Bruce E. Landon, MD, MBA, MSc

 As US healthcare spending increases, insurers are focusing attention on decreasing potentially avoidable specialist care. Little recent research has assessed whether the design of modern health maintenance organization (HMO) insurance is associated with lower utilization of outpatient specialty care versus less restrictive preferred provider organization (PPO) plans.

 Observational study of Massachusetts residents aged 21 to 64 years with any HMO or PPO insurance coverage from 2010 to 2013.

 We examined rates and patterns of primary care visits, new specialist visits, and specialist spending among HMO versus PPO enrollees. We estimated multivariable regression models for each outcome, adjusting for patient and insurance characteristics.

 From 2010 to 2013, 546,397 and 295,427 individuals had continuous HMO or PPO coverage, respectively. HMO patients had fewer annual new specialist visits per member versus PPO patients (unadjusted, 0.37 vs 0.43), a difference after adjustment of 0.05 annual visits, or a 12% relative decrease among HMO members (

 <.001). These visits were more likely to be with a specialist in the same health system as the patient’s primary care physician (44.9% vs 40.7%; adjusted difference, 2.8 percentage points; 

 <.001). Mean annual spending on new specialist visits and subsequent follow-up per member was lower in HMO versus PPO patients (unadjusted, $104.10 vs $128.10), translating to 12% lower annual spending (adjusted difference, —$16.26; 

 Having HMO insurance was associated with lower rates of new specialist visits and lower spending on specialist visits, and these visits were less likely to occur across multiple health systems. The impact of this change on overall spending and clinical outcomes remains unknown.

Is specialist “gatekeeping” in modern health maintenance organization (HMO) insurance associated with differences in outpatient patterns of care?

Compared with preferred provider organization insurance, HMO insurance was associated with lower rates of new specialist visits.

These visits were less likely to occur across multiple health systems.

HMO insurance was also associated with 12% lower specialist visit spending, which was largely driven by lower use, not lower price.

HMO gatekeeping may meaningfully reduce specialist utilization, although the impact of this change on overall spending and clinical outcomes remains unknown.The majority of private health insurance options in the United States fall into 1 of 2 broad categories: preferred provider organization (PPO) plans and health maintenance organization (HMO) plans. PPO plans generally have broadly inclusive physician networks with little barrier to self-referral, whereas lower-premium HMO plans generally have less inclusive networks and rely on provider-facing managed care strategies, such as requiring primary care physicians (PCPs) to approve referrals to access specialists (ie, “gatekeeping”).

 As of 2014, across the United States, HMO insurance products accounted for 24% of enrollment in the employer-sponsored insurance market, whereas PPO products accounted for nearly 50% of the market.

 In Massachusetts, the setting for this analysis, HMO insurance was 39% of the commercial market, with an additional 19% with point-of-service (POS) coverage and 37% with PPO coverage in 2016.

 One concept underpinning the provider-facing strategies used by HMOs is that PCPs can reduce the use of low-value specialty care and ensuing downstream utilization by either treating a condition in the primary care setting or directing specialty referrals to higher-value providers within their health system. This is in contrast to the consumer-facing cost-control strategies seen in PPO plans that may impose higher cost sharing but retain open provider choice and access at the discretion of the enrollee.

In general, HMO plans require enrollees to identify a PCP to help direct their downstream utilization. Conceptually and empirically, a single PCP with overall responsibility for a patient’s care could lead to better-coordinated care while reducing potentially avoidable outpatient specialist referrals.

 Such gatekeeping arrangements, however, can also create problems for both patients and PCPs when disagreements arise over whether a specialist is needed.

The extent to which such gatekeeping affects specialist utilization or costs has mixed results in studies that are largely more than 15 years old.

 Little recent research has assessed whether the design of modern HMO insurance is associated with lower utilization of outpatient specialty care compared with modern PPO plans. Moreover, broadly understanding the effect of insurance design on specialty utilization across a large heterogeneous payer market is an important policy-relevant question recently made feasible through the availability of statewide all-payer claims databases (APCDs).

We used the Massachusetts APCD to examine the association between insurance design and outpatient care utilization patterns in multiple HMO and PPO arrangements. In Massachusetts, the primary distinction between HMO and PPO plans relates to gatekeeping requirements (ie, a referral requires PCP approval), in contrast to the existence of “closed-model” HMOs like Kaiser Permanente (ie, regardless of PCP approval, referrals outside the internal network are largely unavailable), which do not currently have market share in the state. We hypothesized that HMO enrollment would be associated with more PCP visits and fewer new specialist visits, with a larger share of specialty care received within the PCP’s health system.

The Massachusetts APCD contains detailed data on healthcare utilization, insurance eligibility, and provider credentialing across all commercial payers and public health insurance programs in Massachusetts, representing approximately 90% of the nonelderly population in Massachusetts.

Our study cohort included all Massachusetts residents aged 21 to 64 years who had 4 calendar years of continuous enrollment with any commercial HMO or PPO product from an APCD-participating commercial insurance provider from 2010 to 2013; public insurance was not included. We focused on members with this length of continuous enrollment because we were interested in the long-term association between HMO or PPO membership and specialist use among patients with stable benefit design. Enrollees may change from an HMO to a PPO plan or vice versa because of their plans for specialist use, which we wanted to reduce as a source of bias. We did not include children given the unique characteristics of the pediatric healthcare market in the state. The key exposure of interest was whether an individual was a continuous member of an HMO versus a PPO product, as defined by the insurer submitting data in the APCD member eligibility file. These designations are used by the insurers themselves to distinguish HMO from PPO plans for administering health plan benefits and requirements. We defined an HMO product as any product designated as HMO or POS (9.6% of all members, or 15.2% of HMO members in the final study sample), which have very similar benefit designs among the large insurers in Massachusetts.

 PPO products were defined as those with a designation of PPO or exclusive provider organization (EPO; 1.5% of all members, or 4.3% of PPO members in the final study sample), which have similar flexibility to PPO plans.

 We used diagnoses from all outpatient and inpatient claims in 2010 to measure baseline comorbidities for risk adjustment. We then performed cross-sectional analyses that focused on the period from 2011 to 2013. We defined continuous enrollment as 11 or more months of insurance coverage in a year, such that those with a short period of discontinuous enrollment, often due to an administrative error, were not excluded from the analyses. This project was approved by the Committee on Human Subjects at Harvard Medical School.

To attribute patients to PCPs, we assigned enrollees to the PCP with the plurality of office visits in a calendar year (“empirical” PCP assignment).

 Nonphysician primary care providers, such as nurse practitioners or physician assistants, were not included because they were not consistently identified in provider classifications provided by insurers. We identified visits to primary care based on claim lines with evaluation and management (E&M) Current Procedural Terminology (CPT) codes to physicians classified as general internal medicine, general practice, or family medicine. If there were no primary care visits, we assigned enrollees to the non—primary care office-based specialist (eg, excluding radiology, pathology) with the plurality of visits. Any ties between physicians were broken randomly, and enrollees with no visits were not assigned to a PCP or another physician. We also examined patterns of PCP assignment as recorded by insurers, which we labeled “insurance” PCP assignment.

We also assigned physicians to larger physician organizations or health systems in Massachusetts using a publicly available database of Massachusetts hospital and physician group national provider identifiers,

 grouped into physician organizations using the affiliations defined by the Massachusetts Health Care Delivery System Map, a manually curated listing of hospital and clinic affiliations within larger provider organizations (eg, which satellite clinics are part of Partners HealthCare) published by the Blue Cross Blue Shield of Massachusetts Foundation.

 We assigned 76% of physicians to organizations using this approach, and the remainder were largely unaffiliated independent physicians.

We assessed several dimensions of outpatient care delivery potentially influenced by type of insurance coverage. For primary care utilization, we calculated the average annual number of PCP visits of any kind per member from 2011 to 2013 and the proportion of patients with no PCP visits in a year. For primary care assignment, we assessed the proportion of patients ever assigned a PCP by their insurer and the percentage of all PCP visits occurring with those providers, as well as the consistency of empirically attributed and insurer-defined PCP assignment measured by the number of assigned PCPs per member over the 3-year study period.

We defined a new specialist visit as any claim containing a new visit E&M CPT code with a non-PCP, which can only be billed if a patient has not been seen by any physician within that practice in the previous 3 years.

 We then measured characteristics of new specialist visits that could be associated with higher likelihood of PCP-initiated referral, as opposed to patient self-referral. We did not attempt to identify specialist-to-specialist referrals because HMO referral policy focuses on PCPs as the key provider for approving referrals. We examined whether a new specialist visit’s claim designated a referring physician and whether that referring physician was the patient’s assigned PCP. We also measured the time in days between a new specialist visit and the most recent PCP visit for the patient in the past 365 days. When we had information on organizational affiliation for both the PCP and the specialist (“health system match” was available for 59% of new specialist visits), we assessed whether the PCP for the most recent visit or the assigned PCP was affiliated with the same provider organization as the specialist for the new visit.

We measured the percentage of patients with any new specialist visit from 2011 to 2013 and the average number of new specialist visits per patient in that period. Because HMO gatekeeping may have its most pronounced effect on the highest utilizers, we also examined the proportion of patients seeing 3 or more new specialists or seeing new specialists in 2 or more provider organizations, which we identified as approximately the top decile of utilization for the entire study population.

Lastly, we measured the average cost per new specialist visit and all subsequent visits with that specialist over the subsequent year by summing all costs (including insurer reimbursement and any out-of-pocket spending) from all E&M claims for new specialist visits for members, as well as E&M costs from any downstream established office visits with the same specialist seen in the following year. To test whether differences in spending by insurance type were driven by differences in price versus quantity of specialist visits, we standardized specialist visit prices using the average HMO cost per evaluation and management CPT code as a standard price for all encounters.

To measure comorbidities, we calculated Hierarchical Condition Category (HCC) scores from all diagnosis codes in the baseline year using software available from CMS. The HCC score represents the predicted risk of healthcare spending beyond average in the next year; a score of 1.0 corresponds to a risk profile for a Medicare enrollee with average expected annual spending.

 To derive a proxy measure of socioeconomic status, we linked individuals’ 5-digit zip code, the most granular geographic unit available, to the publicly available area deprivation index (ADI), a measure derived using a weighted combination of 17 Census-level indicators of socioeconomic disadvantage.

 We also included the median co-pays for primary care and specialist physician visits for each separate insurance product in 2010 as a measure of plan generosity. Other variables included member age as of December 31, 2010; sex; fixed effects for insurance carrier (eg, Blue Cross Blue Shield vs UnitedHealthcare); insurance plan risk type (fully insured vs self-insured); and employer size (ranging from individual to ≥500 employees; see 

We compared the characteristics of HMO and PPO patients and the unadjusted rates of the outcomes using χ

 tests. To estimate adjusted comparisons between HMO and PPO patients, we fitted a series of multivariable regression models for each outcome, with the key explanatory variable being an indicator for whether a patient was a member of an HMO versus a PPO insurance product. We fitted models with different link functions depending on the outcome: linear models for visit counts, costs, and visit intervals; Poisson models for counts of unique physicians or health systems; and logistic models for all other outcomes, which were binary or fractional in nature. All models adjusted for patient age, sex, HCC score, ADI, median primary care and specialist co-pays in that plan (with the exception of spending outcomes, which excluded co-pay covariates), and employer size. We also included a fixed effect for insurance carrier to control for unmeasured characteristics of carriers’ networks or policies that might also influence the outcomes of interest. As a sensitivity analysis, we also estimated logistic regression models predicting the propensity to have HMO versus PPO insurance and then used propensity score weighting to more robustly address imbalance on observable variables between HMO and PPO patients and replicated all of our model results (see 

Because of the mix of functional forms (eg, linear, Poisson, logistic), we used bootstrap methods within each model to present the effects of HMO membership as the adjusted difference in each individual outcome between the average HMO member and the average PPO member. For each effect, we drew 1000 simulations of the set of regression coefficients using the asymptotic multivariate normal approximation to the log-likelihood function, and we present the mean and 95% CI from these simulations.

 Analyses were performed in R version 3.1.3 and SAS version 9.4 (SAS Institute; Cary, North Carolina). The 95% CI around reported estimates reflects 0.025 in each tail, or 

The study sample included 546,397 and 295,427 individuals with 4 years of continuous coverage in an HMO or PPO insurance product, respectively, from 2010 to 2013. HMO enrollees were younger than PPO patients (45.7 vs 46.1 years, respectively; 

 <.001), had slightly lower HCC scores (0.34 vs 0.36; 

 = .01), had higher ADI scores (ie, were more socioeconomically disadvantaged; 83.8 vs 80.4; 

 <.001), and were significantly less likely to be employed by an employer with 500 or more employees (51.7% vs 81.1%; 

 <.001; Table 1). Sample characteristics after propensity score weighting were identically balanced (

HMO patients had fewer PCP visits annually than PPO patients (2.31 vs 2.45, respectively), but there was only a marginal difference after adjustment (adjusted difference, —0.02; 

). As expected, HMO patients were significantly more likely to ever have a PCP assigned by their insurer (90.4% vs 39.7%; adjusted difference, 46.2 percentage points; 

 <.001) and had a higher proportion of visits with those PCPs than PPO patients did with their designated PCPs (57.0% vs 19.0%; adjusted difference, 38.3 percentage points; 

Over the 3-year period, 56.2% of HMO patients had any new specialty visit versus 62.4% of PPO patients (adjusted difference, —5.3 percentage points; 

), with HMO patients having 14% fewer annual new specialist visits per member (0.37 vs 0.43; adjusted difference, —0.05; 

 <.001). Consistent with this lower utilization, fewer HMO patients were likely to have seen 3 or more specialists from 2011 to 2013 (13.3% vs 16.2%; adjusted difference, —2.5 percentage points; 

 <.001) or specialists in 2 or more health systems (21.8% vs 24.3%; adjusted difference, —2.2 percentage points; 

The interval between the most recent PCP visit and a new specialist visit was 7.7 fewer days for HMO versus PPO patients (

). A lower proportion of PPO patients had no PCP visits in the 60 days preceding a new specialist visit (55.0% vs 49.4%; adjusted difference, 5.3 percentage points; 

 <.001). HMO patients were more likely to have a referring physician for a new specialty visit (53.3% vs 24.5%; adjusted difference, 36.7 percentage points; 

 <.001) and had a modestly higher rate of health system match (44.9% vs 40.5%; adjusted difference, 2.8 percentage points; 

The average total cost per individual new specialty visit was $12.29 lower for HMO versus PPO patients, a relative difference of 6.1% (mean cost, $189.05 vs $201.34; 

). On average, office visit co-pays for both primary care and specialist visits were higher for HMO patients than PPO patients ($14.83 vs $13.48 for primary care; $20.78 vs $19.55 for specialists; 

 <.001 for both). Mean annual spending on new specialist visits plus subsequent downstream visits per member was also lower in HMO versus PPO patients ($104.06 vs $128.14; adjusted difference, —$16.26; 

 <.001) (Table 5), even after using a standardized price ($104.10 vs $122.98; adjusted difference, —$11.74; 

 <.001) (Table 5), suggesting that most of the difference in spending was related to quantity of new specialist visits and not to price.

There were minimal differences in our estimates for these models using propensity score weighting to balance the observable characteristics of HMO- versus PPO-enrolled individuals (

Using statewide data from Massachusetts, we found that HMO insurance was associated with modestly lower rates of new specialist visits and lower specialist visit costs compared with PPO insurance. New specialty visits within HMOs also were more likely to have closely followed a PCP visit, included a referral from a PCP in the claim, and been within the PCP’s health system. Because this study was observational, we cannot interpret our conclusions as causal, and unobserved differences between the HMO and PPO populations may explain some of the effects we observed. However, these results are consistent with the concept that modest provider-facing requirements of HMO coverage may have the effect of PCPs for HMO patients taking a more active role in managing referrals.

Given the emphasis on gatekeeping within the HMO model design and policies, the relatively modest effect size on specialty utilization is notable. There may be few consequences for PCPs not actively coordinating referrals for HMO patients, such that they might simply have administratively approved all patient-initiated referrals without trying to influence them. Consistent with this fact, we found that the referring physician matched the assigned PCP only 30.8% of the time for HMO patients, despite the fact that 90% of HMO enrollees were assigned a PCP by their insurer. This suggests that current HMO policies may not be implemented consistently to maximize referral coordination. Supporting this observation, we find contrasting patterns of use between patients’ insurance-assigned versus empirically assigned PCPs, with 57% of HMO patients’ PCP visits occurring with insurance-assigned PCPs. It is not clear whether this proportion is desirable from the insurer’s perspective or too low. However, the match between patients’ assigned PCP and the PCP actually seen could be an important determinant of the effectiveness of HMO design.

The modest effect size we find is consistent with the “norms hypothesis,” which postulates that physicians have a relatively uniform approach to care consistent with overall financial incentives as opposed to the payment arrangement for an individual patient they are treating.

 As accountable care organization (ACO) and other risk-based contracts become more prevalent, and as physician organizations take a more active role in managing new specialty use, we will likely see more pressure on PCPs to actively coordinate and approve specialty care (especially within ACO networks to which they belong) in the future.

We also found that reduced spending on new specialist visits was driven largely by differences in utilization, rather than price.

 Although HMOs frequently negotiate lower prices in return for a more restricted network, our results show that the bulk of spending differences was driven by lower utilization of specialty care in HMOs. This strategy, likely influenced by PCP decision making, is a strong contrast to consumer-facing approaches, such as high deductibles, reference pricing, or price transparency, that aim to achieve reduced spending by increasing consumer sensitivity to prices, rather than reducing utilization overall. It is possible that a combination of consumer- and provider-facing approaches could be more effective at reducing medical spending than either alone.

Our findings are consistent with those of some prior research on the association of gatekeeping with specialist utilization. In 1979, a randomized controlled trial of a gatekeeping benefit design with more than 2000 participants found decreased rates of all specialist visits among patients with a gatekeeping arrangement versus those without.

 In contrast, a 1998 natural experiment in a large multispecialty physician group where gatekeeping arrangements were eliminated found no change in total specialist visits, although rates of new specialist visits increased modestly after gatekeeping elimination.

 In that study, however, the group also had constrained the availability of specialists within its system, which might have served to limit new specialist visits. Both of these studies are now decades old, making them less applicable to the current healthcare landscape.

This study has several important limitations. First, this is a cross-sectional and observational analysis, so our results cannot be interpreted as causal. We addressed this limitation in part by using propensity score weighting to address the impact of observable patient characteristics on the outcomes of interest. Second, our sample is restricted to Massachusetts residents, which may not generalize to other states, although HMO and PPO insurance are the most commonly used types of plans throughout the country. In addition, we relied on insurers’ designations of HMO versus PPO plan types to identify exposure to PCP gatekeeping patterns. We also combined similar plan types, like POS and EPOs, into these categorizations. This coarse categorization may misclassify the level of PCP control over referrals in a plan because we can only observe the HMO/PPO designation. Third, we are unable to capture potential mechanisms that providers or patients might use to mitigate the influence of gatekeeping tactics, such as automated approval processes or delegation of referral approvals to nonclinical administrative staff, which could bias our findings toward the null. Fourth, our ability to capture socioeconomic status, an important confounder, is limited by relatively coarse data on 5-digit zip code alone. Fifth, our cross-sectional analysis assumes that any changes in benefit design over time were similar between HMO and PPO plans. Due to missing data in the APCD, we were unable to confirm this assumption, which may bias our results. Sixth, we did not have access to accurate and complete data on some aspects of cost sharing, such as deductibles, which could have influenced our findings. Lastly, we were limited in our ability to capture the appropriateness of a referral using claims data alone, which lack the detailed clinical information necessary to assess a referral’s clinical necessity.

We found that HMO coverage was associated with similar PCP utilization but reduced specialist utilization compared with PPO coverage. To our knowledge, this analysis is among the first examinations of HMO benefit design in the current era across a large multipayer population. Not only was specialist utilization reduced, but the referrals that did happen among HMO patients appeared to be more physician-directed and coordinated within health systems. It is not clear whether this pattern is associated with improved outcomes, patient satisfaction, or population health, a set of questions that deserve further research. However, in light of the growing use of benefit designs such as high-deductible health plans that shift decision making to patients who might be poorly equipped to make treatment choices, modern implementation of HMO methods may present an alternative provider-facing strategy for cost control worth revisiting. However, care must be taken to avoid previous errors in overly onerous gatekeeping designs that were not sensitive to patient needs and appropriate clinical decision making.

 Department of Health Policy and Management, Harvard T.H. Chan School of Public Health (MLB), Boston, MA; Division of General Internal Medicine and Primary Care, Department of Medicine, Brigham and Women’s Hospital (MLB, AB), Boston, MA; Department of Medicine, Massachusetts General Hospital (ZS), Boston, MA; Department of Health Care Policy, Harvard Medical School (ZS, AB, SR, BEL), Boston, MA; Ariadne Labs (AB), Boston, MA; Division of General Medicine and Primary Care, Beth Israel Deaconess Medical Center (BEL), Boston, MA.

 Supported by a grant from the National Institute for Health Care Management.

Articles

Is specialist “gatekeeping” in modern health maintenance organization (HMO) insurance associated with differences in outpatient care? The study finds that HMO gatekeeping may meaningfully reduce specialist utilization.

Gatekeeping and Patterns of Outpatient Care Post Healthcare Reform

Prior analyses of Medicare health plans have exam-ined either utilization of services or quality of care, but not both jointly. Our objective was to compare utilization and quality for Medicare Advantage (MA) enrollees with diabetes or cardiovascular disease to that for similarly defined traditional Medicare (TM) beneficiaries.

Cross-sectional matched observational study using data for 2007.

 We obtained individual-level Healthcare Effectiveness Data and Information Set (HEDIS) relative resource use (RRU) and quality data for patients enrolled in MA, and then developed comparable claims-based measures for matched samples of TM beneficiaries. Main outcome measures: utilization levels for inpatient care, evaluation and management services, and surgery; number of emergency department (ED) and inpatient visits; and quality of ambulatory care measures.

We studied approximately 680,000 MA health maintenance organization (HMO) enrollees with diabetes and 270,000 HMO enrollees with cardiovascular conditions. For both conditions and almost all major strata, the RRU was lower for those enrolled in MA than for those in TM. Spending for those with diabetes was $5223 for MA HMO enrollees compared with $6413 for those in TM (cost ratio, 0.81; 

 <.001). ED utilization rates were consistently lower in MA than TM (567 vs 719 visits/1000 enrollees; rate ratio, 0.79; 

 <.001). Health plans that are more established, nonprofit, and/or larger generally had lower resource use and better relative quality than did smaller, newer, for-profit HMOs or preferred provider organizations.

RRU for those with diabetes or cardiovascular disease is lower in MA, while quality of care is higher. Better MA plans may add value to the care of these major chronic medical conditions.

We compared price-standardized utilization and quality of care for Medicare Advan-tage (MA) enrollees with diabetes or cardiovascular disease with matched beneficiaries from traditional Medicare in geographic areas. We found that:

		For both conditions, relative resource use was lower for those enrolled in MA than for those in traditional Medicare.

		Quality of care for diabetes and cardiovascular disease measures was higher in MA for the 4 measures examined, although plans varied greatly in their performance.

		Health plans that are more established, nonprofit, and/or larger generally had lower relative resource use and better relative quality than did HMOs or PPOs that were smaller, newer, and/or for-profit.

Under managed competition, health plans are expected to compete to provide health services of high quality while also controlling the overall costs of care. A key element of a health plan’s ability to contain cost is its ability to control utilization. Although some transparency initiatives are now publicizing the costs that health plans pay for specific services,

such approaches have not assessed differences in the full range of medical services used by patients with particular health conditions. Moreover, many have concerns that health plans that are more successful at controlling utilization might also deliver care of lower quality, though existing data from commercial health plans show no clear relationship between spending and quality.

Medicare’s managed care program, Medicare Advantage (MA), currently provides care to 15.7 million Americans, representing 30% of beneficiaries.

 Relative to traditional Medicare (TM), MA plans may be able to treat patients with particular diagnoses with greater efficiency while attaining equal or superior quality through their flexibility in enrollee benefits, network contracting, and coordination of care, but whether they do so is not known.

 Since 1997, health plans participating in Medicare have been required to report annually on measures of the quality of preventive care and of the management of chronic diseases such as diabetes.

 In 2006 and 2007, CMS also required MA plans to report a set of relative resource use (RRU) measures focused on the care of patients with diabetes (2006 and 2007) and cardiovascular disease (2007 only).

 Such measures of price-adjusted utilization allow for direct comparisons of utilization among health plans as well as between MA and TM.

In this study, we evaluated both the utilization of services and the quality of ambulatory care provided by MA health plans by comparing annual standardized spending and quality of care for 2 specific medical conditions for MA health plan enrollees with corresponding measures among TM beneficiaries.

From CMS and the National Committee for Quality As-surance (NCQA), we obtained health-plan-level Healthcare Effectiveness Data and Information Set (HEDIS) measures of RRU and individual-level HEDIS data on quality of ambulatory care for patients with diabetes and cardiovascular disease enrolled in MA health plans. We focused on care delivered in 2007, the most recent year in which MA plans were required to report these measures before reporting became voluntary and much less consistent.

 We constructed similar measures within TM from Medicare claims data for a random 20% sample of beneficiaries. Given the well-known geographic variation in Medicare services, we created matched samples for each health plan based on its geographic region and enrollee demographic characteristics.

. The RRU measures use standardized pricing applied uniformly to services delivered within MA and TM, thereby accounting for variation in prices due to geography and negotiated prices in MA.

 Spending on all services for all patients with qualifying diagnoses is calculated over the entire year. The RRUs are risk-stratified by age (categorized into 5-year intervals for patients aged 65-85 years), sex, type of diabetes (type 1 or type 2) or cardiovascular disease (acute myocardial infarction, congestive heart failure, angina, or coronary artery disease) and the presence or absence of 1 or more major comorbidities (ie, cardiovascular conditions, chronic obstructive pulmonary disease [COPD], depression, hypertension, or chronic kidney disease for diabetes, and asthma, COPD, diabetes, and hypertension for cardiovascular disease). We then aggregated these strata using nationally determined weights to create a standardized measure of overall resource use. Spending categories include inpatient care, surgery and other procedures, and evaluation and management services, and we also sum spending across these categories. Rates of emergency department (ED) visits and inpatient admissions are also reported. Beneficiaries with concomitant specified dominant medical conditions including active cancer, end-stage renal disease, human immunodeficiency virus/AIDS, and organ transplants are excluded.

HEDIS quality data are collected from administrative billing or encounter records, or by using a hybrid approach in which medical records are also reviewed for services that may not be recorded in administrative data.

CMS has audited HEDIS quality measures reported by Medicare health maintenance organizations (HMOs) and found them to be highly accurate.

 In order to compare MA and TM, we focused on measures that can be constructed from Medicare claims for TM enrollees, including low-density lipoprotein (LDL) cholesterol testing in the current year for enrollees aged 65 to 75 years with cardiovascular disease, and 3 services for enrollees aged 65 to 75 years with diabetes: glycated hemoglobin (A1C) testing in the current year; LDL cholesterol testing in the current year; and a diabetic retinal exam in the current or prior year.

We defined health plans as CMS contracts, meaning a health plan unit operating in a single state, or in a few cases, up to 3 adjoining states, and we included both HMOs and preferred provider organizations (PPOs). We focused on beneficiaries 65 years or older who were enrolled for the entire calendar year. We excluded beneficiaries in legacy health plans that were reimbursed on a cost basis rather than by capitation. In addition, we excluded beneficiaries in private fee-for-service plans because these plans are not required to report HEDIS data to CMS, as well as those enrolled in special needs plans because such plans serve nonrepresentative beneficiaries. Finally, we excluded HMOs with fewer than 500 enrollees (accounting for <0.2% of enrollees).

To create a comparison sample for each health plan for both the RRU and the diabetes quality analyses, we used the TM enrollment file and Part A and Part B claims files for a random 20% of beneficiaries to identify all persons who were continuously enrolled in Medicare Part A and Part B for the entire reporting year and were 65 years or older as of January 1, 2007.

We excluded residents of long-stay nursing homes—identified using a validated algorithm&mdash;because these beneficiaries rarely enroll in MA; however, we had no similar method to exclude these beneficiaries from the MA data.

We applied NCQA specifications to identify the eligible populations for the measures, assigned a standardized priceobtained from NCQA to each delivered service identified in the claims, and aggregated spending across groups using the exact specifications from NCQA.

Medicare Beneficiary Summary files provided demographic data (age, sex, race/ethnicity, zip code, county, and state of residence), vital status, and health plan and Medicare enrollment information for each beneficiary.

We categorized health plans as large (>25,000 enrollees) versus small, and identified health plans new to MA since 2006. CMS provided the tax status.

We compared RRUs and quality of care in each MA plan with a TM sample matched by geographic distribution (RRUs and quality measures) and demographic characteristics (quality measures only) and then aggregated these results to obtain national estimates. For the RRU measures, the control TM population was weighted to match the exact distribution of health plan enrollees across all zip codes in which it operated. Matching on geography controlled for variation in practice patterns within Medicare across regions.

By matching at the zip code level where possible, we also controlled for unmeasured socioeconomic characteristics associated with residence at this level of geography.

To provide nationally representative estimates of over-all HMO and PPO performance relative to traditional Medicare, we averaged HEDIS RRUs and quality mea-sures, respectively, over MA plans and matched cohorts of traditional Medicare enrollees, weighted by MA enrollment. To assess variation in performance of health plans based on particular characteristics relative to matched traditional Medicare in local areas, we used hierarchical regression models with correlated bivariate random effects for each health plan and its matched traditional Medicare sample and fixed effects for the health plan characteristics noted above, including an indicator vari-able for PPOs, with separate coefficients for the MA and traditional Medicare samples. Because more than 80% of PPOs were small, new, and for-profit, the PPO measure pertains to just this category of PPOs (the few other PPOs were not included in these models).

Finally, for each HMO plan, we created a composite of the 3 diabetes quality measures by taking the mean across the measures, and similarly created an aggregate measure of spending by summing over the 3 spending categories: inpatient care, surgery and other procedures, and evaluation and management services. We constructed similar summary measures of quality and spending for the matched TM cohort for each health plan. We then plotted mean quality relative to the local TM comparison group against mean spending relative to the same comparison group to provide a visual representation of the marginal contribution versus TM for each health plan. We also created a similar plot for the cardiovascular cohort using the single quality measure available for this group.

Analyses were conducted with SAS version 9.2 (SAS Institute, Cary, North Carolina). Two-tailed P values are reported for statistical tests. Our study protocol was approved by the Human Studies Committee of Harvard Medical School and the CMS Privacy Board.

Characteristics of the Medicare HMOs and PPOs in our study appear in 

. We studied data from 190 HMOs and 67 PPOs that included 4,207,433, and 318,293 enrollees, respectively, in 2007. About two-thirds of enrollees were in for-profit health plans. Although 75% of the health plans were small (<25,000 members), these HMOs represented only about 25% of enrollment. Most HMOs had participated in Medicare prior to 2006, but only 11 PPOs had done so.

The MA HMOs in our study enrolled approximately 680,000 beneficiaries with diabetes (and the PPOs approximately 50,000) and approximately 268,000 enrollees with 1 of the 4 cardiovascular conditions (and approximately 12,000 PPO enrollees; sample sizes vary slightly by measure). For the diabetes cohort, just under half were male (48.8%) and more than 80% were white. The largest proportion was from the south and more than 85% had at least 1 comorbid condition (

). After weighting the TM sample to match the MA distribution, the 2 samples had identical demographic characteristics.

Utilization in MA and TM by Types of Diabetes and Cardiovascular Disease

With 1 exception, total standardized spending, as well as each of the 3 categories of spending, was markedly lower for MA HMO enrollees than for matched TM enrollees (

). For instance, total spending was 19% less for diabetics enrolled in MA than for those enrolled in TM ($5223 vs $6413; 

 <.001). The single exception was for diabetics without comorbidity (eg, inpatient spending of $1309 for MA vs $925 for TM; 

 <.001). Visits to the ED were consistently lower in MA, as were rates of hospital inpatient discharges. Similar results were observed for those enrolled in PPOs (

Spending and Utilization by Type of Health Plan 

Patterns of comparative price-standardized utilization by type of plan are presented in

 shows the detailed results). For large, established, nonprofit HMOs, all 3 categories of spending were lower than for the matched TM sample, with differences ranging from 16% (evaluation and man-agement services for diabetics) to 70% (surgery rates for those with cardiovascular disease). In contrast, results were mixed for new, for-profit, small HMOs, where aggregate spending was higher in some categories in MA (eg, inpatient spending for cardiovascular disease, 16% higher), and lower for others (eg, surgery and procedures 43% lower for diabetics in MA). Results for PPO health plans were similar to those for new, for-profit, small HMOs.

Spending and Quality of Care by Health Plan Characteristics 

The mean plan-weighted rates of A1C testing, LDL cholesterol testing, and diabetic retinal exams were 89.7%, 87.6%, and 65.2%, respectively. 

 presents scatter plots of HMO health plan spending and the composite measure of quality for the diabetes cohort by health plan age (established prior to 2006 or not), size (>25,000 members), and tax status. Several findings are apparent: first, while members of most health plans experienced higher quality than the matched TM population in their area and therefore are plotted above the solid horizontal axis, the MA-TM difference varied substantially and for some plans was negative (MA worse than TM). Similarly, MA spending was lower than that in TM for most health plans, but the magnitude of the difference varied and was sometimes reversed (plotted to right of solid vertical axis). Second, there was little association between the spending and quality effects (Pearson correlation coefficient, 0.16), as manifested by the nearly equal distribution of plans across the 4 quadrants formed by median splits (dashed lines) on the 2 variables. Finally, although HMO health plans of each type are present in all 4 quadrants of each plot, the upper left hand quadrant (higher quality/lower spending) contains the most established, large, nonprofit HMO plans and the lower right hand quadrant (lower quality/higher spending) contains a higher proportion of new, small, for-profit HMO plans. A similar pattern was seen for the cardiovascular measures (

This study provides the first compre-hensive comparison between MA and TM of price-standardized utilization and quality of care for those with diabetes and cardiovascular disease, 2 prevalent and costly chronic medical conditions. We found several notable results: first, for both cohorts, RRU—which is a measure of total utilization using a standardized set of prices&mdash;was lower in MA health plans than in TM in each of the main categories of spending examined. Moreover, MA plans achieved higher performance on measures of ambulatory quality.

 Second, marked heterogeneity was evident among MA plans. Most older, larger, nonprofit health plans were able to achieve substantial reductions in service utilization while delivering care of high quality, whereas many newer, smaller, for-profit plans had similar or greater utilization when compared with TM. Finally, utilization among PPOs—an alternative arrangement to HMOs that is generally less managed and coordinated&mdash;showed patterns that were similar to new, smaller HMOs and to TM.

Our findings have important implications for policy. With the 2010 passage of the Affordable Care Act (ACA), the attention of policy makers has now shifted to controlling the seemingly inexorable growth in healthcare costs.

Delivering high-value care requires decreasing utilization of services of low value while simultaneously maintaining or increasing the delivery of services of high value. We show that substantial numbers of larger, nonprofit HMO MA plans appear to be delivering care of high quality, while doing so with substantially fewer resources. Whether this is more due to their actions, as payers perhaps linked to their size and ability to influence provider behavior, or to use of more limited and selected networks of providers, or both, is unknown. Although there has been much focus on payment reform in TM—such as the launching of accountable care organizations (ACOs)&mdash;30% of Medicare enrollees are in MA health plans, a far larger proportion than are currently in ACOs.

MA plans are currently paid more than TM on average.

Because of Medicare regulations and competitionamong plans, many of the savings from these extra payments and the reduced utilization we documented in this study are passed through to beneficiaries in the form of lower premiums, less cost sharing, and benefits for noncovered services.

 However, providers may also profit from these excess payments, as they may be able to negotiate higher prices from MA plans. To finance its expansion of health insurance, the ACA reduced reimbursement for MA plans; how these reductions will impact plan and beneficiary participation and the future growth of the MA program remains an open question.

Compared with traditional Medicare, relative resource use for those with diabetes or cardiovascular disease is lower in Medicare Advantage, while quality of care is higher.

Bruce E. Landon, MD, MBA, MSc

Articles

Gatekeeping and Patterns of Outpatient Care Post Healthcare Reform

A Comparison of Relative Resource Use and Quality in Medicare Advantage Health Plans Versus Traditional Medicare