Abstract
Objective: Continuity of primary care reduces costs and hospitalizations over 1-year periods, but its long-term effects remain unclear, a key concern for value-based payment. We examined associations between physician-level continuity and health care expenditures/utilization across single- and multi-year measurement periods.
Design: Retrospective cohort study using 2011 to 2017 Medicare Fee-for-Service (FFS) claims. We constructed physician-level continuity measures over 1- to 5-year lookback periods. Generalized linear models estimated associations with total Medicare Part A & B expenditures, and logit models assessed hospitalization and emergency department (ED) visits, adjusting for patient and physician characteristics.
Setting & Participants: Nationally representative sample of 4,940 primary care practices, including 1.1 to 2.5 million Medicare FFS beneficiaries seen by 6,758-14,949 physicians.
Results: Beneficiaries in the highest continuity quintile had 7.4%-10.4% lower total expenditures than those in the lowest quintile, with the greatest difference in the 1-year lookback. Hospitalization and ED visit odds were 5.5%-8.6% and 4.9%-6.3% lower, respectively, for high-continuity physicians. Effects attenuated slightly with longer lookbacks.
Conclusion: Physician continuity is consistently associated with lower costs, hospitalizations, and ED visits across lookback periods. Given this stability and the complexity of multi-year claims measurement, a 1-year assessment may be sufficient for physician continuity evaluation and value-based payment programs.
- Continuity of Care
- Health Expenditures
- Hospitalization
- Linear Models
- Medicare
- Patient-Centered Care
- Physicians
- Primary Health Care
- Process Measures
- Quality Improvement
- Retrospective Studies
Introduction
Continuity is a defining characteristic of primary care, 1 of 4 conceptualized to explain its salutary benefits. These benefits are thought to result from longitudinal knowledge and relationships shared between clinicians and their patients and the captured history and records that result. Continuity of care (COC) has been associated with positive impacts on health equity, patient and clinician satisfaction cost reduction, and improved quality of care.1–4 The National Academies of Sciences, Engineering, and Medicine 2021 report Implementing High-Quality Primary Care: Rebuilding the Foundation of Health Care reviewed this evidence and capture it in its primary care definition as providing care “through sustained relationships.”5
Continuity of primary care has specifically been associated at the patient level with significantly lower costs, fewer hospitalizations, reduced emergency department (ED) visits, and even mortality differences for a variety of patient populations and in many countries.6–15 Our prior research tested 4 existing continuity measures and selected one for implementation a claims-based physician-level continuity measure based on validated patient-continuity measures.15,16 It advanced understanding of measurement of primary care physician level continuity and its importance to policy outcomes, revealing high clinician-level continuity association with lower costs and fewer hospitalizations. This finding is essential knowledge in an age of value-based payment, and to a US payor audience exploring clinician versus practice-level measures as it reforms and simplifies value-based purchasing.
However, while past studies typically assessed COC over a 1-year measurement period, including 2 recent studies that examined physician-level measures,17–18 little is known about the importance of longitudinality in physician-level continuity measures. An international study suggests that continuity over time, also called longitudinality, is associated with even better outcomes, approaching a dose-dependent pattern. That study explored longitudinality at the patient level, not the physician or physician panel level and we could not find any studies exploring whether or how the established relationship between clinician-level continuity measures and patient outcomes perform when the continuity measured is based on multiple years. This is a critical question for both payors and clinicians as they ponder the value, stability, feasibility and reliability of clinician-level COC measurement and optimal measurement periods in predicting & improving patient outcomes and the value of care provided. With this question in mind, we studied the behavior of physician-level, claims-based continuity measures and tested their associations with health care expenditures and utilization for Medicare beneficiaries across single and multiyear periods.
Methods
Sample
We used 2011 to 2017 Fee-for-Service Medicare claims of beneficiaries who obtained care from primary care physicians practicing in a nationally representative stratified random sample of 4,940 primary care practices to construct multiyear-lookback, physician-level COC scores. There were several established COC measures, but we focused on the Bice-Boxerman Continuity of Care (BB-COC) index in this study, following past physician-level COC studies.16–18 Primary care practices were identified using Tax Identification Numbers for office-based practices and Organizational National Provider Identification Numbers for safety-net clinics – federally qualified health centers and rural health clinics. The multiyear lookback refers to the number of years of past visit information used to create the continuity of care scores. The longitudinality of our data allowed us to vary the lookback periods from 1-year to 5-years. The unit of analysis was the beneficiary, assigned to primary care physicians (PCPs) in our sample practices. We assigned beneficiaries annually to PCPs based on the plurality of Evaluation and Management (E&M) visits in the outpatient setting. We restricted our analysis to assigned beneficiaries who were continuously enrolled in traditional Medicare Part A and Part B for 2 consecutive years unless they died during the second year and if they had no end-stage renal disease in both years. For each multi-year lookback analysis, we restricted to physicians who were in all years during the period and assigned specialty using 2011 to 2017 Medicare Data on Provider Practice and Specialty. We excluded first-year residents and hospitalists, who provided more than 90% of their services in an inpatient setting. The small number of residents who stayed in practice could be included in multiyear analyses, while others only contributed to single-year scores. We also excluded physicians who had fewer than 30 Medicare patients. We dropped beneficiaries with missing values in any of the beneficiary or physician characteristics.
Continuity Measures
We closely followed the method described in our previous study in creating multi-year lookback physician-level continuity of care scores.15,16 For each physician in the sample, we first created a patient panel consisting of patients who had full-year FFS Part A and Part B coverage and at least one E&M visit to the physician during the multiyear period. We then calculated patient-level BB-COC scores for all the patients in the panel based on visits to PCPs during the multiyear period. We then averaged patient-level BB-COC scores across the panel using the share of patient’s E&M visits in the total number of E&M visits as weights to derive physician-level BB-COC scores, which had a possible range of 0 to 100. The multiyear lookback period allowed us to take into account potential longitudinal patient-physician relationships in the physician’s patient panel. The previous years BB-COC scores were used in the analyses.
Outcomes
We used 3 patient outcome measures in this study. The primary outcome was total Part A and Part B expenditures, which included Medicare reimbursements, beneficiary out-of-pocket, and third-party payments. Part A included inpatient, skilled nursing, hospice, and home health expenditures, while Part B included outpatient visits and durable medical equipment expenditures. The other 2 outcomes were whether the beneficiary was hospitalized in a short-term acute care hospital during the year and whether the beneficiary had an ED visit that did not lead to hospitalization.
Covariates
Regression models were adjusted for beneficiary and physician characteristics in the previous year. Beneficiary characteristics were determined from the claims data and included age, sex, race/ethnicity, disabled as reason for Medicare entitlement, dual eligible, Charlson score, and number of primary care visits. Physician demographic and biographic characteristics were determined using the 2017 American Medical Association Masterfile and included sex, years since medical school graduation, and international medical school graduation status. Other physician characteristics including PC specialty, care provided in inpatient setting, Medicare patient panel size, rurality based on Rural Urban Commuting Area (RUCA) Code, Census Region of practice location, practice type (office-based vs safety-net), and practice size were constructed using the claims data. In particular, rurality and Census Region of practice location were based on the 5-digit ZIP code where the PCP provided most of her services. Rurality was then determined using the ZIP-RUCA crosswalk version 3.1.16,19
Analysis
We first assessed the distributions of physician-level BB-COC scores among assigned beneficiaries with varying lookback periods. For each year available, we divided assigned beneficiaries into physician-level BB-COC score quintiles. We used descriptive statistics to compare beneficiary and physician characteristics for PCPs with continuity scores in the highest quintile versus PCPs in the lowest quintile. In assessing patient outcomes, we used the prior year’s physician-level BB-COC scores from 2011 to 2016 in assessing their associations with patient outcomes in 2012 to 2017. Specifically, we estimated generalized linear models with a γ distribution and a log link function to determine whether having a PCP with higher continuity scores in the previous year was associated with lower overall health care expenditure, adjusting for calendar year effects (time trend) and beneficiary and physician characteristics. We estimated logistic models for hospitalization and ED visit. We separately estimated across 5 different lookback periods (1-year to 5-years). All analyses used practice weights to account for stratified sampling design, and the standard errors in regression analyses were clustered at the physician level. In our sensitivity analyses, we constructed physician-level COC scores using the other 3 established continuity measures in the literature – Usual Provider Continuity (UPC) index, Modified Modified Continuity Index (MMCI), and Herfindahl Index (HI) – and repeated the analysis for all lookback periods.
All statistical analyses were performed in Stata version 16.1 (StataCorp LLC). All statistical tests of significance were 2-sided and a P < .05 was considered statistically significant. The American Academy of Family Physicians Institutional Review Board approved this study.
Results
The final sample varied with the lookback period between 1.1–2.5 million beneficiaries assigned to 6,758–14,949 PCPs (see Appendix Table 1 for details). Figure 1 presents the distribution of physician-level BB-COC scores for 1-year to 5-year lookback periods among assigned beneficiaries in 2016. As in our previous study, the distributions were left-skewed but exhibited a wide variation in continuity scores.15,16 Continuity distributions were relatively stable across lookback periods, with the average score ranging between 71.7–72.8.
Distribution of physician-level Bice-Boxerman Continuity of Care (BB-COC) scores for various lookback periods among primary care physicians.
Bivariate and descriptive analyses showed that beneficiary and physician characteristics for PCPs in the top BB-COC score quintile were different from those in the bottom quintile. Beneficiaries in the top quintile were on average older, sicker, and had more primary care visits, while their PCPs were on average more experienced, had fewer Medicare patients, were more likely to be male, provide some hospital care, practice in a rural area, in an office-based practice and in much larger practices, compared with those in the bottom quintile (see Table 1).
Comparison of Measure, Outcomes, and Characteristics Between Assigned Beneficiaries in the Bottom and Top Quintiles of Physician-Level Bice-Boxerman Continuity of Care (BB-COC) Score by Lookback Period in 2016
The adjusted total expenditures were on average 7.4%∼10.4% lower for beneficiaries in the top quintile compared with those in the bottom BB-COC quintile during the previous year across lookback periods (Figure 2). The difference in the expenditure was largest in the 1-year lookback, 10.4% (β = −0.104; 95% CI, −0.118 to −0.09), and attenuated with the lookback period, to 7.4% (β = −0.074; 95% CI, −0.097 to −0.052) for the 5-year lookback. The magnitudes of these estimates were nontrivial given the average expenditure was more than $12,000. Hospitalization and ED visit without hospitalization showed similar patterns: patients of PCPs in the top quintile had 5.5% (OR = 0.945; 95% CI, 0.918 to 0.973) ∼ 8.6% (OR = 0.914; 95% CI, 0.898 to 0.931) lower odds of being admitted to a short-term acute care hospital in the following year and had 4.9% (OR = 0.951; 95% CI, 0.918 to 0.985) ∼ 6.3% (OR = 0.937; 95% CI, 0.914 to 0.961) lower odds of an ED visit that did not lead to hospitalization (Figure 3 and Figure 4) in the following year across lookback periods. All the coefficient estimates were statistically significant (see Appendix Table 2–6 for details). Findings from the other 3 continuity measures in our sensitivity analyses were similar to those of the BB-COC findings. They can be found in the Appendix.
Associations between physician-level Bice-Boxerman Continuity of Care (BB-COC) and patient total part a and part b expenditures. Abbreviation: Q, Quintile.
Associations between physician-level Bice-Boxerman Continuity of Care (BB-COC) and patient hospitalization. Abbreviation: Q, Quintile.
Associations between physician-level Bice-Boxerman Continuity of Care (BB-COC) and patient non-hospitalized ED visit. Abbreviation: Q, Quintile.
Discussion
This study’s findings are consistent with prior findings that higher individual-level patient-to-clinician continuity is associated with lower costs and fewer hospitalizations and ED visits.6–11 In addition, these findings strengthen the feasibility of measuring PCP-level continuity and affirm the ability of one such measure (BB-COC) to consistently differentiate PCPs across multiple lookback periods. They also extend our previous findings to a multiyear setting where potential longitudinal patient-physician relationships in the physician’s patient panel are taken into account.15,16 It is now clear that this relationship holds whether physicians are measured in a 1-, 2-, 3-, 4-, or 5-year lookback periods, a critical finding if we are to consider this result in a measurement and value-based payment paradigm. It is also clear that the relationship is the strongest and most robust across multiyear lookback periods for those whose assigned PCP was in the highest COC quintile group, providing a natural threshold for COC quality measures in assessing PCP performance. However, even though extending the lookback period allowed for a multiyear patient-physician relationship in constructing the physician-level COC measure, this winnowing of the patient panel over time—due to death, disenrollment, or changing physicians—diminishes the size and representativeness of the sample and contributes to the attenuated associations we observed in longer lookbacks. These findings suggest that a 1-year continuity measure is not only more feasible but also more predictive of near-term outcomes. Given this consistency and the complexity of multiyear claims measurement, as well as panel attrition in beneficiary capture across multiyear claims assessment, our results support using the most recent 1-year continuity performance as a reliable and efficient proxy for physician-level continuity. Compared with our prior study,16 effect sizes here are smaller, likely due to inclusion of beneficiaries under age 65 with disabilities. Though models adjust for disability status, this group may respond differently to continuity. Their inclusion, however, enhances generalizability to the full Medicare Fee-for-Service population, of which about 15% are under 65.
Numerous studies have shown that continuity of care in primary care leads to better patient outcomes, including reduced mortality, and lower health care costs.6–15 A patient-level BB-COC measure of continuity was endorsed by the National Quality Forum (NQF) as a quality measure for children with complex care (NQF #3153), though it was later withdrawn by the developer.17,20,21 In addition, the clinician-level BB-COC measure has been endorsed as measuring the value functions of primary care first by the NQF in 2021 and currently by its successor - the Partnership for Quality Measurement (PQM, CBE #3617).21 It is also included as a Qualified Clinical Data Registry (QCDR) quality measure available for the Quality Payment Program (QPP) Medicare Incentive Payment System (MIPS) reporting through PRIME Registry, the largest CMS-approved QCDR for primary care. Importantly, the BB-COC measure is based on the physician, and our findings confirm that the BB-COC continuity index is useful as a high-value, physician-level measure for quality and/or resource-use in primary care and it is sufficient to use 1-year measurement, the most recent lookback, in constructing the index. A single year, claims-based continuity measure like BB-COC could specifically be an important potential addition to a parsimonious suite of measures for the Center for Medicare & Medicaid Services MIPS Value Pathways program given the strength of association with lower costs and utilization.18,22 Notably, under the QPP MIPS program, primary care has the largest number of available measures, most of which are intermediate, disease-focused and process oriented. Measures like BB-COC capture core primary care functions, explains its real value to patients and populations, and creates market incentives to enhance their measurement and improvement in primary care.
Several limitations of these analyses and their implications merit acknowledgment. We cannot generalize our findings to non-Medicare populations, but our approach could be easily replicable in other data environments. We also restricted our analysis to physicians due to lack of consistent capture of specialty information for nonphysicians (ie, nurse practitioners and physician assistants), despite their increasing contribution to the primary care workforce. This article does not address how continuity measurement would impact and alter physician behavior in ways that might change discovered associations, an important question for future work and measure implementation. It is important to remind the reader that our unit of measure is the clinician, and that these findings inform the value of clinician-level continuity across a panel of patients, not the value of longitudinal individual continuity itself. Moreover, our approach cannot assess how longitudinal continuity affects the (average) level of continuity in the current patient panel, which is the basis of our physician continuity since the observability of patients in the panel is closely linked to panel attrition. However, the stability of the measure distribution across lookback period and the observed consistent declines in effect with increasing lookback period reinforce the validity of single year measurement, a reassuring finding given the efficiency and increased reporting feasibility of single year versus multiyear measurement for clinicians already stressed by increasing reporting burden. We advise caution in using this study to surmise that longitudinality is not as beneficial as 1-year continuity given the complexities of patient panel attrition in our sample.
Conclusions
Continuity of care is an essential function and feature of PC that differentiates it from other types of care, such as specialty care, and hospital care. Given the magnitude of the association between continuity and outcomes, it should be included in any suite of PC measures to incentivize PC clinicians and practices to provide advanced/quality primary care to their patients, which is the goal of value-based payment. This study adds more clinician-specific evidence to a robust body of evidence supporting value of continuity in primary care. It reinforces the specific value of single year claims-based physician continuity measurement as a priority component of a new MIPS MVP primary care measures suite.22 Future studies should include nonphysician primary care clinicians and investigate whether there are differences in the effects of continuity, and further refine calculations of continuity to take into account patient transitions across settings, such as inpatient to outpatient.
Appendix
Notes
This article was externally peer reviewed.
Funding: This study received support from the American Board of Family Medicine Foundation.
Conflict of interest: None.
Contributors: All Authors (YC, SP, MD, RP, and AB) of the paper contributed to the conception and design of the work. YC performed the statistical analyses. All authors contributed to the interpretation of results. YC, SP, and AB contributed to the drafting of the manuscript. All authors read, critically reviewed, and approved the manuscript. YC is the guarantor of the work and accepts full responsibility for all aspects of the work.
- Received for publication February 11, 2025.
- Revision received June 2, 2025.
- Accepted for publication June 16, 2025.
