Primary Care Practices Providing a Broader Range of Services Have Lower Medicare Expenditures and Emergency Department Utilization

Background

Comprehensiveness is a key element of primary care [1, 2, 3], and is highlighted in several national primary care transformation initiatives [4]. It has been defined as the extent to which a patient’s primary care provider (practitioner, practice, or team) recognizes and meets the large majority of the patient’s physical and common mental health care needs, including prevention and wellness, and acute, chronic, and comorbid condition management [3]. This concept invokes several aspects of patient care [5]. These include the depth and breadth of conditions managed by the primary care practitioner (PCP), as well as the extent to which the PCP can effectively address the many relatively common problems their patients may experience [5, 6]. Recently, Medicare claims have been used to develop reliable and valid measures of these two dimensions of primary care comprehensiveness [6, 7].

The range of services primary care practices provide is a third key aspect of comprehensive primary care [8], and perhaps the easiest to understand. Delivering commonly needed services like minor procedures is within the competencies of PCPs and offering such services can avoid unnecessary visits to other providers. Unlike the ability to correctly diagnose new problems or integrate pharmacotherapy plans across a patient’s multiple conditions, individual PCPs may not need to personally maintain technical competence with specific procedures if someone at their practice site does. Survey assessments of primary care comprehensiveness developed by Barbara Starfield and her team [3] have focused on the range of services (such as minor procedures, immunizations, and routine counseling) primary care practices offer [8], as has more recent work describing variations in primary care in other countries [9].

However, surveys are costly to implement and burdensome to practitioners [5]. Therefore, we sought to develop a reliable and valid claim-based approach to measure the range of services provided by a primary care practice site. We describe the variations in this measure across primary care practice sites, and test the predictive validity of the measure, investigating whether it was associated with total costs of care as well as hospital admissions and emergency department (ED) visits. We also explore the practice characteristics associated with meaningful variations in the range of services offered by primary care practice sites.

Methods

This work was conducted in the context of the 4-year Comprehensive Primary Care (CPC) initiative launched in October 2012 by the Centers for Medicare & Medicaid Services (CMS), joined by 39 other private and public payers [10]. CPC tested whether providing financial and technical support for particular enhanced primary care activities reduced costs and improved quality in 502 practices across seven US regions. Additional details regarding the CPC initiative are included in the Appendix.

Measure of Practice Site Range of Services (Practice-ROS)

The Practice-ROS measure assesses the comprehensiveness of services that the PCPs at a practice site provided to Medicare fee-for-service (FFS) beneficiaries. We consulted several sources to identify the potential services (beyond face-to-face office visits) that practice sites should provide to Medicare beneficiaries. These sources included Primary Care Assessment Tool surveys developed by Starfield and colleagues [8], a refinement of this practice survey developed under technical expert guidance for the Agency for Healthcare Research and Quality [11], and a list of services documented in assessments of comprehensiveness of primary care practices in other developed countries [9]. From this combined list, we identified a subgroup of primary care services relevant to Medicare beneficiaries over age 65 that could be detected through Medicare claims. The 5 categories of services identified consisted of [1] immunizations, [2] counseling for behavioral or mental health problems, [3] treatment of a minor laceration, [4] cryotherapy and/or skin excision, and [5] joint or tendon injection. To identify the specific Healthcare Common Procedure Coding System (HCPCS) codes relevant to these 5 categories of services PCPs deliver, we identified the codes that represented at least 0.5% of the services in each category billed by PCPs in the observed practices. Appendix Table B presents details on the categories and HCPCS codes.

To calculate the Practice-ROS score for each primary care practice site, we then reviewed all the Medicare claims billed by each PCP (primary care physician, nurse practitioner [NP], physician assistant [PA], or certified clinical nurse specialist [CCNS]) practicing at that site during 2013. We used the practitioner billings under the Tax Identification Number (TIN) of the practice organization for the primary care practice site to avoid counting services practitioners provided in unrelated practice settings (such as PCPs “moonlighting” in emergency rooms). If any of the PCPs at the practice site’s TIN billed for at least one instance of providing a Medicare FFS beneficiary a specific service in a category, the practice site was scored as being able to provide services in that category. Thus, the Practice-ROS measure ranges from 0 to 5, with a 5 indicating PCPs at the practice site collectively provided at least one instance of each of the 5 categories of service (immunizations, counseling for behavioral or mental health problems, treatment of a minor laceration, cryotherapy and/or skin excision, and joint or tendon injection) to beneficiaries seen at that practice site that year. A score of 3 indicated the practice provided at least 1 instance of each of 3 service categories, and 0 indicated that practitioners at the practice site never billed for any of these service categories during the year.

Data Sources

Data

To construct our measure, we used Medicare FFS claims in the Research Identifiable Files from the CMS Virtual Research Data Center. To calculate the Practice-ROS, we analyzed 2013 Medicare Part B claims submitted by CPC and comparison practice sites included in the CPC evaluation. We observed 6050 primary care physicians, NPs, PAs, and CCNSs in 1383 CPC and comparison practices in 2013, using claims of all 1,232,940 beneficiaries they had seen. In all our analyses, we combined CPC and comparison practices since our focus was to measure development rather than investigate differences between these groups. To participate in the CPC initiative, practice sites needed to have at least 100 Medicare FFS beneficiaries and 1 primary care clinician.

We examined cost and utilization outcomes in 2014 for beneficiaries attributed to all these practices in 2013 and still alive and enrolled in fee-for-service Medicare in 2014. We constructed beneficiary-level outcomes using beneficiaries’ 2014 Medicare claims for services received from all providers.

Control variables for patient, practice, and market characteristics came from a range of sources, including the Medicare enrollment database (EDB), CPC application data, the Area Resource File, SK&A, the Medicare Data on Provider Practice and Specialty (MD-PPAS), the Health Resources and Services Administration (HRSA), and the National Committee for Quality Assurance (NCQA) (Table 4 provides details on data sources).

Primary Care Practitioners

We identified PCPs in practice sites in 2013 using the following practitioner taxonomy codes in the MD-PPAS: internal medicine, family practice, pediatric medicine, geriatric medicine, general practice, NPs, and PAs. We excluded hospitalists.

Beneficiary Attribution

For CPC, CMS attributed beneficiaries to CPC and comparison practices where they had received the largest share of their primary care in the prior 2 years. On a quarterly basis, attribution started with all Medicare FFS professional services and outpatient claims for office/outpatient evaluation and management visits, nursing home and home care visits, or Medicare and annual wellness visits as determined by the HCPCS codes on the claim. Using these claims, we identified the practice that accounted for the greatest share of these services for each beneficiary [6].

Statistical Analysis

Reliability Testing

We assessed reliability of the Practice-ROS measure using the split-half method to observe the repeatability and reproducibility for the same population at the same time [12]. We randomly split each practice site’s professional services claims in half, computed the measure score for each practice site in each split sample, and then assessed correlations between practice site scores for the two samples. We calculated Spearman’s rho to quantify the repeatability and reproducibility of the measured result across the two half-samples.

Associations with Practice Characteristics

We explored various practice site covariates potentially relevant to the range of services the site might offer. These included practice site size (1–3 practitioners, in contrast to larger sites). We observed the dominant primary care physician specialty at the practice site (if there was one), by constructing dummy variables indicating if more than 50% of physicians at the practice site were in a particular primary care specialty (such as internal medicine) as well as when no specialty represented more than 50% of the physicians (>50% of physicians in family medicine was the reference group). Because the ease of providing diverse services at a practice site might be mediated by the presence of advanced practice clinicians, we included a covariate indicating whether there were PAs or NPs at the practice site. Since the type of setting (and thus availability of resources, and incentives to refer) may vary based on the practitioner’s proximity to and affiliation with a hospital-based health system, we included an indicator for whether the physicians at the practice site billed at least 95% of outpatient claims (defined as those in either a hospital outpatient department or non-hospital-based office setting) from a hospital outpatient department (9% of the sample). Additional characteristics of interest included whether the practice site was owned by a health system, was recognized as a medical home, or met meaningful use criteria for electronic health records. In multivariate modeling, we also included a variable to control for whether the practice site was in the CPC intervention or comparison group for the CPC initiative (note the CPC intervention was not designed to specifically require or reward expanded practice site range of services). To better isolate associations with the covariates of interest, we also controlled for other demographic characteristics of the physician’s patient panel and for county-level characteristics of the practice site. Finally, we included two other measures of PCP physician comprehensiveness, New Problem Management and Involvement in Patient Conditions, [6] to assess the relationship of the Practice-ROS to these dimensions of primary care comprehensiveness.

For the bivariate comparisons of practice and market characteristics with Practice-ROS, we calculated Pearson correlations for the continuous variables and T-tests of independent group means for the categorical variables. To estimate the independent association between practice site-level covariates and Practice-ROS count for the site, we used multivariate Poisson regression analysis. Each model included the covariates of interest—physician specialty, service location, and characteristics of the practice site—as well as controls for Medicare claims per PCP, practice site physician demographics, the practice site patient panel, and county-level factors. We did not adjust p-values to account for multiple comparisons, but in our analyses, we used a conservative significance level of p ≤ 0.01 for all measures, to avoid type I errors.

Association with Beneficiary Cost and Utilization Outcomes

We used a lagged analysis to test for associations between the Practice-ROS measure (in year 2013) and 3 beneficiary-level cost and utilization outcomes (measured in year 2014): total Medicare FFS expenditures per beneficiary per month, and annualized measures of the number of hospitalizations and ED visits per 1000 beneficiaries. Hospitalizations and ED visits were modeled using zero-inflated negative binomial regression (to account for possible over-dispersion in utilization counts and the large percentage of zeroes for beneficiaries with no use during a year). Total Medicare expenditures were modeled using ordinary least squares regression. Values above the 99th percentile were Winsorized (reset) to the 99th percentile to avoid the potentially distorting effects of extreme outliers. Each regression model controlled for the same beneficiary, practice, and market characteristics identified in 2012, before the start of CPC. All models accounted for clustering of patient outcomes within practices. Each observation (beneficiary) was weighted to account for the percentage of the year the beneficiary was eligible for Medicare. For beneficiaries in the comparison practices, the weights were adjusted for practice-level matching. Analyses were conducted using STATA 14.

Beneficiary-level control variables included Hierarchical Condition Category score, race, age, sex, original reason for Medicare eligibility (disability or end-stage renal disease, versus age), and dual status (eligible for both Medicare and Medicaid). We also included control variables for the beneficiary’s practice and local market as described above.

To ease interpretation of the associations between practice comprehensiveness measures and beneficiary-level outcomes, we report both the magnitude and percentage difference in the adjusted mean outcomes for an increase in the comprehensiveness score from the 25th to 75th percentile among all practice sites in the analysis.

Results

Variations in Services Provided Across Practice Sites

The provision of the different service categories varied substantially across practice sites (Table 1). Ninety-four percent of practice sites provided immunizations (1300 of 1383), but only 1.7% of practice sites had PCP billing for mental health counseling services. Many more practice sites provided joint or tendon injections (73.2%) than repair of minor lacerations (37.6%). Using the information regarding practice site billings for each category of service, we constructed the Practice-ROS score for each site. Only 1.4% of the 1383 practice sites provided none of the 5 categories of services (score of 0) and only 0.6% of the practice sites provided Medicare beneficiaries with all 5 categories of services in 2013 (score of 5).

Table 1 Variation in Services Provided by Practice Sites

Reliability of Practice-ROS Measure

Spearman’s rho of 0.84 indicates good repeatability and reproducibility of the Practice-ROS measure across the two half-samples.

Practice-ROS Association with Practice Characteristics

Bivariate analyses found a variety of practice characteristics associated with Practice-ROS at p<0.01 (Tables 2 and 3). Practice sites serving urban populations and in areas with higher median income provided lower Practice-ROS. Practice sites had higher Practice-ROS if they had patient-centered medical home certification, had a practice site of ≥ 4 physicians, had one or more NPs/PAs, or had >50% of the PCPs in family medicine.

Table 2 Bivariate Association of Practice Site Characteristics with Practice Range of Services (ROS)—Continuous Variables (Pearson Correlations)

Table 3 Bivariate Association of Practice Site Characteristics with Practice Range of Services (ROS)—Categorical Variables (T-Test of Independent Group Means)

Multivariate analysis confirmed that practices with more than 50% of primary care physicians in internal medicine provide a narrower range of services than do those with more than 50% of primary care physicians in family medicine (marginal effect on Practice-ROS −0.68, p < 0.001) (Table 4). Larger practice sites with 4 or more physicians provide broader ROS than practice sites with 1–3 physicians (+0.40, p < 0.001). After controlling for other practice characteristics, other measures of primary care comprehensiveness (Involvement in Patient Conditions or New Problem Management [6]) were not associated with Practice-ROS (p> 0.50 and 0.34, respectively).

Table 4 Multivariate Associations of Practice Site Characteristics, Including PCP Comprehensiveness Measures, with the Practice Range of Services

Practice-ROS Associations with Medicare FFS Beneficiary Outcomes

In multivariate analysis, Practice-ROS was associated with cost and utilization outcomes of Medicare beneficiaries observed in the following year. As shown in Table 5, Medicare beneficiaries served by primary care practice sites that offered Practice-ROS at the 75th percentile of practices had 3.2% lower rates of ED visits (p<0.01) yielding 22.27 fewer ED visits per 1000 beneficiaries. Likewise, beneficiaries cared for by sites that offered Practice-ROS at the 75th percentile of practices had 3.1% lower Medicare spending (p < 0.01), and therefore $25.03 less monthly spending per beneficiary compared to beneficiaries served by primary care practice sites with Practice-ROS at the 25th percentile. There was no association between Practice-ROS and hospitalization rates (p = 0.119).

Table 5 Practice-ROS Associations with Medicare FFS Beneficiary Outcomes

Discussion

This paper describes a new claim-based approach to measuring the range of services provided by a primary care practice site. We find that this measure has high reproducibility and detects substantial variation across primary care practice sites. We demonstrate that Medicare FFS beneficiaries served by primary care practice sites that offered a broader range of services had 3% lower ED visit rates and Medicare spending in the subsequent year.

Like prior survey approaches to measuring the range of services primary care practices offer, the Practice-ROS measure evaluates whether these services were provided to Medicare FFS beneficiaries by any PCP at the primary care practice site, not just by specific PCPs. In the USA, primary care practice sites can involve several different types of primary care practitioners, so the CPC initiative included physicians in internal medicine, family medicine, pediatric medicine, geriatric medicine, and general practice, as well as nurse practitioners and physician assistants. PCPs in these disciplines varied in the rate at which they billed for the different categories of services (see Appendix Table 1). Not surprisingly, therefore, practice-ROS varied with PCP characteristics. Specifically, practices with more than 50% of primary care physicians in internal medicine are more likely to offer a narrower range of services. This is consistent with differences in training between internal medicine and family medicine [13, 14, 15, 16], as well as with other recent observations of services provided in primary care practices [17]. These findings suggest the internal medicine community may wish to encourage primary care internal medicine physicians to ensure their skills related to providing these services are current, or otherwise make these services readily available at their practice sites. Unfortunately, procedure skills acquired during training may not reflect those regularly used in practice [15, 18].

We find that smaller practice sites, on average, offer a narrower ROS than larger ones. This may not be surprising since each additional PCP may bring some skills with specific services that could benefit all beneficiaries cared for in the practice. However, Practice-ROS was not broader when primary care sites were owned by larger entities such as hospitals or health systems.

Variations in two other claim-based measures of primary care comprehensiveness, New Problem Management (NPM) and Involvement in Patient Conditions (IPC), are not associated with Practice-ROS. IPC measures the degree to which the individual PCP [6], or PCPs at the practice [7], is involved in the broad range of conditions for which their patients receive care, while NPM measures the extent to which the PCPs effectively address the various common new problems their patients present to them [5, 6, 7]. Therefore, as intended, these three distinct measures (Practice-ROS, IPC, and NPM) capture different aspects of primary care comprehensiveness [5, 6]. Variations in each are associated with important variations in patient cost and utilization outcomes, but further research will be required to clarify the relative marginal benefit of improvements on each dimension of comprehensiveness. Furthermore, effective primary care also requires accessibility, interpersonal continuity, and care coordination, in addition to comprehensiveness [3]. Conceptually, improvements in each of these primary care dimensions might facilitate delivering more comprehensive care to a complex Medicare beneficiary [6]. However, we lacked sufficient claim-based measures of these primary care dimensions to assess the independent impact of these other aspects of primary care.

Additional work will be required to confirm that improvement within practices in performance on Practice-ROS can yield important benefits for patients. Given our observational study design, we cannot exclude the possibility that better outcomes for patients whose practices provide a broader range of services may be due to other unmeasured features of the practice setting that are correlated with the measure. Another limitation of any claim-based measure of range of services is that it can only describe services billed by the practice site, which will inevitably be influenced by the mix of patients seeking care. During some years, the covered beneficiaries a practice sees may not require the full range of services available, and thus the observed Practice-ROS may be narrower than the range of services the practice could provide. In addition, unless the measure uses all payer claims, the precision of the Practice-ROS might be influenced by the proportion of practice patients seen who are covered by the relevant payer (in this case, Medicare FFS). It is possible for Medicare beneficiaries to receive no relevant services from the practice, while patients covered by other insurers may receive such needed services during the year (thus causing the Medicare claim-based Practice-ROS to be a falsely low measure of what the practice delivers). Nonetheless, we note that the Practice-ROS measure proved highly reliable in the “split half” analysis. This indicates our claim-based Practice-ROS measure can be informative in practices that achieved the minimum of 100 Medicare FFS beneficiaries required for inclusion in the CPC evaluation.

The observed association of Practice-ROS with cost and utilization outcomes is not only statistically significant but potentially policy-relevant. Beneficiaries cared for by more comprehensive practices (at the 75th percentile on the Practice-ROS measure) have on average 3.1% lower total Medicare spending relative to beneficiaries cared for by less-comprehensive practices (at the 25th percentile on Practice-ROS). The stronger association of Practice-ROS with ED visits than with hospitalizations reinforces the face validity of the measure. Practices that provide additional on-site services like procedures for treating lacerations, skin lesions, or joint and tendon complaints may avert some patient visits to the ED, as observed in our analysis, while office-based treatment of these patient concerns may be less likely to prevent hospitalizations. In this era of growing value-based payment, public policy-makers, private payers, and many practice organizations are interested in reducing unneeded ED visits and overall Medicare spending. Practice-ROS can be readily measured using diverse data sources from Medicare claims to electronic health records, and various strategies could encourage primary care practice sites to provide a broader range of the services most needed by their Medicare beneficiaries.

In conclusion, we describe a promising new claim-based measure of practice site ROS. Our analysis suggests that Medicare beneficiaries experience fewer ED visits and lower total Medicare spending when their primary care practices provide a broader range of services.