribution to practice with NP(s) on staff
Non-FPA states
OLS estimate (SE)	–0.0116 (0.00617)	0.00109 (0.00404)	–0.00317* (0.00154)
2SLS (IV) estimate (SE)	0.0137 (0.0793)	0.0488 (0.0512)	–0.00996 (0.0212)
Mean of outcome	0.6535	0.1042	0.0158
N	36 682	36 682	36 682
FPA states
OLS estimate (SE)	–0.0107 (0.0167)	0.00223 (0.0103)	0.00118 (0.00451)
2SLS (IV) estimate (SE)	–0.0745 (0.159)	–0.0504 (0.103)	0.0644 (0.0380)
Mean of outcome	0.6708	0.0865	0.0170
N	3062	3062	3062

Note. Differential distance = ln(dist_NP)—ln(dist), where dist_NP is the distance to the nearest primary care practice with one or more NPs, and dist is the distance to the nearest primary care practice. Table reports coefficients (standard errors) on differential distance in first-stage models corresponding to 2SLS estimates reported in Table 3 of manuscript (First stage is identical for all outcomes). Regressions control for all covariates used in regression models.

^*P < .05. ***P < .001.

Table A10.

Fee-for-Service Cost Outcomes: Adults With Diabetes.

Total annualized costs	Total (N = 3317)			NP-attributed patients (N = 181)			Physician-attributed patients (N = 3136)			Standardized difference in mean cost +
	N non-zero	Mean	SD	N non-zero	Mean	SD	N non-zero	Mean	SD	Unweighted	Weighted
Outpatient
Overall	2059	$1902	$4795	123	$1483	$3311	1936	$1927	$4868	−0.11	−0.03
Non-FPA	1475	$2112	$5304	44	$1753	$4057	1431	$2123	$5338	−0.078	−0.065
FPA	584	$1269	$2635	79	$1304	$2718	505	$1264	$2624	0.015	0.002
Inpatient
Overall	501	$3862	$11 986	23	$2253	$7386	478	$3957	$12 198	−0.17	−0.02
Non-FPA	419	$4567	$13 171	11	$3523	$10 172	408	$4599	$13 253	−0.091	−0.038
FPA	82	$1733	$6911	12	$1411	$4600	70	$1783	$7207	−0.062	−0.018
Prescription
Overall	1330	$1653	$2798	71	$1704	$2796	1259	$1650	$2799	0.02	0.06
Non-FPA	989	$1730	$2896	29	$2104	$2879	960	$1718	$2897	0.134	0.109
FPA	341	$1423	$2468	42	$1439	$2724	299	$1420	$2428	0.007	−0.016
Total
Overall	2145	$7978	$16 350	127	$5525	$10 174	2018	$8123	$16 634	−0.19	−0.03
Non-FPA	1548	$9143	$18 055	47	$7579	$13 570	1501	$9191	$18 175	−0.101	−0.061
FPA	597	$4462	$8609	80	$4163	$6848	517	$4508	$8857	−0.044	−0.021

Note. Total costs are based on the sum of inpatient, outpatient, and prescription costs.

SD = standard deviation.

⁺Standardized difference calculated as average rate of outcome for NP-attributed patients minus average rate of outcome for physician-attributed patients.

Table A11.

Fee-for-Service Cost Outcomes: Children With Asthma.

Total annualized costs	Total (N = 1818)			NP-attributed patients (N = 96)			Physician-attributed patients (N = 1722)			Standardized difference in mean cost +
	N non-zero	Mean	SD	N non-zero	Mean	SD	N non-zero	Mean	SD	Unweighted	Weighted
Outpatient
Overall	901	$402	$1241	36	$147	$322	865	$417	$1273	−0.29	−0.27***
Non-FPA	619	$448	$1409	15	$130	$370	604	$464	$1439	−0.318	−0.33***
FPA	282	$266	$458	21	$177	$217	261	$273	$472	−0.263	−0.22
Inpatient
Overall	58	$319	$1755	1	$45	$404	57	$335	$1801	−0.22	−0.19**
Non-FPA	44	$341	$1838	1	$71	$506	43	$354	$1879	−0.206	−0.19*
FPA	14	$256	$1481	0	$0	$0	14	$278	$1541	−0.255	−0.21***
Prescription
Overall	921	$600	$962	39	$345	$583	882	$615	$978	−0.34*	−0.37***
Non-FPA	626	$635	$1053	16	$293	$627	610	$652	$1067	−0.410	−0.52***
FPA	295	$497	$610	23	$436	$493	272	$503	$619	−0.119	−0.07
Total
Overall	1161	$1531	$3724	48	$588	$990	1113	$1586	$3816	−0.36*	−0.35***
Non-FPA	812	$1692	$4100	23	$573	$1177	789	$1747	$4184	−0.382*	−0.43***
FPA	349	$1059	$2223	25	$613	$541	324	$1097	$2306	−0.289	−0.24*

Note. Total costs are based on the sum of inpatient, outpatient, and prescription costs.

⁺Standardized difference calculated as average rate of outcome for NP-attributed patients minus average rate of outcome for physician-attributed patients.

^*P < .05. **P < .01. ***P < .001.

Figure A1.

Patient flow diagram: Adults with diabetes.

Note. PCMD = primary care physician; PCNP = primary care nurse practitioner.

Figure A2.

Patient flow diagram: Children with persistent asthma.

Note. PCMD = primary care physician; PCNP = primary care nurse practitioner.

Appendix B: Methods

Appendix B-1: Calculation of Differential Distance

We used Google Maps Application Programming Interfaces to calculate the driving distance between patient ZIP code centroids and practice coordinates (latitude and longitude) obtained from SK&A. Because distance to practices is skewed and nonnegative, we took the natural log of the distance to each practice before calculating differential distance. For instance, if a patient’s distance to the nearest practice was 3 km and distance to the nearest practice with a nurse practitioner (NP) was 5 km, then the differential distance measure would be ln(5)-ln(3). This functional form for differential distance captures the percentage increase in travel distance to a practice with an NP relative to the nearest practice, allowing us to analyze patients with very different distances to practices (eg, metropolitan and non-metropolitan patients) in the same model. For example, a patient with a distance of 30 km to the nearest practice and 50 km to the nearest NP practice would have the same differential distance measure as a patient with a distance of 3 km to the nearest practice and 5 km to the nearest NP practice. Without the log transformation, the differential distance measure would contain some large outliers that could be excessively influential on our estimates.

Appendix B-2: Additional Discussion of Instrumental Variables Methods and Instrument Validity

The IV analysis in this study uses a differential distance strategy similar to that used in other papers. ⁴⁶ This approach requires three key assumptions for the IV estimates to identify the causal effect of NP attribution on outcomes: relevance, exclusion, and monotonicity. ⁵⁶ First-stage F-statistics on the excluded instruments, which can be used to test relevance, were discussed in the manuscript. Here, we provide further discussion of the validity of the relevance, exclusion, and monotonicity IV assumptions.

The relevance assumption is testable, and we confirmed that differential distance to an NP practice is a strong predictor of NP attribution. The first-stage F-statistics on the differential distance instruments are 116 for adults with diabetes and 50 for children with asthma and, both well above the threshold of 10 customarily viewed as a warning sign of weak instruments (see Appendix for first-stage regression tables and discussion). ^iv First-stage F-statistics for practice-level attribution were even higher, equaling 289 for children with asthma and 540 for adults with diabetes. We note that many more patients were attributed to a practice with NPs on staff than are attributed to an NP at the individual level. For children with asthma, 24.7% of patients are attributed to a practice with an NP on staff, versus 3.5% of patients attributed to an NP at the individual level. For adults with diabetes, 23.5% of patients are attributed to a practice with an NP on staff, versus 4.55% of patients attributed to an NP at the individual level.

For children with asthma, a 10% increase in the differential distance to an NP practice is associated with an 0.04 percentage point reduction in the probability of NP attribution, a 1.2% reduction relative to the sample average NP attribution rate of 3.51%. For adults, a 10% increase in the differential distance to an NP practice is associated with an 0.07 percentage point reduction in the probability of NP attribution, a 1.5% reduction relative to the sample average NP attribution rate of 4.55%.

The exclusion restriction necessary for IV to identify causal effects is that differential distance does not directly affect cost or quality outcomes, so that any causal effect of differential distance on outcomes operates only through its effect on the treatment variable (i.e., patient attribution to an NP practice). This assumption cannot be tested, but we think it is unlikely that the differential distance from a patient residence to an NP practice would be associated with unobserved factors that directly affect quality or cost outcomes independently of the included covariates (which include controls for health status, residence in a metropolitan area, and Medicaid eligibility category).

The monotonicity assumption necessary for IV to identify causal effects is that the effect of differential distance on NP attribution has the same sign for all patients. A violation of this assumption would require there to be patients for whom higher differential distance from an NP practice increases the patient’s probability of NP attribution. The monotonicity assumption is also untestable, but we do not think it is likely that this assumption would be violated in our setting.