Published by Oxford University Press in association with The London School of Hygiene and Tropical Medicine ß The Author 2014; all rights reserved. Advance Access publication 30 April 2014
Health Policy and Planning 2015;30:423–431 doi:10.1093/heapol/czu024
Curing over-use by prescribing fees: an evaluation of the effect of user fees’ implementation on healthcare use in the Czech Republic Lucie Kalousova1,2* 1
Department of Sociology, University of Michigan and 2Department of Health Management and Policy, University of Michigan
Accepted
15 March 2014 In 2008, the Czech Republic instituted a new policy that requires most patients to pay a small fee for some inpatient and outpatient healthcare services. Using the Survey of Health Aging and Retirement in Europe, this article examines the changes in healthcare utilization of Czechs 50 years and older following the new fee requirement by constructing difference-in-differences regression models focusing on four outcome measures: any visits to primary care physician, any hospitalization, number of visits to the primary care physician and number of nights hospitalized. For this population, I find that the likelihood of having any primary care visit decreased after the policy was instituted. The likelihood of reporting any hospitalization was not significantly changed. The predicted number of primary care visits per person declined, but the predicted number of nights spent in a hospital did not. I find only mixed evidence of greater effect of the user fees on some subpopulations compared with others. Those 65 or older reduced their use more than those between 50 and 64, and so did those who consider their health to be good, and the less educated.
Keywords
User fee, Czech Republic, primary care, access, inequality
KEY MESSAGES
Czech Republic implemented small user fees for healthcare use in 2008. This article evaluates their impact on healthcare use.
Difference-in-differences regression strategy reveals that healthcare use decreased after the fees were implemented.
This article finds mixed evidence of a disproportionate effect of the user fees on more vulnerable subpopulations.
Introduction Czechs have enjoyed rapid improvements in population health following the country’s democratization in 1989. Today, they face lower infant mortality than most developed countries, and the rate of communicable disease transmission is among the lowest in the region also (Antonova et al. 2010). However, the Czech healthcare system continues to encounter unique challenges as it
transforms from being completely state-dominated to a market orientation. In particular, it has been troubled by very frequent use of apparently cost-free healthcare services by citizens– patients. One such service is outpatient care. In 2006, Czechs had the highest number of outpatient contacts of all countries in the European WHO (World Health Organization) region (Holcik and Koupilova 2000, Bryndova et al. 2009).
423
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
*Corresponding author. Department of Sociology, University of Michigan, 500 S State Street, Ann Arbor, MI 48103, USA. E-mail: [email protected]
424
HEALTH POLICY AND PLANNING
Methods Data and analytical sample The second (2006/07) and fourth wave (2010/11) of longitudinal data collected by the Survey of Health, Retirement and Ageing in Europe (SHARE II and SHARE IV) provided an ideal opportunity for evaluating the impact of the user fees policy.4 SHARE II is a cross-national longitudinal dataset with survey responses collected in-person from more than 30 000 noninstitutionalized adults 50 years or older residing in 13 European countries: Austria, Belgium, Czech Republic, Denmark, France, Germany, Greece, Italy, Netherlands, Poland, Spain, Sweden and Switzerland. At SHARE IV, the survey expanded to 19 countries by adding Estonia, Hungary, Portugal, and Slovenia and interviewing more than 85 000 individuals in total. During both its second and fourth interviews, SHARE collected detailed self-reports of healthcare utilization in the past year. The study period, 2006–11, saw a number of heterogeneous transformations in the healthcare systems of the SHARE countries. In order to assess whether the enactment of the user fees impacted utilization in the Czech Republic, it is important to compare the changes observed there with the developments in utilization in another country that did not enact policies aimed at increasing or decreasing healthcare use during the same period. For this purpose, I selected Poland as an analytical control group. Like the Czech healthcare system, also the Polish one has a legacy of the centralized Semashko model and it is primarily funded through mandatory social insurance contributions. Both the Czech Republic and Poland generally spend a low proportion of their gross domestic product on healthcare compared with other European countries. This has been reflected in the low salaries and dissatisfaction of healthcare professionals and the material deterioration of some healthcare facilities. But, unlike the Czech Republic, Poland does not have a flat user fee for visiting a physician or staying in a hospital and did not institute any major changes to healthcare delivery that could impact healthcare use between 2006 and 2011 (Sagan et al. 2011). This makes it ideal for a comparison of trends in healthcare utilization. I constructed an analytical sample that includes all Czech and Polish SHARE respondents who participated in either one of the relevant data collection waves (Czech Republic 2006: 2666; Poland 2006: 2388; Czech Republic 2011: 5736; Poland 2011: 1514). In order to take advantage of the largest possible sample size in both countries and minimize the impact of attrition between waves, I treated the dataset as a repeated cross-section of the populations and applied survey weights appropriately.
Measures of healthcare use The impact of the user fees policy was evaluated by focusing on four measures of healthcare use among older patients: any primary care physician visit in the last year (Czech Republic 2006: 85%, Poland 2006: 76%; Czech Republic 2011: 84%, Poland 2011: 80%); number of primary care physician visits in the last year (Czech Republic 2006: 4.7, Poland 2006: 5.3; Czech Republic 2011: 3.8, Poland 2011: 5.2); any hospitalization in the last year (Czech Republic 2006: 15%, Poland 2006: 17%; Czech Republic 2011: 17%, Poland 2011: 19%); and the number of
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
Outpatient care provided by general practitioners, most of whom operate private practices, is the cornerstone of the Czech healthcare system. Patients are free to choose their primary care provider and, if they are not satisfied, change their doctor every 3 months. General practitioners may provide referrals to specialists, but they do not have a strong gatekeeping position because patients can access secondary care on their own, even without a primary care provider’s recommendation. However, primary care providers remain responsible for a significant number of administrative tasks, such as certifying workers’ ability or inability to perform jobs. Moreover, inpatient treatment is in most instances only available after a written recommendation by a physician (Bryndova et al. 2009). A large average number of visits by Czech patients combined with the physicians’ administrative duties resulted in full waiting rooms and dissatisfaction on the side of both patients and providers. In an attempt to deal with this issue, the Czech government instituted healthcare user fees in 2008. Their intention was not only to raise additional funds for the healthcare system but also to incentivize patients to think twice about seeking outpatient care. Effective 1 January 2008, every Czech patient1 became required to pay 30 Crowns ($1.60)2 for a visit to a doctor’s office, 90 Crowns ($4.80) for a trip to an emergency room, 30 Crowns ($1.60) for a prescription for medications and 60 Crowns ($3.18)3 per day spent in a hospital. In order to prevent disproportionate impact on the most vulnerable citizens, special provisions in the law exempt people in dire financial need, and, as a protective measure for chronically ill patients, the annual amount of fees paid by any one person was capped at 5000 Crowns ($266.00) for the general adult population and at 2500 Crowns ($133.00) for those older than 65 years. Routine preventative care, including a biennial physical exam, an annual dental exam and annual gynecological exam was exempt from fees payments. Median monthly earnings in the Czech Republic were $1163 in 2008. The user fees were not expected to constitute a substantial financial burden to most ˚ tilova´ families, but signal that healthcare is a valued good (Kru and Yaya 2012). The enactment of the fees sparked a wave of popular protests across the country, fueled impassioned political speeches by elected officials and, in some instances, escalated into collective action among healthcare providers who refused to charge the fees. Opponents of the policy argued that the fees would deter the neediest from seeking care and magnify the extant health disparities between the rich and the poor. They pointed to studies of healthcare user fees in middle and low income countries and emphasized their adverse consequences for population health (McIntyre et al. 2006; Ponsar et al. 2011). The promoters of the fees countered with examples of similar Western European provisions for regulating healthcare use that did not result in the widening of health disparities (Chiappori et al. 1998; Cockx and Brasseur 2003). The purpose of this article is to evaluate whether the enactment of the user fees diminished the use of care among older residents of the Czech Republic and if so, whether the observed effect was limited to more socioeconomically vulnerable subpopulations who were burdened by the fees to a greater extent.
CURING OVER-USE BY PRESCRIBING FEES
Analytic strategy Difference-in-differences regression models, a quasi-experimental strategy, were estimated to assess the impact of the fees on healthcare use. This approach is based on the assumption that if no intervention occurred, healthcare utilization would have increased or decreased at the same rate for the intervention group—here the Czechs exposed to a broad scale policy change instituting user fees—as for the control group—Poland, which did not experience such a change. The regression models are set up to test the statistical significance of the deviation from the hypothesized common trend. The choice of a suitable control group is of fundamental importance to this strategy. A comparison of the Czech population and the Polish control group on key sociodemographic characteristics and utilization reports for both survey waves is provided in the Appendix.
First, I examined the impact of the fees’ implementation on the population overall. Then, I evaluated the influence of the fees implementation in a multivariate framework. Logistic regression models were estimated for binary utilization measures (any primary care visit and any hospitalization) and negative binomial regression models conditional on having at least one event were constructed for continuous utilization measures (number of visits to a primary care physician and number of nights spent in a hospital). In order to increase the precision of these models and rule out additional underlying variation that could influence model estimates, I added individual-level controls for sex, mean-centered age, poor or fair self-reported health (SRH) and net worth (asset minus debts). Findings from these models are presented in Table 1 in predicted probabilities of an event and Table 2 in predicted mean numbers of events. Then, I evaluated the impact of the policy on several subpopulations, namely: people from 50 to 64 years and those 65 or older; men and women; those with good, very good or excellent self-reported health status vs those with poor or fair self-reported health status; highly-educated people and those with secondary education or less; and people with medium or high net worth and those with low net worth (designated separately within both groups). Again, all models included controls for sex, mean-centered age, poor or fair selfreported health and net worth. Findings for these models are presented in Table 3 in predicted probabilities of a utilization event and Table 4 in predicted mean numbers of utilization events. To generate population representative results, all findings presented in this article were estimated using the official survey weights created by SHARE along with survey estimation commands in Stata/SE 13.0 (Stata Corporation 2013).
Table 1 Predicted probabilities of any primary care visit and any hospitalization before and after user fees’ introduction Before fees’ introduction Average effects
Czech Republic (2006)
Poland (2006)
Predicted probability of any primary care visit
0.85
0.76
Predicted probability of any hospitalization
0.15
0.17
Difference in probabilities at baseline
0.09*** 0.02
After fees’ introduction Czech Republic (2011)
Poland (2011)
0.84
0.80
0.17
0.19
Difference in probabilities at follow up
0.04*** 0.02y
Difference in percentage-point change (2006–11)
Significance of the difference in differences
0.05
*
0.00
NS
Note: Results of weighted logistic regression models with binary dependent variables. Models controlled for sex, mean-centered age, poor/fair health, and networth. N ¼ 12 304. Difference in percentage-point change is the difference-in-differences estimate of the net policy effect. NS, not statistically significant, y P < 0.1, *P < 0.05, ***P < 0.001.
Table 2 Predicted number of primary care visits and nights spent in a hospital before and after user fees’ introduction Before fees’ introduction Average effects—conditional
No. of primary care visits No. of nights spent in a hospital
Czech Republic (2006)
Poland (2006)
Difference between means at baseline
After fees’ introduction Czech Republic (2011)
Poland (2011)
5.47
6.98
1.51***
4.49
6.58
17.39
15.98
1.41
14.23
14.74
Difference between means at follow up
Difference in differences between means (2006–11)
Significance of the difference in difference
2.09***
0.58
*
0.51
1.92
NS
Note: Results of weighted binomial regression models with continuous dependent variables, conditional on having at least one primary care visit or night spent in a hospital. Models controlled for sex, mean-centered age, poor/fair health, and net-worth. N ¼ 10 302 for primary care visits; N ¼ 2120 for hospital nights. Difference in percentage-point change is the difference-in-differences estimate of the net policy effect. NS, not statistically significant, P < 0.1, *P < 0.05, ***P < 0.001.
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
nights spent in the hospital during the past twelve months (Czech Republic 2006: 2.6, Poland 2006: 2.8; Czech Republic 2011: 2.4, Poland 2011: 2.8). Each survey item listed above measures a separate care use decision; some of these decisions are more likely than others to be affected by the incentive structure of the user fee policy. Although patients have some discretion over whether and how often they visit their primary care physician, the decisions over hospitalization and the number of nights spent in a hospital are more driven by their physician’s recommendations and the severity of their health issues. Physicians may take hospital user fees and cost incurred by their patients into consideration when prescribing care and patients may express a preference for a shorter stay due to cost, but, in general, we would expect hospital care to be impacted by the user fees considerably less than primary care.
425
0.93 0.13 0.20
Predicted probability any hospitalization (50–64)
Predicted probability any hospitalization (65þ)
0.16 0.14
Predicted probability any primary care visit (women)
Predicted probability any hospitalization (men)
Predicted probability any hospitalization (women)
0.91 0.10 0.22
Predicted probability any primary care visit (poor/fair SRH)
Predicted probability any hospitalization (good/very good/excellent SRH)
Predicted probability any hospitalization (poor/fair SRH)
0.86 0.15 0.15
Predicted probability any primary care visit (secondary education or less)
Predicted probability any hospitalization (higher education)
Predicted probability any hospitalization (secondary education or less)
0.79 0.15 0.18
Predicted probability any primary care visit (low)
Predicted probability any hospitalization (medium or high)
Predicted probability any hospitalization (low)
0.19
0.17
0.75
0.77
0.18
0.13
0.77
0.72
0.23
0.09
0.83
0.65
0.18
0.16
0.81
0.70
0.22
0.14
0.81
0.73
Poland (2006)
0.01
0.02
0.04
0.09***
0.03
0.02
0.09***
0.10*
0.01
0.01
0.08***
0.16***
0.04*
0.00
0.06***
0.14***
0.03
0.01
0.12***
0.08***
Difference in probabilities at baseline
0.16
0.17
0.85
0.84
0.17
0.16
0.85
0.81
0.25
0.11
0.90
0.81
0.18
0.16
0.86
0.82
0.22
0.15
0.89
0.81
Czech Republic (2011)
0.22
0.18
0.75
0.82
0.20
0.17
0.80
0.77
0.27
0.09
0.88
0.69
0.18
0.20
0.83
0.76
0.28
0.14
0.86
0.76
Poland (2011)
After fees’ introduction
y
NS NS
0.09 0.02 0.05
0.08 0.03 0.04 0.03
0.01 0.08**
0.06** 0.03y 0.04* 0.00
NS NS
0.06 0.01 0.01
0.07
0.02 0.02 0.02
0.03 0.05**
NS
0.03 0.00
0.07 0.06
0.03
0.02* 0.10***
NS NS
0.01 0.05
0.01 0.05
NS
***
* NS
0.05 0.01
NS
**
0.04
0.12***
NS
NS
*
*** NS
0.03
NS
Significance of the difference in differences
0.03y
Difference in percentage-point change (2006–11)
0.05*
Difference in probabilities at follow up
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
Note: This table shows results of weighted logistic regression models with binary dependent variables stratified by subgroups. Models controlled for sex, mean-centered age, poor/fair health, and net-worth, unless used as a stratifying variable. N ¼ 12 304. Difference in percentage-point change is the difference-in-differences estimate of the net policy effect. Small imprecisions in additions and subtractions due to the rounding of the estimates. NS, not statistically significant, yP < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001.
0.86
Predicted probability any primary care visit (medium or high)
Net worth
0.82
Predicted probability any primary care visit (higher education)
Education
0.81
Predicted probability any primary care visit (good/very good/excellent SRH)
Self-reported health status
0.84 0.87
Predicted probability any primary care visit (men)
Sex
0.81
Predicted probability any primary care visit (65þ)
Czech Republic (2006)
Predicted probability any primary care visit (50–64)
Age
Average effects
Before fees’ introduction
Table 3 Predicted probabilities of any primary care visit and any hospitalization before and after user fees’ introduction stratified by sociodemographic categories
426 HEALTH POLICY AND PLANNING
18.04
Predicted mean of nights spent in a hospital (65þ)
Predicted mean of nights spent in a hospital (women)
20.15
Predicted mean of nights spent in a hospital (poor/fair SRH)
17.03
17.05
15.62
7.33
6.88
1.68**
0.02
1.64
0.76
1.56***
1.46
2.73
1.45***
17.55
13.29
4.93
4.31
14.99
9.07
4.63
3.73
18.03
9.10
3.77y 2.06
5.83
3.47
13.29
15.74
4.71
4.21
16.69
12.28
5.01
4.05
Czech Republic (2011)
18.24
12.76
7.08
6.37
9.96
9.26
6.72
5.63
15.84
9.75
7.40
5.09
13.91
15.40
6.68
6.41
13.77
15.98
7.25
6.06
Poland (2011)
After fees’ introduction
1.21***
0.37
1.35
1.16
1.48***
1.55***
0.60
1.94
1.09**
1.72***
Difference in probabilities at baseline
y
NS
NS *
4.42 0.13
0.22 0.64
0.65 2.19
1.90*** 2.09***
NS
NS y
NS NS
2.54 1.22
0.51 1.39 1.11 0.68
0.19 0.24
2.06*** 2.15*** 0.53 0.70
NS
**
NS
1.97
NS
NS
0.82
0.34 0.62
1.24
NS NS
0.66 0.50
2.21*** 1.97***
0.36
NS 2.32
2.92
1.61***
* NS
5.64
3.70
1.56***
NS
0.29 1.15
Significance of the difference in differences
2.01***
Difference in percentage-point change (2006–11)
2.24***
Difference in probabilities at follow up
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
Note: This table shows results of weighted binomial regression models with continuous dependent variables stratified by subgroups. Models controlled for sex, mean-centered age, poor/fair health, and net-worth, unless used as a stratifying variable N ¼ 10 302 for primary care visits; N ¼ 2120 for hospital nights. Difference in percentage-point change is the difference-indifferences estimate of the net policy effect. Small imprecisions in additions and subtractions due to the rounding of the estimates. NS, not statistically significant, yP < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001.
17.26
Predicted mean of nights spent in a hospital (low)
6.57
Predicted mean of primary care visits (low)
Predicted mean of nights spent in a hospital (medium or high)
5.32
Predicted mean of primary care visits (medium or high)
Net worth
17.23
14.50 17.05
Predicted mean of nights spent in a hospital (higher education)
Predicted mean of nights spent in a hospital (secondary education or less)
18.50
7.07
4.57 5.62
Predicted mean of primary care visits (higher education)
6.25
18.09
8.07
8.16
4.62
15.35
17.18
7.21
6.64
17.44
14.38
7.87
6.29
Poland (2006)
Predicted mean of primary care visits (secondary education or less)
Education
11.84
6.96
Predicted mean of nights spent in a hospital (good/very good/excellent SRH)
4.25
Predicted mean of primary care visits (good/very good/excellent SRH)
Predicted mean of primary care visits (poor/fair SRH)
Self-reported health status
18.34 16.70
Predicted mean of nights spent in a hospital (men)
5.09 5.73
Predicted mean of primary care visits (men)
Predicted mean of primary care visits (women)
Sex
16.31
6.78
Predicted mean of nights spent in a hospital (50–64)
4.57
Predicted mean of primary care visits (50–64)
Czech Republic (2006)
Predicted mean of primary care visits (65þ)
Age
Average effects
Before fees’ introduction
Table 4 Predicted number of primary care visits and nights spent in a hospital before and after user fees’ introduction stratified by sociodemographic categories
CURING OVER-USE BY PRESCRIBING FEES
427
428
HEALTH POLICY AND PLANNING
Results
(a) 100% 95%
% Reporting a Primary Care Visit
30%
Observed Poland Observed Czech Republic
25%
Expected Czech Republic 89%
90%
(b)
% Reporting a Hospitalization Observed Poland Observed Czech Republic Expected Czech Republic 19%
20% 17%
85%
85%
80% 75%
84%
15%
80%
10%
0%
2006
7.5
15%
5%
76%
70%
(c)
17%
Conditional Number of Primary Care Visits
(d)
2011
Conditional Number of Nights Spent in a Hospital
18.0
7.0
7.0
2006
2011
6.5
17.0 17.0
6.5 16.0
6.0
16.2
5.5 5.5 5.0 4.5 4.0
15.4 5.0
15.0 Observed Poland
Observed Poland Observed Czech Republic
14.0 4.5
Expected Czech Republic
14.6 14.3
Observed Czech Republic 13.0
3.5 12.0
3.0 2006
2011
Expected Czech Republic 2006
2011
Figure 1 Expected utilization shows differences between Poland and the Czech Republic, if their differences remained unchanged after policy implementation. Number of primary care visits and nights hospitalized conditional on having at least one visit or hospitalization.
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
Figure 1a presents the unadjusted percentage of the population 50 years or older in both Czech Republic and Poland that reported a primary care visit in 2006 and 2011, as well as the expected prevalence in 2011 if no policy change took place in the Czech Republic and the difference between Czech healthcare use and Poland healthcare remained constant. In 2006, 85% of the Czech population reported having seen a primary care physician compared with 76% of the Poles in the sample. In 2011, we find that although the percentage of the Czech population with a visit decreased to 84%, the percentage of the control group reporting a visit increased to 80%. Had the prevalence of visits among the Czechs mirrored the increase in the Polish population, we would have anticipated 89% of them reporting a primary care visit. Figure 1b presents the unadjusted percentages of the Czech and Polish populations who reported being hospitalized in 2006 and 2011. The Polish control group reported a hospitalization more often than the Czechs in 2006 (17% vs 15%), and the rate of hospitalizations increased at about the same rate in the control group compared with the Czech Republic in 2011 (19% vs 17%).
Figure 1c shows the changes in the unadjusted conditional mean number of primary care visits between 2006 and 2011. The conditional number of visits at the first time point was lower in the Czech Republic than in the Polish control (5.5 vs 7.0). At the second time point, the conditional number of visits decreased in the Czech Republic to 4.5 and in Poland to 6.5. If the observed prevalence at the second time point in the Czech Republic was consistent with rate of change in the control group, we would have expected to find a less steep decrease, to 5.0 visits. Figure 1d displays the unadjusted conditional mean numbers of nights spent in a hospital. The observed mean number of nights spent in a hospital was lower in Poland than in the Czech Republic in 2006 (16.2 vs 17.0) but higher in 2011 (14.6 vs 14.3). Had the number of nights spent in a hospital decreased at the same rate in the Czech Republic as in the Poland, we would anticipate seeing a less steep decrease, to a mean of 15.4 nights. Table 1 shows the predicted probabilities of any primary care visit in each year obtained from logistic regression models with survey weights and controls. The difference in probabilities column refers to the amount of change in the difference between Czech Republic and Poland in 2006 and in 2011. For example, the predicted probability of any primary care visit for
CURING OVER-USE BY PRESCRIBING FEES
be valid, we must assume that Poland is a good comparator to the Czech Republic. It is possible that the change in healthcare utilization of the Polish sample was not typical, and that it was driven by other changes that took place in Poland during the study period. As a test of the sensitivity of the results to the choice of the control country, I conducted similar analysis with all other countries included in the SHARE II and IV dataset. Overall, the trends I observed were similar to those reported here. These additional results are available on request. Second, relying on only two data points may introduce inaccurate understanding of the policy effect, if these two time points were not reflective of the prior trend in healthcare use. Third, SHARE is a survey of adults age 50 or older and their partners. This is both an advantage and a disadvantage. Older people are the most frequent users of healthcare. Therefore, a policy aimed at reducing the number of contacts with the healthcare system should have the strongest effects on this population. However, it could also be the case that their high care demand will make them less responsive to any price changes than other age groups. The results may therefore not be generalizable to the remainder of the population, especially because people older than 65 are subject to lower maximum payments. However, some of these concerns may be alleviated by the age-stratified analysis presented among the results. Moreover, the findings are based on self-reported data of healthcare use. It is possible that these reports may not always be accurate, especially when the number of primary care visits or nights spent in a hospital is high (Bhandari and Wagner 2006). However, it is unlikely that patients’ tendency to under-report or over-report utilization would change between 2006 and 2011. Consequently these data are suitable for the analysis of change that is presented here, but may be less accurate when used for drawing conclusions about the precise amount of healthcare used. An additional concern for any analysis of healthcare use is that it is generally subject to ceiling and floor effects. Assuming an increase or a decrease following a policy change may not always be realistic if, for example, a proportion of the population will not see a primary care provider under any circumstances. Some, but not all, of these concerns should be lessened by using a non-linear modelling strategy. An important matter to consider when interpreting the results presented here is the likely increase of the general level of economic hardship in the population due to the onset of the Great Recession that coincided with the enactment of the fees. The impact of the policy may have been compounded by the unfavourable economic climate, and the number of visits will return to its pre-recession levels once the economy recovers. I recommend that the findings presented here are re-examined at a later date, with data collected during a less economically troubled period. Finally, the analyses presented in this article offer no insight into whether the policy change had any effect on population health. A much longer time-frame will be necessary to evaluate the potential health effects of this policy.
Limitations Although the SHARE data provide a unique opportunity to examine the influence of the new fee policy in the first years after its enactment, the results of this analysis should be interpreted with some caveats. First, in order for these results to
Discussion The purpose of this study was to evaluate whether the enactment of the user fees diminished the use of care as the policy intended,
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
Czechs was 0.85 and for Poles was 0.76 in 2006, reflecting a 0.09 lower predicted probability for Poland. The predicted probability of having had a primary care visit in 2011 was 0.84 for Czechs and 0.80 for Poles in 2011; the difference was 0.04. The difference in differences between the two time points was therefore 0.05 and statistically significant. This suggests that the implementation of the fees influenced the overall likelihood of seeking care. However, there appear to be no similar effects on the predicted probability of hospitalization. The probability of hospitalization increased at the same rate in both countries and therefore appears to not have been affected by the policy. Table 2 presents the predicted conditional mean number of primary care visits and nights spent in a hospital obtained from negative binomial regression models with survey weights and controls. The difference in differences between means refers to the absolute magnitude of change in the differences between the two groups in 2006 and 2011. I observe a significant change in difference for the number of primary care visits. Czechs had a significantly lower predicted mean number of visits in 2006 than that of the Polish control group in 2006 (5.47 vs 6.98), and this difference became larger over time. In 2011, the predicted mean number of visits for Czechs dropped to 4.49 whereas it dropped only slightly in Poland to 6.58. The overall difference in differences, 0.58, is statistically significant. It thus appears that the user fees had the intended effect of lowering the number of primary care visits. At the same time, the difference between the numbers of nights spent in a hospital in the two countries at the two time points was not statistically significant. Table 3 displays predicted probabilities from stratified logistic difference-in-differences models with survey weights and controls. I find a significant decrease in the predicted probability of having any primary care visit for those who were 65 or older, for men, for people who finished secondary education or less, and people with medium or higher net worth. The probability of a primary care visit for people with poor or fair health remained approximately stagnant in the Czech Republic (0.91 in 2006 and 0.90 in 2011), but increased in Poland from 0.83 to 0.88, and the resulting difference is statistically significant. There was only one marginally statistically significant difference in the probability of hospitalization: although Czech men’s probability increased slightly from 0.20 to 0.22 over time, the probability of hospitalization for Polish men increased from 0.22 to 0.28 during the same time period. Table 4 presents results of stratified negative binomial regression models predicting mean number of visits and nights spent in a hospital for each subpopulation in 2006 and 2011. I observe a significant decrease in the number of visits for Czechs older than 65, people with good self-reported health, and those with secondary education or less. For number of nights spent in a hospital, the only marginally statistically significant decrease can be seen in the healthier group, where the predicted number of nights dropped in the Czech Republic, while it increased in Poland.
429
430
HEALTH POLICY AND PLANNING
It has been well documented that the perceived cost of healthcare and demand for it are tightly bound to one other. In some circumstances, an unlimited access to a valuable good, such as care, may lead to the distortion of its real price in the minds of the recipients, and possibly overuse (Arrow 1963; Manning et al. 1987). From this perspective, instituting healthcare user fees should improve allocative efficiency of the system by reducing the moral hazard associated with the availability of free care (Winkelmann 2004; Culyer 1971). The analysis presented in this article suggests that the enactment of the user fees created effects consistent with their intentions. Other neighbouring countries of the Czech Republic have also experimented with instituting user fees, but they met with insurmountable public and political opposition. In Hungary, small user fees were introduced in 2007 and abolished just one year later following a public referendum (Baji et al. 2011). In Slovakia, healthcare user fees were instituted in 2003, but most of them were abolished after a change in governmental leadership in 2006 (Szalay et al. 2011). Similarly, the newly elected (autumn 2013) leading Czech political coalition has taken a strong stand on the healthcare user fees and made their abolishment one of its main goals in the eyes of the public (Kopecky 2013). In all three countries, the driving political argument behind the policy reversal has been protecting the disadvantaged. Although this article fully agrees with the intention of providing maximum equity and access to healthcare to all people, further studies are needed exploring specifically who is at most risk of losing access to care on account of the user fees. Such research would also have the potential of helping us erect additional safety nets for those who are identified as most disadvantaged. In light of the high healthcare use found in the Czech Republic and the relatively low level of resources devoted to healthcare in this country, the findings presented in this article suggest that implementing and maintaining a strong safety net rather than the wholesale abolishment of the user fees appears to be the more promising policy direction.
Acknowledgements This article uses data from SHARE wave 4 release 1, as of 30 November 2012 and SHARE wave 1 and 2 release 2.5.0, as of 24 May 2011. The SHARE data collection has been primarily funded by the European Commission through the Fifth Framework Program (project QLK6-CT-2001-00360 in the thematic program Quality of Life), through the Sixth Framework Program (projects SHARE-I3, RII-CT-2006-062193, COMPARE, CIT5-CT-2005-028857, and SHARELIFE, CIT4-CT2006-028812) and through the Seventh Framework Program (SHARE-PREP, No. 211909, SHARE-LEAP, No. 227822 and SHARE M4, No. 261982). Additional funding from the U.S. National Institute on Aging (U01 AG09740-13S2, P01 AG005842, P01 AG08291, P30 AG12815, R21 AG025169, Y1AG-4553-01, IAG BSR06-11 and OGHA 04-064) and the German Ministry of Education and Research as well as from various national sources is gratefully acknowledged (see www. share-project.org for a full list of funding institutions). Conflict of interest statement. None declared.
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
and if so, whether this happened at the expense of the more vulnerable subpopulations. The results show that the overall frequency of visits to primary care physicians decreased after the fees were implemented, as did the likelihood of having any primary care visit, but the probability of any hospitalization, and number of nights spent in a hospital did not. This finding aligns with the assumption that primary care visits are determined by the consumers of care to a greater degree than other types of healthcare use, and are therefore more sensitive to the newly imposed cost. It is important to add that although the decrease in the number of visits can be lauded as a success of the policy, it is less clear whether the decrease in the proportion of the older population who report having seen their primary care physician at least once last year can be described as an entirely positive development. Those 65 and older became significantly less likely to have at least one primary care visit after the policy implementation, and reported cutting back more on the number of primary care visits than the younger group, although they still have more primary care visits overall. This could be indicative either of the greater amount of initial unnecessary use by the older group or of the greater burden placed on them by the fees. One would generally expect that most people in the older age group would have at least one primary care visit a year, and a decrease in the probability of having seen a primary care provider is not necessarily a positive development in this age group. We see a similar picture when examining the extent to which the policy affected the more- and less-educated groups. The introduction of fees was associated with a reduction in number of visits among older Czech citizens with secondary education or less. This difference could again signify either that an unnecessary number of visits was more common among the less-educated group, and they were therefore able to cut back more, or, that the burden placed on this group is larger. However, further investigation will be required to establish whether the observed steeper decline points to emerging health access disparities along educational strata or a signal of convergence of utilization at a more appropriate rate. As hoped for by the designers of the policy, the results showed no significant stratifying effect of having low net worth on any of the outcomes studied. Because no clear and consistent differences in the decrease in utilization between the more and less advantaged were uncovered, this article highlights the importance of the symbolic meaning of the user fees. However, the strength of this conclusion would have been substantially enhanced by a finer measure of income and wealth. Further work is also needed that would be able to distinguish between those who were in material need, but were not exempt from fees payments, and those who were, and consider whether fees became an obstacle to care seeking for the first group. Moreover, I was not able to investigate what types of care exactly were reduced. Evidence from other countries, such as the American RAND Health Insurance Experiment, shows that cost-sharing reduces both unnecessary and necessary use of care (Keeler 1992). Although their findings may not be applicable to the Czech Republic where out of pocket costs of care are generally low,
Health Policy and Planning 2015;30:423–431 doi:10.1093/heapol/czu024
Curing over-use by prescribing fees: an evaluation of the effect of user fees’ implementation on healthcare use in the Czech Republic Lucie Kalousova1,2* 1
Department of Sociology, University of Michigan and 2Department of Health Management and Policy, University of Michigan
Accepted
15 March 2014 In 2008, the Czech Republic instituted a new policy that requires most patients to pay a small fee for some inpatient and outpatient healthcare services. Using the Survey of Health Aging and Retirement in Europe, this article examines the changes in healthcare utilization of Czechs 50 years and older following the new fee requirement by constructing difference-in-differences regression models focusing on four outcome measures: any visits to primary care physician, any hospitalization, number of visits to the primary care physician and number of nights hospitalized. For this population, I find that the likelihood of having any primary care visit decreased after the policy was instituted. The likelihood of reporting any hospitalization was not significantly changed. The predicted number of primary care visits per person declined, but the predicted number of nights spent in a hospital did not. I find only mixed evidence of greater effect of the user fees on some subpopulations compared with others. Those 65 or older reduced their use more than those between 50 and 64, and so did those who consider their health to be good, and the less educated.
Keywords
User fee, Czech Republic, primary care, access, inequality
KEY MESSAGES
Czech Republic implemented small user fees for healthcare use in 2008. This article evaluates their impact on healthcare use.
Difference-in-differences regression strategy reveals that healthcare use decreased after the fees were implemented.
This article finds mixed evidence of a disproportionate effect of the user fees on more vulnerable subpopulations.
Introduction Czechs have enjoyed rapid improvements in population health following the country’s democratization in 1989. Today, they face lower infant mortality than most developed countries, and the rate of communicable disease transmission is among the lowest in the region also (Antonova et al. 2010). However, the Czech healthcare system continues to encounter unique challenges as it
transforms from being completely state-dominated to a market orientation. In particular, it has been troubled by very frequent use of apparently cost-free healthcare services by citizens– patients. One such service is outpatient care. In 2006, Czechs had the highest number of outpatient contacts of all countries in the European WHO (World Health Organization) region (Holcik and Koupilova 2000, Bryndova et al. 2009).
423
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
*Corresponding author. Department of Sociology, University of Michigan, 500 S State Street, Ann Arbor, MI 48103, USA. E-mail: [email protected]
424
HEALTH POLICY AND PLANNING
Methods Data and analytical sample The second (2006/07) and fourth wave (2010/11) of longitudinal data collected by the Survey of Health, Retirement and Ageing in Europe (SHARE II and SHARE IV) provided an ideal opportunity for evaluating the impact of the user fees policy.4 SHARE II is a cross-national longitudinal dataset with survey responses collected in-person from more than 30 000 noninstitutionalized adults 50 years or older residing in 13 European countries: Austria, Belgium, Czech Republic, Denmark, France, Germany, Greece, Italy, Netherlands, Poland, Spain, Sweden and Switzerland. At SHARE IV, the survey expanded to 19 countries by adding Estonia, Hungary, Portugal, and Slovenia and interviewing more than 85 000 individuals in total. During both its second and fourth interviews, SHARE collected detailed self-reports of healthcare utilization in the past year. The study period, 2006–11, saw a number of heterogeneous transformations in the healthcare systems of the SHARE countries. In order to assess whether the enactment of the user fees impacted utilization in the Czech Republic, it is important to compare the changes observed there with the developments in utilization in another country that did not enact policies aimed at increasing or decreasing healthcare use during the same period. For this purpose, I selected Poland as an analytical control group. Like the Czech healthcare system, also the Polish one has a legacy of the centralized Semashko model and it is primarily funded through mandatory social insurance contributions. Both the Czech Republic and Poland generally spend a low proportion of their gross domestic product on healthcare compared with other European countries. This has been reflected in the low salaries and dissatisfaction of healthcare professionals and the material deterioration of some healthcare facilities. But, unlike the Czech Republic, Poland does not have a flat user fee for visiting a physician or staying in a hospital and did not institute any major changes to healthcare delivery that could impact healthcare use between 2006 and 2011 (Sagan et al. 2011). This makes it ideal for a comparison of trends in healthcare utilization. I constructed an analytical sample that includes all Czech and Polish SHARE respondents who participated in either one of the relevant data collection waves (Czech Republic 2006: 2666; Poland 2006: 2388; Czech Republic 2011: 5736; Poland 2011: 1514). In order to take advantage of the largest possible sample size in both countries and minimize the impact of attrition between waves, I treated the dataset as a repeated cross-section of the populations and applied survey weights appropriately.
Measures of healthcare use The impact of the user fees policy was evaluated by focusing on four measures of healthcare use among older patients: any primary care physician visit in the last year (Czech Republic 2006: 85%, Poland 2006: 76%; Czech Republic 2011: 84%, Poland 2011: 80%); number of primary care physician visits in the last year (Czech Republic 2006: 4.7, Poland 2006: 5.3; Czech Republic 2011: 3.8, Poland 2011: 5.2); any hospitalization in the last year (Czech Republic 2006: 15%, Poland 2006: 17%; Czech Republic 2011: 17%, Poland 2011: 19%); and the number of
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
Outpatient care provided by general practitioners, most of whom operate private practices, is the cornerstone of the Czech healthcare system. Patients are free to choose their primary care provider and, if they are not satisfied, change their doctor every 3 months. General practitioners may provide referrals to specialists, but they do not have a strong gatekeeping position because patients can access secondary care on their own, even without a primary care provider’s recommendation. However, primary care providers remain responsible for a significant number of administrative tasks, such as certifying workers’ ability or inability to perform jobs. Moreover, inpatient treatment is in most instances only available after a written recommendation by a physician (Bryndova et al. 2009). A large average number of visits by Czech patients combined with the physicians’ administrative duties resulted in full waiting rooms and dissatisfaction on the side of both patients and providers. In an attempt to deal with this issue, the Czech government instituted healthcare user fees in 2008. Their intention was not only to raise additional funds for the healthcare system but also to incentivize patients to think twice about seeking outpatient care. Effective 1 January 2008, every Czech patient1 became required to pay 30 Crowns ($1.60)2 for a visit to a doctor’s office, 90 Crowns ($4.80) for a trip to an emergency room, 30 Crowns ($1.60) for a prescription for medications and 60 Crowns ($3.18)3 per day spent in a hospital. In order to prevent disproportionate impact on the most vulnerable citizens, special provisions in the law exempt people in dire financial need, and, as a protective measure for chronically ill patients, the annual amount of fees paid by any one person was capped at 5000 Crowns ($266.00) for the general adult population and at 2500 Crowns ($133.00) for those older than 65 years. Routine preventative care, including a biennial physical exam, an annual dental exam and annual gynecological exam was exempt from fees payments. Median monthly earnings in the Czech Republic were $1163 in 2008. The user fees were not expected to constitute a substantial financial burden to most ˚ tilova´ families, but signal that healthcare is a valued good (Kru and Yaya 2012). The enactment of the fees sparked a wave of popular protests across the country, fueled impassioned political speeches by elected officials and, in some instances, escalated into collective action among healthcare providers who refused to charge the fees. Opponents of the policy argued that the fees would deter the neediest from seeking care and magnify the extant health disparities between the rich and the poor. They pointed to studies of healthcare user fees in middle and low income countries and emphasized their adverse consequences for population health (McIntyre et al. 2006; Ponsar et al. 2011). The promoters of the fees countered with examples of similar Western European provisions for regulating healthcare use that did not result in the widening of health disparities (Chiappori et al. 1998; Cockx and Brasseur 2003). The purpose of this article is to evaluate whether the enactment of the user fees diminished the use of care among older residents of the Czech Republic and if so, whether the observed effect was limited to more socioeconomically vulnerable subpopulations who were burdened by the fees to a greater extent.
CURING OVER-USE BY PRESCRIBING FEES
Analytic strategy Difference-in-differences regression models, a quasi-experimental strategy, were estimated to assess the impact of the fees on healthcare use. This approach is based on the assumption that if no intervention occurred, healthcare utilization would have increased or decreased at the same rate for the intervention group—here the Czechs exposed to a broad scale policy change instituting user fees—as for the control group—Poland, which did not experience such a change. The regression models are set up to test the statistical significance of the deviation from the hypothesized common trend. The choice of a suitable control group is of fundamental importance to this strategy. A comparison of the Czech population and the Polish control group on key sociodemographic characteristics and utilization reports for both survey waves is provided in the Appendix.
First, I examined the impact of the fees’ implementation on the population overall. Then, I evaluated the influence of the fees implementation in a multivariate framework. Logistic regression models were estimated for binary utilization measures (any primary care visit and any hospitalization) and negative binomial regression models conditional on having at least one event were constructed for continuous utilization measures (number of visits to a primary care physician and number of nights spent in a hospital). In order to increase the precision of these models and rule out additional underlying variation that could influence model estimates, I added individual-level controls for sex, mean-centered age, poor or fair self-reported health (SRH) and net worth (asset minus debts). Findings from these models are presented in Table 1 in predicted probabilities of an event and Table 2 in predicted mean numbers of events. Then, I evaluated the impact of the policy on several subpopulations, namely: people from 50 to 64 years and those 65 or older; men and women; those with good, very good or excellent self-reported health status vs those with poor or fair self-reported health status; highly-educated people and those with secondary education or less; and people with medium or high net worth and those with low net worth (designated separately within both groups). Again, all models included controls for sex, mean-centered age, poor or fair selfreported health and net worth. Findings for these models are presented in Table 3 in predicted probabilities of a utilization event and Table 4 in predicted mean numbers of utilization events. To generate population representative results, all findings presented in this article were estimated using the official survey weights created by SHARE along with survey estimation commands in Stata/SE 13.0 (Stata Corporation 2013).
Table 1 Predicted probabilities of any primary care visit and any hospitalization before and after user fees’ introduction Before fees’ introduction Average effects
Czech Republic (2006)
Poland (2006)
Predicted probability of any primary care visit
0.85
0.76
Predicted probability of any hospitalization
0.15
0.17
Difference in probabilities at baseline
0.09*** 0.02
After fees’ introduction Czech Republic (2011)
Poland (2011)
0.84
0.80
0.17
0.19
Difference in probabilities at follow up
0.04*** 0.02y
Difference in percentage-point change (2006–11)
Significance of the difference in differences
0.05
*
0.00
NS
Note: Results of weighted logistic regression models with binary dependent variables. Models controlled for sex, mean-centered age, poor/fair health, and networth. N ¼ 12 304. Difference in percentage-point change is the difference-in-differences estimate of the net policy effect. NS, not statistically significant, y P < 0.1, *P < 0.05, ***P < 0.001.
Table 2 Predicted number of primary care visits and nights spent in a hospital before and after user fees’ introduction Before fees’ introduction Average effects—conditional
No. of primary care visits No. of nights spent in a hospital
Czech Republic (2006)
Poland (2006)
Difference between means at baseline
After fees’ introduction Czech Republic (2011)
Poland (2011)
5.47
6.98
1.51***
4.49
6.58
17.39
15.98
1.41
14.23
14.74
Difference between means at follow up
Difference in differences between means (2006–11)
Significance of the difference in difference
2.09***
0.58
*
0.51
1.92
NS
Note: Results of weighted binomial regression models with continuous dependent variables, conditional on having at least one primary care visit or night spent in a hospital. Models controlled for sex, mean-centered age, poor/fair health, and net-worth. N ¼ 10 302 for primary care visits; N ¼ 2120 for hospital nights. Difference in percentage-point change is the difference-in-differences estimate of the net policy effect. NS, not statistically significant, P < 0.1, *P < 0.05, ***P < 0.001.
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
nights spent in the hospital during the past twelve months (Czech Republic 2006: 2.6, Poland 2006: 2.8; Czech Republic 2011: 2.4, Poland 2011: 2.8). Each survey item listed above measures a separate care use decision; some of these decisions are more likely than others to be affected by the incentive structure of the user fee policy. Although patients have some discretion over whether and how often they visit their primary care physician, the decisions over hospitalization and the number of nights spent in a hospital are more driven by their physician’s recommendations and the severity of their health issues. Physicians may take hospital user fees and cost incurred by their patients into consideration when prescribing care and patients may express a preference for a shorter stay due to cost, but, in general, we would expect hospital care to be impacted by the user fees considerably less than primary care.
425
0.93 0.13 0.20
Predicted probability any hospitalization (50–64)
Predicted probability any hospitalization (65þ)
0.16 0.14
Predicted probability any primary care visit (women)
Predicted probability any hospitalization (men)
Predicted probability any hospitalization (women)
0.91 0.10 0.22
Predicted probability any primary care visit (poor/fair SRH)
Predicted probability any hospitalization (good/very good/excellent SRH)
Predicted probability any hospitalization (poor/fair SRH)
0.86 0.15 0.15
Predicted probability any primary care visit (secondary education or less)
Predicted probability any hospitalization (higher education)
Predicted probability any hospitalization (secondary education or less)
0.79 0.15 0.18
Predicted probability any primary care visit (low)
Predicted probability any hospitalization (medium or high)
Predicted probability any hospitalization (low)
0.19
0.17
0.75
0.77
0.18
0.13
0.77
0.72
0.23
0.09
0.83
0.65
0.18
0.16
0.81
0.70
0.22
0.14
0.81
0.73
Poland (2006)
0.01
0.02
0.04
0.09***
0.03
0.02
0.09***
0.10*
0.01
0.01
0.08***
0.16***
0.04*
0.00
0.06***
0.14***
0.03
0.01
0.12***
0.08***
Difference in probabilities at baseline
0.16
0.17
0.85
0.84
0.17
0.16
0.85
0.81
0.25
0.11
0.90
0.81
0.18
0.16
0.86
0.82
0.22
0.15
0.89
0.81
Czech Republic (2011)
0.22
0.18
0.75
0.82
0.20
0.17
0.80
0.77
0.27
0.09
0.88
0.69
0.18
0.20
0.83
0.76
0.28
0.14
0.86
0.76
Poland (2011)
After fees’ introduction
y
NS NS
0.09 0.02 0.05
0.08 0.03 0.04 0.03
0.01 0.08**
0.06** 0.03y 0.04* 0.00
NS NS
0.06 0.01 0.01
0.07
0.02 0.02 0.02
0.03 0.05**
NS
0.03 0.00
0.07 0.06
0.03
0.02* 0.10***
NS NS
0.01 0.05
0.01 0.05
NS
***
* NS
0.05 0.01
NS
**
0.04
0.12***
NS
NS
*
*** NS
0.03
NS
Significance of the difference in differences
0.03y
Difference in percentage-point change (2006–11)
0.05*
Difference in probabilities at follow up
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
Note: This table shows results of weighted logistic regression models with binary dependent variables stratified by subgroups. Models controlled for sex, mean-centered age, poor/fair health, and net-worth, unless used as a stratifying variable. N ¼ 12 304. Difference in percentage-point change is the difference-in-differences estimate of the net policy effect. Small imprecisions in additions and subtractions due to the rounding of the estimates. NS, not statistically significant, yP < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001.
0.86
Predicted probability any primary care visit (medium or high)
Net worth
0.82
Predicted probability any primary care visit (higher education)
Education
0.81
Predicted probability any primary care visit (good/very good/excellent SRH)
Self-reported health status
0.84 0.87
Predicted probability any primary care visit (men)
Sex
0.81
Predicted probability any primary care visit (65þ)
Czech Republic (2006)
Predicted probability any primary care visit (50–64)
Age
Average effects
Before fees’ introduction
Table 3 Predicted probabilities of any primary care visit and any hospitalization before and after user fees’ introduction stratified by sociodemographic categories
426 HEALTH POLICY AND PLANNING
18.04
Predicted mean of nights spent in a hospital (65þ)
Predicted mean of nights spent in a hospital (women)
20.15
Predicted mean of nights spent in a hospital (poor/fair SRH)
17.03
17.05
15.62
7.33
6.88
1.68**
0.02
1.64
0.76
1.56***
1.46
2.73
1.45***
17.55
13.29
4.93
4.31
14.99
9.07
4.63
3.73
18.03
9.10
3.77y 2.06
5.83
3.47
13.29
15.74
4.71
4.21
16.69
12.28
5.01
4.05
Czech Republic (2011)
18.24
12.76
7.08
6.37
9.96
9.26
6.72
5.63
15.84
9.75
7.40
5.09
13.91
15.40
6.68
6.41
13.77
15.98
7.25
6.06
Poland (2011)
After fees’ introduction
1.21***
0.37
1.35
1.16
1.48***
1.55***
0.60
1.94
1.09**
1.72***
Difference in probabilities at baseline
y
NS
NS *
4.42 0.13
0.22 0.64
0.65 2.19
1.90*** 2.09***
NS
NS y
NS NS
2.54 1.22
0.51 1.39 1.11 0.68
0.19 0.24
2.06*** 2.15*** 0.53 0.70
NS
**
NS
1.97
NS
NS
0.82
0.34 0.62
1.24
NS NS
0.66 0.50
2.21*** 1.97***
0.36
NS 2.32
2.92
1.61***
* NS
5.64
3.70
1.56***
NS
0.29 1.15
Significance of the difference in differences
2.01***
Difference in percentage-point change (2006–11)
2.24***
Difference in probabilities at follow up
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
Note: This table shows results of weighted binomial regression models with continuous dependent variables stratified by subgroups. Models controlled for sex, mean-centered age, poor/fair health, and net-worth, unless used as a stratifying variable N ¼ 10 302 for primary care visits; N ¼ 2120 for hospital nights. Difference in percentage-point change is the difference-indifferences estimate of the net policy effect. Small imprecisions in additions and subtractions due to the rounding of the estimates. NS, not statistically significant, yP < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001.
17.26
Predicted mean of nights spent in a hospital (low)
6.57
Predicted mean of primary care visits (low)
Predicted mean of nights spent in a hospital (medium or high)
5.32
Predicted mean of primary care visits (medium or high)
Net worth
17.23
14.50 17.05
Predicted mean of nights spent in a hospital (higher education)
Predicted mean of nights spent in a hospital (secondary education or less)
18.50
7.07
4.57 5.62
Predicted mean of primary care visits (higher education)
6.25
18.09
8.07
8.16
4.62
15.35
17.18
7.21
6.64
17.44
14.38
7.87
6.29
Poland (2006)
Predicted mean of primary care visits (secondary education or less)
Education
11.84
6.96
Predicted mean of nights spent in a hospital (good/very good/excellent SRH)
4.25
Predicted mean of primary care visits (good/very good/excellent SRH)
Predicted mean of primary care visits (poor/fair SRH)
Self-reported health status
18.34 16.70
Predicted mean of nights spent in a hospital (men)
5.09 5.73
Predicted mean of primary care visits (men)
Predicted mean of primary care visits (women)
Sex
16.31
6.78
Predicted mean of nights spent in a hospital (50–64)
4.57
Predicted mean of primary care visits (50–64)
Czech Republic (2006)
Predicted mean of primary care visits (65þ)
Age
Average effects
Before fees’ introduction
Table 4 Predicted number of primary care visits and nights spent in a hospital before and after user fees’ introduction stratified by sociodemographic categories
CURING OVER-USE BY PRESCRIBING FEES
427
428
HEALTH POLICY AND PLANNING
Results
(a) 100% 95%
% Reporting a Primary Care Visit
30%
Observed Poland Observed Czech Republic
25%
Expected Czech Republic 89%
90%
(b)
% Reporting a Hospitalization Observed Poland Observed Czech Republic Expected Czech Republic 19%
20% 17%
85%
85%
80% 75%
84%
15%
80%
10%
0%
2006
7.5
15%
5%
76%
70%
(c)
17%
Conditional Number of Primary Care Visits
(d)
2011
Conditional Number of Nights Spent in a Hospital
18.0
7.0
7.0
2006
2011
6.5
17.0 17.0
6.5 16.0
6.0
16.2
5.5 5.5 5.0 4.5 4.0
15.4 5.0
15.0 Observed Poland
Observed Poland Observed Czech Republic
14.0 4.5
Expected Czech Republic
14.6 14.3
Observed Czech Republic 13.0
3.5 12.0
3.0 2006
2011
Expected Czech Republic 2006
2011
Figure 1 Expected utilization shows differences between Poland and the Czech Republic, if their differences remained unchanged after policy implementation. Number of primary care visits and nights hospitalized conditional on having at least one visit or hospitalization.
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
Figure 1a presents the unadjusted percentage of the population 50 years or older in both Czech Republic and Poland that reported a primary care visit in 2006 and 2011, as well as the expected prevalence in 2011 if no policy change took place in the Czech Republic and the difference between Czech healthcare use and Poland healthcare remained constant. In 2006, 85% of the Czech population reported having seen a primary care physician compared with 76% of the Poles in the sample. In 2011, we find that although the percentage of the Czech population with a visit decreased to 84%, the percentage of the control group reporting a visit increased to 80%. Had the prevalence of visits among the Czechs mirrored the increase in the Polish population, we would have anticipated 89% of them reporting a primary care visit. Figure 1b presents the unadjusted percentages of the Czech and Polish populations who reported being hospitalized in 2006 and 2011. The Polish control group reported a hospitalization more often than the Czechs in 2006 (17% vs 15%), and the rate of hospitalizations increased at about the same rate in the control group compared with the Czech Republic in 2011 (19% vs 17%).
Figure 1c shows the changes in the unadjusted conditional mean number of primary care visits between 2006 and 2011. The conditional number of visits at the first time point was lower in the Czech Republic than in the Polish control (5.5 vs 7.0). At the second time point, the conditional number of visits decreased in the Czech Republic to 4.5 and in Poland to 6.5. If the observed prevalence at the second time point in the Czech Republic was consistent with rate of change in the control group, we would have expected to find a less steep decrease, to 5.0 visits. Figure 1d displays the unadjusted conditional mean numbers of nights spent in a hospital. The observed mean number of nights spent in a hospital was lower in Poland than in the Czech Republic in 2006 (16.2 vs 17.0) but higher in 2011 (14.6 vs 14.3). Had the number of nights spent in a hospital decreased at the same rate in the Czech Republic as in the Poland, we would anticipate seeing a less steep decrease, to a mean of 15.4 nights. Table 1 shows the predicted probabilities of any primary care visit in each year obtained from logistic regression models with survey weights and controls. The difference in probabilities column refers to the amount of change in the difference between Czech Republic and Poland in 2006 and in 2011. For example, the predicted probability of any primary care visit for
CURING OVER-USE BY PRESCRIBING FEES
be valid, we must assume that Poland is a good comparator to the Czech Republic. It is possible that the change in healthcare utilization of the Polish sample was not typical, and that it was driven by other changes that took place in Poland during the study period. As a test of the sensitivity of the results to the choice of the control country, I conducted similar analysis with all other countries included in the SHARE II and IV dataset. Overall, the trends I observed were similar to those reported here. These additional results are available on request. Second, relying on only two data points may introduce inaccurate understanding of the policy effect, if these two time points were not reflective of the prior trend in healthcare use. Third, SHARE is a survey of adults age 50 or older and their partners. This is both an advantage and a disadvantage. Older people are the most frequent users of healthcare. Therefore, a policy aimed at reducing the number of contacts with the healthcare system should have the strongest effects on this population. However, it could also be the case that their high care demand will make them less responsive to any price changes than other age groups. The results may therefore not be generalizable to the remainder of the population, especially because people older than 65 are subject to lower maximum payments. However, some of these concerns may be alleviated by the age-stratified analysis presented among the results. Moreover, the findings are based on self-reported data of healthcare use. It is possible that these reports may not always be accurate, especially when the number of primary care visits or nights spent in a hospital is high (Bhandari and Wagner 2006). However, it is unlikely that patients’ tendency to under-report or over-report utilization would change between 2006 and 2011. Consequently these data are suitable for the analysis of change that is presented here, but may be less accurate when used for drawing conclusions about the precise amount of healthcare used. An additional concern for any analysis of healthcare use is that it is generally subject to ceiling and floor effects. Assuming an increase or a decrease following a policy change may not always be realistic if, for example, a proportion of the population will not see a primary care provider under any circumstances. Some, but not all, of these concerns should be lessened by using a non-linear modelling strategy. An important matter to consider when interpreting the results presented here is the likely increase of the general level of economic hardship in the population due to the onset of the Great Recession that coincided with the enactment of the fees. The impact of the policy may have been compounded by the unfavourable economic climate, and the number of visits will return to its pre-recession levels once the economy recovers. I recommend that the findings presented here are re-examined at a later date, with data collected during a less economically troubled period. Finally, the analyses presented in this article offer no insight into whether the policy change had any effect on population health. A much longer time-frame will be necessary to evaluate the potential health effects of this policy.
Limitations Although the SHARE data provide a unique opportunity to examine the influence of the new fee policy in the first years after its enactment, the results of this analysis should be interpreted with some caveats. First, in order for these results to
Discussion The purpose of this study was to evaluate whether the enactment of the user fees diminished the use of care as the policy intended,
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
Czechs was 0.85 and for Poles was 0.76 in 2006, reflecting a 0.09 lower predicted probability for Poland. The predicted probability of having had a primary care visit in 2011 was 0.84 for Czechs and 0.80 for Poles in 2011; the difference was 0.04. The difference in differences between the two time points was therefore 0.05 and statistically significant. This suggests that the implementation of the fees influenced the overall likelihood of seeking care. However, there appear to be no similar effects on the predicted probability of hospitalization. The probability of hospitalization increased at the same rate in both countries and therefore appears to not have been affected by the policy. Table 2 presents the predicted conditional mean number of primary care visits and nights spent in a hospital obtained from negative binomial regression models with survey weights and controls. The difference in differences between means refers to the absolute magnitude of change in the differences between the two groups in 2006 and 2011. I observe a significant change in difference for the number of primary care visits. Czechs had a significantly lower predicted mean number of visits in 2006 than that of the Polish control group in 2006 (5.47 vs 6.98), and this difference became larger over time. In 2011, the predicted mean number of visits for Czechs dropped to 4.49 whereas it dropped only slightly in Poland to 6.58. The overall difference in differences, 0.58, is statistically significant. It thus appears that the user fees had the intended effect of lowering the number of primary care visits. At the same time, the difference between the numbers of nights spent in a hospital in the two countries at the two time points was not statistically significant. Table 3 displays predicted probabilities from stratified logistic difference-in-differences models with survey weights and controls. I find a significant decrease in the predicted probability of having any primary care visit for those who were 65 or older, for men, for people who finished secondary education or less, and people with medium or higher net worth. The probability of a primary care visit for people with poor or fair health remained approximately stagnant in the Czech Republic (0.91 in 2006 and 0.90 in 2011), but increased in Poland from 0.83 to 0.88, and the resulting difference is statistically significant. There was only one marginally statistically significant difference in the probability of hospitalization: although Czech men’s probability increased slightly from 0.20 to 0.22 over time, the probability of hospitalization for Polish men increased from 0.22 to 0.28 during the same time period. Table 4 presents results of stratified negative binomial regression models predicting mean number of visits and nights spent in a hospital for each subpopulation in 2006 and 2011. I observe a significant decrease in the number of visits for Czechs older than 65, people with good self-reported health, and those with secondary education or less. For number of nights spent in a hospital, the only marginally statistically significant decrease can be seen in the healthier group, where the predicted number of nights dropped in the Czech Republic, while it increased in Poland.
429
430
HEALTH POLICY AND PLANNING
It has been well documented that the perceived cost of healthcare and demand for it are tightly bound to one other. In some circumstances, an unlimited access to a valuable good, such as care, may lead to the distortion of its real price in the minds of the recipients, and possibly overuse (Arrow 1963; Manning et al. 1987). From this perspective, instituting healthcare user fees should improve allocative efficiency of the system by reducing the moral hazard associated with the availability of free care (Winkelmann 2004; Culyer 1971). The analysis presented in this article suggests that the enactment of the user fees created effects consistent with their intentions. Other neighbouring countries of the Czech Republic have also experimented with instituting user fees, but they met with insurmountable public and political opposition. In Hungary, small user fees were introduced in 2007 and abolished just one year later following a public referendum (Baji et al. 2011). In Slovakia, healthcare user fees were instituted in 2003, but most of them were abolished after a change in governmental leadership in 2006 (Szalay et al. 2011). Similarly, the newly elected (autumn 2013) leading Czech political coalition has taken a strong stand on the healthcare user fees and made their abolishment one of its main goals in the eyes of the public (Kopecky 2013). In all three countries, the driving political argument behind the policy reversal has been protecting the disadvantaged. Although this article fully agrees with the intention of providing maximum equity and access to healthcare to all people, further studies are needed exploring specifically who is at most risk of losing access to care on account of the user fees. Such research would also have the potential of helping us erect additional safety nets for those who are identified as most disadvantaged. In light of the high healthcare use found in the Czech Republic and the relatively low level of resources devoted to healthcare in this country, the findings presented in this article suggest that implementing and maintaining a strong safety net rather than the wholesale abolishment of the user fees appears to be the more promising policy direction.
Acknowledgements This article uses data from SHARE wave 4 release 1, as of 30 November 2012 and SHARE wave 1 and 2 release 2.5.0, as of 24 May 2011. The SHARE data collection has been primarily funded by the European Commission through the Fifth Framework Program (project QLK6-CT-2001-00360 in the thematic program Quality of Life), through the Sixth Framework Program (projects SHARE-I3, RII-CT-2006-062193, COMPARE, CIT5-CT-2005-028857, and SHARELIFE, CIT4-CT2006-028812) and through the Seventh Framework Program (SHARE-PREP, No. 211909, SHARE-LEAP, No. 227822 and SHARE M4, No. 261982). Additional funding from the U.S. National Institute on Aging (U01 AG09740-13S2, P01 AG005842, P01 AG08291, P30 AG12815, R21 AG025169, Y1AG-4553-01, IAG BSR06-11 and OGHA 04-064) and the German Ministry of Education and Research as well as from various national sources is gratefully acknowledged (see www. share-project.org for a full list of funding institutions). Conflict of interest statement. None declared.
Downloaded from http://heapol.oxfordjournals.org/ at University of Lethbridge on November 14, 2015
and if so, whether this happened at the expense of the more vulnerable subpopulations. The results show that the overall frequency of visits to primary care physicians decreased after the fees were implemented, as did the likelihood of having any primary care visit, but the probability of any hospitalization, and number of nights spent in a hospital did not. This finding aligns with the assumption that primary care visits are determined by the consumers of care to a greater degree than other types of healthcare use, and are therefore more sensitive to the newly imposed cost. It is important to add that although the decrease in the number of visits can be lauded as a success of the policy, it is less clear whether the decrease in the proportion of the older population who report having seen their primary care physician at least once last year can be described as an entirely positive development. Those 65 and older became significantly less likely to have at least one primary care visit after the policy implementation, and reported cutting back more on the number of primary care visits than the younger group, although they still have more primary care visits overall. This could be indicative either of the greater amount of initial unnecessary use by the older group or of the greater burden placed on them by the fees. One would generally expect that most people in the older age group would have at least one primary care visit a year, and a decrease in the probability of having seen a primary care provider is not necessarily a positive development in this age group. We see a similar picture when examining the extent to which the policy affected the more- and less-educated groups. The introduction of fees was associated with a reduction in number of visits among older Czech citizens with secondary education or less. This difference could again signify either that an unnecessary number of visits was more common among the less-educated group, and they were therefore able to cut back more, or, that the burden placed on this group is larger. However, further investigation will be required to establish whether the observed steeper decline points to emerging health access disparities along educational strata or a signal of convergence of utilization at a more appropriate rate. As hoped for by the designers of the policy, the results showed no significant stratifying effect of having low net worth on any of the outcomes studied. Because no clear and consistent differences in the decrease in utilization between the more and less advantaged were uncovered, this article highlights the importance of the symbolic meaning of the user fees. However, the strength of this conclusion would have been substantially enhanced by a finer measure of income and wealth. Further work is also needed that would be able to distinguish between those who were in material need, but were not exempt from fees payments, and those who were, and consider whether fees became an obstacle to care seeking for the first group. Moreover, I was not able to investigate what types of care exactly were reduced. Evidence from other countries, such as the American RAND Health Insurance Experiment, shows that cost-sharing reduces both unnecessary and necessary use of care (Keeler 1992). Although their findings may not be applicable to the Czech Republic where out of pocket costs of care are generally low,
