Patient cost sharing and medical expenditures for the Elderly

Abstract

Despite the rapidly aging population, relatively little is known about how cost sharing affects the elderly's medical spending. Exploiting longitudinal claims data and the drastic reduction of coinsurance from 30% to 10% at age 70 in Japan, we find that the elderly's demand responses are heterogeneous in ways that have not been previously reported. Outpatient services by orthopedic and eye specialties, which will continue to increase in an aging society, are particularly price responsive and account for a large share of the spending increase. Lower cost sharing increases demand for brand-name drugs but not for generics. These high price elasticities may call for different cost-sharing rules for these services. Patient health status also matters: receiving medical services appears more discretionary for the healthy than the sick in the outpatient setting. Finally, we found no evidence that additional medical spending improved short-term health outcomes.

Introduction

Populations are rapidly aging throughout the world¹. The elderly account for the majority of medical spending in both developed and developing countries, and insurers are under increasing pressure to contain medical spending. Health insurance coverage induces consumers to demand more medical services than they do when they are responsible for the full costs of care, and cost sharing plays an important role in mitigating the moral hazard problem. Although an extensive literature exists on the effects of cost sharing on the demand for medical care, surprisingly few studies have examined the elderly population except for a few recent contributions, including that of Shigeoka (2014). Given that the elderly population is undoubtedly an important volume segment that is rapidly expanding, this is an important gap to be filled. This paper aims to fill this gap by examining the Japanese market, in which the coinsurance rate drastically declines from 30% to 10% at age 70.

We have individual-level claims data, enrollment data, and health checkup data that cover the period between 2005 and 2013. Using the claims data, we first examine the impact of the reduced cost sharing at the aggregate level. An advantage of our data is that we have accurate and comprehensive data on medical spending. Surprisingly, few previous studies have reliably measured medical spending data, as we discuss below. We then delve into details and examine whether demand responses are heterogeneous by service type. From the perspective of developing an optimal insurance policy, cost sharing can be different if price elasticity varies by type of medical service. Indeed, health insurance policies, including Medicare, apply different cost-sharing schedules to different services (Cutler and Zeckhauser, 2000). However, for the elderly population, relatively little is known about whether demand responds heterogeneously. Such information will help policy makers implement better cost-sharing policies for the elderly.

Third, we examine whether the healthy or the chronically ill elderly are more price sensitive. This question has attracted the interest of economists, but to our knowledge, the literature is inconclusive. Manning et al. (1987) argued that one would expect sicker patients to be less price sensitive because receiving medical services is less discretionary for these patients and their out-of-pocket expenditures are more likely to reach the stop loss. However, in their study using the data from the RAND Health Insurance Experiment (RAND), these authors found no such differential response. Our longitudinal data allow us to address this question by comparing the same individual's utilization before and after the change in cost sharing.

Finally, using data from annual health checkups, we examine whether the reduced cost sharing and the resulting increases in medical spending lead to better health. Although we can examine only short-term impacts, our analysis nonetheless sheds some light on the important concern that higher coinsurance may negatively affect health outcomes.

We identify the impact of cost sharing by implementing a sharp regression discontinuity (RD) design. For identification purposes, studying the Japanese market has a number of advantages. First, under the universal health insurance coverage in Japan, people of all ages are covered by health insurance and the benefit package is the same for everyone. Thus, there is no issue of selection into insurance. Second, unlike the case of Medicare in the US, in which coverage starts at age 65, only the generosity of the insurance but not access to the insurance changes at age 70 in Japan. Thus we can cleanly identify the impact of cost sharing on the demand for medical services². Third, the timing of retirement and receiving a pension does not coincide with the reduced cost sharing at age 70. This makes identification substantially easier compared with the case of the US, where the timing of Medicare eligibility coincides with retirement at age 65.

We find that at the aggregate level, the reduction of coinsurance from 30% to 10% at age 70 increases medical spending by 11%, or $34 per person-month (approximately $410 per year) on average. The implied price elasticity of demand for medical care is −0.16, which is similar to previous estimates including RAND. Because of the large reduction in cost sharing, the impact of lower cost sharing is substantial: if the reduction in the coinsurance rate were delayed by one year, from age 70 to 71, medical spending would be reduced by $580 million per year, assuming that the elasticity estimates apply to the entire elderly population.

Second, demand responses are heterogeneous, and the impact of reduced cost sharing varies by the type of medical service. In particular, outpatient orthopedic and eye specialties stand out for their high price elasticities of −0.39 and −0.48, respectively. Combined, they account for as much as 37% of the outpatient spending increase owing to the reduced cost sharing. We also find that demand responses differ between brand-name drugs and generics: lower cost sharing increases the demand for brand-name drugs but not for generics. This result contrasts with those of RAND, which did not find that the level of cost sharing differentially affects the share of brand names and generics.

Third, we find that the effect of patient cost sharing on medical spending differs by patient health status. Specifically, reduced cost sharing causes the healthier elderly to utilize more outpatient services, but we did not find the same response for the sicker elderly. This indicates that receiving medical services is less discretionary for the sick, as Manning et al. (1987) hypothesized.

Finally, we found no evidence that the reduced cost sharing improves health outcomes as measured by physical health data.

Our study builds on a large body of literature that examines the effects of cost sharing on medical utilization. A landmark study on this topic is RAND of the 1970s, which found that medical services are moderately price sensitive with a price elasticity of −0.17 (e.g., Manning et al., 1987, Newhouse, 1993). However, RAND did not include people above 62 years old, and thus, the results may not be applicable to the elderly. Additionally, RAND is nearly 40 years old and demand responses may differ now because disease types and medical technologies have substantially changed during the last 40 years.

After RAND, a large number of studies examined the effects of cost sharing on utilization and spending, especially in the areas of prescription drugs (e.g., Goldman et al. 2007; Gaynor et al. 2007), preventative service (e.g., Solanki et al., 2000, Trivedi et al., 2008), and emergency care (e.g., Selby et al., 1996, Wharam et al., 2007). However, as summarized in Swartz (2010) and Baicker and Goldman (2011), these studies share some common limitations. First, as already discussed, very few papers study the elderly population which is the focus of this study. Second, most studies draw data from managed care plans. In these cases, patients self-select into and out of an insurance plan, and dealing with plan selection is inherently difficult. This is not an issue in our study, as everyone is covered by the same insurance. Third, these prior studies examine relatively small changes in copayment (such as $10), and little is known about the impact of a large change in cost sharing such as ours. In fact, a large change in cost sharing would inevitably affect patient insurance plan choice, which in turn would exacerbate the selection problem. Fourth, most studies focus on a few specific services in each study, such as prescription drugs and emergency services. In contrast, we study the impact of cost sharing on nearly all health care services.

Our study is closely related to the analysis by Shigeoka (2014) who exploited the same age discontinuity in Japan. He found that the elasticity of outpatient visits and inpatient admissions are both around −0.2. Additionally, he found that lower coinsurance does not improve health outcomes as measured by mortality and self-reported health but reduces out-of-pocket expenses especially for higher spending patients.

A major difference between his and our study lies in the data used; we use longitudinal claims data, whereas he used survey data. The longitudinal data allow us to examine important history-dependent demand responses, such as whether the healthy or the chronically ill are more responsive to price. Moreover, claims data contain information on all services provided and associated spending, and thus we can thoroughly investigate how demand responds to cost sharing. In contrast, the survey data contain no information on medical spending and cover only a small set of procedures chosen by the government. Additionally, for health outcomes, we use physical exam data, whereas Shigeoka used mortality. Thus, we view our study as complementary to Shigeoka's.

Another closely related study is Chandra et al. (2010). Focusing on the elderly, they studied the “offset effect”, whether changes in cost sharing for prescription drugs and physician visits affect hospital utilization. Although the premise was intriguing, their main focus was to study substitutions between inpatient and outpatient services. In contrast, we study the impact of a large, across-the-board change in cost sharing on medical spending. Chandra et al. (2010) also did not observe actual payments to HMOs and thus had to rely on imputations to derive medical spending.

The rest of the paper proceeds as follows. In Section 2, we provide background information on the institution and the change in cost sharing at age 70. Sections 3 Data, 4 Empirical strategy discuss the data and the empirical models used in the study. Section 5 reports the estimation results. Section 6 provides robustness checks. Section 7 concludes the paper with a discussion of the policy implications of our findings.