Price setting in the Brazilian private health insurance sector

Abstract

Brazil’s private health insurance market is the second largest in the world, behind only the United States, making it a valuable source of real-world evidence. This paper documents how physicians' inpatient reimbursement fees vary in the country and explores the relationship between these fees and the market share of health providers and health insurance companies. We implement a fixed-effects panel regression and take advantage of an unprecedented database that contains national administrative records of inpatient procedures paid by health insurance companies in 2016. We find a positive correlation between reimbursement for ICU procedures and provider market share. Conversely, we observe a negative correlation with insurers' market share. Additionally, we document substantial variation in procedure prices, both across and within Brazilian states, and observe that more competitive markets in Brazil tend to have higher population and GDP levels. Overall, our research enhances our understanding of the price setting dynamics of physician reimbursement fees in the context of a developing country. The insights gained from this study can assist policymakers in formulating appropriate regulations to ensure appropriate access to healthcare services.

Appendices

Appendix

Data details

The National Health Agency established TISS (Troca de Informações na Saúde Suplementar) as a mandatory standard for the electronic exchange of healthcare data and expenditures between private health insurance companies and health providers. In this study, we used a confidential version of TISS received directly from NHA in 2018. It contains records of inpatient procedures delivered and financed by private health plans in Brazil, from July 2015 to December 2016.⁹¹⁰ More specifically, the data includes information related to the procedures performed (e.g. date, quantity, the amount paid to the provider), the patient (age, gender, and fake id), the insurer (company’ size and a fake id) and the provider (municipality and a fake id).

Even though sending this information to NHA is mandatory, we were concerned about TISS representativeness since we obtained its identified version for the first 18 months of existing records. We decided to exclude 2015 from the study and focus on 2016, the period for which we knew that the amount informed at TISS represented, on average, 64.5% of all private healthcare expenditures in the year (32).

Our database started with 6,932,687 observations or 55,535,644 procedures performed in 2016. We kept only the procedures related to hospital medical visits: daily ICU consultant ward round per patient (TISS code 10104011); 12-h, on-site, ICU intensivist shift per patient (TISS code 10104020); hospital consultations (TISS code 10102019); newborn appointment in baby nursery (TISS code 10103015); and newborn appointment in the delivery room (TISS code 10103023). This amounted to 8,109,493 procedures¹¹ in total.

Because TISS is an administrative database, imputation errors are present and must be addressed. To deal with these measurement errors, we trimmed the bottom and top 2.5% price values per state and procedure type.¹² After the removal of outliers, our final sample comprised 7,632,240 medical visits distributed among 723,503 users, 9180 providers, 499 insurance companies, and 1234 municipalities.

The price analysis focuses on the two types of ICU physician fees (TISS codes 10,104,011 and 10,104,020) for three reasons. First, the only costs related to these events are medical fees; therefore, there is no need to identify any additional items related to these reimbursement codes. In some circumstances, health providers may practice lower fees because they compensate for their receipts augmenting the quantity or prices of other charges (18). Second, the bargaining processes for ICU reimbursements are usually mediated by hospitals, and patients are not able to choose their preferred physician. Hence, we expect bilateral bargaining to be better assessed. Third, ICU fees do not vary by the type of room covered in the health plan (private vs. shared) or by time and weekdays of care, something that happens with the other procedures. In this manner, we assume that that the provider-insurer bargain is the main source of price variation. Table 3 displays, for each type of hospital medical consultation, the total quantity produced, and the total amount paid by health insurance companies to providers in 2016. In quantity terms, ICU visits represented 34% of all medical consultations. As for the value paid by health insurance companies, it amounts to 44%. When we compare the two ICU procedures, we see that the intensivist fee is the most representative, accounting for nearly 86% of all ICU medical reimbursement.

Table 3 Quantity and total reimbursement fees (R$) for each type of hospital medical consultation

It is important to note that a provider in our dataset may be a physician or an association of physicians. Unfortunately, the identified version of TISS dataset made available by NHA does not detail information about the type of provider (corporate entity or individual), health insurance contract characteristics and neither allows identification of the hospital where the inpatient care occurred. Also, as TISS displays information regarding utilization, only users that were hospitalized, and not all beneficiaries of health plans, are inserted in the analysis.

Besides the TISS, other publicly available datasets were also used to obtain socioeconomic information at the local level. Data for private health insurance coverage for each municipality were obtained from the NHA website (24); population projections for each Brazilian municipality and the per capita Gross Domestic Product were obtained from the Instituto Brasileiro de Geografia e Estatística (35–36).

Empirical strategy and results

The following section explains how the market shares and HHI were defined and details the regression model used to produce Exhibits 3 and 4 in the main text, together with their results.

Relevant market, market share and HHI

One challenge of antitrust analysis is the definition of the relevant market. The product dimension is determined by the procedure itself, so our defiance is the definition of the geographic market. Since we are investigating only medical reimbursement codes during hospitalizations, it is reasonable to consider that the geographic market for these procedures is local and determined by doctors' availability to displace to deliver inpatient care. In our database, the only available information regarding the geographic market is the provider's municipality. In this context, one ideal definition of a geographic market should consider a maximum radius of displacement and then include all municipalities inside this area (9, 11). Some difficulties arise with this definition, specifically in Brazil. Firstly, it is hard to uniquely define a parameter for the radius definition since Brazil presents a large heterogeneity in terms of population density and urbanization. Secondly, some municipalities could not be totally inside the area determined by the radius and might belong to more than one market.

In this paper, we considered the geopolitical boundaries of Brazilian municipalities as the geographic definition of the relevant market. Since we are analyzing only ICU procedures, our sample encompasses only municipalities that have hospitals with ICU units and private users. Because of that, although our geographic markets are municipalities, we are working with larger and more populated areas. This definition also accords well with the literature, especially for the United States, where many papers consider geopolitical boundaries to define the geographical market (10), (11), (12), (13), (14), (15), (20), (21). Table 4 displays the characteristics of the relevant markets for both the Southeast region (considered in the regression models) and the entire country.

Table 4 Size of relevant markets, 2016

The market shares are computed for both providers and health insurance companies in each geographical market. It considers their participation in the total number of medical procedures in each municipality per year. The HHI was estimated for both insurance companies and provider markets. According to the Brazilian Administrative Council for Economic Defense (CADE) and the U.S. Department of Justice, an HHI of less than 1500 represents an industry with low market concentration; an HHI ranging between 1500 and 2500 represents moderate concentration; and HHI values of more than 2500 represent a highly concentrated industry.

Regression model and results

We estimated linear regression models with fixed effects to analyze the association between procedure prices and market competition. We considered the 2016 monthly data for the Brazilian Southeast region, which comprises the majority of private health insurance contracts in Brazil. Our variables of interest are the market share of both providers and insurance companies. Three main specifications were estimated for each ICU procedure, with the dependent variable being the natural log of the reimbursement fee.

In all three models, we included month and market fixed effects. Month fixed effects are important in accounting for price adjustments between health insurance and providers that only occur once a year. Market fixed effects were included to account for socioeconomic heterogeneity and the differences in the management of health care services in SUS that may affect opportunity costs for private health providers. The inclusion of market fixed effects also captures other crucial market characteristics, including both the providers’ and insurers’ HHI. Our specifications cluster standard errors at the municipality level to account for the heteroscedastic and serially correlated errors.

In the first specification, or the provider’s model, we analyzed the relationship between the provider's market share and the ICU medical reimbursement fees, controlling for insurance companies fixed effects. The inclusion of health insurance fixed effects allows us to capture the variation in procedure prices that arises from different providers, each with different market shares, bargaining with the same health insurance company. This means that health insurance companies’ time-invariant characteristics, such as size, market share, and overall quality were accounted for in the model. Specifically, we estimated the following regression:

$$Ln\left({P}_{ihmt}\right)=\alpha +\beta {S}_{im}+{\tau }_{h}+{\lambda }_{m}+{\phi }_{t}+{\varepsilon }_{ihm}$$

$Ln\left({P}_{ihmt}\right)$: reimbursement fee received by provider i from health insurance company h in market m during month t; ${S}_{im}$: market share of provider i in market m; ${\tau }_{h}$: health insurance company fixed effects; ${\lambda }_{m}$: market fixed effect;${\phi }_{t}$: month fixed effect.

In the second specification, or the insurers' model, we focused on the association between insurance companies' market shares and the ICU medical reimbursement fees, controlling for providers' fixed effects. The variable of interest captures the variation in procedure prices due to different insurance companies bargaining with the same provider. Therefore, provider characteristics stable throughout the year, including overall quality, are controlled for in the model. More specifically, the regression estimated was:

$$Ln\left({P}_{ihmt}\right)=\alpha +\beta {S}_{hm}+{\mu }_{i}+{\lambda }_{m}+{\phi }_{t}+{\varepsilon }_{ihm}$$

$Ln\left({P}_{ihmt}\right)$: reimbursement fee received by provider i from health insurance company h in market m during month t; ${S}_{hm}$: market share of health insurance company h in market m; ${\mu }_{i}$: provider fixed effects; ${\lambda }_{m}$: market fixed effect; ${\phi }_{t}$: month fixed effect.

The advantage of estimating separate models (the first and second specifications) is the inclusion of provider fixed effects (in the insurers model) and insurer fixed effects (in the providers model) in the regressions. The fixed effects of both insurers and providers are crucial to our analysis. They are used to control for differences in time-invariant non observable characteristics that could affect both shares and prices, including the quality of these players. The use of these fixed effects significantly reduces the potential bias of omitted variables in our results.

Additionally, the bargaining process between health agents may vary depending on the size of the counterpart market. To account for this heterogeneity, we conducted an additional analysis by re-estimating both the provider and insurer models for different subsamples. These subsamples were based on three levels of concentration indices (HHI) that corresponded to the counterpart player in the bargaining process. This approach allowed us to gain further insights into the dynamics of the bargaining process in different market contexts.

In the final specification, we adapted the model proposed by Halbersma et al. (2011) to our specific context and available data. We aimed to jointly estimate the relationship of health insurance and provider market shares with prices. We directly regressed reimbursement fees on the market shares of both providers and health insurance companies. Unfortunately, due to data variability constraints, we were unable to include all fixed effects in this model. Instead, we included the provider fixed effect, which captures the many provider-level information considered in Halbersma et al. (2011). This integrated or combined model allows us to analyze the association between market shares and prices in a more comprehensive manner. The model specification is described below:

$$Ln\left({P}_{ihmt}\right)=\alpha +{\beta }_{1}{S}_{im}+ {\beta }_{2}{S}_{hm}+{\mu }_{i}+{\lambda }_{m}+{\phi }_{t}+{\varepsilon }_{ihm}$$

$Ln\left({P}_{ihmt}\right)$: reimbursement fee received by provider i from health insurance company h in market m during month t; ${S}_{im}$: market share of provider i in market m; ${S}_{hm}$: market share of health insurance company h in market m; ${\mu }_{i}$: provider fixed effects; ${\lambda }_{m}$: market fixed effect; ${\phi }_{t}$: month fixed effect.

Table 5 displays the regression results shown in Exhibit 3 in the paper, together with their F-statistic and R-squared.

Table 5 Results: regression of reimbursement fees on market shares