Persistent shortages and uneven geographic distribution of primary care physicians (PCMDs)1,2 have stimulated proposals to increase the capacity of the nation’s primary care workforce.3–5 Although expanding the use of nurse practitioners (NPs) is frequently advocated,6–9 this proposal has evoked controversies centered largely on questions about the cost and quality of services provided by NPs.10–12
Recently we analyzed the costs of primary care provided to Medicare beneficiaries by primary care nurse practitioners (PCNPs) and by PCMDs.13 Results showed the costs of PCNP-managed care were between 11% and 29% lower than the costs of PCMDs, even after adjusting for beneficiaries’ severity of illness and other characteristics. With respect to the quality of care, while numerous studies conclude NP-provided care is comparable to physicians,14–17 many of these studies did not adequately control for patient selection biases and disease severity, analyzed a limited number of clinical conditions, and assessed quality measures over brief time periods making it difficult to generalize results to broader populations.
The present study addresses these shortcomings by using national Medicare claims data to assess 16 indicators of the quality of care provided by PCNPs and PCMDs to Medicare beneficiaries across 4 domains of primary care. To capture beneficiaries who may have received team care, our analyses went beyond a simple classification of PCNP-managed versus PCMD-managed beneficiaries to include a third category of beneficiaries who received primary care services by both clinicians over a 12-month period. Given evidence from previous research showing NPs produce high quality of care, and because many stakeholders believe physicians provide the gold standard of primary care, we hypothesized the quality of primary indicators we examined would be equivalent or better among beneficiaries managed by PCNPs compared with beneficiaries who were managed by PCMDs or managed by both clinicians.
Data came from 2012 and 2013 Medicare enrollment files and part A and part B billing records for all aged, disabled, and dual Medicaid and Medicare beneficiaries. The area resource file was used to identify beneficiaries’ geographical residence.
As described below, the sample was constructed by selecting random samples of NPs [Centers for Medicare and Medicaid Services (CMS) specialty code=50] and physicians (specialty code=8 or 11) from all clinicians billing Medicare in 2012–2013; gathering all claims submitted to Medicare for 1,000,000 beneficiaries treated by these clinicians; and by applying strategies described by Mehrotra et al18 to attribute these beneficiaries to primary care clinicians.
Starting with clinicians, we oversampled NPs at a rate of 4:1 to account for their smaller beneficiary panel sizes relative to physicians.13 In total, 4065 NPs were randomly selected into the sample until 800,000 beneficiaries with at least 1 billed service from an NP were captured; 549 physicians were randomly selected until 200,000 beneficiaries were captured in the sample.
Next, we gathered all claims for each of these beneficiaries. Because beneficiaries see an average of 6.4 clinicians in a year, gathering all of their claims resulted in a substantial expansion in the number of NPs (51,595) and physicians (160,000) pulled into the sample.
The final step was to exclude 425,356 beneficiaries whose care was provided by medical specialists and institutions (eg, kidney dialysis center), 161,058 who were not continuously enrolled in fee-for-service Medicare during 2012–2013, 72,587 who had no Medicare paid service in one of the study years, and 18,071 beneficiaries with end-stage renal disease, leaving 322,928 beneficiaries for attribution into 3 groups: PCNPs alone, PCMDs alone, and both PCNPs and PCMDs.
To ensure beneficiaries had an established relationship with their primary care clinician, we used 2012 data to attribute beneficiaries to clinicians. The attribution strategy was limited to new (M1A: 99201-99205) and established office visits (M1B: 99211-99215), home (M4A:99340-99345; 99347-99350), and nursing home visits (M4B: 99304-99306; 99307-99310) from the Berenson-Eggers Type of Service Codes. Evaluation and management (E&M) services from inpatient or specialty care setting were excluded so that attribution would focus only on primary care. The proportion of primary care E&M claims was calculated for each beneficiary who was subsequently attributed to the PCNP or PCMD who submitted at least 25% of the E&M services paid by Medicare. In cases where a PCMD and PCNP each provided at least 25% of the E&M claims, the beneficiary was jointly attributed to both clinicians to create the group of beneficiaries who may have received team care.
We examined attribution thresholds ranging between 10% and 35% and selected 25% because it provided a strong relationship between clinician and beneficiary while also capturing a large sample of primary care services jointly attributed to both clinicians (Appendix 1, Supplemental Digital Content 1, http://links.lww.com/MLR/B552). This threshold resulted in attributing 47,814 beneficiaries to 25,384 unique PCNPs, 250,117 beneficiaries to 83,805 unique PCMDs, and 24,997 beneficiaries uniquely attributed to 16,450 PCNPs and PCMDs.
We used claims-based primary care quality measures from the Accountable Care Organization measures set, Healthcare Effectiveness Data and Information Set (HEDIS), Agency for Healthcare Research and Quality (AHRQ), and the New York University (NYU) algorithm for classifying emergency department (ED) utilization.19 With the exception of the NYU algorithm, all measures are endorsed by the National Quality Forum. We grouped measures into 4 primary care domains: chronic disease management, preventable hospitalizations, adverse outcomes, and cancer screening (Appendix 2, Supplemental Digital Content 2, http://links.lww.com/MLR/B553, describes the construction of measures).
Chronic Disease Management
For the chronic disease management domain, measures were constructed for 3 high-cost high-prevalence conditions: diabetes, coronary artery disease (CAD), and chronic obstructive pulmonary disease (COPD).
For beneficiaries with diabetes, we used the HEDIS comprehensive diabetes measure set20 that includes the rates of hemoglobin A1c testing at least twice a year, annual low density lipoprotein screenings, medical attention for nephropathy, and annual eye examination. The presence of each test or service was determined by procedure codes on paid claims from 2013.
Rates of lipid screening were calculated for beneficiaries with CAD20 and spirometry testing for beneficiaries with COPD.21 For each measure, the denominator was the disease cohort and the numerator was the presence of ≥1 target procedure codes (Appendix 2, Supplemental Digital Content 2, http://links.lww.com/MLR/B553). For these measures, higher rates suggest better quality of care.
Preventable Hospitalizations—AHRQ Prevention Quality Indicators (PQIs)
PQIs are population-based measures identifying hospitalizations believed to be avoidable with appropriate primary care. PQIs comprise 13 individual measures which we combined into 3 distinct composite groups: overall, acute care, and chronic disease preventable hospitalizations based on the type of condition requiring hospitalization.22 The original measures are calculated as a ratio, expressed as the number of conditions per 100,000 population aged 18 and older. Our measure captures preventable hospitalizations as 13 0/1 flags grouped into 3 composite measures. Lower rates suggests better quality of care.
One-year incidence rates were calculated for hospital readmissions, inappropriate ED visits, and low-value imaging for low back pain. The HEDIS all-cause readmissions measure was used to identify hospitalizations within 30 days of discharge, excluding transfers, planned surgeries, and maternity care.20 The NYU ED algorithm was used to identify overall ED visits and visits categorized as nonemergent or emergent, but primary care sensitive.19 The metric for low-value imaging identifies the proportion of lumbar magnetic resonance imaging (MRI) studies in the context of diagnosis of low back pain with no claims-based evidence of previous conservative therapy.23 For these 3 measures, lower rates suggest better quality care of care.
Measures relevant to Medicare beneficiaries included the rate of mammography screening for women aged 50–7520,24 and colorectal cancer screening.20,25 Women with average cancer risk should receive a mammogram every 2 years and colorectal screening is recommended every 10 years in healthy adults. Because these measures were assessed over a single year (2013), screening rates will be lower than had they been assessed over their respective longer time horizons. Higher screening rates suggest better quality of care.
To control for the effects of geographic variation in clinician practice patterns, we used beneficiaries’ zip codes from the Medicare enrollment file to identify the CMS region a beneficiary resided in 2012. Because of small sample size, regions 8 and 9 were combined, leaving 7 regions in the model with CMS Region 4 (Alabama, Florida, Georgia, Kentucky, Mississippi, North Carolina, South Carolina, Tennessee) as the reference group. The area resource file was used to identify beneficiaries’ who resided in a rural county. Demographic control variables included age, female sex, and white race versus all other race/ethnicity. Dual Medicaid-Medicare status, a proxy for poverty, was measured as ≥1 months during the calendar year when the state subsidized the beneficiary’s Medicare premium.26 To assess treatment settings, the HEDIS utilization of ambulatory care measure was used to identify whether care occurred in an outpatient, home, nursing facility, or long-stay nursing home setting.20 Finally, because our analysis of state restrictions on NPs’ scope of practice (SoP) showed no evidence that state-level SoP restrictions were related in any consistent or discernable pattern to the quality of care provided by PCNPs, a state’s SoP environment was not included as a control variable.27
We assessed the complexity of beneficiaries’ clinical conditions using the Elixhauser index, a list of 30 comorbid conditions. Comorbidities were identified by diagnostic codes on beneficiaries’ paid claims in 2012.28 Because CMS excluded substance abuse-related claims in research files starting in 2012, we were unable to calculate measures of alcohol and substance dependency.29
To account for beneficiary differences on variables potentially related to quality measures, propensity score weighted regression was used where the weight was constructed from 2 probabilities: the predicted probability a beneficiary would have been treated by a PCNP, and the predicted probability a beneficiary would have been treated by both a PCNP and a PCMD as defined by our attribution algorithm.30 We used multinomial logistic regression including the demographic variables shown in Table 1, care provided in a nursing home as a proxy for frailty, CMS region, and Elixhauser comorbidities to estimate these probabilities. The predicted probabilities (1 for PCNP and 1 for PCNP-PCMD) were then used to create a weight with the PCMD group as the reference (weight=1). For PCNP-attributed beneficiaries the weight was the probability of having been treated by a PCMD/probability of having been seen by a PCNP. Similarly, for beneficiaries jointly attributed to both PCNPs and PCMDs the weight was the probability of having been treated by a PCMD/probability of having been treated by both types of clinician. This generalized propensity score method is designed to create balance when there are multiple treatment groups.31,32
All multivariate models for quality measures used logistic regression and controlled for demographics, region, and comorbidities. As a sensitivity test, the models were estimated using 0.20 and 0.35 attribution thresholds to assess whether results are sensitive to alternative threshold levels.
Beneficiaries attributed to PCNPs were significantly more likely than those attributed to PCMDs to be younger, reside in rural areas, dually eligible for Medicaid and Medicare, and to have originally qualified for Medicare due to a disability rather than aging into the program (Table 1). Beneficiaries attributed to PCMDs had the highest severity of illness, indicated by the mean number of Elixhauser comorbidities, whereas those attributed to PCNPs had the lowest severity, with beneficiaries jointly attributed to both clinicians in between.
Unweighted results for each quality measure in the 4 primary care domains are shown in Table 2. For chronic disease management, significantly higher percentages of beneficiaries attributed to PCMDs received diabetes care (hemoglobin A1c and low density lipoprotein testing, medical attention for nephropathy, and an annual eye examination), and lipid screening for CAD and spirometry testing for COPD. Similarly, in the screening domain the rate of breast cancer screening was 7 percentage points higher for beneficiaries attributed to PCMDs compared with beneficiaries attributed to PCNPs.
Beneficiaries attributed to PCNPs had significantly fewer preventable hospitalizations (measured by PQIs), ED visits, and all-cause hospital readmissions than did beneficiaries attributed to PCMDs or to those jointly attributed to both clinicians. Further, although beneficiaries jointly attributed to both clinicians had higher screening rates than did beneficiaries attributed to PCNPs, they had lower screening rates than beneficiaries attributed to PCMDs (PCNP-PCMD=16%; PCNP=13% and PCMD=18%).
Table 3 reports results of the propensity score weighted logistic regression analysis using PCMDs as the reference group. After adjusting for demographics and comorbidities, beneficiaries attributed to PCNPs and beneficiaries jointly attributed to both clinicians continued to have significantly lower rates of chronic disease management for diabetes, CAD, and COPD, less medical attention for nephropathy among beneficiaries with diabetes, and lower rates of cancer screening. In contrast, beneficiaries attributed to PCNPs had significantly lower odds of having been hospitalized for a preventable condition and these beneficiaries also had lower rates of inappropriate ED use, fewer 30-day all-cause readmissions and lower rates of low-value MRIs for low back pain compared with beneficiaries attributed to PCMDs. Jointly attributed beneficiaries also had significantly lower rates of preventable hospitalizations compared with PCMDs, but higher odds than PCNP-attributed beneficiaries. In particular, the overall PQI composite score are 22% lower for PCNP but 15% lower for jointly attributed beneficiaries.
Finally, sensitivity tests using a 20% and 30% threshold for attribution did not change the relative size, direction, and significance of any of the above results.
This study sought to assess whether the quality of care received by Medicare beneficiaries over a 12-month period differed across 4 domains of primary care when provided predominantly by PCNPs, PCMDs, or by both clinician types. Overall, results suggest that specific types of indicators are stronger in PCNPs and others seem stronger in PCMDs.
Beneficiaries with a disability, dually eligible for Medicaid and Medicare, and living in rural areas were significantly more likely to be attributed to PCNPs than to PCMDs. These beneficiaries were also more likely to have fewer preventable hospital admissions, all-cause readmissions within 30 days, inappropriate ED visits, and fewer low-value MRIs associated with lower back pain. These findings may reflect differences in practice styles and philosophies of care relative to PCMDs, and to differences in patient characteristics and preferences for provider type that could not be assessed in our data. Whatever the explanation, these outcomes have large financial consequences as Medicare spends roughly $100 billion per year on these outcomes.33–35
Beneficiaries attributed to PCMDs were more likely than those attributed to PCNPs to receive more recommended chronic disease management services and cancer screenings. These findings could be explained by PCNPs having fewer organizational resources (eg, support staff and examination rooms),36 requiring a physician to order the screening, differences in beneficiary access to screening technology, particularly for those living in rural areas,37 or differences in clinician incentives. A 2012 national survey found 50% of PCMDs versus 14% of PCNPs reported receiving salary adjustments tied to productivity or quality performance.38 Because the number of beneficiaries is increasing and 3 of 4 have multiple chronic conditions.39,40 PCMDs have a financial incentive to submit claims for services provided to these beneficiaries even when the services were actually provided by PCNPs. This type of billing, termed “incident to,” is allowed under certain circumstances by Medicare.41
The third group of beneficiaries, those jointly attributed to both clinician types, allowed us to examine the possibility of having receiving team care in which quality measures presumably would be stronger than when care was managed by either a PCNP or PCMD. However, because findings indicate that in only 1 measure was care improved, cancer screening, the care received by jointly attributed beneficiaries seems to reflect care fragmentation. Before concluding these findings contradict the belief that teamwork improves care and outcomes, more information on organizational and clinical practice characteristics, such as continuity and comprehensiveness of services, is needed to understand the specifics of patient management when received from both PCNPs and PCMDs.
Study results should be interpreted in light of well-known limitations of claims data for measuring quality of care.42 Such data do not include factors affecting the delivery of primary care, including geographic access, language, culture, organizational, and clinician characteristics.37,43 Our primary care quality measures omit patient education and care coordination, which are increasingly emphasized in delivery system reforms and are recognized hallmarks of PCNPs’ approaches to care.8,44,45
Our attribution strategy assumed all claims paid by Medicare were for services provided by the clinician who actually provided the service. Yet, this assumption is undermined because incident to billing (when a physician bills for services that were actually provided by an NP) cannot be identified in Medicare claims data. Consequently, this means the frequency of incident to billing is not know nor can the direction and magnitude of any bias be ascertained. Incident to billing creates limitations for our study and others claims-based studies analyzing the types, quantities, costs, and quality of care provided by clinicians billing Medicare.13,44 If nothing else, the results of our study illustrate the need for Medicare to change its coding procedures to identify incident to billing in its claims data.46
Our findings have implications for alternative delivery organizations seeking to improve population outcomes by increasing access to primary care.47 A 2017 national survey of accountable care organizations reported reducing utilization of costly services and addressing chronic diseases were among commonly implemented population health management initiatives.48 Decreasing preventable hospital admissions, readmissions, and inappropriate ED use were key strengths of PCNPs found in our study, as was PCMDs’ chronic disease management, suggesting the judicious use of both type of clinicians will yield favorable results for these organizations.
Finally, while some of our results await explanation in future studies, we suggest they may be beneficial in reframing the substitution and competition discourse that often dominates discussions in practice and in education. In particular, faculty in interprofessional education programs can use our results to stimulate discussions about how PCNPs and PCMDs can work together to respond to growing demands for primary care. Discussing how PCNPs and PCMDs achieve the contributions found in our study can enrich mutual appreciation and understanding, and help break down communications and other barriers among practitioners and between nursing and medical students.
2. National Center for Health Workforce Analysis. Projecting the Supply and Demand for Primary Care Practitioners through 2020. Rockville, MD: US Department of Health and Human Services; 2013.
3. Bodenheimer TS, Smith MD. Primary care: proposed solutions to the physician shortage without training more physicians. Health Aff (Millwood). 2013;32:1881–1886.
4. Green LV, Savin S, Lu Y. Primary care physician shortages could be eliminated through use of teams, nonphysicians, and electronic communication. Health Aff (Millwood). 2013;32:11–19.
5. Auerbach D, Chen P, Friedberg M, et al. Nurse-managed health centers and patient-centered medical homes could mitigate expected primary care physician shortage. Health Aff (Millwood). 2013;32:1933–1941.
6. Institute of Medicine. The Future of Nursing: Leading Change, Advancing Health. Washington, DC: National Academies Press; 2010.
8. Naylor MD, Kurtzman ET. The role of nurse practitioners in reinventing primary care. Health Aff (Millwood). 2010;29:893–899.
13. Perloff J, DesRoches CM, Buerhaus P. Comparing the cost of care provided to Medicare
beneficiaries assigned to primary care nurse practitioners and physicians. Health Serv Res. 2016;51:1407–1423.
14. Newhouse RP, Bass EB, Stanik-Hutt J, et al. Advanced practice nurse outcomes: 1990-2008: a systematic review. Nurs Econ. 2011;29:1–21.
15. Kurtzman ET, Barnow BS. A comparison of nurse practitioners, physician assistants, and primary care physicians’ patterns of practice and quality of care in health centers. Med Care. 2017;55:615–622.
16. Jennings N, Clifford S, Fox AR, et al. The impact of nurse practitioner
services on cost, quality of care, satisfaction and waiting times in the emergency department: a systematic review. Int J Nurs Stud. 2015;52:421–435.
17. Martínez-González NA, Djalali S, Tandjung R, et al. Substitution of physicians by nurses in primary care: a systematic review and meta-analysis. BMC Health Serv Res. 2014;14:1–17.
18. Mehrotra A, Adams JL, Thomas JW, et al. The effect of different attribution rules on individual physician cost profiles. Ann Intern Med. 2010;152:649–654.
20. National Committee for Quality Assurance. HEDIS 2013 Volume 2: Technical Specifications. Washington, DC: National Committee for Quality Assurance; 2013.
21. Institute for Clinical Systems Improvement. Diagnosis and Management of Chronic Obstructive Pulmonary Disease (COPD). Bloomington, MN: Institute for Clinical Systems Improvement; 2013.
22. Agency for Healthcare Research and Quality. Prevention Quality Indicators, Technical Specifications. Rockville, MD: Agency for Healthcare Research and Quality; 2013.
24. Nelson HD, Fu R, Cantor A, et al. Effectiveness of breast cancer screening: systematic review and meta-analysis to update the 2009 US Preventive Services Task Force Recommendation. Ann Intern Med. 2016;164:244–255.
25. National Quality Measures Clearinghouse. Colorectal cancer screening: percentage of members 50 to 75 years of age who had appropriate screening for colorectal cancer. 2015. Available at: www.qualitymeasures.ahrq.gov
. Accessed September 5, 2016.
26. Research Data Assistance Center. Identifying dual eligible Medicare
beneficiaries in the Medicare
beneficiary enrollment files. 2012. Available at: www.resdac.org/resconnect/articles/149
. Accessed September 5, 2016.
27. Perloff J, Clarke S, DesRoches C, et al. Association of state-level scope of practice restrictions with the quality of primary care provided to Medicare
beneficiaries by nurse practitioners. Med Care Res Rev. 2017. Available at: https://doi.org/10.1177/1077558717732402
. Accessed September 14, 2017.
28. Elixhauser A, Steiner C, Harris DR, et al. Comorbidity measures for use with administrative data. Med Care. 1998;36:8–27.
29. Rough K, Bateman BT, Patorno E, et al. Suppression of substance abuse claims in Medicaid data and rates of diagnoses for non-substance abuse conditions. JAMA. 2016;315:1164–1166.
30. Weintraub WS, Grau-Sepulveda MV, Weiss JM, et al. Comparative effectiveness of revascularization strategies. N Engl J Med. 2012;366:1467–1476.
31. Spreeuwenberg M, Bartak A, Croon M, et al. The multiple propensity score as control for bias in the comparison of more than two treatment arms: an introduction from a case study in mental health. Med Care. 2010;48:166–174.
32. Yang S, Imbens GW, Cui Z, et al. Propensity score matching and subclassification in observational studeis with multi-level treatments. Biometrics. 2016;72:1005–1065.
33. Joynt KE, Gawande AA, Orav E, et al. Contribution of preventable acute care spending to total spending for high-cost Medicare
patients. JAMA. 2013;309:2572–2578.
34. Jiang J, Russo C, Barrett M. Nationwide frequency and costs of potentially preventable hospitalizations, 2006. Statistical brief #72. 2009. Available at: www.ncbi.nlm.nih.gov/books/NBK53971/
. Accessed September 1, 2017.
36. Poghosyan L, Nannini A, Smaldone A, et al. Revisiting scope of practice facilitators and barriers for primary care nurse practitioners: a qualitative investigation. Policy Polit Nurs Pract. 2013;14:6–15.
37. Graves JA, Mishra P, Dittus RS, et al. Role of geography and nurse practitioner
scope-of-practice in efforts to expand primary care system capacity: health reform and the primary care workforce. Med Care. 2016;54:81–89.
38. Buerhaus PI, DesRoches CM, Dittus R, et al. Practice characteristics of primary care nurse practitioners and physicians. Nurs Outlook. 2015;63:144–153.
39. Gerteis J, Izrael D, Deitz D, et al. Multiple Chronic Conditions Chartbook. Rockville, MD: Agency for Healthcare Research and Quality; 2014.
42. Iezzoni LI. Assessing quality using administrative data. Ann Intern Med. 1997;127:666–674.
43. Donelan K, DesRoches CM, Dittus RS, et al. Perspectives of physicians and nurse practitioners on primary care practice. N Engl J Med. 2013;368:1898–1906.
44. DesRoches CM, Gaudet J, Perloff J, et al. Using Medicare
data to assess nurse practitioner
-provided care. Nurs Outlook. 2013;61:400–407.
45. Xue Y, Intrator O. Cultivating the role of nurse practitioners in providing primary care to vulnerable populations in an era of health-care reform. Policy Polit Nurs Pract. 2016;17:24–31.
48. Hsu J, Price M, Vogeli C, et al. Bending the spending curve by altering care delivery patterns: the role of care management within a pioneer ACO. Health Aff (Millwood). 2017;36:876–884.