Primary care and specialty providers: an assessment of continuity of care, utilization, and expenditures.

ORIGINAL ARTICLE

Primary Care and Specialty Providers An Assessment of Continuity of Care, Utilization, and Expenditures Melissa A. Romaire, PhD, MPH,*wz Susan G. Haber, ScD,*wz Suzanne G. Wensky, PhD,y and Nancy McCall, ScD*wz

Background: Little is known as to whether medical home principles, such as continuity of care (COC), would have the same effect on health service use for individuals whose primary (or predominant) provider is a specialist instead of a primary care provider (PCP). Objective: To test associations between health service use and expenditures and (1) beneficiaries’ predominant provider type (PCP or specialist) and (2) COC among beneficiaries who primarily see a PCP and those who primarily see a specialist. Research Design: This is a cross-sectional analysis of Medicare fee-for-service claims data from July 2007 to June 2009. Negative binomial and generalized linear models were used in multivariate regression modeling. Subjects: The study cohort comprised 613,471 community-residing Medicare fee-for-service beneficiaries. Measures: Beneficiaries’ predominant provider type and COC index during a baseline period (July 2007–June 2008) were studied. All-cause and ambulatory care sensitive condition (ACSC) hospitalizations and emergency department (ED) visits and related expenditures and total expenditures in a 1-year follow-up period (July 2008–June 2009) were also reported. Results: Twenty-five percent of beneficiaries primarily saw a specialist. Having a specialist predominant provider was associated with 9% fewer ED visits, 14% fewer ACSC ED visits, and 8% fewer ACSC hospitalizations (all P < 0.001). Regardless of whether the beneficiary’s predominant provider was a specialist or a PCP, higher continuity was associated with fewer all-cause hospitalizations and ED visits and lower expenditures for these services. Higher continuity was also associated with lower total expenditures. From the *RTI International, Research Triangle Park, NC; wRTI International, Waltham, MA; zRTI International, Washington, DC; and yThe Centers for Medicare & Medicaid Services, Baltimore, MD. Funded by The Centers for Medicare & Medicaid Services Contract Number 500-2005-00029I. The statements contained herein are those of the authors and do not necessarily reflect the views or policies of the Centers for Medicare & Medicaid Services. The authors declare no conflict of interest. Reprints: Melissa A. Romaire, PhD, MPH, RTI International, 3040 E. Cornwallis Rd., Research Triangle Park, NC 27709. E-mail: [email protected]. Supplemental Digital Content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Website, www.lwwmedicalcare.com. Copyright r 2014 by Lippincott Williams & Wilkins ISSN: 0025-7079/14/5212-1042

1042 | www.lww-medicalcare.com

Conclusions: Regardless of the predominant provider’s specialty, greater continuity was associated with less use of high-cost services and lower expenditures for these services. Key Words: primary care, continuity of care, Medicare, health utilization, health expenditures (Med Care 2014;52: 1042–1049)

S

ince its inception, the medical home concept has gained momentum as a promising framework for providing patient-centered care through a long-term relationship between patients and their health care providers.1–3 Numerous policy statements, clinical practice demonstration programs, including some sponsored in part by Medicare, and medical home recognition processes operationalize the medical home in the context of a primary care provider (PCP), recognizing that a key tenet of primary care is to provide first-contact, whole-person care that extends beyond specific episodes of illness or particular medical conditions.1,4–7 However, many individuals, particularly seniors, with high clinical need rely on medical specialists to treat and manage a range of chronic conditions.8 Among Medicare enrollees, Starfield et al9 found that the number of visits to specialists increases with increasing comorbidity and that specialist visits were common, even for high-prevalence conditions commonly treated within primary care. Pham et al10 estimated that 34% of Medicare beneficiaries’ primary physician is a medical specialist or surgeon on the basis of plurality of visits. As the prevalence of chronic conditions increases and the availability of PCPs steadily decreases, there may be an increased reliance on specialists.11 For individuals with chronic conditions, including seniors, a specialist may serve as the primary physician by providing first-line care for the patient’s chronic condition while also addressing patients’ other general health care needs.12,13 In these instances, questions have been raised as to whether or not specialists are suitable and equipped to serve as patients’ medical home providers.2 Continuity of care (COC) is a cornerstone of the medical home model.1 COC is defined as the presence and use of a regular provider over time, and increased COC with PCPs has been shown to improve patient satisfaction, medication compliance, and receipt of preventive services and to reduce hospitalizations and emergency department (ED) visits.14–18 Little is known as to whether medical home Medical Care

Volume 52, Number 12, December 2014

Medical Care


principles, such as COC, would have the same effect on health service use for individuals whose primary (or predominant) provider is a specialist instead of a PCP. In this study, we (1) describe the type of provider, primary care or specialist, predominantly seen in a sample of community-residing Medicare fee-for-service (FFS) beneficiaries; (2) examine health service use and expenditures by beneficiaries’ predominant provider type; and (3) test associations between COC and health service use and expenditures among beneficiaries with PCPs as their predominant provider compared with beneficiaries with specialists as their predominant provider.

METHODS Data and Study Population We conducted a cross-sectional analysis of Medicare enrollment data and Part A and Part B FFS claims for a 5% sample of community-residing beneficiaries enrolled in Medicare at some point between July 2007 through June 2009 (N = 1,398,284). Beneficiaries were defined as communityresiding if they had 0 days in a nursing facility over the study period. Beneficiaries were excluded from the analysis if: (1) they were not enrolled in Medicare Parts A and B for at least 1 month during the baseline period (July 2007–June 2008) and during the follow-up period (July 2008–June 2009) (N excluded = 189,491); (2) their predominant provider could not be identified because the plurality of their ambulatory claims were missing provider identification data, that is, missing the National Provider Identifiers (NPI) (N excluded = 348,934); (3) their predominant provider did not have a primary care or medical specialty of interest (described in greater detail below) (N excluded = 33,282); (4) they did not have complete covariate data available (N excluded = 33,354); or (5) they did not have at least 3 ambulatory care visits (described in greater detail below) (N excluded = 179,752). The final study sample included 613,471 beneficiaries. We used the Area Resource File to ascertain information on availability of generalist and specialist providers in the county in which a beneficiary resides, county level per capita income, and percent of the county residents 25 years of age or older with a high school education.19 To ascertain the specialty of a beneficiary’s predominant provider, we used the National Plan and Provider Enumeration System (NPPES).20

Independent Variables: Predominant Provider, COC Index, and Usual Provider Continuity (UPC) Index Predominant Provider The predominant provider was defined as the provider who the beneficiary saw for the plurality of their FFS evaluation and management ambulatory visits during the baseline year, July 1, 2007 through June 30, 2008.10 A list of the Current Procedural Terminology codes used to identify outpatient evaluation and management claims in the carrier and institutional outpatient Medicare claims files can be found in Supplemental Digital Content 1, http://links. r

2014 Lippincott Williams & Wilkins

Provider Type: Continuity of Care

lww.com/MLR/A811. The beneficiary’s predominant provider was identified as either a PCP or a medical specialist. We identified the predominant provider’s NPI and assigned that NPI to his/her primary specialty as recorded in primary taxonomy code variable of the National Plan and Provider Enumeration System (NPPES) Downloadable File.20 A list of the taxonomy codes that were assigned as primary care or medical specialty can be found in Supplemental Digital Content 1 (http://links.lww.com/MLR/A811). If a beneficiary’s predominant provider was not on our list of specialties (eg, optometrist, podiatrist), the beneficiary was excluded from the analysis. In addition, beneficiaries whose predominant provider typically does not have responsibility for ongoing patient care were excluded (eg, anesthesiologist, radiologist, emergency medicine or critical care physician, pathologist, surgeon) because there is little expectation that COC could be established over the long term. For the 5% of the study sample who had a PCP and a specialist tied as the predominant provider, the PCP was chosen. In a sensitivity analysis, we changed the predominant provider to a specialist, and the results presented here were unchanged. If there were ties between PCPs or between specialists, the provider with the most charges on their ambulatory visit claims was chosen.

COC Index The Bice-Boxerman index was used as the primary measure of care continuity (the formula can be found in Supplemental Digital Content 1, http://links.lww.com/MLR/ A811). This index is a measure of dispersion of visits across different providers; scores range from 0 (no continuity) to 1 (high level of continuity).21 The index was constructed using all ambulatory visits during the baseline period, but because the index can be uninformative for beneficiaries with few clinical visits, we restricted this analysis to beneficiaries who had Z3 PCP or specialist ambulatory visits.22 We categorized beneficiaries into 3 groups on the basis of tertiles: low COC (score between 0 and 0.286); medium COC (score between 0.287 and 0.533), and high COC (score between 0.534 and 1). In a secondary analysis, we also constructed the index using only visits to the provider type of interest, for example, visits to a PCP for those beneficiaries whose predominant provider was a PCP and specialty visits for beneficiaries whose predominant provider was a specialist. We did this to explore whether results were sensitive to the visit type used to construct the index.

UPC Index Because COC is conceptualized across a range of different clinical practice and visit patterns, we also examined an alternative measure, the UPC Index, which is a measure of concentration of visits to the predominant provider (the formula can be found in Supplemental Digital Content 1, http://links.lww.com/MLR/A811).16,23 We categorized beneficiaries into 3 groups on the basis of tertiles: low UPC (score between 0 and 0.417); medium UPC (score between 0.418 and 0.615), and high UPC (score between 0.616 and 1). Similar to the sensitivity analysis described above, we constructed 2 additional indices, the first using www.lww-medicalcare.com |

1043

Romaire et al

only ambulatory visits to a PCP and the second using only those ambulatory visits to a specialist.

Dependent Variables: Utilization and Expenditures Our first set of dependent variables included the following utilization measures during the follow-up period (July 2008–June 2009): all-cause hospitalizations, hospitalizations for ambulatory care sensitive conditions (ACSC), all-cause ED visits, and ACSC ED visits. ACSCs are those conditions for which timely and effective outpatient care could potentially prevent the need for the admission; a list of ACSC conditions included in this analysis can be found in Supplemental Digital Content 1 (http://links.lww.com/MLR/A811). Our second set of dependent variables included expenditures, defined as the amount paid by Medicare (out-of-pocket payments, third-party payments, or Part D expenditures are not included), for the following services during the follow-up period: all-cause hospitalizations, ACSC hospitalizations, all-cause ED visits, ACSC ED visits, and total expenditures. Total expenditures were defined as the summation of paid amounts on all Part A and Part B claims incurred during the follow-up period. We annualized visits and expenditures for those beneficiaries not enrolled for the full 12 months of the follow-up period, and we adjusted expenditures by the 2008 hospital wage index to account for geographic variation in Medicare payment.24

Statistical Analysis Multivariate, negative binomial regression was used to assess the effect of the independent variables on health care visits. Multivariate, generalized linear models (GLM) with gamma family and log link were used to assess the effect of the independent variables on health care expenditures.25 GLM accounts for the characteristically right skewed distribution of the expenditure data.25 We also tested the models using ordinary least squares on log-transformed expenditures, and there were no substantive differences. All analyses controlled for baseline sex, age, race, enrollment in Medicaid, original reason for Medicare eligibility (aged, disabled, or have end-stage renal disease), comorbidity (using the Charlson Index26 and the Hierarchical Condition Category risk score27), residence in an urban area, state of residence, number of PCPs per 100,000 residents in county of residence, number of specialists per 100,000 residents in county of residence, per capita income in county of residence, percent of residents 25 years of age or older with a high school diploma in the county of residence, and death during the follow-up period. The expenditure models also controlled for annualized expenditures during the baseline period. In the models examining the effect of having a specialist provider, we included 2 additional covariates, number of unique providers seen by the beneficiary and number of ambulatory visits during the baseline period. These covariate were not included in the COC regression models because they are components of the COC index. The COC analyses were stratified a priori by the specialty type of the predominant provider (primary care vs. specialist). We examined how measures of association between continuity and utilization and expenditures differed in


Medical Care


the stratified models to draw conclusions about the consistency of these relationships across specialty type of the predominant provider. All analyses were weighted by the fraction of the follow-up period the beneficiary was enrolled in FFS Medicare. Stata was used for all analyses (version 12; StataCorp LP, College Station, TX). Study procedures were approved by RTI’s Institutional Review Board.

RESULTS Select characteristics of the study sample at baseline are summarized in Table 1. One quarter of beneficiaries had a specialist as their predominant provider in the baseline period. Among beneficiaries with a specialist as their predominant provider, cardiologists (23%), hematologists/oncologists (9%), urologists (9%), and dermatologists (8%) were the most frequently reported. Among those with a PCP as their predominant provider, family practice (45%) and general internal medicine (45%) physicians were the most frequently reported. On average, 57% of a beneficiary’s total ambulatory visits were to their assigned predominant provider. The mean Bice-Boxerman COC score was 0.473 across all beneficiaries. Those with a primary care predominant provider had a higher mean score (COC score = 0.517) compared with those with a specialist predominant provider (COC score = 0.340). The mean UPC score across all beneficiaries was 0.550, again with beneficiaries with a PCP predominant provider having a higher score (UPC score = 0.572) compared with those who had a specialist (UPC score = 0.481). When the Bice-Boxerman COC and UPC indices were calculated using only PCP visits or only specialist visits, continuity was higher for all beneficiaries, regardless of their predominant provider type (Table 1).

Association Between Predominant Provider and Utilization and Expenditures In multivariate models, beneficiaries whose predominant provider was a specialist had 2% fewer hospitalizations [incidence rate ratio (IRR) = 0.98 (95% confidence interval (CI): 0.96–0.99)], 8% fewer ACSC hospitalizations [IRR = 0.92 (95% CI: 0.90–0.94)], 9% fewer ED visits [IRR = 0.91 (95% CI: 0.90–0.92)], and 14% fewer ACSC ED visits [IRR = 0.86 (95% CI: 0.84–0.87)] compared with beneficiaries whose predominant provider was a PCP. In an exploratory analysis of these findings, this negative association between predominant provider and less use was most pronounced among beneficiaries with high comorbidity, that is, Charlson Index Z2 (Table 2). In regression analyses, beneficiaries whose predominant provider was a specialist had significantly greater all-cause hospitalization expenditures [b = 0.028 (95% CI: 0.008–0.047)] and total expenditures [b = 0.058 (95% CI: 0.047–0.070)] and lower ACSC ED expenditures [b = 0.087 (95% CI: 0.130 to 0.045)] compared with beneficiaries whose predominant provider was a PCP (Table 2). There was no association between predominant provider and expenditures for ACSC hospitalizations or all-cause ED visits. When exponentiated, the b coefficient from the GLM model can be interpreted as the percent change in costs associated with having a predominant r


Medical Care



TABLE 1. Select Characteristics of Study Sample Predominant Provider Generalist N = 461,446 Baseline period predominant provider — Specialist — Primary care — No. baseline ambulatory care visits, median Specialist visits 2 Generalist visits 5 All visits* 8 No. unique providers, median 3 Baseline continuity of care Bice-Boxerman Index: all visits, mean 0.517 Bice-Boxerman Index: generalist visits, mean 0.790 Bice-Boxerman Index: specialist visits, mean — Usual Provider Continuity Index: all visits, mean 0.572 Usual Provider Continuity Index: generalist visits, 0.859 mean Usual Provider Continuity Index: specialist visits, mean — Sociodemographic characteristics Female, % 56.1 Age in years, mean 71 White, % 87.5 Black, % 8.4 Hispanic, % 1.4 Other, % 2.6 Original reason for Medicare enrollment: disabled, % 23.1 Original reason for Medicare enrollment: end-stage 0.20 renal disease, % Enrolled in Medicaid, % 14.7 Died, % 3.8 Geographic characteristics Urban residence, % 69.0 PCPs/100,000 residents 68 Specialists/100,000 residents 195 Per capita income in county of residence, mean 37,832 Residents age Z25 y with high school diploma, % 84.5 Baseline health status Charlson Index, mean 0.87 Concurrent HCC Score, mean 1.24 Follow-up period mean annual health care utilization, rate per 100 beneficiaries All-cause hospitalizations 35.6 ACSC hospitalizations 14.0 All-cause emergency department visits 66.0 ACSC emergency department visits 38.3 Follow-up period mean annual health care payments, $ All-cause hospitalizations 4111 ACSC hospitalizations 1262 All-cause emergency department payments 287 ACSC emergency department payments 85 Total payments 9759

Specialist N = 152,025

Total Sample N = 613,471

P for Difference Between Generalist and Specialist

— — —

25% 75%

— — —

7 1 9 4

4 3 9 4

< 0.001 < 0.001 < 0.001 < 0.001

0.340 — 0.516 0.481 —

0.473 — — 0.550 —

< 0.001 — — < 0.001 —

0.657

—

—

47.2 71 88.3 7.8 1.4 2.5 21.3 0.67

53.8 71 87.7 8.2 1.4 2.6 22.7 0.3

< 0.001 < 0.001 < 0.001 < 0.001 0.108 0.006 < 0.001 < 0.001

10.6 5.0

13.7 4.1

< 0.001 < 0.001

78.4 71 220 40,800 84.7

71.2 69 201 38,568 84.5

< 0.001 < 0.001 < 0.001 < 0.001 < 0.001

1.18 1.44

0.94 1.29

< 0.001 < 0.001

40.5 15.0 65.9 35.9

36.8 14.3 66.0 37.7

< 0.001 < 0.001 0.845 < 0.001

5033 1479 280 74 12,429

4339 1315 285 83 10,419

< 0.001 < 0.001 0.021 < 0.001 < 0.001

Notes: Health care visit and expenditure estimates are weighted by fraction of the follow-up period year the beneficiary was enrolled in Medicare Part A and Part B. P-values testing means and proportions are derived from t tests, w2 tests, and regression analyses not adjusted for covariates. *All visits include ambulatory, evaluation, and management visits to providers that included primary care and specialty physicians as well as ancillary providers such as optometrists or podiatrists. These nonphysician providers were not counted as a generalist or specialist for purposes of this analysis. ACSC indicates ambulatory care sensitive condition.

provider who is a specialist. For example, e0.058 equates to 6% higher total expenditures associated with having a specialist predominant provider compared with having a PCP predominant provider. The average marginal effects from these regression models can be found in Supplemental Digital Content 2 (http://links.lww.com/MLR/A812). r


Association Between COC and Utilization and Expenditures As shown in Table 3, in multivariate models, we observed a significant, negative association between higher COC measured using the Bice-Boxerman Index and rates of all-cause hospitalizations (P < 0.001), all-cause ED visits www.lww-medicalcare.com |

1045

Medical Care

Romaire et al


TABLE 2. Association Between Type of Predominant Provider and Health Care Visits and Expenditures Specialist Provider (Referent: PCP) Total Sample Stratiﬁed by Comorbidity Index

Visitsw All-cause hospitalizations ACSC Hospitalizations All-cause ED ACSC ED

Total Sample N = 613,471

Charlson Index = 0 N = 322,076

Charlson Index = 1 N = 149,781

Charlson Index Z2 N = 141,614

Incidence Rate Ratio (95% CI)




0.98 0.92 0.91 0.86

(0.96–0.99)*** (0.90–0.94)*** (0.90–0.92)*** (0.84–0.87)*** b (95% CI)

Expendituresz All-cause Hospitalizations ACSC Hospitalizations All-cause ED ACSC ED Total

0.028 0.006 0.023 0.087 0.058

(0.008–0.047)* ( 0.044, 0.033) ( 0.045, 0.000) ( 0.130, 0.045)*** (0.047–0.070)***

1.04 0.99 0.98 0.94

(1.01–1.06)** (0.95–1.02) (0.95–0.99)* (0.91–0.96)***

b (95% CI) 0.075 0.022 0.016 0.033 0.066

(0.045–0.105)*** ( 0.059, 0.065) ( 0.017, 0.050) ( 0.097, 0.031) (0.048–0.083)***

1.07 1.08 0.96 0.92

(1.04–1.10)*** (1.03–1.13)*** (0.93–0.98)*** (0.89–0.95)***

0.88 0.81 0.82 0.76

b (95% CI) 0.074 0.121 0.008 0.047 0.073

(0.87–0.90)*** (0.79–0.84)*** (0.81–0.84)*** (0.74–0.78)*** b (95% CI)

(0.033–0.114)** (0.053–0.189)*** ( 0.056, 0.039) ( 0.132, 0.038) (0.048, 0.097)***

0.053 0.122 0.085 0.204 0.031

(0.082, 0.024)*** (0.168, 0.077)*** (0.124, 0.047)*** (0.274, 0.134)*** (0.012–0.048)*

w Negative binomial models controlled for baseline sex, age, race, enrollment in Medicaid, original reason for Medicare eligibility (aged, disabled, or have end-stage renal disease), comorbidity (using the Charlson Index and the Hierarchical Condition Category risk score), residence in an urban area, state of residence, number of PCPs per 100,000 residents in county of residence, number of specialists per 100,000 residents in county of residence, per capita income in county of residence, percent of the county of residence with a high school education, number of unique providers seen by the beneficiary, number of ambulatory visits, and death in the follow-up period. z Generalized linear models controlled for baseline sex, age, race, enrollment in Medicaid, original reason for Medicare eligibility (aged, disabled, or have end-stage renal disease), comorbidity (using the Charlson Index and the Hierarchical Condition Category risk score), residence in an urban area, state of residence, number of PCPs per 100,000 residents in county of residence, number of specialists per 100,000 residents in county of residence, per capita income in county of residence, percent of the county of residence with a high school education, number of unique providers seen by the beneficiary, number of ambulatory visits, death in the follow-up period, and baseline annualized costs. When exponentiated, the b coefficient from the generalized linear model can be interpreted as the percent change in costs associated with having a predominant provider who is a specialist. For example, e0.058 equates to 6% higher total expenditures associated with having a specialist predominant provider compared with having a PCP predominant provider. *P < 0.05. **P < 0.001. ***P < 0.0001. ACSC indicates ambulatory care sensitive condition; CI, confidence interval; ED, emergency department; PCP, primary care provider.

(P < 0.001), and ACSC ED visits (P < 0.001) for both beneficiaries whose predominant provider was a PCP and those whose predominant provider was a specialist. We also noted a small gradient effect with the highest continuity category showing a larger association between continuity and fewer hospitalizations and ED visits. In contrast, higher continuity was associated with higher rates of ACSC hospitalizations among beneficiaries whose predominant provider was a PCP. Similar results were observed when the UPC index was used (Table 4). When the Bice-Boxerman and the UPC indices were calculated using only PCP visits for beneficiaries with a PCP as the predominant provider and using only specialist visits for beneficiaries with a specialist as the predominant provider, conclusions did not differ (results not shown). Regardless of predominant provider type, higher COC, measured using either the Bice-Boxerman Index or the UPC Index, was also significantly associated with lower all-cause hospitalizations expenditures, all-cause ED expenditures, and total expenditures (all P < 0.001) (Tables 3 and 4). Patterns of association were less clear for ACSC hospitalizations and ED expenditures. There was a trend toward a positive association between ACSC hospitalization expenditures and higher continuity, regardless of the COC index examined, but statistical significance was not consistent. In contrast, there was a general trend toward a negative association between ACSC ED expenditures and higher continuity regardless of the in-


dex, but again statistical significance was not consistent. (Tables 3 and 4). The average marginal effects from these regression models can be found in Supplemental Digital Content 2 (http://links.lww.com/MLR/A812). Overall conclusions did not differ when the Bice-Boxerman and the UPC indices were calculated using only PCP visits and using only specialist visits, although in some models the measures of effect were seen in the highest COC category and not in the medium COC category (results not shown).

CONCLUSIONS In summary, we found that one quarter of beneficiaries saw a specialist for the plurality of their ambulatory care during this study period, which was a little lower than findings reported by Pham et al.10 Beneficiaries who had a specialist predominant provider had fewer hospitalizations and ED visits compared with those who primarily saw a PCP. Notably, the presence of multiple chronic conditions appears to be a factor. Beneficiaries with a Charlson Index score Z2, that is, beneficiaries with conditions such as cardiovascular disease, chronic obstructive pulmonary disease, diabetes, renal failure, and cancer had higher rates of hospital and ED use compared with beneficiaries with less comorbidity (data not shown), but among this high morbidity group, beneficiaries with a specialist predominant provider r


Medical Care



TABLE 3. Association Between Continuity of Care (Bice-Boxerman Index) and Health Care Visits and Expenditures Bice-Boxerman Index PCP Predominant Provider N = 461,446

Visitsw All-cause hospitalizations ACSC hospitalizations All-cause ED ACSC ED

Medium (Referent: Low)

High (Referent: Low)







0.96*** (0.94–0.97)

0.91*** (0.90–0.93)

0.91*** (0.89–0.94)

0.91*** (0.88–0.95)

1.05** (1.02–1.07)

1.08*** (1.05–1.11)

1.01 (0.97–1.05)

1.04 (0.99–1.10)

0.91*** (0.90–0.93) 0.95*** (0.93–0.96)

0.85*** (0.84–0.86) 0.90*** (0.89–0.92)

0.91*** (0.89–0.93) 0.91*** (0.89–0.94)

0.85*** (0.83–0.88) 0.85*** (0.82–0.88)

b (95% CI) Expendituresz All-cause hospitalizations ACSC hospitalizations All-cause ED ACSC ED Total

Specialist Predominant Provider N = 152,025

b (95% CI)

b (95% CI)

b (95% CI)

0.079*** (0.103, 0.054) 0.141*** (0.165, 0.118) 0.107*** (0.144, 0.069) 0.115*** (0.164, 0.067) 0.067* (0.019–0.115)

0.109*** (0.062–0.156)

0.091* (0.017–0.166)

0.070 (0.025, 0.164)

0.073*** (0.100, 0.046) 0.177*** (0.203, 0.151) 0.131*** (0.177, 0.085) 0.240*** (0.299, 0.180) 0.032 (0.080, 0.016) 0.075* (0.122, 0.028) 0.006 (0.089, 0.102) 0.122 (0.244, 0.001) 0.071*** (0.085, 0.056) 0.158*** (0.172, 0.144) 0.079*** (0.100, 0.058) 0.131*** (0.158, 0.104)

w Negative binomial models controlled for baseline sex, age, race, enrollment in Medicaid, original reason for Medicare eligibility (aged, disabled, or have end-stage renal disease), comorbidity (using the Charlson Index and the Hierarchical Condition Category risk score), residence in an urban area, state of residence, number of PCPs per 100,000 residents in county of residence, number of specialists per 100,000 residents in county of residence, per capita income in county of residence, percent of the county of residence with a high school education, and death in the follow-up period. z Generalized linear models controlled for baseline sex, age, race, enrollment in Medicaid, original reason for Medicare eligibility (aged, disabled, or have end-stage renal disease), comorbidity (using the Charlson Index and the Hierarchical Condition Category risk score), residence in an urban area, state of residence, number of PCPs per 100,000 residents in county of residence, number of specialists per 100,000 residents in county of residence, per capita income in county of residence, percent of the county of residence with a high school education, death in the follow-up period, and baseline annualized costs. When exponentiated, the b coefficient from the generalized linear model can be interpreted as the percent change in costs. For example, e 0.071 equates to 6.8% lower total expenditures associated with reporting medium continuity of care (measured by the Bice-Boxerman index) compared with reporting low continuity of care. *P < 0.05. **P < 0.001. ***P < 0.0001. ACSC indicates ambulatory care sensitive condition; CI, confidence interval; ED, emergency department; PCP, primary care provider.

had lower rates of hospital and ED use compared with beneficiaries with a PCP predominant provider. There could be multiple explanations for this finding; one being that management of these conditions typically necessitates receipt of ongoing specialty care and that specialists may be better positioned to prevent inpatient and ED use associated with the chronic condition they are treating. Although outside the scope of this analysis, this could be empirically examined for select groups (eg, examine diabetes-related admissions for diabetics primarily treated by endocrinologists vs. those primarily treated by a PCP). It may also be that beneficiaries with greater morbidity or more advanced disease progression may transition to a PCP for more whole-person monitoring and commensurate with this transition comes higher rates of health care utilization related to their morbidity. Although we could not assess how disease progression influenced findings, as a proxy we did examine, in an exploratory analysis, how our results differed if we excluded beneficiaries who died in the follow-up period. The results can be found in Supplemental Digital Content 3 (http://links.lww.com/MLR/A813); regression point estimates are similar in magnitude and direction to the results presented in Tables 2–4. r


The relationship between provider specialty and expenditures was less pronounced; relatively small changes in use may not be associated with appreciable changes in cost. Nevertheless, our results show that beneficiaries whose predominant provider was a specialist had lower expenditures for ACSC ED visits and higher expenditures for all-cause hospitalizations, which generally aligned with the observed patterns of utilization for these services. In addition, total expenditures were higher for beneficiaries with a specialist as the predominant provider compared with beneficiaries whose predominant provider was a PCP. Inpatient expenditures comprise a significant proportion of total expenditures, so the finding of higher all-cause hospitalization expenditures associated with a specialist may explain, in part, the higher total expenditures. Given that specialty services, whether in the inpatient or outpatient setting, are typically paid at higher amounts than visits to PCPs28 and that the clinical needs of beneficiaries who frequent specialists may dictate a greater volume of outpatient and/or inpatient procedures,8 the higher total costs associated with having a specialist predominant provider is not surprising. Regardless of specialty type of the predominant provider, higher continuity was associated with lower rates of allcause hospitalization and all-cause and ACSC ED use, and www.lww-medicalcare.com |

1047

Medical Care

Romaire et al


TABLE 4. Association Between Continuity of Care (Usual Provider Continuity Index) and Health Care Visits and Expenditures Usual Provider Continuity Index PCP Predominant Provider N = 461,446

Visitsw All-cause hospitalizations ACSC hospitalizations All-cause ED ACSC ED









0.90*** (0.89–0.92)

0.82*** (0.81–0.84)

0.88*** (0.85–0.90)

0.84*** (0.82–0.87)

1.01 (0.99–1.04)

1.05*** (1.02–1.08)

0.95* (0.91–0.99)

0.97 (0.92–1.01)

0.88*** (0.87–0.89) 0.92*** (0.91–0.94)

0.80*** (0.78–0.81) 0.85*** (0.84–0.87)

0.88*** (0.86–0.90) 0.88*** (0.86–0.91)

0.81*** (0.79–0.83) 0.80*** (0.77–0.82)

b (95% CI) Expendituresz All-cause hospitalizations ACSC hospitalizations All-cause ED ACSC ED Total

Specialist Predominant Provider N = 152,025

b (95% CI)

b (95% CI)

b (95% CI)

0.146*** (0.170, 0.122) 0.274*** (0.298, 0.251) 0.158*** (0.196, 0.120) 0.175*** (0.220, 0.131) 0.009 (0.038, 0.056)

0.075* (0.028, 0.121)

0.068 (0.142, 0.006)

0.067 (0.020, 0.155)

0.135*** (0.162, 0.109) 0.257*** (0.283, 0.231) 0.143*** (0.189, 0.096) 0.268*** (0.323, 0.213) 0.029 (0.076, 0.018) 0.084*** (0.131, 0.037) 0.019 (0.114, 0.076) 0.233** (0.345, 0.120) 0.127*** (0.141, 0.113) 0.267*** (0.281, 0.253) 0.124*** (0.145, 0.103) 0.169*** (0.194, 0.144)

w Negative binomial models controlled for baseline sex, age, race, enrollment in Medicaid, original reason for Medicare eligibility (aged, disabled, or have end-stage renal disease), comorbidity (using the Charlson Index and the Hierarchical Condition Category risk score), residence in an urban area, state of residence, number of PCPs per 100,000 residents in county of residence, number of specialists per 100,000 residents in county of residence, per capita income in county of residence, percent of the county of residence with a high school education, and death in the follow-up period. z Generalized linear models controlled for baseline sex, age, race, enrollment in Medicaid, original reason for Medicare eligibility (aged, disabled, or have end-stage renal disease), comorbidity (using the Charlson Index and the Hierarchical Condition Category risk score), residence in an urban area, state of residence, number of PCPs per 100,000 residents in county of residence, number of specialists per 100,000 residents in county of residence, per capita income in county of residence, percent of the county of residence with a high school education, death in the follow-up period, and baseline annualized costs. When exponentiated, the b coefficient from the GLM model can be interpreted as the percent change in costs. For example, e 0.127 equates to 12% lower total expenditures associated with reporting medium continuity of care (measured by the UPC index) compared with reporting low continuity of care. *P < 0.05. **P < 0.001. ***P < 0.0001. ACSC indicates ambulatory care sensitive condition; CI, confidence interval; ED, emergency department; PCP, primary care provider.

commensurate with these utilization patterns, lower expenditures for these services. These findings align with literature suggesting that higher COC is associated with less utilization and lower expenditures among adults and the elderly with chronic conditions.29–31 Study findings suggest that effecting change in high-dollar health care use may require a focus not only on increasing the number of visits to a single provider that serves as a patient’s primary provider but decreasing the number of unique practitioners seen for care. The extent to which this can be done in a real-world setting remains to be seen; clinical practice change and patient preference in provider choice are challenging at best to address.32,33 Yet, several of the large-scale medical home or medical home-related initiatives, such as accountable care organizations, underway in the United States may provide the opportunity to investigate how a patient’s dispersion of visits changes under alternative models of health care delivery.6,7 The results of this study provide some early evidence that the value of increased COC may not necessarily reside only within the context of primary care. Many unanswered questions remain about what role specialists should assume in alternative models of medical care delivery, including the medical home. How well a specialist-oriented model can efficiently and effectively meet the health care needs of a


population may be dictated, in part, by provider, patient, and payer preferences in how large a role a specialist versus a PCP should have in ongoing clinical care. It may also be dictated, in part, by ongoing research to better understand how well specialists can deliver on other components of the medical home, including expectations for patient-centered, coordinated, comprehensive, and high-quality care. There are important considerations in interpreting our results. This study sample is generalizable only to communityresiding, FFS Medicare beneficiaries who had 3 or more PCP or specialist visits. Visit patterns may substantially differ in younger populations, managed care enrollees, those with less comorbidity or clinical need, and those with less access to ambulatory care (eg, low-income or uninsured populations), limiting generalizability of the COC scores seen in this study. Furthermore, a significant portion of the initial study sample did not meet inclusion criteria for this study. Compared with the beneficiaries excluded, the study sample was older, more likely to be female, less likely to be a nonwhite racial/ethnic minority, had higher levels of morbidity, and higher rates of health service utilization and expenditures. This is the kind of higher needs group that may reap more benefits from greater COC, but it does limit the ability to draw conclusions on the basis of a less resource intensive Medicare population. There r


Medical Care


are also limitations to the COC measures used in this analysis. Both measures capture visit patterns at the individual provider level; however, visits to a medical practice is likely to become a more relevant point of reference as the medical home model is organized around the practice. In addition, this was an observational study. Causal inference cannot be made, and unmeasured confounding may bias results. We used 2 years of Medicare claims data, and there may be temporal trends in clinical practice among PCPs and specialists that could account for changes in rates of utilization and expenditures for which we could not control. In addition, with 2 years of data we could not examine whether a beneficiary’s predominant provider type and the association between higher continuity and utilization/expenditures remain stable over time. In conclusion, this analysis demonstrates that among beneficiaries with a specialist predominant provider, greater claims-based continuity is associated with fewer all-cause hospitalizations and ED visits and related costs and less total cost. As long as policy-makers, provider groups, and payers continue to explore the role of specialists in the medical home framework, additional research will be needed to examine how well specialists provide care aligned with other medical home principles, such as comprehensive care, coordinated, and patient-centered care, and which Medicare beneficiaries might be best served in a specialist-oriented medical home. REFERENCES 1. American Academy of Family Physicians, American Academy of Pediatrics, American College of Physicians, et al. Joint Principles of the Patient-Centered Medical Home. 2007. Available at: http://www. aafp.org/dam/AAFP/documents/practice_management/pcmh/initiatives/ PCMHJoint.pdf. Accessed May 1, 2014. 2. Berenson RA. Is there room for specialists in the patient-centered medical home? Chest. 2010;137:10–11. 3. Fishman PA, Johnson EA, Coleman K, et al. Impact on seniors of the patient-centered medical home: evidence from a pilot study. Gerontologist. 2012;52:703–711. Available at: http://www.ncqa.org/Programs/ Recognition/Practices/PatientCenteredMedicalHomePCMH.aspx. Accessed October 1, 2013. 4. National Committee on Quality Assurance. Patient-Centered Medical Home Program. 2013. 5. Starfield B. Primary Care: Balancing Health Needs, Services, and Technology. New York: Oxford University Press; 1998. 6. The Centers for Medicare & Medicaid Services. Comprehensive Primary Care Initiative. Available at: http://innovation.cms.gov/initiatives/ Comprehensive-Primary-Care-Initiative. Accessed May 27, 2014. 7. The Centers for Medicare & Medicaid Services. Multi-Payer Advanced Primary Care Practice. Available at: http://innovation.cms.gov/initiatives/ Multi-Payer-Advanced-Primary-Care-Practice. Accessed May 27, 2014. 8. Thorpe KE, Ogden LL, Galactionova K. Chronic conditions account for rise in Medicare spending from 1987 to 2006. Health Aff (Millwood). 2010;29:718–724. 9. Starfield B, Lemke KW, Herbert R, et al. Comorbidity and the use of primary care and specialist care in the elderly. Ann Fam Med. 2005; 3:215–222. 10. Pham HH, Schrag D, O’Malley AS, et al. Care patterns in Medicare and their implications for pay for performance. N Engl J Med. 2007;356:1130–1139. 11. Council on Graduate Medical Education. Advancing Primary Care. 2010. This is a report. Published by the Council on Graduate Medical

r



12. 13. 14. 15. 16. 17. 18. 19. 20.

21. 22. 23. 24.

25. 26. 27. 28. 29. 30. 31. 32. 33.

Education in December 2010. Available at: www.hrsa.gov/advisory committees/bhpradvisory/cogme/reports/twentiethreport.pdf. Accessed October 1, 2013. Kirschner N, Barr MS. Specialists/subspecialists and the patientcentered medical home. Chest. 2010;137:200–204. Rosenblatt RA, Hart LG, Baldwin LM, et al. The generalist role of specialty physicians: is there a hidden system of primary care? JAMA. 1998;279:1364–1370. Gill JM, Mainous AG III, Diamond JJ, et al. Impact of provider continuity on quality of care for persons with diabetes mellitus. Ann Fam Med. 2003;1:162–170. Haggerty JL, Reid RJ, Freeman GK, et al. Continuity of care: a multidisciplinary review. BMJ. 2003;327:1219–1221. Jee SH, Cabana MD. Indices for continuity of care: a systematic review of the literature. Med Care Res Rev. 2006;63:158–188. Reid RJ, Barer ML, McKenry R, et al. Patient-focused care over time: issues related to measurement, prevalence, and strategies for improvement among patient populations. Can Health Serv Res Found. 2003. Reid RJ, Haggerty J, McKendry R. Defusing the Confusion: Concepts and Measures of Continuity of Healthcare; Canadian Health Services Research Foundation; 2002. Health Resources and Services Administration. Area Resource File Available at: http://ahrf.hrsa.gov/. Accessed August 24, 2013. Centers for Medicare & Medicaid Servcies. National Plan and Provider Enumeration System. Available at: http://www.cms.gov/Regulationsand-Guidance/HIPAA-Administrative-Simplification/NationalProvIdent Stand/DataDissemination.html. Accessed August 1, 2013. Bice TW, Boxerman SB. A quantitative measure of continuity of care. Med Care. 1977;15:347–349. Liss DT, Chubak J, Anderson ML, et al. Patient-reported care coordination: associations with primary care continuity and specialty care use. Ann Fam Med. 2011;9:323–329. Flores AI, Bilker WB, Alessandrini EA. Effects of continuity of care in infancy on receipt of lead, anemia, and tuberculosis screening. Pediatrics. 2008;121:e399–e406. The Centers for Medicare & Medicaid Services. FY 2008 Wage Index. 2014. Available at: http://www.cms.gov/Medicare/Medicare-Fee-forService-Payment/AcuteInpatientPPS/Wage-Index-Files-Items/FY2008 WageIndexHomePage.html Accessed May 5, 2014. Buntin MB, Zaslavsky AM. Too much ado about two-part models and transformation? Comparing methods of modeling Medicare expenditures. J Health Econ. 2004;23:525–542. Charlson ME, Pompei P, Ales KL, et al. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40:373–383. Ash AS, Ellis RP, Pope GC, et al. Using diagnoses to describe populations and predict costs. Health Care Financ Rev. 2000;21: 7–28. Reschovsky J, Ghosh A, Stewart K, et al. Paying More for Primary Care: Can It Help Bend the Medicare Cost Curve?. New York City: The Common wealth Fund; 2012. Knight JC, Dowden JJ, Worrall GJ, et al. Does higher continuity of family physician care reduce hospitalizations in elderly people with diabetes? Popul Health Manag. 2009;12:81–86. McCusker J, Tousignant P, Borges Da Silva R, et al. Factors predicting patient use of the emergency department: a retrospective cohort study. CMAJ. 2012;184:E307–E316. Liu CW, Einstadter D, Cebul RD. Care fragmentation and emergency department use among complex patients with diabetes. Am J Manag Care. 2010;16:413–420. Bitton A, Schwartz GR, Stewart EE, et al. Off the hamster wheel? Qualitative evaluation of a payment-linked patient-centered medical home (PCMH) pilot. Milbank Q. 2012;90:484–515. Carrier E, Gourevitch MN, Shah NR. Medical homes: challenges in translating theory into practice. Med Care. 2009;47:714–722.

www.lww-medicalcare.com |

1049

Primary Care Providers Report Challenges to Cirrhosis Management and Specialty Care Coordination.

The causal effects of home care use on institutional long-term care utilization and expenditures.

Health Care Resource Utilization for Outpatient Cardiovascular Disease and Diabetes Care Delivery Among Advanced Practice Providers and Physician Providers in Primary Care.

Continuity: A Central Principle of Primary Care.

Young adults' health care utilization and expenditures prior to the Affordable Care Act.

Association of Continuity of Primary Care and Statin Adherence.

Cost, utilization, and quality of care: an evaluation of illinois' medicaid primary care case management program.

Specialty pediatric transport in primary care or urgent care settings.

Improving primary care: Continuity is about relationships.

Continuity and coordination in primary care: their achievement and utility.

Impact of cost sharing on utilization of primary health care Services: providers versus household perspectives.

Care continuity and care coordination: what counts?

Patients' experience of Chinese Medicine Primary Care Services: Implications on Improving Coordination and Continuity of Care.

Rectal cancer: An evidence-based update for primary care providers.

Primary care providers' bereavement care practices: recommendations for research directions.

Effect of continuity of care on hospital utilization for seniors with multiple medical conditions in an integrated health care system.

The Value of Continuity between Primary Care and Surgical Care in Colon Cancer.

Electronic communication between providers of primary and secondary care.

Continuity of care after percutaneous coronary intervention: The patient's perspective across secondary and primary care settings.

Migraine diagnosis and treatment: A knowledge and needs assessment among primary care providers.

Primary care appointment availability and preventive care utilization: evidence from an audit study.

Financial implications of the continuity of primary care.

Awareness, Interest, and Preferences of Primary Care Providers in Using Point-of-Care Cancer Screening Technology.

Proximity of providers: Colocating behavioral health and primary care and the prospects for an integrated workforce.