Nurse Practitioners and Their Effects on Visits to Primary Care Physicians

1 Introduction

Shortages of primary care physicians (PCPs) have been a growing public health concern in the United States. Current health workforce projections indicate that the shortage of PCPs is likely to worsen, and by 2025 the country will require 52,000 additional PCPs (Petterson et al. 2012). Some possible explanations for the projected mismatch of PCPs vis-à-vis expected increases in demand for primary health care services are population growth, disproportionate ageing of the population, and the passage of the Patient Protection and Affordable Care Act (2010) – which is expected to extend health insurance coverage to an estimated 34 million people in the United States. Given that physicians’ time is already highly constrained and may not be efficiently allocated across patients with different health care needs (Tai-Seale and McGuire 2012), a PCP shortage could impede patients’ access to care and further strain physicians’ workloads.

Several proposals have been put forward to help alleviate the current PCPs shortage. These include increasing the inflow of PCPs from abroad, consolidating physician practices from solo to shared, expanding the use of technology (e.g., electronic records and the internet), and expanding the use of other allied health care professionals such as nurse practitioners (NPs) and physician assistants (PAs).^[1] In fact, simulations from Green, Savin, and Lu (2013) show that if some of these proposals are implemented effectively, the problem of PCP shortages could be mitigated or even eliminated.

The proposal to expand the role of NPs in primary care settings merits special attention for several reasons. First, the NP labor force in the United States has grown considerably. In the past decade, the growth of primary care physicians has been declining, but the number of NPs increased from 81,000 in 1999 to 180,000 in 2011 (Pearson 2011) and now constitutes about 19% of the U.S. primary care labor force (Stange 2013). Second, compared to PCPs, the cost of training NPs is relatively lower and the length of training NPs is relatively shorter (Dierick-van Daele et al. 2010). Research also shows that the quality of care and health outcomes achieved by the two professions is comparable. ^[2]

There are several of studies that pertain to our paper. Studies by Stange (2013) and Traczynski and Udalova (2013) identified the effects of NPs’ supply and regulations on various measures of utilization. Stange (2013) applies fixed effects and IV estimators to identify the county level supply of NPs on individual health care use and finds insignificant relationships between these variables. Traczynski and Udalova (2013) use variation in state legislature on NPs scope of practice to identify whether NPs independence from physician’s oversight affected individual’s utilization. The authors found that expansion of NPs’ independence increased patients’ utilization measures and reduced visits to emergency rooms.

A number of studies have examined the effects of states’ NP regulations on physicians’ incomes (Perry 2009) and enrollment in NP programs (Kalist and Spurr 2004). Other studies have examined the substitution patterns between private and public providers (Atella and Deb 2008), hygienists and dentists (Kleiner and Park 2010), PAs and overall office-based visits (Morgan et al. 2008), and primary and specialty care (Fortney et al. 2005). These studies find significant evidence of substitution between these related health professions. However, substitution patterns between patients’ NP visits and patients’ PCP visits have not been examined in the current literature.

This paper contributes to the literature by examining whether patients’ visits to NPs affect the demand for care from PCPs using nationally representative data from Medical Expenditures Panel Survey (MEPS) on 25–64 years old adults. Specifically, we examine the effects of patients’ NP visits on the probability and number of visits to PCPs. We also examine the effects of patients’ NP visits on the probabilities of receiving the following specific services from PCPs: prescriptions, laboratory tests, medical check-ups, diagnoses, and vaccinations. A key challenge in identifying these effects stems from the endogeneity associated with the non-random selection of patients’ visits to NPs. Patients with visits to NPs are often healthier than those with visits to PCPs. Also, patients with visits to NPs may be disproportionately located in medically underserved areas where preferences for health and health care differ considerably. The presence of these omitted factors can bias our estimates downward.

To account for the endogeneity associated with the non-random selection of visits to NPs, we employ a system of simultaneous equations where states’ educational requirement for obtaining an NP license is used as an identifying exclusion. We conducted several robustness checks and applied alternative estimation technique using panel data. The results suggest that the care provided by the two professions has significant overlap and NPs could serve as a potential option for filling the shortage in the supply of primary care services.

2 Conceptual model

To frame the analysis, we consider a simplified version of Grossman’s (1972) model of the demand for health where patient’s health is determined by a health production function with two types of inputs. Specifically, the utility function UC,H is defined over health, H and consumption C. The health production function H is determined by two types of health care inputs which consists of care received from a PCP, mPCP, and an NP, mNP

We assume that

The individual’s budget constraint is

where I is income, pPCP and pNP are prices for a unit of care received from a PCP and an NP while the price of consumption C is normalized to one. The individual’s problem is to choose mPCP, mNP, and C to maximize the utility function subject to health production function and the budget constraint. Specifically, the individual maximizes

The first-order conditions are

where pj/Hj represents the cost to a patient of producing an extra unit of health from a provider j. At the optimum, patient’s marginal rate of substitution between health and a consumer good equals their price ratio. The comparative statics analysis that identifies the effect of visiting one provide type on the demand from another yields the following expression

where ∂mj/∂mk is the change in demand from provider j in response to a change in the demand for care from provider k, and ∂Ω/∂mj is the second-order conditions of the optimization problem and it is assumed to be satisfied, i.e. ∂Ω/∂mj<0. In this framework, NPs and PCPs are considered substitutes (complements) if ∂mj∂mk<0>0. The sign in eq. [5] will depend on the sign of Hjk and the relative magnitude of UHHjk and pjpkUCC−UCHHk+Hj−pkUHC+UHHHk=A. If Hjk>0 and UHHjk>A, then NPs and PCPs are complements, otherwise, the two provider types are substitutes.

If Hjk=∂∂H/∂mj∂mk>0, an increase in health care input mk not only shifts health production function upward but also causes each unit of mj to have a greater effect on health. This can occur if provider k supplements a work of a provider j, for instance, if an NP measures a patient’s blood pressure while a PCP diagnoses a patient, in which case the two provider types are considered as technical complements. On the other hand, if Hjk<0, the two provider types are considered technical substitutes. The exact nature of the relationship between NPs and PCPs will depend on the severity of patient’s illness. A review by Bauer (2010) suggests that the portion of health services that can be provided by NPs range from 25% in some specialty health services to 90% in primary care. Treatment of patients with severe health problems that are beyond the range of NP’s practice would allocate NP’s services to complement those provided by PCPs. Therefore, the sign of expression [5] can only be identified empirically which is an objective of our analysis. Given that our focus is on primary care setting where an NP can perform about 90% of the services of a PCP, we expect substitution between these two types of providers.

3 Econometric methods

Our primary objective is to examine the effects of patients’ NP visits on the probability of PCP visits and the number of PCP visits. As mentioned above, patients who visit NPs are likely to have unobserved characteristics that affect the demand for care provided from PCPs. For our dichotomous dependent variables, we estimate bivariate probit models of the following form:

where the dichotomous variables PCP and NP equal one if person i had a PCP visit and an NP visit any time during the year, respectively; X and Z are vectors of observed characteristics hypothesized to influence health care use; b1, B1B2 are coefficients, and F and G are the cumulative normal distributions. The error terms e1 and e2 have bivariate normal distributions with zero mean, unit variance, and correlation coefficient ρ that captures unobservables common to both equations. Equations [6] and [7] are also used to identify the effects of patients’ NP visits on the probabilities of receiving a prescription, a laboratory test, a vaccination, a diagnosis, and a medical check-up during a visit to a PCP.

The estimate of interest b1 measures the extent to which patients’ visits to NPs affect their likelihood of visiting PCPs. A negative sign on b1 suggests that services provided by NPs offset the demand for care provided by PCPs, implying that the two professions provide substitutable services. We are also interested in whether visits to NPs affect the number of visits to PCPs. To accommodate this outcome variable, we estimate systems of equations similar to eqs [6] and [7] but instead of a probit in eq. [6] we estimate a negative binomial using the following specification:

where μi represents the conditional mean of PCP visits yi. The density for yi follows conventional negative binomial distribution:

where ψ=1α, and θ=11+αμi and α is the dispersion parameter identified by the model. Equations [8]–[9] are correlated through the residuals e3 and e4 which are decomposed into:

In eq. [10] δi is normally distributed and represents the unobserved component common to both equations; and ρ1 and ρ2 are factor loadings that measure the extent of unobserved component in each of the equations. The system of equations in eqs [8] and [9] is estimated jointly as a single likelihood function by integrating over the unobserved component δi for each individual. The identification of the system of equations in [6, 7] and [8, 9] is achieved through nonlinearity and the exclusion restriction described below. The system of equations in [8] and [9] is estimated using aML (Lillard and Panis 2003).

4 Data

Our primary data source is the Household Component of Medical Expenditure Panel Survey (MEPS), which is a nationally representative survey of the U.S. non-institutionalized population. In each year, the MEPS consists of two panels drawn from the National Health Interview Survey where a new panel begins and the old one ends participation in the survey. For the cross-sectional analysis, we pooled the three most recent years of available data (2008–2010) and analyze the data as a repeated cross section. We limit our sample to individuals 25 to 64 years of age who were in scope the entire period of our analysis. We use information from the MEPS office visits and outpatient visits files which are person-visit level files to create our dependent variables. For each visit, information is available on the type of provider, the specialty of the provider, and the month of the visit. We created a measure that captures visits to PCPs if a patient had a reported visit to a physician specializing in a family, a general practice, or internal medicine.

To create an indicator for visits to NPs, we use the measure indicating whether a patient visited a nurse or NP. The use of this measure alone can introduce measurement error and bias our results toward complementarities between NPs and PCPs. We address this issue by requiring that in addition to patients’ reported visit to a nurse or NP, the patient must also receive a medical check-up, a diagnosis, or physiotherapy/mental counseling services. We use this approach because nurses who do not have an NP license are not authorized to provide any of the above-mentioned services. To check the accuracy of this definition, we resorted to the National Ambulatory Medical Care Survey (NAMCS), which explicitly asks question about visits to NPs. According to the National Health Statistics Report (Hsiao et al. 2010), about 1.9% of office visits constitute visits to NPs. In our sample, about 2.6% of office visits were visits to NPs.

Additional concern arises that MEPS first asks if respondents saw a physician and, if not, the survey asks about seeing other providers. This is likely to result in under-reporting of visits to other types of providers. In a comprehensive review of this issue, Morga et al. (2007) assess representativeness of NPs in MEPS, NAMCS, and Community Tracking Survey (CTS). The study suggests that a collection methodology of CTS seems to be more representative than in MEPS and NAMCS. According to CTS, visits to PA and NP constitute about 8% of total outpatient visits. Since NPs represent about 64% of NP and PA labor force (Stange 2013), we should expect NPs to account for about 4.8% of total physician visits. Based on our definition of NP, we find that NP visits constitute about 3% of total physician visits and 9% of PCP visits. These numbers are clearly lower but do not seem to be exceedingly off the CTS data.

In either case, the existing data issues can introduce several sources of biases which merit attention. First, during a visit, if a patient sees both an NP and a PCP but reports seeing only a PCP, our results will be biased toward substitutes. Although possible, this scenario does not seem to be likely purely because it is unusual for a patient to see both an NP and a PCP in one visit in a primary care setting. ^[5] Second, if visits to NPs are remembered for their high quality of care (thus removing the need to visit a PCP), our results will be biased toward substitutes. Third, respondents’ misreporting may be purely random which is likely to introduce a classical measurement error, biasing results toward zero. Although the econometric approach in our study accounts for the presence of measurement error, we acknowledged these issues need in our analysis.

We linked the MEPS data to the ARF to extract county-level variables that measure the number of physicians and rural residents per capita. We also use the Pearson (2009) report to create an indicator that captures states’ educational requirements for practicing as a licensed NP. Given differences in health care needs and health preferences, the concentration of physicians and NPs are likely to be different across rural and urban centers and in areas with predominantly elderly populations. These area-level variables, therefore serve as controls in all models to help capture local demand factors that may be correlated with visits to NPs and PCPs.

In a cross-sectional analyses, we also examine the effects of NP visits on specific services provided by PCPs such as check-ups, diagnoses, prescriptions, and laboratory exams. In our simultaneous equation models, the dependent variables measure patients’ utilization during the year. Rates of specific service use are rare in the quarterly data, so we examine the effects of NP visits on specific PCP services only in the simultaneous equation models.

Table 1 provides information on the names, definitions, and descriptive statistics of the remaining variables among those who visited PCPs and NPs. Overall, individuals with NP visits are similar to those with PCP visits along many dimensions, such as education, poverty, and employment. Individuals with NP visits are more likely to be female, live in rural areas, and report to have poor or fair health. In all models, we control for all the covariates presented in Table 1. The analytic sample comprises 48,316 person-year observations in a cross-sectional sample. For the dynamic models, we expand this sample to quarterly periods where the sample is further restricted to individuals with non-missing observation on visits in each quarter. Because we lose the first quarter, the final sample results in 143,795 person-quarters of observations.

5 Results

6 Sensitivity analysis

7 Conclusion

In the next decade, the United States will need an additional 50,000 physicians to match the rising demand for primary care services (Petterson et al. 2012). An expanded use of NPs in primary care is one of many proposed solutions to the shortfall in PCPs. However, the previous literature provides limited evidence about the role of NPs in the current market place, particularly, the extent to which the two professions provide similar services. This is the first study that fills this void.

Our findings indicate that the primary care services provided by the two professions are similar. In particular, visits to NPs significantly reduce the likelihood of visits to PCPs by 22.2 percentage points, decreases the number of PCP visits by 27.0 percentage points, and reduces the probability of patients receiving prescriptions, check-ups, diagnoses, and laboratory tests from PCPs. Although the marginal effects are relatively large, the elasticity of NP visit is only 0.07% due to the smaller share of the population with visits to NPs. This result also indicates that a 10% increase in patients’ NPs visits reduce visits to PCPs by only 0.7%.

To check for robustness of our results, we conducted a falsification test and found that the reduced form associations between a visit to NPs and dentists, psychologists, and optometrist are significantly positive, but in the simultaneous systems these effects were not significant. As an alternative estimation technique, we estimate dynamic models, testing whether visits to NPs in the past period affects the demand for PCP services in the current period. The results from these models produced negative and significant estimates, suggesting that NPs provide sufficient care to patients – reducing the need for visits to PCPs during the specified period.

There are several limitations in this study that could be addressed in a future research. First, the choice of the exclusion restriction and the presence of measurement errors in the data are important limitation in our empirical methods. Although our findings are robust, future research using other data sources and alternative estimation techniques could shed additional light on the robustness of our results. Second, we did not incorporate patients’ interactions with other professions such as PAs and registered nurses into our analysis. The inclusion of these allied health professions in future research may enhance our understanding of other options of meeting the expected increases in demand for primary care services.

Notwithstanding these caveats, the findings in this study provide valuable information about the types of health care services supplied by NPs and how services provided by NPs can help mitigate the current shortfall in PCP services. The findings from this study also highlight the importance of expanding the use of other allied health profession in primary care delivery.