The Value of Health Insurance: A Field Experiment in India

Overview

In this study, the investigators use a randomized field experiment in Karnataka, India, to measure the effects of a free inpatient public health insurance plan, Rashtriya Swasthya Bima Yojana (RSBY), on health and poverty. RSBY is India's first large-scale national public health insurance program, and was adopted in 2008 to cover below poverty line (BPL) households. This study examines the impact of expanding RSBY to cover above poverty line (APL) households that are not covered by RSBY or other secondary hospital care insurance plans. The study will examine the impact of a public insurance program on a range of health and economic outcomes to capture the full range of benefits from insurance. The goal of the study is to quantify how the RSBY program benefits health and reduces poverty. Differences in the outcomes between the treatment and control groups over the duration of the project will measure the causal effect of health insurance on health and financial outcomes. The investigators believe this will serve as a platform for the future of development of RSBY, and provide further insight into the effectiveness of health insurance in low-income countries.

The study measures the impact of health insurance on healthcare utilization, health expenditures, nonmedical consumption, and financial behavior of participant households. The study consists of an evaluation involving households in two Indian districts of Karnataka - Mysore and Gulbarga. To measure the impact, households were randomized into one of three treatment groups and a control group that did not receive any intervention. Comparing households' health and financial status across the control and treatment groups will identify the impact of the RSBY coverage. INNOVATIVE ASPECTS OF THE STUDY The study has a number of innovative features that distinguish it from prior health insurance studies. Spillovers from insurance: The study attempts to estimate the impact of financial and health spillovers from insurance. For example, formal health insurance may crowd out informal health insurance. The investigators study such spillovers by randomizing villages to different fractional allocations of households to study arms. Varying fractional allocations to each study group across villages allows them to randomly vary the fraction of households in each village that has insurance. This in turn yields different spillovers across villages that the investigators estimate using knowledge of the fractional allocation in each village. Cognitive impacts of sickness and insurance: The investigators study the cognitive benefits of health insurance and better health. Recent research suggests that economic shocks, perhaps because they cause stress, reduce the cognitive capacity of individuals. This in turn leads to poor economic decision making and, perhaps, poverty traps. The investigators examine whether the same logic applies to health shocks and whether health insurance ameliorates negative cognitive effects associated with illness. Intra-household distribution of benefits: The investigators examine the effect of health insurance on the distribution of resources within the family. By ignoring that health insurance products are often designed for a household rather than for an individual (in their case, with a household-level cap), existing studies are missing potentially important distributional effects. The investigators also test whether health insurance is more likely to benefit women and children when it is provided for free relative to when households decide to purchase it. Such a behavior may arise because household willingness-to-pay for health insurance and for health care may differ depending on the health status of the different members of a household. This study's theoretical framework is based on the collective model of household decision making, which allows the investigators to model how a policy affects the distribution of resources within a household and the well-being of different household members. STUDY DESIGN Study Location: The study will be conducted in 2 districts in the state of Karnataka - Mysore and Gulbarga. The study locations were determined based on the following criteria. First, because the treatment group is to receive an RSBY insurance plan, the investigators selected districts in which the RSBY scheme has already been rolled out. This ensures there is a local insurance company able to provide RSBY coverage and hospitals with the infrastructure required to process payment from RSBY insurance cards. Second, because limited hospital supply can undermine the value of insurance, the investigators selected districts with substantial peri-urban regions. Peri-urban areas are sufficiently developed that they have adequate numbers of secondary and tertiary hospitals and a transportation network to enable access by patients. The study will take place exclusively in peri-urban regions of each district. Third, because study participants are drawn from the middle and upper tercile of the asset distribution, the investigators selected districts in which no more than 50% of the population (26% in Mysore, 48% in Gulbarga) is on the state's BPL or National Rural Employment Guarantee Act (NREGA) list, which comprises primarily individuals from the bottom tercile of the nation's income distribution. Avoiding the poorest districts ensures a reasonable yield of eligible households per district. Finally, the investigators selected a state in which the local government offered substantial cooperation with their study. The Minister of Labor and the chief executive officer of the local RSBY program in Karnataka are enthusiastic about the study. Their hope is to learn about what works in RSBY and what does not so that they can improve the program going forward. Detailed description of Intervention: After the baseline survey, roughly 11,000 households in 2 districts (Mysore and Gulbarga) were randomized to either be in the control (no intervention) group or one of three treatments groups. The goal was to quantify how the RSBY program benefits health and reduces poverty. Differences in the outcomes between the treatment and control groups over time measures the causal effect of health insurance on health and financial outcomes. Experimental assignment process: Subjects were randomized to the four study groups using a two-stage matched randomization process. First, the investigators matched villages based on criteria such as village population and then randomized matched villages to different fractional allocations to each study arm. Second, the investigators matched households within the village based on a range of health and financial measures and then randomized matched households to different arms based on the fractional allocation assigned to that village in the first stage. The first stage was intended to allow the investigators to measure the spillover effects of insurance. The second stage was intended to increase power. Matching households based on characteristics before randomizing increases the statistical power of the study because it increases the statistical similarity of households in a group above and beyond that from randomization itself. It makes the sample even more conservative with respect to power than implied by power calculations. DATA SOURCES The surveys were designed to gather information on not only how proving access to insurance influences households' health and financial standing, but also on how households interact with RSBY. The investigators will measure health-related behaviors, healthcare utilization, as well as household consumption and finances, including the cost of alternative methods of financing available to households. The investigators will use these outcomes to inform future design of RSBY expansion and to influence the design of other health insurance schemes. The study will obtain data from household surveys, special surveys following serious health events, and insurance claims data. Questionnaires containing anthropometrics health measurements such as height, weight, blood pressure, skin fold, lung function will also be administered. Before any data collection, informed consent was (and will be) sought from all participants in future surveys. Household listing exercise: The investigators first identified and recruited households from selected villages to participate in the study. To identify these households, their survey partner, the Centre for Microfinance (CMF), used the "participatory rural appraisal" method to conduct a house-listing exercise (a social map) with residents of the village. This map identified household names, locations, and assets. Households in both treatment and control groups were then selected on the basis of two criteria which are discussed in the 'Eligibility Criteria' section. Household surveys: Several surveys will be conducted on each household, including a Baseline Survey, Annual Survey, Ethnographic Survey, and Post Health Event Survey. Each survey contains a common set of core questions on employment status, income, health utilization (inpatient and outpatient), health expenditures (inpatient and outpatient), health status (symptoms and activities of daily living (ADLs)), nonmedical consumption, productive and nonproductive assets, borrowing and savings, and other investments, including education. In addition, the baseline survey contained questions on household demographics and both the baseline and annual surveys will ask about knowledge of and attitude toward the healthcare system and facilities. Appropriate questions have been drawn from several Indian surveys (e.g., the NSS socioeconomic and consumption surveys and the National Family Health and District Level Household surveys), along with prior international health insurance experiments. Each of these questionnaires has been previously pilot tested. The Post Health Event Survey is an innovative, high-frequency survey methodology designed to capture hospitalization data more efficiently. An incentive of Rs. 50 cash will be provided to encourage households to cooperate with each survey. Administrative claims data from RSBY servers: Insofar as is possible, the investigators will access data on claims filed by households and paid by the insurance company, procedures performed, and date and location of those procedures. These data are available only for households in the treatment group. The ratio of claims filed to claims paid is a measure of the effective "dosage" of insurance provided to the treatment group. POWER CALCULATIONS The investigators powered the existing study to ensure that they could detect a 0.1 standard deviation (25%) change in the primary outcome, hospitalization, and a 1/3 standard deviation (26.5%) change in secondary outcome, financial status as proxied by a composite measure of asset levels drawn from the Indian National Sample Survey (NSS) survey. The investigators also padded the sample to allow for 10% attrition. The power calculations required each group have roughly 2,250 households. The large sample size requirement is driven primarily by the rarity of the hospitalization, roughly 2.5 per 100 persons per year in Karnataka. Because the investigators doubled the size of the first treatment group to account for its centrality in their analysis, the sample size is 11,250 households (5 x 2,250), roughly 56,000 people, drawn equally from Gulbarga and Mysore districts. The investigators believe that their sample makes the study the second largest health insurance experiment ever conducted, the largest being the Seguro Popular study in 2009. DATA MANAGEMENT The study design will generate a longitudinal dataset of treatment and control group households. There will be at least two waves of the data - a baseline and a follow-up annual survey. In addition, those households that experience at least one serious health event will be captured in a separate dataset from the Post Health Event Survey. This longitudinal framework will allow the investigators to measure the effect of RSBY on the outcomes mentioned in specific aims. Randomization of households to the treatment and control groups eliminates the risk of household-specific confounding effects on outcomes. The use of a baseline survey minimizes the risk of time varying confounders that are common across households, and improves precision. Regression and statistical analysis will be used to analyze the data. Data will be stored in encrypted files that only key research team members have access to. Personally identifying information will be delinked completely from the dataset before analysis.