Social Science & Medicine 124 (2015) 115e120
Contents lists available at ScienceDirect
Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed
Cost sharing and hospitalizations for ambulatory care sensitive conditions Alejandro Arrieta a, *, Ariadna García-Prado b a b
Florida International University, Department of Health Policy and Management, USA Universidad Pública de Navarra, Departamento de Economía, Spain
a r t i c l e i n f o
a b s t r a c t
Article history: Available online 13 November 2014
During the last decade, Chile's private health sector has experienced a dramatic increase in hospitalization rates, growing at four times the rate of ambulatory visits. Such evolution has raised concern among policy-makers. We studied the effect of ambulatory and hospital co-insurance rates on hospitalizations for ambulatory care sensitive conditions (ACSC) among individuals with private insurance in Chile. We used a large administrative dataset of private insurance claims for the period 2007e8 and a final sample of 2,792,662 individuals to estimate a structural model of two equations. The first equation was for ambulatory visits and the second for future hospitalizations for ACSC. We estimated the system by Two Stage Least Squares (2SLS) corrected by heteroskedasticity via Generalized Method of Moments (GMM) estimation. Results show that increased ambulatory visits reduced the probability of future hospitalizations, and increased ambulatory co-insurance decreased ambulatory visits for the adult population (19e65 years-old). Both findings indicate the need to reduce ambulatory co-insurance as a way to reduce hospitalizations for ACSC. Results also showed that increasing hospital co-insurance does have a statistically significant reduction on hospitalizations for the adult group, while it does not seem to have a significant effect on hospitalizations for the children (1e18 years-old) group. This paper's contribution is twofold: first, it shows how the level of co-insurance can be a determinant in avoiding unnecessary hospitalizations for certain conditions; second, it highlights the relevance for policy-making of using data on ACSC to improve the efficiency of health systems by promoting ambulatory care as well as population health. © 2014 Elsevier Ltd. All rights reserved.
Keywords: Chile Avoidable hospitalizations Ambulatory care sensitive conditions Co-insurance
1. Introduction It is well known that lack of timely, appropriate ambulatory care may lead to complications that require hospitalization, creating unnecessary costs in economic and human terms. There is considerable international evidence which shows that better ambulatory care can decrease the need for hospitalization (Starfield, 1991; Fleming, 1995; Caminal et al., 2004; Macinko et al., 2010). This is especially true in the case of ambulatory care sensitive conditions (ACSC). These conditions can be easily managed with timely and effective outpatient care (Rizza et al., 2007). Rates of avoidable hospitalizations have been used as a tracer to assess access, quality, and performance of the primary care delivery system (Bindman et al., 1995; Ansari et al., 2006). Most research on ACSC
* Corresponding author. Florida International University, Department of Health Policy and Management, 11200 S.W. 8th St., Miami, FL 33199, USA. E-mail address: alejarri@fiu.edu (A. Arrieta). http://dx.doi.org/10.1016/j.socscimed.2014.11.026 0277-9536/© 2014 Elsevier Ltd. All rights reserved.
has focused on determining the socioeconomic and medical conditions associated with hospitalized ACSC (Prospective Studies Collaboration, et al., 2007; Roos et al., 2005; Bliziotis et al., 2012). Such studies have found that potential avoidable hospitalizations vary by socioeconomic status. Middle and lower income population are less likely to receive preventive services, more likely to experience delays in healthcare, and less likely to have a regular source of care. The access to health insurance is another factor that can affect rates of avoidable hospitalizations, a relationship which has been studied in the U.S. through the Medicaid program (Kaestner et al., 2001; Bermudez and Baker, 2005; Dafny and Gruber, 2005). Kaestner et al. (2001) and Dafny and Gruber (2005) estimated the impact of Medicaid eligibility expansions on child hospitalizations. While Kaester et al. found that Medicaid eligibility expansions moderately improved the health of low-income children and reduced hospitalizations, Dafny and Gruber showed that the number of hospitalizations increased, but there was a much smaller increase in avoidable hospitalizations. Aizer (2007) estimated the
116
A. Arrieta, A. García-Prado / Social Science & Medicine 124 (2015) 115e120
impact of Medicaid take-up rather than insurance eligibility expansion. She estimated the impact of early Medicaid take-up on access to primary care and avoidable hospitalizations, finding that a 15% increase in Medicaid enrollment would lead to a 2.7% decline in avoidable hospitalizations. Little is known, however, about the impact of insurance cost-sharing (through co-insurance or deductibles) on hospitalizations for ACSC. In this paper, we studied the effect of ambulatory and hospital co-insurance rates on ACSC among individuals with private insurance in Chile. During the last decade, Chile's private health sector has experienced a dramatic increase in its hospitalization rates, growing at four times the rate of ambulatory visits (see Fig. 1). Such evolution has raised concern among policy-makers interested in promoting more preventive services and higher use of ambulatory care. The increased prevalence of chronic diseases has also raised alarm. A study made in 2007 shows that 84% of the total diseases in Chile were non-communicable diseases (Ministerio de Salud, 2007). The 2003 National Health Survey showed that only a small fraction of those affected by a chronic disease had their condition n et al., 2010). In this context, co-insurance can under control (Bitra be a valuable tool for dealing with cost-escalating problems in the health system while promoting more ambulatory visits and preventive services and less ACSC. Cost sharing is a common feature of insurance contracts. It is useful to reduce patient moral hazard, and, therefore, overconsumption of medical care. By raising co-insurance rates, insurers can reduce unnecessary care and control costs. However, high co-insurance rates may also produce losses due to reduced financial risk protection. The search for an optimal co-insurance rate that balances this trade-off has been broadly studied both theoretically and empirically (Newhouse, 2006; Ellis and Manning, 2007; Pauly and Blavin, 2008). We analyzed cost sharing in the context of an inter-temporal relationship between primary care visits and avoidable hospitalizations. Such relationship presents an additional feature that increases the losses of an un-optimally high co-insurance rate: a high ambulatory co-insurance rate could reduce consumption of effective care, and, as a result, lead to a potential increase of hospitalizations for ACSC in the future. What is the most appropriate co-insurance that would help to reduce hospital admissions for ACSC? What are the effects of different coinsurance levels on hospital and ambulatory visits for ACSC? Our response to these questions is a structural model that describes the inter-temporal correlation between ambulatory visits and hospital admissions for ACSC. We used a large administrative dataset of private insurance claims in Chile to examine the effects of ambulatory and hospital co-insurance rates on both ambulatory and hospital visits. Since hospitalizations for ACSC are avoidable through timely outpatient treatment, policy-makers must ensure that co-insurance for hospital and ambulatory care promotes adequate primary care coverage.
Fig. 1. Evolution of ambulatory versus hospital visits in the Chilean private sector.
Our results show that increasing ambulatory visits reduced the probability of future hospitalizations. For adults, we also found that increasing ambulatory co-insurance decreased ambulatory visits. Both findings indicate the need to reduce ambulatory co-insurance as a way to reduce hospitalizations for ACSC. This paper highlights the need to introduce co-insurance rates that are differentiated according to type of diseases (ACSC) so as to promote more ambulatory care and reduce health costs. It also focuses attention on an area that has been barely researched in less developed countries. 2. The data Data was provided by the health regulator, the Superintendencia de Salud, which validates and consolidates information provided on a quarterly basis1 by all the private insurers (ISAPREs or Instituciones de Salud Provisional) operating in Chile. Health insurance in Chile is dual; i.e., employed individuals must purchase insurance for a minimum of 7% of their taxable income up to a specified threshold in order to enroll in the public insurer (Public Health Fund, PHF) or purchase a health plan from a private insurer (ISAPRE). A total of 2.8 million individuals, or 16.8% of the population, were covered by a contract in one of the 14 ISAPREs that operated in the market by the end of 2007. Our data cover the period from January 1 to December 31, 2007. The administrative data used in this study was exempt from ethical review. All claim records were de-identified by the owner of the data, the Superintendencia de Salud. Information provided by the ISAPRES includes characteristics of plan holders, as well as all the beneficiaries, including sex, age, income, and earnings. We also have extensive information on all claims made by these individuals, including ambulatory visits, recording diagnoses using the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10), codes for hospitalizations, and plan characteristics including co-insurance rates. A health plan from ISAPRES typically specifies co-insurance rates for both outpatient and hospital services, together with caps on coverage by unit of service. Deductibles are not used in insurance plans. Data collected for 2007 include a total of 3,004,102 observations on the insured and their beneficiaries. From this dataset, we excluded two groups e infants (up to 1 year-old) and people over age 65 e for the following reasons: first, both clearly show different patterns of disease and behavior compared to other groups; second, the population over 65 years old represents only about 5% of the total population covered by the ISAPRES, third, prices for the seniors increase more than for others, so that many end up moving to FONASA after retirement. As a result of these exclusions, the sample was reduced to a total of 2,792,662 individuals. For each individual, we constructed indicators for ambulatory visits and subsequent hospital visits within a 30-day window. To check the robustness of the results, we also considered 60- and 90-day windows. We classified hospitalizations as ACSC, using ICD-10 codes, and following the definitions of Alfradique et al. (2009) for Brazil. The list of ACSC inclusions is reported in the Appendix. We chose the Brazilian ACSC list because Brazil is the only country in Latin America that has made a systematic effort to adapt the ACSC lists from the US, Canada and Spain to its own circumstances. Like most Latin American countries, Chile does not have its own ACSC list, and so we decided to use the Brazilian ACSC list as the closest available approximation for a Latin American context. However, it is important to note that there is no international consensus on how best to compose ACSC lists and that several alternative lists are used among and within different countries (Ansari et al., 2006).
1 Starting in 2008, ISAPRES are required to provide information on a monthly basis.
A. Arrieta, A. García-Prado / Social Science & Medicine 124 (2015) 115e120
Table 1 provides descriptive statistics by age group of all variables used in the study. Co-insurance, on average, is higher for ambulatory visits than for hospital visits, according to age group. While ambulatory coinsurance accounts for about 28% of the total expenditures in healthcare made by both age groups, hospital co-insurance accounts for 14% and 17% for children and adults, respectively. The incidence of hospitalizations for ACSC is not high (compared to other countries in the region), but accounts for about 10 and 8 hospitalizations for every 10,000 insured children and adults, respectively. Finally, the percentage of ambulatory visits for children and adults are 75.5% and 71.6%, respectively. 3. The empirical strategy Since the main goal of this study is to assess the impact of coinsurance on hospitalizations for ACSC, a simple estimation strategy is to model the probability of being hospitalized as a function of co-insurance and ambulatory visits. As described in equation (1), if we consider two periods, the probability that patient i is hospitalized for an ACSC in period 1 (hi1) depends on the hospital coinsurance rate in patient i's plan (chi ), a variable indicating if patient i had an ambulatory visit in the past (ai0), and an a contemporaneous error term. We expected that for ACSC, a patient who had an ambulatory visit would have a lower probability of being hospitalized (q < 0). However, if a patient's health status was not controlled, an estimation strategy such as the one in equation (1) could result in biased estimates. First, health status may have driven both ambulatory and hospital visits. Therefore, if health status is not fully controlled, it would bias the estimate of q. Second, the insurance plan selection, and consequently the observed coinsurance rate, could also depend on health status if individuals are free to choose among plans or if insurance companies select customers on the basis of health conditions. In this case, if health status is not fully controlled, it would bias the estimate of gh.
Prðhi1 ¼ 1Þ ¼ gh chi þ qai0 þ εhi1
Table 1 Descriptive statistics by age group. Age group
Hospitalizations for ACSC (fraction)% Co-insurance (ambulatory)% Co-insurance (hospitalization)% Age (in years) Gender (female ¼ 1)% Income (in thousands of Chilean pesos) Sample size (individuals)
was based on co-insurance for ambulatory and hospital visits, and also on health status in each period. Health status evolves stochastically between period 0 and period 1, and we assumed that having an ambulatory visit could affect future health status. This is easily justifiable, since we focused on ambulatory care sensitive conditions, which are those that can improve health and avoid future hospitalizations. The dynamic of health status is as follows:
~si1 ¼ ~si0 þ ~ qai0 þ mi1
(2)
where ~si1 is the (unobserved) health status of patient i in period 1, ai0 is a binary variable that indicates if patient i visited an ambulatory setting in period 0, and mi1 is an independent error term. We modeled a system of two equations, one for ambulatory visits and the second for hospital visits. For identification purposes we made the reasonable assumption that hospital visits depended only on hospital co-insurance, while ambulatory visits depended on both ambulatory and hospital co-insurances. We believed this assumption was justified because patients may avoid ambulatory care if hospital visits are relatively cheap. The system of simultaneous equations is represented by equations (3)e(4).
Prðai0 ¼ 1Þ ¼ g0a cai þ g0h chi þ aa si þ ba~si0 þ εai0
(3)
Prðhi1 ¼ 1Þ ¼ g1h chi þ ah si þ bh~si1 þ εhi1
(4)
A binary variable, hi1, indicates that patient i visits the hospital for chronic ACSC in period 1. Co-insurance for ambulatory visits (cai ) and hospital visits (chi ) are assumed to be time independent; i.e., defined when the individual selected an insurance plan. si are patients' observable characteristics such as age, gender, income, etc. Many other patient's characteristics remain unobservable, and because they affect both, ambulatory and hospital visits, the error terms become correlated. Replacing (2) in (4), we can re-write the simultaneous equation model (3e4) as:
(1)
Although we controlled for patients' characteristics, a significant component of patients' health status remained unobservable, thus, creating potentially omitted variable bias. To overcome the problem of simultaneous dependency between ambulatory and hospital visits (endogeneity), we considered a model of two periods where individuals decided to visit an ambulatory setting in the first period and a hospital setting in the second period. The patient's decision
Ambulatory visits (fraction)%
117
1e18 years old
19e65 years old
75.453 (43.036) 0.098 (3.132) 28.006 (17.479) 14.479 (22.2) 9.684 (5.081) 51.034 (50.005) 69.426 (26.998)
71.634 (45.077) 0.078 (2.789) 28.778 (19.648) 16.508 (25.648) 38.12 (12.125) 51.929 (49.963) 63.84 (29.545)
875,735
1,916,927
Averages and standard deviations (in parenthesis). It only considers hospitalizations for ACSC that occurred after the first 30 days of any ambulatory visit.
Prðai0 ¼ 1Þ ¼ g0a cai þ g0h chi þ aa si þ εai0
(5)
Prðhi1 ¼ 1Þ ¼ qahi0 þ g1h chi þ ah si þ εhi1
(6)
where q≡bh ~ q, εai0 ≡ba~si0 þ εai0 , and εhi1 ≡bh~si1 þ bh mi1 þ εhi1 Although this system corrects for the simultaneous ambulatory and hospital visits, co-insurance rates could still be correlated with error terms in equations (5)e(6). If insurance selection is endogenous and driven by unobserved health status, the impact of coinsurance on ambulatory and hospital visits would be biased. In the selection of an insurance plan, two factors may make insurance choice dependent on health status. First, patients may select plans based on their health status (adverse selection). In that case, sickly patients may choose better plans, with lower co-insurance because more care visits are expected. This may downward bias the parameter estimates for co-insurance on hospital and ambulatory visits. However, a second possibility is that insurance companies can discriminate based on health status (cream skimming). In that case the insurance pool will be formed by healthy patients, which use fewer care visits. In Chile there is evidence of cream skimming by ISAPRES (Sapelli, 2004). This may upward bias the parameter estimates for co-insurance on hospital and ambulatory visits. To reduce these problems, we include two instrumental variables, one for the co-insurance rate of ambulatory visits and the other for hospital visits. Our instruments are the average co-insurance rates corresponding to all plans available in the market at a price equivalent to the member's minimum legal insurance premium. By
118
A. Arrieta, A. García-Prado / Social Science & Medicine 124 (2015) 115e120
law, Chileans in the ISAPRE system have to purchase insurance whose minimum premium is equivalent to 7% of personal income up to a maximum limit of 4.2 UF (inflation-adjusted legal unit of accounting commonly used in Chile). Chileans can opt to pay higher premiums in order to obtain more generous plans. The plans priced at the minimum legal insurance premium can be considered as default plans, as they represent the set of plans that individuals are mandated to obtain without extra payment. To obtain our instruments, we first calculated the minimum legal insurance premium for each member based on reported personal income. Then, we identified the default plans in the market as those with prices close to the minimum premium. Finally, we obtained the average co-insurance rates for the identified default plans. Because default plans are not chosen but set by law, and the average co-insurance rates are not tied to a particular individual, we expect our instruments to not be directly correlated to health status. However, they could be indirectly correlated to health status through income. Because we explicitly control for personal income, our instrument remains valid (Angrist and Pischke, 2009). Notice that although the system of equations (5)e(6) is recursive, identification cannot rest on recursivity because both error terms are dependent (through the unobserved health status). For the sake of simplicity, we assumed a linear probability model, which produced a linear system of equations. We found that ambulatory and hospital co-insurance rates are highly correlated in our data, creating potential distortions in equation (5) due to multicollinearity. To avoid this problem we decided to exclude the hospital co-insurance rates from equation (5). We estimated the system by 2SLS corrected by heteroskedasticity via GMM estimation (Greene, 2008).
4. Results Table 2 presents results for hospitalizations by age group. Those results correspond to the two equations of interest, hospital visits and ambulatory visits. The equation of hospital visits estimates the probability of having a hospital visit. This equation confirms the basic hypothesis underneath ACSC studies: ambulatory visits tend to reduce the probability of future hospitalizations. In particular, the probability of having a hospital admission for ACSC after having had one or more ambulatory visits, decreases by 0.43% and 0.72% for both children and adults, respectively. The hospital equation offers a second result: increasing the hospital co-insurance rate does not reduce the probability of having more hospital visits in the case of children. For adults, this effect is statistically negative, suggesting Table 2 Results for ambulatory and hospital visits. Age group
Ambulatory visits Intercept Ambulatory co-insurance Age Gender (male) Income (in millions of Chilean pesos) Hospital visits Intercept Ambulatory visit (1 or more) Hospital co-insurance Age Gender (male) Income (in millions of Chilean pesos)
1e18 years old
19e65 years old
0.13123* 0.02750* 0.01061* 0.01841* 4.29600*
0.98418* 0.00816* 0.00230* 0.12742* 0.56040*
0.00352* 0.00431* 0.00004 0.00015* 0.00004 0.02040*
0.00751* 0.00717* 0.00012* 0.00003* 0.00088* 0.00361
Simultaneous equation linear model estimated by 2SLS. Ambulatory and Hospital visits are binary variables. The heteroskedasticity produced by the linear probabilities are adjusted using GMM. *Significant at 1%, **Significant at 5%.
that 1 percentage point increment in hospital co-insurance rate reduces the probability of a hospital visit by 0.012 percentage points. The ambulatory visits equation estimates the probability of having ambulatory visits. Again, our results differ by age group. For adults, the effect of ambulatory co-insurance rate on ambulatory visits is statistically negative, indicating that a 1 percentage point increment in ambulatory co-insurance rate reduces the probability of an ambulatory visit by 0.82 percentage points. For children, this effect is positive. For adults, a reduction in ambulatory co-insurance rates increases ambulatory visits, and consequently reduces the probability of future hospitalizations. By combining the results from both equations to obtain the overall marginal effect, our findings suggest that a 1 percentage point reduction in ambulatory co-insurance rates reduces the probability of future hospitalizations by 0.006%. The effect is not meaningless considering that hospitalization rates among adults are 0.078% (see Table 1). Although out-of-sample extrapolation is questionable, our results imply that to fully avoid hospitalizations for ACSC (probability of hospital visit equal to zero), co-insurance rates for ambulatory visits should be reduced from the current average of 28.8% to around 17%. Notice that this last result only applies under the linearity assumption of our estimation strategy. Furthermore, other variables, in addition to coinsurance rates should be considered in the analysis to determine the most appropriate co-insurance rates for ambulatory visits. 5. Discussion In this paper we have applied a structural model to a large administrative dataset of private insurance claims in Chile in order to estimate the effects of ambulatory visits on avoidable hospitalizations, as well as the effect of ambulatory and hospital coinsurance rates on both ambulatory visits and hospitalizations for ACSC. Our evidence supports the theory that ambulatory visits can reduce hospitalizations in both adults and children. For the average Chilean adult with private insurance, having one or more ambulatory visits can reduce the probability of a future hospitalization for ACSC by 0.72 percentage points. The same effect in children is estimated as 0.43 percentage points. We also found that cost sharing reduces hospital and ambulatory visits only in the case of adults. In children, altering hospital co-insurance does not seem to have a significant effect on hospitalizations, while the relationship between ambulatory coinsurance and ambulatory visits is unexpectedly positive. We believe the unexpected results for children reflect the lower sensitivity to prices observed in this age group. For instance, results from the RAND Health Insurance Experiment showed low sensitivity to cost sharing and mixed results for older children compared to adults (Leibowitz et al., 1985), suggesting that cost-sharing behavior differs when individuals decide for their children. For adults, the two main findingsdthat lower ambulatory coinsurance increases ambulatory visits and that more ambulatory visits reduces avoidable hospitalizationsdimply that ambulatory co-insurance should be decreased substantially in order to reduce the high rate of hospitalizations in the private sector. How much should ambulatory co-insurance be diminished so as to reduce hospitalizations for ACSC to zero? We show that for the 2007 average hospital co-insurance rate of 16.5%, the ambulatory coinsurance should be 17% for adults, a figure substantially lower than the ambulatory co-insurance of Chile in 2007 (28%). Our study has several limitations that could be addressed in future investigations. First, our analysis of the effect of co-insurance rates on hospital and ambulatory visits rests on the validity of our instrumental variables. As is common in observational studies that assess the effect of health insurance on healthcare demand, finding valid instruments is challenging. Although our proposed
A. Arrieta, A. García-Prado / Social Science & Medicine 124 (2015) 115e120
instruments are theoretically reasonable, we could not test their validity. Nevertheless, a comparison between the model estimated with and without instrumental variables (results not shown) reveals that the instrumental variable estimation produces the expected effects of co-insurance among adults, suggesting a bias correction of the instrumental variable approach. Second, we considered co-insurance rates to define cost-sharing, omitting other characteristics of the plan that, unfortunately, are not available in the database used for this study. Typically, in addition to co-insurance, insurance plans in Chile have associated caps that limit the maximum value to be paid by the ISAPRES. Plans also differ in terms of preferred health providers and reduced copayment for catastrophic illness. However, since co-insurance rates represent the most important component of insurance coverage in Chile, we think that any potential bias is relatively small. Third, we did not include in the analysis the incentives to hospitalize patients that may exist among physicians working for ISAPRES. While decreasing ambulatory co-insurance seems to be a good policy, the Government should analyze the incentives behind the dramatic increase of hospitalizations in the private sector of Chile. We offer a demand side solution, through co-insurance, but future research should also focus on analyzing the effects from the supply side. It seems, for instance, that some uncovered vertical integration between providers and insurers within the ISAPRE system may be pushing hospitalization rates up. This clearly calls for further research and analysis. Finally, we could not perform a more detailed analysis considering different groups of ACSC. Since we only had data for one year, 2007, we would compromise the statistical robustness of the estimation by adding more groups. Unfortunately, data availability was a constraint when this research was carried out so we could not expand the analysis to include other years. Such analysis would have been extremely interesting, particularly because the AUGE law was passed in Chile in 2005. This law prioritized the health services related to 56 medical conditions, most of them chronic; promoted preventive services; and introduced new treatment guidelines and new co-sharing rules, among other measures. In particular, the AUGE law mandated coverage for a set of interventions for children and adults aimed at prevention and early detection of hypertension, types 1 and 2 diabetes, HIV/AIDS, cervical cancer, and other chronic diseases. The new law also established that insurers, both Public Health Funds (PHF) and ISAPRES, must reimburse a specific amount for each guaranteed health intervention, so that the beneficiary's out-of-pocket spending will not exceed a predefined share of household income. In addition, when seeking care for a medical problem under the AUGE reforms, both PHF and ISAPRES beneficiaries could make one of two care choices: (1) select a closed mode, choosing providers from the AUGE network and paying lower or no co-insurance, or (2) obtain care outside of AUGE through a free choice mode, paying higher co-insurance, but receiving faster, higher-quality care, better amenities, and n et al., 2010). AUGE law continued access to a personal doctor (Bitra also promoted the systematic application of treatment guidelines in order to lower the incidence of complications, and, therefore, hospitalizations from inadequate disease management. Having the specific information for certain ACSCs would be essential to understanding the new cost-sharing rules under the AUGE program as well as its effects on particular ACSCs. It would also be interesting to see the impact of the introduction of AUGE program on hospitalization rates versus hospitalization rates under non-AUGE condin et al. (2010) analyzed the impact on hospitalization tions. Bitra rates for six AUGE conditions but it would be worthwhile to expand this analysis to more conditions and over a greater timespan. Chile is in a process of reform and debate about the escalating costs and the provision of quality healthcare, with a focus on the role
119
of ISAPRES. After a decade of discussions about ways to reform ISAPRES failed to lead to any major changes, the new government that took power in March 2014 will finally have to implement reforms. As our results are still valid for the current situation, we hope that they will contribute to the upcoming policy discussion. Other countries in Latin America, e.g., Colombia and Peru, are facing similar debates, dealing with dramatic increases of hospitalization rates. Since many developing countries are facing the epidemiological transition that more developed economies experienced decades ago, and a dramatic increase of chronic conditions (Glassman et al., 2010), the analysis of the interrelation between co-insurance rates and hospitalizations for ACSCs is going to become more and more relevant. The solution offered here, based on introducing a different co-insurance for those diseases that are more easily preventable, could be a useful strategy for many of these countries in their combat against cost escalation and in favor of primary care promotion. The widely reported U.S. experience of Pitney Bowes, who provided copay relief for drugs used to treat asthma and diabetes, has demonstrated that such an approach is feasible and produces both clinical and economical returns (Mahoney, 2005; Fendrick and et al., 2006). Acknowledgment The authors thank the Inter-American Development Bank (IADB) for funding this research (contract number 521318-0004) and providing useful comments. We also thank the comments received by participants at the Seminar “Chronic Diseases, Primary care and Health Systems Performance” organized by the IADB, the University of Bahía and Ministry of Health in Brazil, and those at the Spanish Health Economics Association meeting that took place in Santander, Spain. Ariadna García Prado also acknowledges support from the project ECO2012-36480: “Analisis de Economía Pública y Bienestar” of the Spanish Ministry of Science and Innovation (2013e2015). Appendix Table A List and Definition of Ambulatory Care Sensitive Conditions (ACSC) Group Diagnostic 1 2
3 4 5
6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Vaccine-preventable
ICD-10 (primary condition only)
G00.0, A17.0, A33, A34, A35, A36, A37, A95, B16, B05, B06, B26, A19 A17.1, A17.2, A17.3, A17.4, A17.5, A17.6, Not vaccine-preventable A17.7, A17.8, A17.9, A15, A16, A18, I00, (rheumatic fever, syphilis, I01, I02, A51, A52, A53, B50, B51, B52, tuberculosis) B53, B54, B77 Dehydration and E86, A00, A01, A01, A03, A04, A05, A06, gastroenteritis A07, A08, A09 Iron deficiency anemia D50 Nutritional deficiencies E40, E41, E42, E43, E44, E45, E46, E50, E51, E52, E53, E54, E55, E56, E57, E58, E59, E60, E61, E62, E63, E64 Ear, nose and throat infections H66, J00, J01, J02, J03, J06, J31 Bacterial pneumonia J15.3, J15.4, J15.8, J15.9, J18.1, J13, J14 Asthma J45, J46 Chronic obstructive pulmonary J20, J21, J40, J41, J42, J43, J44, J47 disease Hypertension I10, I11 Angina pectoris I20 Congestive heart failure I50, J81 Cerebrovascular disease I63, I64, I65, I66, I67, I69, G45, G46 Diabetes E10, E11, E12, E13, E14 Convulsions and epilepsy G40, G41 Pyelonephritis N39.0, N10, N11, N12, N30, N34 Skin and subcutaneous tissue A46, L01, L02, L03, L04, L08 infection Pelvic inflammatory disease N70, N71, N72, N73, N75, N76 Perforated/bleeding ulcer K92.0, K92.1, K92.2, K25, K26, K27, K28 Prenatal and natal P35.0, O23, A50 complications
120
A. Arrieta, A. García-Prado / Social Science & Medicine 124 (2015) 115e120
References Aizer, A., 2007. Public health insurance, program take-up, and child health. Rev. Econ. Stat. 89 (3), 400e415. Alfradique, M.E., et al., 2009. Ambulatory care sensitive hospitalizations: elaboration of Brazilian list as a tool for measuring health system performance (Project ICSAPeBrazil). Cadernos de Saúde Pública 25 (6), 1337e1349. Angrist, J.D., Pischke, J.r.-S., 2009. Mostly Harmless Econometrics: an Empiricist's Companion. Princeton University Press, Princeton, p. 373 xiii. Ansari, Z., Laditka, J.N., Laditka, S.B., 2006. Access to health care and hospitalization for ambulatory care sensitive conditions. Med. Care Res. Rev. 63 (6), 719e741. Bermudez, D., Baker, L., 2005. The relationship between SCHIP enrollment and hospitalizations for ambulatory care sensitive conditions in California. J. Health Care Poor Underserved 16 (1), 96e110. Bindman, A.B., et al., 1995. Preventable hospitalizations and access to health care. JAMA 274 (4), 305e311. n, R., Escobar, L., Gassibe, P., 2010. After Chile's health reform: increase in Bitra coverage and access, decline in hospitalization and death rates. Health Aff. 29 (12), 2161e2170. Bliziotis, I.A., Destounis, A., Stergiou, G.S., 2012. Home versus ambulatory and office blood pressure in predicting target organ damage in hypertension: a systematic review and meta-analysis. J. Hypertens. 30 (7), 1289e1299. Caminal, J., et al., 2004. The role of primary care in preventing ambulatory care sensitive conditions. Eur. J. Public Health 14 (3), 246e251. Dafny, L., Gruber, J., 2005. Public insurance and child hospitalizations: access and efficiency effects. J. Public Econ. 89 (1), 109e129. Ellis, R.P., Manning, W.G., 2007. Optimal health insurance for prevention and treatment. J. Health Econ. 26 (6), 1128e1150. Fendrick, A.M., et al., 2006. Value-based insurance design: a “clinically sensitive” approach to preserve quality of care and contain costs. Am. J. Manag. Care 12 (1), 18. Fleming, S.T., 1995. Primary care, avoidable hospitalization, and outcomes of care: a literature review and methodological approach. Med. Care Res. Rev. MCRR 52 (1), 88e108.
Glassman, A., et al., 2010. Confronting the chronic disease burden in Latin America and the Caribbean. Health Aff. 29 (12), 2142e2148. Greene, W.H., 2008. Econometric Analysis, sixth ed. Prentice Hall, Upper Saddle River, N.J., p. 1177 xxxvii. Kaestner, R., Joyce, T., Racine, A., 2001. Medicaid eligibility and the incidence of ambulatory care sensitive hospitalizations for children. Soc. Sci. Med. 52 (2), 305e313. Leibowitz, A., et al., 1985. Effect of cost-sharing on the use of medical services by children: interim results from a randomized controlled trial. Pediatrics 75 (5), 942e951. Macinko, J., et al., 2010. Major expansion of primary care in Brazil linked to decline in unnecessary hospitalization. Health Aff. Millwood 29 (12), 2149e2160. Mahoney, J.J., 2005. Reducing patient drug acquisition costs can lower diabetes health claims. Am. J. Manag. Care 11 (5 Suppl. l), S170eS176. Ministerio de Salud, Estudio de Carga de Enfermedad y Carga Atribuible: Chile 2007, 2008, Estudios AUGE; Santiago de Chile, 101. Newhouse, J.P., 2006. Reconsidering the moral hazard-risk avoidance tradeoff. J. Health Econ. 25 (5), 1005e1014. Pauly, M.V., Blavin, F.E., 2008. Moral hazard in insurance, value-based cost sharing, and the benefits of blissful ignorance. J. health Econ. 27 (6), 1407e1417. Prospective Studies Collaboration, Lewington, S., Whitlock, G., Clarke, R., Sherliker, P., Emberson, J., Halsey, J., Qizilbash, N., Peto, R., Collins, R., 2007. Blood cholesterol and vascular mortality by age, sex, and blood pressure: a meta-analysis of individual data from 61 prospective studies with 55,000 vascular deaths. Lancet 370 (9602), 1829e1839. Rizza, P., et al., 2007. Preventable hospitalization and access to primary health care in an area of Southern Italy. BMC Health Serv. Res. 7, 134. Roos, L.L., et al., 2005. Physician visits, hospitalizations, and socioeconomic status: ambulatory care sensitive conditions in a Canadian setting. Health Serv. Res. 40 (4), 1167e1185. Sapelli, C., 2004. Risk segmentation and equity in the Chilean mandatory health insurance system. Soc. Sci. Med. 58 (2), 259e265. Starfield, B., 1991. Primary care and health: a cross-national comparison. JAMA 266 (16), 2268e2271.
Contents lists available at ScienceDirect
Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed
Cost sharing and hospitalizations for ambulatory care sensitive conditions Alejandro Arrieta a, *, Ariadna García-Prado b a b
Florida International University, Department of Health Policy and Management, USA Universidad Pública de Navarra, Departamento de Economía, Spain
a r t i c l e i n f o
a b s t r a c t
Article history: Available online 13 November 2014
During the last decade, Chile's private health sector has experienced a dramatic increase in hospitalization rates, growing at four times the rate of ambulatory visits. Such evolution has raised concern among policy-makers. We studied the effect of ambulatory and hospital co-insurance rates on hospitalizations for ambulatory care sensitive conditions (ACSC) among individuals with private insurance in Chile. We used a large administrative dataset of private insurance claims for the period 2007e8 and a final sample of 2,792,662 individuals to estimate a structural model of two equations. The first equation was for ambulatory visits and the second for future hospitalizations for ACSC. We estimated the system by Two Stage Least Squares (2SLS) corrected by heteroskedasticity via Generalized Method of Moments (GMM) estimation. Results show that increased ambulatory visits reduced the probability of future hospitalizations, and increased ambulatory co-insurance decreased ambulatory visits for the adult population (19e65 years-old). Both findings indicate the need to reduce ambulatory co-insurance as a way to reduce hospitalizations for ACSC. Results also showed that increasing hospital co-insurance does have a statistically significant reduction on hospitalizations for the adult group, while it does not seem to have a significant effect on hospitalizations for the children (1e18 years-old) group. This paper's contribution is twofold: first, it shows how the level of co-insurance can be a determinant in avoiding unnecessary hospitalizations for certain conditions; second, it highlights the relevance for policy-making of using data on ACSC to improve the efficiency of health systems by promoting ambulatory care as well as population health. © 2014 Elsevier Ltd. All rights reserved.
Keywords: Chile Avoidable hospitalizations Ambulatory care sensitive conditions Co-insurance
1. Introduction It is well known that lack of timely, appropriate ambulatory care may lead to complications that require hospitalization, creating unnecessary costs in economic and human terms. There is considerable international evidence which shows that better ambulatory care can decrease the need for hospitalization (Starfield, 1991; Fleming, 1995; Caminal et al., 2004; Macinko et al., 2010). This is especially true in the case of ambulatory care sensitive conditions (ACSC). These conditions can be easily managed with timely and effective outpatient care (Rizza et al., 2007). Rates of avoidable hospitalizations have been used as a tracer to assess access, quality, and performance of the primary care delivery system (Bindman et al., 1995; Ansari et al., 2006). Most research on ACSC
* Corresponding author. Florida International University, Department of Health Policy and Management, 11200 S.W. 8th St., Miami, FL 33199, USA. E-mail address: alejarri@fiu.edu (A. Arrieta). http://dx.doi.org/10.1016/j.socscimed.2014.11.026 0277-9536/© 2014 Elsevier Ltd. All rights reserved.
has focused on determining the socioeconomic and medical conditions associated with hospitalized ACSC (Prospective Studies Collaboration, et al., 2007; Roos et al., 2005; Bliziotis et al., 2012). Such studies have found that potential avoidable hospitalizations vary by socioeconomic status. Middle and lower income population are less likely to receive preventive services, more likely to experience delays in healthcare, and less likely to have a regular source of care. The access to health insurance is another factor that can affect rates of avoidable hospitalizations, a relationship which has been studied in the U.S. through the Medicaid program (Kaestner et al., 2001; Bermudez and Baker, 2005; Dafny and Gruber, 2005). Kaestner et al. (2001) and Dafny and Gruber (2005) estimated the impact of Medicaid eligibility expansions on child hospitalizations. While Kaester et al. found that Medicaid eligibility expansions moderately improved the health of low-income children and reduced hospitalizations, Dafny and Gruber showed that the number of hospitalizations increased, but there was a much smaller increase in avoidable hospitalizations. Aizer (2007) estimated the
116
A. Arrieta, A. García-Prado / Social Science & Medicine 124 (2015) 115e120
impact of Medicaid take-up rather than insurance eligibility expansion. She estimated the impact of early Medicaid take-up on access to primary care and avoidable hospitalizations, finding that a 15% increase in Medicaid enrollment would lead to a 2.7% decline in avoidable hospitalizations. Little is known, however, about the impact of insurance cost-sharing (through co-insurance or deductibles) on hospitalizations for ACSC. In this paper, we studied the effect of ambulatory and hospital co-insurance rates on ACSC among individuals with private insurance in Chile. During the last decade, Chile's private health sector has experienced a dramatic increase in its hospitalization rates, growing at four times the rate of ambulatory visits (see Fig. 1). Such evolution has raised concern among policy-makers interested in promoting more preventive services and higher use of ambulatory care. The increased prevalence of chronic diseases has also raised alarm. A study made in 2007 shows that 84% of the total diseases in Chile were non-communicable diseases (Ministerio de Salud, 2007). The 2003 National Health Survey showed that only a small fraction of those affected by a chronic disease had their condition n et al., 2010). In this context, co-insurance can under control (Bitra be a valuable tool for dealing with cost-escalating problems in the health system while promoting more ambulatory visits and preventive services and less ACSC. Cost sharing is a common feature of insurance contracts. It is useful to reduce patient moral hazard, and, therefore, overconsumption of medical care. By raising co-insurance rates, insurers can reduce unnecessary care and control costs. However, high co-insurance rates may also produce losses due to reduced financial risk protection. The search for an optimal co-insurance rate that balances this trade-off has been broadly studied both theoretically and empirically (Newhouse, 2006; Ellis and Manning, 2007; Pauly and Blavin, 2008). We analyzed cost sharing in the context of an inter-temporal relationship between primary care visits and avoidable hospitalizations. Such relationship presents an additional feature that increases the losses of an un-optimally high co-insurance rate: a high ambulatory co-insurance rate could reduce consumption of effective care, and, as a result, lead to a potential increase of hospitalizations for ACSC in the future. What is the most appropriate co-insurance that would help to reduce hospital admissions for ACSC? What are the effects of different coinsurance levels on hospital and ambulatory visits for ACSC? Our response to these questions is a structural model that describes the inter-temporal correlation between ambulatory visits and hospital admissions for ACSC. We used a large administrative dataset of private insurance claims in Chile to examine the effects of ambulatory and hospital co-insurance rates on both ambulatory and hospital visits. Since hospitalizations for ACSC are avoidable through timely outpatient treatment, policy-makers must ensure that co-insurance for hospital and ambulatory care promotes adequate primary care coverage.
Fig. 1. Evolution of ambulatory versus hospital visits in the Chilean private sector.
Our results show that increasing ambulatory visits reduced the probability of future hospitalizations. For adults, we also found that increasing ambulatory co-insurance decreased ambulatory visits. Both findings indicate the need to reduce ambulatory co-insurance as a way to reduce hospitalizations for ACSC. This paper highlights the need to introduce co-insurance rates that are differentiated according to type of diseases (ACSC) so as to promote more ambulatory care and reduce health costs. It also focuses attention on an area that has been barely researched in less developed countries. 2. The data Data was provided by the health regulator, the Superintendencia de Salud, which validates and consolidates information provided on a quarterly basis1 by all the private insurers (ISAPREs or Instituciones de Salud Provisional) operating in Chile. Health insurance in Chile is dual; i.e., employed individuals must purchase insurance for a minimum of 7% of their taxable income up to a specified threshold in order to enroll in the public insurer (Public Health Fund, PHF) or purchase a health plan from a private insurer (ISAPRE). A total of 2.8 million individuals, or 16.8% of the population, were covered by a contract in one of the 14 ISAPREs that operated in the market by the end of 2007. Our data cover the period from January 1 to December 31, 2007. The administrative data used in this study was exempt from ethical review. All claim records were de-identified by the owner of the data, the Superintendencia de Salud. Information provided by the ISAPRES includes characteristics of plan holders, as well as all the beneficiaries, including sex, age, income, and earnings. We also have extensive information on all claims made by these individuals, including ambulatory visits, recording diagnoses using the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10), codes for hospitalizations, and plan characteristics including co-insurance rates. A health plan from ISAPRES typically specifies co-insurance rates for both outpatient and hospital services, together with caps on coverage by unit of service. Deductibles are not used in insurance plans. Data collected for 2007 include a total of 3,004,102 observations on the insured and their beneficiaries. From this dataset, we excluded two groups e infants (up to 1 year-old) and people over age 65 e for the following reasons: first, both clearly show different patterns of disease and behavior compared to other groups; second, the population over 65 years old represents only about 5% of the total population covered by the ISAPRES, third, prices for the seniors increase more than for others, so that many end up moving to FONASA after retirement. As a result of these exclusions, the sample was reduced to a total of 2,792,662 individuals. For each individual, we constructed indicators for ambulatory visits and subsequent hospital visits within a 30-day window. To check the robustness of the results, we also considered 60- and 90-day windows. We classified hospitalizations as ACSC, using ICD-10 codes, and following the definitions of Alfradique et al. (2009) for Brazil. The list of ACSC inclusions is reported in the Appendix. We chose the Brazilian ACSC list because Brazil is the only country in Latin America that has made a systematic effort to adapt the ACSC lists from the US, Canada and Spain to its own circumstances. Like most Latin American countries, Chile does not have its own ACSC list, and so we decided to use the Brazilian ACSC list as the closest available approximation for a Latin American context. However, it is important to note that there is no international consensus on how best to compose ACSC lists and that several alternative lists are used among and within different countries (Ansari et al., 2006).
1 Starting in 2008, ISAPRES are required to provide information on a monthly basis.
A. Arrieta, A. García-Prado / Social Science & Medicine 124 (2015) 115e120
Table 1 provides descriptive statistics by age group of all variables used in the study. Co-insurance, on average, is higher for ambulatory visits than for hospital visits, according to age group. While ambulatory coinsurance accounts for about 28% of the total expenditures in healthcare made by both age groups, hospital co-insurance accounts for 14% and 17% for children and adults, respectively. The incidence of hospitalizations for ACSC is not high (compared to other countries in the region), but accounts for about 10 and 8 hospitalizations for every 10,000 insured children and adults, respectively. Finally, the percentage of ambulatory visits for children and adults are 75.5% and 71.6%, respectively. 3. The empirical strategy Since the main goal of this study is to assess the impact of coinsurance on hospitalizations for ACSC, a simple estimation strategy is to model the probability of being hospitalized as a function of co-insurance and ambulatory visits. As described in equation (1), if we consider two periods, the probability that patient i is hospitalized for an ACSC in period 1 (hi1) depends on the hospital coinsurance rate in patient i's plan (chi ), a variable indicating if patient i had an ambulatory visit in the past (ai0), and an a contemporaneous error term. We expected that for ACSC, a patient who had an ambulatory visit would have a lower probability of being hospitalized (q < 0). However, if a patient's health status was not controlled, an estimation strategy such as the one in equation (1) could result in biased estimates. First, health status may have driven both ambulatory and hospital visits. Therefore, if health status is not fully controlled, it would bias the estimate of q. Second, the insurance plan selection, and consequently the observed coinsurance rate, could also depend on health status if individuals are free to choose among plans or if insurance companies select customers on the basis of health conditions. In this case, if health status is not fully controlled, it would bias the estimate of gh.
Prðhi1 ¼ 1Þ ¼ gh chi þ qai0 þ εhi1
Table 1 Descriptive statistics by age group. Age group
Hospitalizations for ACSC (fraction)% Co-insurance (ambulatory)% Co-insurance (hospitalization)% Age (in years) Gender (female ¼ 1)% Income (in thousands of Chilean pesos) Sample size (individuals)
was based on co-insurance for ambulatory and hospital visits, and also on health status in each period. Health status evolves stochastically between period 0 and period 1, and we assumed that having an ambulatory visit could affect future health status. This is easily justifiable, since we focused on ambulatory care sensitive conditions, which are those that can improve health and avoid future hospitalizations. The dynamic of health status is as follows:
~si1 ¼ ~si0 þ ~ qai0 þ mi1
(2)
where ~si1 is the (unobserved) health status of patient i in period 1, ai0 is a binary variable that indicates if patient i visited an ambulatory setting in period 0, and mi1 is an independent error term. We modeled a system of two equations, one for ambulatory visits and the second for hospital visits. For identification purposes we made the reasonable assumption that hospital visits depended only on hospital co-insurance, while ambulatory visits depended on both ambulatory and hospital co-insurances. We believed this assumption was justified because patients may avoid ambulatory care if hospital visits are relatively cheap. The system of simultaneous equations is represented by equations (3)e(4).
Prðai0 ¼ 1Þ ¼ g0a cai þ g0h chi þ aa si þ ba~si0 þ εai0
(3)
Prðhi1 ¼ 1Þ ¼ g1h chi þ ah si þ bh~si1 þ εhi1
(4)
A binary variable, hi1, indicates that patient i visits the hospital for chronic ACSC in period 1. Co-insurance for ambulatory visits (cai ) and hospital visits (chi ) are assumed to be time independent; i.e., defined when the individual selected an insurance plan. si are patients' observable characteristics such as age, gender, income, etc. Many other patient's characteristics remain unobservable, and because they affect both, ambulatory and hospital visits, the error terms become correlated. Replacing (2) in (4), we can re-write the simultaneous equation model (3e4) as:
(1)
Although we controlled for patients' characteristics, a significant component of patients' health status remained unobservable, thus, creating potentially omitted variable bias. To overcome the problem of simultaneous dependency between ambulatory and hospital visits (endogeneity), we considered a model of two periods where individuals decided to visit an ambulatory setting in the first period and a hospital setting in the second period. The patient's decision
Ambulatory visits (fraction)%
117
1e18 years old
19e65 years old
75.453 (43.036) 0.098 (3.132) 28.006 (17.479) 14.479 (22.2) 9.684 (5.081) 51.034 (50.005) 69.426 (26.998)
71.634 (45.077) 0.078 (2.789) 28.778 (19.648) 16.508 (25.648) 38.12 (12.125) 51.929 (49.963) 63.84 (29.545)
875,735
1,916,927
Averages and standard deviations (in parenthesis). It only considers hospitalizations for ACSC that occurred after the first 30 days of any ambulatory visit.
Prðai0 ¼ 1Þ ¼ g0a cai þ g0h chi þ aa si þ εai0
(5)
Prðhi1 ¼ 1Þ ¼ qahi0 þ g1h chi þ ah si þ εhi1
(6)
where q≡bh ~ q, εai0 ≡ba~si0 þ εai0 , and εhi1 ≡bh~si1 þ bh mi1 þ εhi1 Although this system corrects for the simultaneous ambulatory and hospital visits, co-insurance rates could still be correlated with error terms in equations (5)e(6). If insurance selection is endogenous and driven by unobserved health status, the impact of coinsurance on ambulatory and hospital visits would be biased. In the selection of an insurance plan, two factors may make insurance choice dependent on health status. First, patients may select plans based on their health status (adverse selection). In that case, sickly patients may choose better plans, with lower co-insurance because more care visits are expected. This may downward bias the parameter estimates for co-insurance on hospital and ambulatory visits. However, a second possibility is that insurance companies can discriminate based on health status (cream skimming). In that case the insurance pool will be formed by healthy patients, which use fewer care visits. In Chile there is evidence of cream skimming by ISAPRES (Sapelli, 2004). This may upward bias the parameter estimates for co-insurance on hospital and ambulatory visits. To reduce these problems, we include two instrumental variables, one for the co-insurance rate of ambulatory visits and the other for hospital visits. Our instruments are the average co-insurance rates corresponding to all plans available in the market at a price equivalent to the member's minimum legal insurance premium. By
118
A. Arrieta, A. García-Prado / Social Science & Medicine 124 (2015) 115e120
law, Chileans in the ISAPRE system have to purchase insurance whose minimum premium is equivalent to 7% of personal income up to a maximum limit of 4.2 UF (inflation-adjusted legal unit of accounting commonly used in Chile). Chileans can opt to pay higher premiums in order to obtain more generous plans. The plans priced at the minimum legal insurance premium can be considered as default plans, as they represent the set of plans that individuals are mandated to obtain without extra payment. To obtain our instruments, we first calculated the minimum legal insurance premium for each member based on reported personal income. Then, we identified the default plans in the market as those with prices close to the minimum premium. Finally, we obtained the average co-insurance rates for the identified default plans. Because default plans are not chosen but set by law, and the average co-insurance rates are not tied to a particular individual, we expect our instruments to not be directly correlated to health status. However, they could be indirectly correlated to health status through income. Because we explicitly control for personal income, our instrument remains valid (Angrist and Pischke, 2009). Notice that although the system of equations (5)e(6) is recursive, identification cannot rest on recursivity because both error terms are dependent (through the unobserved health status). For the sake of simplicity, we assumed a linear probability model, which produced a linear system of equations. We found that ambulatory and hospital co-insurance rates are highly correlated in our data, creating potential distortions in equation (5) due to multicollinearity. To avoid this problem we decided to exclude the hospital co-insurance rates from equation (5). We estimated the system by 2SLS corrected by heteroskedasticity via GMM estimation (Greene, 2008).
4. Results Table 2 presents results for hospitalizations by age group. Those results correspond to the two equations of interest, hospital visits and ambulatory visits. The equation of hospital visits estimates the probability of having a hospital visit. This equation confirms the basic hypothesis underneath ACSC studies: ambulatory visits tend to reduce the probability of future hospitalizations. In particular, the probability of having a hospital admission for ACSC after having had one or more ambulatory visits, decreases by 0.43% and 0.72% for both children and adults, respectively. The hospital equation offers a second result: increasing the hospital co-insurance rate does not reduce the probability of having more hospital visits in the case of children. For adults, this effect is statistically negative, suggesting Table 2 Results for ambulatory and hospital visits. Age group
Ambulatory visits Intercept Ambulatory co-insurance Age Gender (male) Income (in millions of Chilean pesos) Hospital visits Intercept Ambulatory visit (1 or more) Hospital co-insurance Age Gender (male) Income (in millions of Chilean pesos)
1e18 years old
19e65 years old
0.13123* 0.02750* 0.01061* 0.01841* 4.29600*
0.98418* 0.00816* 0.00230* 0.12742* 0.56040*
0.00352* 0.00431* 0.00004 0.00015* 0.00004 0.02040*
0.00751* 0.00717* 0.00012* 0.00003* 0.00088* 0.00361
Simultaneous equation linear model estimated by 2SLS. Ambulatory and Hospital visits are binary variables. The heteroskedasticity produced by the linear probabilities are adjusted using GMM. *Significant at 1%, **Significant at 5%.
that 1 percentage point increment in hospital co-insurance rate reduces the probability of a hospital visit by 0.012 percentage points. The ambulatory visits equation estimates the probability of having ambulatory visits. Again, our results differ by age group. For adults, the effect of ambulatory co-insurance rate on ambulatory visits is statistically negative, indicating that a 1 percentage point increment in ambulatory co-insurance rate reduces the probability of an ambulatory visit by 0.82 percentage points. For children, this effect is positive. For adults, a reduction in ambulatory co-insurance rates increases ambulatory visits, and consequently reduces the probability of future hospitalizations. By combining the results from both equations to obtain the overall marginal effect, our findings suggest that a 1 percentage point reduction in ambulatory co-insurance rates reduces the probability of future hospitalizations by 0.006%. The effect is not meaningless considering that hospitalization rates among adults are 0.078% (see Table 1). Although out-of-sample extrapolation is questionable, our results imply that to fully avoid hospitalizations for ACSC (probability of hospital visit equal to zero), co-insurance rates for ambulatory visits should be reduced from the current average of 28.8% to around 17%. Notice that this last result only applies under the linearity assumption of our estimation strategy. Furthermore, other variables, in addition to coinsurance rates should be considered in the analysis to determine the most appropriate co-insurance rates for ambulatory visits. 5. Discussion In this paper we have applied a structural model to a large administrative dataset of private insurance claims in Chile in order to estimate the effects of ambulatory visits on avoidable hospitalizations, as well as the effect of ambulatory and hospital coinsurance rates on both ambulatory visits and hospitalizations for ACSC. Our evidence supports the theory that ambulatory visits can reduce hospitalizations in both adults and children. For the average Chilean adult with private insurance, having one or more ambulatory visits can reduce the probability of a future hospitalization for ACSC by 0.72 percentage points. The same effect in children is estimated as 0.43 percentage points. We also found that cost sharing reduces hospital and ambulatory visits only in the case of adults. In children, altering hospital co-insurance does not seem to have a significant effect on hospitalizations, while the relationship between ambulatory coinsurance and ambulatory visits is unexpectedly positive. We believe the unexpected results for children reflect the lower sensitivity to prices observed in this age group. For instance, results from the RAND Health Insurance Experiment showed low sensitivity to cost sharing and mixed results for older children compared to adults (Leibowitz et al., 1985), suggesting that cost-sharing behavior differs when individuals decide for their children. For adults, the two main findingsdthat lower ambulatory coinsurance increases ambulatory visits and that more ambulatory visits reduces avoidable hospitalizationsdimply that ambulatory co-insurance should be decreased substantially in order to reduce the high rate of hospitalizations in the private sector. How much should ambulatory co-insurance be diminished so as to reduce hospitalizations for ACSC to zero? We show that for the 2007 average hospital co-insurance rate of 16.5%, the ambulatory coinsurance should be 17% for adults, a figure substantially lower than the ambulatory co-insurance of Chile in 2007 (28%). Our study has several limitations that could be addressed in future investigations. First, our analysis of the effect of co-insurance rates on hospital and ambulatory visits rests on the validity of our instrumental variables. As is common in observational studies that assess the effect of health insurance on healthcare demand, finding valid instruments is challenging. Although our proposed
A. Arrieta, A. García-Prado / Social Science & Medicine 124 (2015) 115e120
instruments are theoretically reasonable, we could not test their validity. Nevertheless, a comparison between the model estimated with and without instrumental variables (results not shown) reveals that the instrumental variable estimation produces the expected effects of co-insurance among adults, suggesting a bias correction of the instrumental variable approach. Second, we considered co-insurance rates to define cost-sharing, omitting other characteristics of the plan that, unfortunately, are not available in the database used for this study. Typically, in addition to co-insurance, insurance plans in Chile have associated caps that limit the maximum value to be paid by the ISAPRES. Plans also differ in terms of preferred health providers and reduced copayment for catastrophic illness. However, since co-insurance rates represent the most important component of insurance coverage in Chile, we think that any potential bias is relatively small. Third, we did not include in the analysis the incentives to hospitalize patients that may exist among physicians working for ISAPRES. While decreasing ambulatory co-insurance seems to be a good policy, the Government should analyze the incentives behind the dramatic increase of hospitalizations in the private sector of Chile. We offer a demand side solution, through co-insurance, but future research should also focus on analyzing the effects from the supply side. It seems, for instance, that some uncovered vertical integration between providers and insurers within the ISAPRE system may be pushing hospitalization rates up. This clearly calls for further research and analysis. Finally, we could not perform a more detailed analysis considering different groups of ACSC. Since we only had data for one year, 2007, we would compromise the statistical robustness of the estimation by adding more groups. Unfortunately, data availability was a constraint when this research was carried out so we could not expand the analysis to include other years. Such analysis would have been extremely interesting, particularly because the AUGE law was passed in Chile in 2005. This law prioritized the health services related to 56 medical conditions, most of them chronic; promoted preventive services; and introduced new treatment guidelines and new co-sharing rules, among other measures. In particular, the AUGE law mandated coverage for a set of interventions for children and adults aimed at prevention and early detection of hypertension, types 1 and 2 diabetes, HIV/AIDS, cervical cancer, and other chronic diseases. The new law also established that insurers, both Public Health Funds (PHF) and ISAPRES, must reimburse a specific amount for each guaranteed health intervention, so that the beneficiary's out-of-pocket spending will not exceed a predefined share of household income. In addition, when seeking care for a medical problem under the AUGE reforms, both PHF and ISAPRES beneficiaries could make one of two care choices: (1) select a closed mode, choosing providers from the AUGE network and paying lower or no co-insurance, or (2) obtain care outside of AUGE through a free choice mode, paying higher co-insurance, but receiving faster, higher-quality care, better amenities, and n et al., 2010). AUGE law continued access to a personal doctor (Bitra also promoted the systematic application of treatment guidelines in order to lower the incidence of complications, and, therefore, hospitalizations from inadequate disease management. Having the specific information for certain ACSCs would be essential to understanding the new cost-sharing rules under the AUGE program as well as its effects on particular ACSCs. It would also be interesting to see the impact of the introduction of AUGE program on hospitalization rates versus hospitalization rates under non-AUGE condin et al. (2010) analyzed the impact on hospitalization tions. Bitra rates for six AUGE conditions but it would be worthwhile to expand this analysis to more conditions and over a greater timespan. Chile is in a process of reform and debate about the escalating costs and the provision of quality healthcare, with a focus on the role
119
of ISAPRES. After a decade of discussions about ways to reform ISAPRES failed to lead to any major changes, the new government that took power in March 2014 will finally have to implement reforms. As our results are still valid for the current situation, we hope that they will contribute to the upcoming policy discussion. Other countries in Latin America, e.g., Colombia and Peru, are facing similar debates, dealing with dramatic increases of hospitalization rates. Since many developing countries are facing the epidemiological transition that more developed economies experienced decades ago, and a dramatic increase of chronic conditions (Glassman et al., 2010), the analysis of the interrelation between co-insurance rates and hospitalizations for ACSCs is going to become more and more relevant. The solution offered here, based on introducing a different co-insurance for those diseases that are more easily preventable, could be a useful strategy for many of these countries in their combat against cost escalation and in favor of primary care promotion. The widely reported U.S. experience of Pitney Bowes, who provided copay relief for drugs used to treat asthma and diabetes, has demonstrated that such an approach is feasible and produces both clinical and economical returns (Mahoney, 2005; Fendrick and et al., 2006). Acknowledgment The authors thank the Inter-American Development Bank (IADB) for funding this research (contract number 521318-0004) and providing useful comments. We also thank the comments received by participants at the Seminar “Chronic Diseases, Primary care and Health Systems Performance” organized by the IADB, the University of Bahía and Ministry of Health in Brazil, and those at the Spanish Health Economics Association meeting that took place in Santander, Spain. Ariadna García Prado also acknowledges support from the project ECO2012-36480: “Analisis de Economía Pública y Bienestar” of the Spanish Ministry of Science and Innovation (2013e2015). Appendix Table A List and Definition of Ambulatory Care Sensitive Conditions (ACSC) Group Diagnostic 1 2
3 4 5
6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Vaccine-preventable
ICD-10 (primary condition only)
G00.0, A17.0, A33, A34, A35, A36, A37, A95, B16, B05, B06, B26, A19 A17.1, A17.2, A17.3, A17.4, A17.5, A17.6, Not vaccine-preventable A17.7, A17.8, A17.9, A15, A16, A18, I00, (rheumatic fever, syphilis, I01, I02, A51, A52, A53, B50, B51, B52, tuberculosis) B53, B54, B77 Dehydration and E86, A00, A01, A01, A03, A04, A05, A06, gastroenteritis A07, A08, A09 Iron deficiency anemia D50 Nutritional deficiencies E40, E41, E42, E43, E44, E45, E46, E50, E51, E52, E53, E54, E55, E56, E57, E58, E59, E60, E61, E62, E63, E64 Ear, nose and throat infections H66, J00, J01, J02, J03, J06, J31 Bacterial pneumonia J15.3, J15.4, J15.8, J15.9, J18.1, J13, J14 Asthma J45, J46 Chronic obstructive pulmonary J20, J21, J40, J41, J42, J43, J44, J47 disease Hypertension I10, I11 Angina pectoris I20 Congestive heart failure I50, J81 Cerebrovascular disease I63, I64, I65, I66, I67, I69, G45, G46 Diabetes E10, E11, E12, E13, E14 Convulsions and epilepsy G40, G41 Pyelonephritis N39.0, N10, N11, N12, N30, N34 Skin and subcutaneous tissue A46, L01, L02, L03, L04, L08 infection Pelvic inflammatory disease N70, N71, N72, N73, N75, N76 Perforated/bleeding ulcer K92.0, K92.1, K92.2, K25, K26, K27, K28 Prenatal and natal P35.0, O23, A50 complications
120
A. Arrieta, A. García-Prado / Social Science & Medicine 124 (2015) 115e120
References Aizer, A., 2007. Public health insurance, program take-up, and child health. Rev. Econ. Stat. 89 (3), 400e415. Alfradique, M.E., et al., 2009. Ambulatory care sensitive hospitalizations: elaboration of Brazilian list as a tool for measuring health system performance (Project ICSAPeBrazil). Cadernos de Saúde Pública 25 (6), 1337e1349. Angrist, J.D., Pischke, J.r.-S., 2009. Mostly Harmless Econometrics: an Empiricist's Companion. Princeton University Press, Princeton, p. 373 xiii. Ansari, Z., Laditka, J.N., Laditka, S.B., 2006. Access to health care and hospitalization for ambulatory care sensitive conditions. Med. Care Res. Rev. 63 (6), 719e741. Bermudez, D., Baker, L., 2005. The relationship between SCHIP enrollment and hospitalizations for ambulatory care sensitive conditions in California. J. Health Care Poor Underserved 16 (1), 96e110. Bindman, A.B., et al., 1995. Preventable hospitalizations and access to health care. JAMA 274 (4), 305e311. n, R., Escobar, L., Gassibe, P., 2010. After Chile's health reform: increase in Bitra coverage and access, decline in hospitalization and death rates. Health Aff. 29 (12), 2161e2170. Bliziotis, I.A., Destounis, A., Stergiou, G.S., 2012. Home versus ambulatory and office blood pressure in predicting target organ damage in hypertension: a systematic review and meta-analysis. J. Hypertens. 30 (7), 1289e1299. Caminal, J., et al., 2004. The role of primary care in preventing ambulatory care sensitive conditions. Eur. J. Public Health 14 (3), 246e251. Dafny, L., Gruber, J., 2005. Public insurance and child hospitalizations: access and efficiency effects. J. Public Econ. 89 (1), 109e129. Ellis, R.P., Manning, W.G., 2007. Optimal health insurance for prevention and treatment. J. Health Econ. 26 (6), 1128e1150. Fendrick, A.M., et al., 2006. Value-based insurance design: a “clinically sensitive” approach to preserve quality of care and contain costs. Am. J. Manag. Care 12 (1), 18. Fleming, S.T., 1995. Primary care, avoidable hospitalization, and outcomes of care: a literature review and methodological approach. Med. Care Res. Rev. MCRR 52 (1), 88e108.
Glassman, A., et al., 2010. Confronting the chronic disease burden in Latin America and the Caribbean. Health Aff. 29 (12), 2142e2148. Greene, W.H., 2008. Econometric Analysis, sixth ed. Prentice Hall, Upper Saddle River, N.J., p. 1177 xxxvii. Kaestner, R., Joyce, T., Racine, A., 2001. Medicaid eligibility and the incidence of ambulatory care sensitive hospitalizations for children. Soc. Sci. Med. 52 (2), 305e313. Leibowitz, A., et al., 1985. Effect of cost-sharing on the use of medical services by children: interim results from a randomized controlled trial. Pediatrics 75 (5), 942e951. Macinko, J., et al., 2010. Major expansion of primary care in Brazil linked to decline in unnecessary hospitalization. Health Aff. Millwood 29 (12), 2149e2160. Mahoney, J.J., 2005. Reducing patient drug acquisition costs can lower diabetes health claims. Am. J. Manag. Care 11 (5 Suppl. l), S170eS176. Ministerio de Salud, Estudio de Carga de Enfermedad y Carga Atribuible: Chile 2007, 2008, Estudios AUGE; Santiago de Chile, 101. Newhouse, J.P., 2006. Reconsidering the moral hazard-risk avoidance tradeoff. J. Health Econ. 25 (5), 1005e1014. Pauly, M.V., Blavin, F.E., 2008. Moral hazard in insurance, value-based cost sharing, and the benefits of blissful ignorance. J. health Econ. 27 (6), 1407e1417. Prospective Studies Collaboration, Lewington, S., Whitlock, G., Clarke, R., Sherliker, P., Emberson, J., Halsey, J., Qizilbash, N., Peto, R., Collins, R., 2007. Blood cholesterol and vascular mortality by age, sex, and blood pressure: a meta-analysis of individual data from 61 prospective studies with 55,000 vascular deaths. Lancet 370 (9602), 1829e1839. Rizza, P., et al., 2007. Preventable hospitalization and access to primary health care in an area of Southern Italy. BMC Health Serv. Res. 7, 134. Roos, L.L., et al., 2005. Physician visits, hospitalizations, and socioeconomic status: ambulatory care sensitive conditions in a Canadian setting. Health Serv. Res. 40 (4), 1167e1185. Sapelli, C., 2004. Risk segmentation and equity in the Chilean mandatory health insurance system. Soc. Sci. Med. 58 (2), 259e265. Starfield, B., 1991. Primary care and health: a cross-national comparison. JAMA 266 (16), 2268e2271.
