Anna D. Sinaiko, PhD
Tiered physician networks are a managed care network design used by payers to contain healthcare costs and improve value in the healthcare system. Physicians are sorted into strata within the network based on their performance on quality and cost-efficiency measures, and patients are assessed lower co-payments for visits with physicians in preferred (eg, better-performing) tiers. This network design aims to channel patients to preferred physicians by virtue of the quality and value signal provided through the rating system, the financial incentive provided through the lower co-payment, or both simultaneously. Motivating providers to improve their performance so as to improve their tier ranking, either to prevent a loss in market share of patients or simply for its own sake, is a second aim of this network design.
Increases in patient out-of-pocket medical costs are known to decrease the amount of medical care that patients use,1
but the majority of evidence on patient response to changes in physician cost sharing is based on price changes that affect all physician services, not just a select, or tiered, group of physicians. Tiered network plans are relatively common in private health insurance—19% of all firms report offering a tiered network in the health plan with the largest enrollment2
—but the evidence on the impact of tiered physician networks on patient choices is sparse. In previous work, I found that physicians with nonpreferred tier rankings and a $10- to $20-higher co-payment for an office visit earned a 12% lower market share of new patients than their top- or average-tiered colleagues, but were not more likely to have their existing patients switch away.3
Tiered networks can also focus on hospitals, and inter-tier cost-sharing differences are usually much larger. Scanlon et al found that workers in one union were more likely to select a preferred hospital for medical visits, which allowed them to avoid 5% coinsurance (equivalent to approximately $400 in out-of-pocket payments), while workers in a second union and all patients admitted for a surgical diagnosis were no more likely to choose within the preferred tier of hospitals.4
More recently, Frank et al evaluated the Blue Cross Blue Shield of Massachusetts tiered hospital network, where patients saved $900 out of pocket ($690) if they chose a top-tiered (middle-tiered) hospital relative to the worst-tiered hospitals, finding increased patient use of hospitals on the preferred and middle tiers relative to the nonpreferred tier.5
A related form of physician network design are “narrow networks” (also called “limited networks”) that exclude physicians with the worst performance on quality and cost-efficiency measures from the network, and provide little to no reimbursement to patients for healthcare services received from out-of-network providers. Following the narrowing of a PPO network in Las Vegas, Nevada, that increased patient out-of-pocket costs to see excluded physicians from a $10 co-payment to 40% of allowed charges, nearly 9 out of 10 patients of affected physicians switched away.6
However, sicker patients were observed to be more loyal to excluded physicians.
There is no evidence on whether the impact of tiered physician networks varies across patients, and the expected impact is uncertain. On one hand, older and sicker patients are likely to use more medical care and, thus, would have a greater net benefit from receiving care from better-tiered physicians. However, physician concerns about reliability of measurement (eg, if the measures used to evaluate performance do not adequately account for additional complexities and resources needed to treat sicker populations) and misclassification risk may make avoidance of higher-risk patients a potential unintended consequence of tiering.7
Information about response gradients across patients is also important if tiered networks are to achieve their greatest impact, as they will need to channel sicker patients—who are associated with the highest proportion of medical spending—to higher-value providers. In contrast, savings due to tiered networks will be less if only the youngest, healthiest patients respond. More evidence can also inform the implementation of tiered networks and the focus of marketing and outreach.
This paper analyzes data from 5 commercial health plans in Massachusetts to test whether the impact of tiered physician networks varies across different groups of patients. Specifically, I assessed whether tier ranking affects physician market share among new patients or the proportion of patients who switch to other physicians for male versus female patients, older versus younger patients, and sicker versus healthier patients.
Group Insurance Commission Tiered Physician Networks
The Massachusetts Group Insurance Commission (GIC) is a quasi-state agency that provides health benefits to over 400,000 state and municipal employees, as well as retirees and their dependents. In fiscal year (FY) 2009 and FY2010, a 3-tiered physician network was included in all GIC non-Medicare health plans, with approximately 20% of physicians in the top tier, 65% classified in the middle, and 15% assigned to the bottom (or worst-performing) tier at each plan. In FY2009, the range of office visit co-payments was from $10 to $15 for the preferred tier, $20 to $25 for the middle tier, and $25 to $35 for the worst-performing tier; co-payments increased by $5 to $10 at all levels in FY2010.
Tier rankings were assigned based on physician quality performance first, and efficiency performance second, using a common, cross-health plan database of performance profiles. Physicians with insufficient data were assigned to the middle tier. Tiered networks in all health plans included cardiologists, endocrinologists, gastroenterologists, obstetricians/gynecologists, orthopedists, and rheumatologists. Plans could choose to implement tiered networks for additional specialties, and several also tiered allergist/immunologists, dermatologists, general surgeons, internal medicine, neurologists, ophthalmologists, and otolaryngology (for additional detail on tiering methodology and specialties, see the eAppendix
[available at www.ajmc.com
Administrative enrollment and claims data, for all non-Medicare individuals enrolled in 5 of the 6 GIC health plans over July 2006 to June 2010 (FY2007-FY2010), were obtained. Data included patient age and gender, diagnosis code, and the providing physician’s name, practice tax identifier, and tier ranking. Data also included a physician identifier variable—unique to each physician—and specialty designation for the providing physician on each claim. This was the same physician identifier used to construct the tiered physician networks by the plans, and allowed for reliable observation of unique physicians across claims data from different health plans. Claims where this identifier was unavailable were dropped (6% of claims).
The analysis sample consisted of physicians in a tiered network in FY2009 and FY2010; the sample included 15,401 physician-year observations, of which 6236 were included in multiple plans’ tiered networks. The main outcomes were the physician’s market share among new patients, stratified by patient characteristics, and the percent of a physician’s patients that switch to another doctor, also stratified by patient characteristics.
Following methods published previously, all office visits with tiered physicians were assessed for whether they were with a new physician or if the patient was a “potential switcher.”3
Patients who switched health plans during this period were excluded, because switches could be based (at least in part) on tiering (eg, to switch to a plan where one’s physician has a better tier ranking) and subsequently bias the results. A patient is defined as “new” if neither the patient nor any other person in the patient’s family had a visit within the same specialty in the previous 2 years. “Potential switchers” are patients who had at least 2 visits in prior years with a specific tiered physician and then had an office visit with any physician of the same specialty type in FY2009 or FY2010. Patients who saw a new, different doctor are classified as switchers. Patients who return to see the previous doctor are classified as non-switchers.
The effect of tiering on these outcomes was assessed within 6 subgroups of patients: 1) “older” patients, who were 46 years or older—the median patient age of patients of tiered physicians; 2) “younger” patients, who were younger than the median age; 3) “sicker” patients, who had a diagnosis in the study year or earlier in one of the 70 condition categories identified for risk adjustment in the Medicare CMS-Hierarchical Condition Categories (HCC) models8
—this method for risk-adjustment was used because it focused on identifying individuals expected to have higher than average medical expenditures, and because it does not require information about the Diagnosis-Related Group assigned to hospitalizations, which was not available in our data; 4) “healthier” patients, who did not have any of these prior diagnoses; 5) males; and 6) females.
Each physician’s market share of new patients by plan and patient group was calculated (eg, Dr Smith’s share of new male patients in health plan X, Dr Smith’s share of new male patients in health plan Y) as the physician’s total number of new patients within a patient group and plan (eg, the number of Dr Smith’s new male patients enrolled in health plan X) divided by the total number of new patients in that group and enrolled in that plan (eg, the total number of new male patients enrolled in health plan X). The percent of a physician’s patients by plan and patient group who switched away was equal to the number of switchers among each group of patients and each plan (eg, the number Dr Smith’s male patients in plan X who switched away) divided by the number of potential switchers in each group and plan (eg, the number Dr Smith’s male patients in plan X who were potential switchers). To avoid misclassifying visits where a patient is seeking a second opinion as a switch, in sensitivity analyses a requirement was imposed that a switch entail a minimum of 2 visits with the new physician in a year.
Following earlier methods, I estimated multivariate models to analyze the impact of tiering. Models included dummy variables for plan, year, and specialty to control for differences in plan benefits, generosity, and tiered-network structures. The key empirical fact is over half (53%) of physicians tiered by at least 2 GIC health plans had different tier rankings across plans (eg, Dr Smith was ranked in the most-preferred tier in Plan A and the middle tier in Plan B in the same year). These differences occurred because the actual cut-points between tiers varied across plans, plans may have considered additional data (beyond the cost and quality ratings calculated from all-payer data aggregated by the GIC) on performance when determining tier rankings, and because plans with more selective or smaller networks may have ranked the same physician lower (in percentile terms) than a broad network simply because they excluded lower-performing physicians from the network. Models included a dummy variable for each physician to control for unmeasured differences among physicians that might differentially attract new patients (eg, reputation). With this control, the coefficient on the variable indicating a physician’s tier ranking is the effect of tiering.
Results are presented as predicted probabilities based on regression models. Additional detail on data and methods is provided in the eAppendix. The Harvard T.H. Chan School of Public Health Institutional Review Board approved this study.
Among patients who had an office visit with a tiered physician in FY2009 or FY2010, 97,896 patients were new patients and 109,270 were potential switchers. New patients were more likely to be female (63%) and 41% had a prior diagnosis of a major medical comorbidity (Table 1
). Potential switchers had similar characteristics, but were sicker (eAppendix Table).
Physicians with the worst tier rankings (eg, they had the lowest performance scores and highest office visit co-payments) earned lower market shares of new patient visits among male patients and new older patients (Table 2
). In contrast, tier rankings did not affect physician market share of new patients who were female or younger. Relative to their average-tiered colleagues, physicians with the worst tier ranking have a predicted market share of new male patients that is 0.08-percentage points lower (market share of 0.76% vs 0.88%), and of new older patients that is 0.12-percentage points lower (market share of 0.81% vs 0.69%). Although the magnitude of these differences is small, it is meaningful on a relative basis as it represents losses in market share of 15% among new male patients (ie, [0.89%-0.76%]/0.89% = 15%) and 15% of older patients.
Physicians with the worst tier rankings earned a lower market share of new patients with and without a major medical comorbidity; however, age and the number of comorbidities are highly correlated. To assess which of these characteristics is driving the results, I looked within the group of sicker patients and analyzed older patients versus younger patients. Physicians with the worst tier ranking did not lose market share among sicker, but rather, younger patients relative to their average-tiered colleagues (Table 2). However, among older and sicker patients who were selecting a new physician, physicians with the worst tier rankings experienced market share that was 0.11-percentage points lower than that earned by their top- and average-tiered colleagues. This is equivalent to a relative loss in market share of 10% among these patients.
There was no effect of physician tier ranking on the proportion of a physician’s patients who switch to other doctors within any of the groups of patients. This result was unchanged in sensitivity analyses where patients were classified as having switched physicians only if they had a minimum of 2 visits with the new physician in a year.
This paper is one of the first to examine whether tiered physician networks have different effects on different types of patients. Patients’ “stickiness” to their own physicians is pervasive, as all patients—including men, women, and patients who are older, younger, sicker, or healthier—were no more likely to switch away from lower-tiered physicians than higher-tiered physicians. When choosing a physician for the first time, however, tiered physician networks channeled new older, sicker patients, and new male patients away from tiered physicians with the worst ranking.
Multiple mechanisms could be at work, as patient choice of new physician could be a result of patients deciding for themselves, physicians using tier-ranking information in their decisions about where to refer their patients, or both. Regardless of whether certain demographic groups or their physicians are more likely to make different choices, the effect of tiering is that the worst-ranked physicians earned lower market share of certain groups of new patients. Another question arising from these findings is why tiering consistently channeled groups away from the worst performers with no movement between the average and the best tier levels. One explanation, is that individuals evaluate options not in terms of absolutes, but relative to reference points. Thus, patients may simply want to avoid physicians with the worst rankings but not move all the way to the top tier. It is also possible that a low number of top-tiered physicians, and capacity constraints in their practices, will prevent patients who want to choose them from doing so.
The most prevalent conditions observed among the “sicker” patient subgroup were diabetes, heart disease, and depression (Table 1)—3 chronic conditions associated with high use of the healthcare system and older age. The finding of the responsiveness among this patient group makes economic sense, as it suggests that tiering is having an effect on choices among the population most likely to consume more care and, thus, who has more to gain from choosing a higher-performing physician and more to save with lower co-payments if they expect to have multiple visits. It is also possible that these patient flows are due to actions on the part of the worst-tiered providers to avoid these patient groups as older patients with these conditions may be more complicated. Further research should investigate this question.
The finding that tiered networks are channeling male patients, but not women, is more surprising. Men and women are known to have different rates of utilization of healthcare services across all types of care due both to healthcare needs and behavioral and attitudinal differences.9,10
Kozhimannil et al found that men were more responsive to the introduction of a high-deductible health plan than women, cutting back on use of care—specifically emergency department visits—more than women, and suggesting that men respond to cost-sharing incentives in health insurance differently than women.11
Within a tiered-network design, patients can continue to see nonpreferred physicians if they are willing to pay a higher co-payment for each visit. Unlike “narrow network” health plans, which provide patients with almost no coverage for services provided by out-of-network providers, tiered networks allow patients to have continued access to a broad network of providers for nominal increases in cost sharing, and, therefore, are likely to be preferable to many consumers who value having a choice of physician. If tiering providers in a network, instead of excluding them, can still channel patients to more efficient, higher-quality physicians, they offer a tool to improve value that is less severe than narrowed networks.
For physicians, the fact that tiered networks could channel patients with the highest medical spending away from certain physicians could be, in some cases, to the advantage of those facing global budgets. However, under new payment models, such as accountable care organizations (ACOs), where physicians are financially accountable for the care their patients receive outside the ACO, as well as within (often called “patient leakage”), these selection effects may be unfavorable. In fact, these findings suggest that tiering is potentially a tool to encourage high-value choice in the context of ACOs and similar accountability models. Tiered network designs could be used to encourage patients to seek care within an ACO, for example, by sorting specialty physicians into tiers according to their ACO affiliation so that patients would pay lower co-payments for visits to physicians within their ACO, thereby aligning patient incentives with those of the ACO providers. Currently, there are no such incentives for patients to seek care within ACOs.
There are a few important limitations. This analysis uses data from the late 2000s, which was a different environment than exists today. However, these data are from a unique natural experiment, and as the prevalence of tiered networks has continued to grow, understanding variations in impact within subpopulations of patients has remained an unanswered question. The focus of this study is a commercially insured, employed population in 1 state. Although GIC beneficiaries consist of a diverse range of workers, the study setting may limit the generalizability of these findings. The financial incentives in the tiered networks studied here are minor, and, thus, this analysis is not a test of the impact of tiered networks when incentives are large; however, many tiered physician networks in commercial plans include financial incentives of this magnitude. The study period focuses on the first few years of the GIC tiered networks initiative, and choices of physician may change as consumers become more familiar with the networks in their health plans, and as healthcare markets change. Finally, data limitations prevent the study of whether tiered networks vary for patients along other important dimensions besides age, gender, and health status, such as income, geography, and racial/ethnic gradients, which are also important in evaluations of the costs and benefits of tiered network designs.
Although there is no easy solution to reducing cost and improving efficiency in healthcare, tiered networks seem to have promise as a part of a set of mechanisms to increase the value of healthcare spending—particularly among those patients associated with the highest proportion of medical spending. Targeting these interventions to encourage patients to make higher-value choices so they reach patients when they are choosing a doctor to see for the first time may be better received by patients, and be more effective than strategies that interrupt existing care relationships. Future work should focus on potential adverse effects of this network design, such as provider avoidance of high-risk patients or patient decisions to stop going to the doctor altogether rather than to switch to a lower-cost one. Such evidence will allow for refinements to the design and implementation of physician networks to maximize their benefits while limiting harm and inequity.