اروپا و آزمایش بزرگ بیمه سلامت اجتماعی پست کمونیستی آسیای مرکزی: اثرات بهم پیوسته بر نتایج بخش سلامت

The post-Communist transition to social health insurance in many of the Central and Eastern European and Central Asian countries provides a unique opportunity to try to answer some of the unresolved issues in the debate over the relative merits of social health insurance and tax-financed health systems. This paper employs regression-based generalizations of the difference-in-differences method on panel data from 28 countries for the period 1990–2004. We find that, controlling for any concurrent provider payment reforms, adoption of social health insurance increased national health spending and hospital activity rates, but did not lead to better health outcomes.

All but two of the OECD's 30 countries—Mexico and the United States—finance the majority of their health spending publicly, with half operating broad-based ‘tax-financed’ health systems (e.g. Canada and the United Kingdom) and half operating payroll-based ‘social health insurance’ (SHI) systems (e.g. Germany and Japan).1 Outside the OECD, the fraction of countries financing the majority of their health spending publicly is smaller (56%), and only one fifth of these countries finance the majority of their government spending through SHI.2 The relative merits of SHI and tax finance is an old debate, but one that has recently resurfaced. In part this is due to the fact that three of the world's oldest SHI countries—France, Germany and the Netherlands—are in the process of reducing their reliance on payroll contributions in favor of a broader financing base.3 But the renewed interest in the SHI vs. tax-finance debate also stems from the current interest in SHI in the developing world.4 Many developing countries that have relied largely on general revenues (and out-of-pocket payments) to finance their health systems have introduced SHI, or are thinking about doing so.5 And countries that have a fledgling SHI scheme in place are redoubling their efforts to expand its reach, especially to the informal sector.6 Despite the topicality and vibrancy of the SHI vs. tax-finance debate, the evidence base is surprisingly thin. Some comparisons have been undertaken, especially on distributional issues: payments for health care tend to be more progressive or less regressive in tax-financed systems than in SHI systems; and tax-financed systems seem to be more successful at ensuring universal coverage within a single health system.7,8 But on aggregate system-wide differences, there appears to be no rigorous evidence. We do not know whether SHI systems spend more on health care, and if they do whether this translates into higher levels of throughput and better health outcomes. Getting at these questions through a cross-country econometric analysis where some systems are financed through SHI contributions and others are financed through general revenues would be problematic because there are likely to be unobservable variables that would be correlated with the type of financing system in place and the outcomes of interest (i.e. SHI is likely to be endogenous). A more promising strategy would be to look for changes in the way countries finance their health care, exploiting the variations in changes across countries to eliminate (time-invariant) unobservable variables. The difficulty with this approach is that in the group of countries that have the best data (those in the OECD), there have been very few switches between the SHI and tax-financed camps (six “old” OECD countries abandoned SHI in the 1970s and 1980s, notably Denmark, Greece, Iceland, Italy, Portugal and Spain) and the transitions occurred some time ago, so the available data are very limited. This paper looks instead to a (mostly) different group of countries where transitions have occurred with greater frequency and more recently, namely the countries of (central and eastern) Europe and Central Asia (ECA).9 Of the 28 ECA countries, 14 abandoned tax-finance and adopted SHI at some stage between 1990 and 2004 (and 4 other countries had adopted SHI prior to 1990). These countries are also data-rich countries, having inherited and largely maintained the Communist tradition of extensive data-gathering, and falling under the most data-rich regional office of the World Health Organization.10 One dimension in which the database we have been able to assemble is especially rich is health outcomes; we have been able to assemble extensive information on mortality and disease incidence by disease. The fact that a sizeable fraction (perhaps 70–80%) of mortality is not amenable to medical care (cf. Nolte and McKee, 2008) probably helps explain why many cross-country regression studies have been unable to find a strong relationship between health spending and health outcomes (cf. e.g. Martin et al., 2008). The same fact might—in the absence of disease-specific mortality data—have made it hard for us to credibly establish whether, by increasing health spending or by raising the efficiency of health spending, countries that switched to SHI have been able to improve health outcomes. 11 The ECA health financing experiment thus affords a valuable “laboratory” to try to shed light on the question of how SHI systems fare vis-à-vis tax-financed systems in spending, throughput and health outcomes. To shed light on these issues, we use regression-based generalizations of the differences-in-differences (DID) method, with data from (up to) 28 countries for 15 years (1990–2004). We pay particular attention to the issue of the possible endogeneity of SHI, since it seems likely that there may be events that occurred around the time SHI was introduced that we implicitly lump into our error term but which may affect outcomes. We employ three different approaches to allowing for this possible endogeneity. The first is a simple individual-specific effects model estimated along the lines of the classic DID model. This allows for the endogeneity of SHI only insofar as the unobservables that are correlated with SHI adoption and with our outcomes are time-invariant. This is the parallel trends assumption that is often considered the Achilles heel of the DID approach (cf. e.g. Blundell and Costa Dias, 2000). Because our database spans a relatively long period of time, we can explore two more flexible—and more robust—approaches to controlling for the potential endogeneity of SHI. The first is a random (linear) trend model: this allows for a country-specific unobserved linear time trend whose growth rate could be correlated with SHI status (i.e. whether the country operates a SHI system in the year in question). The second is a differential trend model: this allows SHI and tax-financed systems to have different trends in unobservables that are not necessarily linear but do depend only on SHI status. This is not the first paper to employ the random trend regression model.12 But it is—to our knowledge—the first to propose and employ a regression version of the differential trend generalization of the DID model.13 We are able, using the two generalizations of the DID approach, to shed light empirically on the validity of the parallel trends assumption. In the event, we find that for most outcomes the data are reasonably consistent with the assumption. We also test for reverse causality in all three models, and find little evidence of it. The organization of the paper is as follows. Section 2 provides a brief history of the SHI reforms in the post-Communist ECA region and discusses the hypothesized effects of SHI adoption on health spending, throughput and health outcomes. Section 3 outlines our methods, Section 4 our data, and Section 5 our empirical results. Section 6 presents our conclusions.

The health system reforms the European and central Asian (ECA) countries implemented during their transition from socialist economies in the 1990s provide a unique opportunity to assess the impacts of social health insurance (SHI) on the health sector. We took advantage of this highly unusual “experiment” in which many ECA countries unequivocally switched from general tax-funded to SHI systems in a relatively short period of time, and on a staggered basis, so as to shed light on a broad set of currently unanswered questions: how does SHI affect national health spending, the way such resources are spent, and population health outcomes? In order to obtain empirical evidence on these issues, we have used regression-based generalizations of the differences-in-differences approach on panel data from 28 ECA countries for the period 1990–2004. In two of our generalizations we relaxed the parallel trends assumption that is seen as a major drawback of the differences-in-differences approach: in one we allow for a country-specific unobserved linear time trend that could be correlated with SHI adoption; and in the other we allow for differential (possibly nonlinear) time trends between SHI adopters and non-adopters. Our tests suggest that the parallel trends assumption is not, in fact, inconsistent with our data. We also find that whichever model we use there is no evidence of reverse causality—SHI adoption being caused by changes in our outcomes. Our estimated SHI impacts are also similar for our three models. Our tests and parameter constancy provide some reassurance that we have identified causal relationships between SHI adoption and health sector outcomes. Our estimates suggest that SHI adoption per se increased government health expenditure per capita. We also obtain some evidence that part of the extra financial resources available in the health sector due to SHI adoption have served to increase the fraction of salaries as percentage of government health spending in SHI countries. This result provides quantitative evidence in support of claims about the process of transition to SHI in some ECA countries being favored and accelerated by pressure from health professionals, who expected to have their income levels driven up by the introduction of a SHI system. 34 Our results also suggest that SHI has impacted on how physical resources are used, by reducing average hospital length of stay and increasing bed occupancy rates and hospital admissions. Despite this, our analysis of several mortality and morbidity indicators showed that transition to SHI has not caused general improvements in health outcomes for ECA countries. This is despite the fact that we have been able to analyze SHI impacts on over 40 health outcome measures, including cause-specific mortality and morbidity indicators. 35 It might be argued that the absence of any beneficial impacts of SHI adoption—with its associated increase in health spending—on general mortality is not surprising, since for many mortality causes it is not reasonable to expect death to be averted by timely or higher quality health care after the condition develops (e.g. many types of malignant neoplasms, heart and circulatory diseases). However, the wide range of mortality measures examined here means that we are not restricted to examining only such “unavoidable” deaths; rather, we have been able to show that SHI adoption did not cause any general improvements in mortality from causes that should not occur in the presence of timely and effective/better quality health care, a concept known as “amenable mortality” which has been used elsewhere to assess the quality of health care systems ( Nolte and McKee, 2008). Compared to tax-funded health systems, SHI systems do not seem to have reduced amenable mortality when this is measured by a variety of mortality indicators containing an important “avoidable” or preventive component for most age ranges, such as standardized death rates by tuberculosis, female breast cancer, cerebrovascular diseases, diabetes, ischemic heart disease, alcohol-related causes and the maternal mortality ratio; the same general conclusion arises as far as children-specific amenable mortality is concerned (judged by measures such as death rates by diarrhea and acute respiratory infections). Thus, in this study, we have been able to perform an investigation of the broad population health impacts that could in theory be brought about by a different organization of a national health system—i.e. the adoption of social insurance—taking into account both morbidity and amenable mortality indicators. Our results are mostly robust to the inclusion of dummy variables capturing shifts in provider payment methods alongside the SHI status dummy; they are therefore pure SHI effects, not provider-payment reform effects. For example, the higher government spending caused by a transition to SHI is not a spurious result attributable to the fact that some countries switched to fee-for-service when they adopted SHI. We are able to estimate separate provider payment effects because SHI adoption did not always lead to provider payment reform and even when it did sometimes did so with a lag, because some non-SHI countries reformed the way they paid hospitals as well, and because some SHI countries switched provider payment methods more than once (some, for example, switched to fee-for-service only to change to a patient-based payment method later on). The question arises: Why did health outcomes not improve as a result of SHI adoption even though it led of itself to higher government health spending and higher inpatient admissions? One might be tempted to explain our results in terms of lack of statistical power. From Table 1 it might appear that this is not a major issue. For example, for total health expenditures, we have 186 observations in the ‘treatment group’ (SHI is in place) and 173 in the ‘control group’ (a tax-financed system). The problem is that our observations are clustered at the country level, and we have a relatively small number of countries (28). It is possible that our effective number of observations (i.e. allowing for intra-cluster correlation) may be too small for us to detect effects that in reality exist. So, some of the insignificant results we obtain could in reality be significant effects. The opposite is not true—the significant effects we obtain, despite low power, are (assuming the model is specified and estimated correctly) real effects. Why might we fail to detect significant effects for some variables and not others (when in reality effects do exist)? The obvious explanation is differences in standardized effect size, i.e. the effect size standardized by the standard error of the parameter in question (which depends, of course, on the standard deviation and sample size). After all, the other key elements of a power calculation—sample size, which has a direct effect on power in addition to its indirect effect via the standard error, and the significance level chosen—are much the same across models (the sample size does vary but only marginally). With limited power, then, we may be able to detect significant impacts where the impacts are large but not where they are small, and with a larger sample size we might have been able to determine with greater certainty whether the effects of SHI on variables such as the standardized mortality rate and the infant mortality rate are indeed significantly different from zero. But the fact remains that their estimated effect is still likely to be small. Lack of statistical power helps us explain the preponderance of high probability values on the SHI coefficient. But it does not help explain the small estimated impacts on health outcomes; these still need explaining. One reason for the small impacts of SHI on health outcomes could be that the percentage increase in admissions due to SHI is much smaller than the percentage increase in spending (3% compared to point estimates of 11% and 15% for total and government health expenditures, respectively). Much of the extra spending caused by SHI adoption would appear, therefore, to have resulted in more costly admissions and/or extra spending elsewhere in the health system. Part of the story seems to be the higher salary share of costs as a result of SHI adoption. But it also seems likely that costs were incurred undertaking new activities (e.g. collecting contributions, writing contracts with providers) or that existing activities became more costly (e.g. more tests being administered on in-patients, more expensive drugs being given, etc.). It is also possible that SHI adoption may have resulted in less comprehensive and less well integrated public health and prevention programs (cf. e.g. Allin et al., 2004), and that the extra admissions and extra costs caused by the transition to SHI were incurred in treating additional patients who would not have otherwise become sick. The fact that SHI adoption appears to have led to increased numbers of infectious disease hospital discharges is consistent with this story. Gaps in coverage may also be part of the explanation. Some groups seem to have fallen through the coverage net, such as the Roma population (cf. e.g. Rechel and McKee, 2003), and there is anecdotal evidence that some formal sector workers wait to enroll until they get sick. Because of lack of coverage, these groups may use primary care less than they would have otherwise done, increasing the likelihood that illness is left untreated until serious enough to warrant hospitalization. Some of the extra hospital caseload associated with SHI may therefore simply be due to people waiting until they get so sick that they require hospitalization. Of course, our results do not necessarily imply that SHI adoption everywhere must necessarily raise health spending without improving health outcomes. These results are likely to hinge in part on the fact that SHI was introduced with costly institutional reforms but ones that did little to stimulate the performance of the health system. Nonetheless, the largely negative results in the paper ought to serve as a warning to those contemplating shifting from general revenue finance to SHI.