Quantifying the contribution of changes in healthcare expenditures and smoking to the reversal of the trend in life expectancy in the Netherlands

Since we included only high-income countries with well-established systems of national statistics that were members of the OECD for at least as long as the study period, we are confident about the comparability and quality of the data in general. Missing information that often casts doubt on international comparison was not an issue, since our sample had almost complete information on the study variables [18]. Data on all-cause mortality and population exposure, as well as healthcare expenditures and GDP, were obtained from harmonized databases. Nevertheless, some caveats apply for certain aspects of the data. The implementation of the OECD system of health accounts framework underlying the data on healthcare expenditures varied among countries and over time limiting the comparability. Thereby, differences in the definition of costs for long-term care were identified as most important issue affecting the share of HCE to GDP up to 1 percentage points [21]. To cope with such inconsistencies, we used different statistical models and sample compositions, suggesting that our results are generally robust against these inconsistencies (see Additional file 1). Information on lung cancer counts is (as is the case with all cause-of-death specific data) subject to uncertainty due to variations in national coding practices and changes in coding behavior over time [22]. However, these issues are of minor relevance for lung cancer data because detection is relatively clear and coding schemes have been established for a long time without drastic changes [23].

The most important part of our study is the quantification of the effect of changes in healthcare expenditures on changes in life expectancy. Although there is a longer tradition of assessing this relationship in empirical research, there is no consensus on the appropriate strategy to estimate it [16]. We believe this is largely due to the absence of appropriate tools to analyze panel data, that became available only a few years ago [18, 24]. A main insight of this new literature is that the relation between healthcare expenditures and life expectancy can be estimated more reliably for a group of countries than for an individual country alone [25]. However, this comes at the cost of providing only “insights regarding the central tendency of the panel” so that we had to apply the same parameter for every country in the main part of our analysis to compute the contribution of healthcare expenditures on life expectancy [26]. Nevertheless, this assumption seems plausible for the countries in our sample, as they share similar political and economic structures, and all provide almost full public coverage of basic healthcare services [27]. If there were substantial differences in the efficiency of the production of health between the Netherlands and the other 18 countries, this would be likely within the relatively large confidence intervals around our main results representing the large uncertainty about the effect of changes in healthcare expenditures.

In our analysis, we assumed that variations in healthcare spending causally explain variations in life expectancy ruling out the opposite direction, i.e., that improvements in life expectancy cause additional costs in the health sector. A study explicitly testing the influences of such reverse causality found that improvements in most health outcomes did not led to higher healthcare costs [28]. Furthermore, it is reported that with postponing death also costs are postponed, deeming increases in life expectancy to be less relevant for growing healthcare expenditures [29]. An exception is spending on long-term care that is to a large extent driven by population ageing [30]. Evaluating the influence of long-term care on healthcare expenditures on our estimates is hardly possible since this aspect of the OECD data is particularly imprecisely measured, as explained above.

Regarding the effect of smoking, we applied a well-established tool that has proven reliable and informative in numerous applications [31, 32]. Although the indirect modeling of the damage from smoking on other causes-of-death partly relies on a set of assumptions, different approaches with different assumption arrived at very similar estimates [20, 33]. Furthermore, the estimated smoking-attributable fractions plausibly describe the variation of the timing in the epidemiological transitions between countries and sexes [32, 34].

An important assumption of our study is that the contribution of healthcare expenditures and smoking on trends in life expectancy could be quantified independently of each other. This might be a simplistic view, given that a well-funded healthcare system certainly mitigates the consequences of smoking, particularly for smoking-induced cardiovascular diseases for which effective treatments exist. However, due to the lack of evidence on that topic, it is difficult to speculate on the impact of our assumption on the study results.

The results of this study underpin the hypothesis that changes in healthcare expenditures were the main driver of the trend reversal of life expectancy in the Netherlands [8]. While previous analyses only demonstrated the presence of a common trend break for these two variables at around 2002, our study shows that: 1) changes in healthcare spending were generally positively associated with changes in life expectancy within high-income countries, 2) the size of the impact of healthcare could plausibly explain the acceleration of the Dutch life expectancy increase, 3) the exceptionally large changes in healthcare spending and the trend reversal in life expectancy at around 2002 was a particular feature of the Netherlands, and 4) that the major alternative explanation for the trend reversal, i.e., changes in smoking, could be ruled out. These four aspects explaining the Dutch trend reversal counter arguments stating: “that there is no observable relationship with changes healthcare funding whatsoever” [10].

The importance of the contribution of healthcare expenditures to the trend reversal in Dutch life expectancy is in line with case studies of other countries with rapid trend reversals in life expectancy. The natural experiment of the separation and subsequent unification of Germany demonstrated that improvements of the healthcare infrastructure could affect life expectancy immediately and with a large impact [35]. In Denmark, a huge investment program to reduce cardiovascular mortality was held partially responsible for the upturn in Danish life expectancy after a longer period of stagnation [36]. The same was noted for the case of Ireland, where particularly cardiovascular mortality suddenly declined after access to primary care and pharmaceuticals was facilitated in 2001 [37]. The results of our main analysis confirm the findings for Denmark and Ireland that increases in healthcare expenditures in these countries contributed to the accelerations of the increase in life expectancy (see Additional file 1).

Nevertheless, the example of Japan shows that a large increase in life expectancy can be achieved without a large increase in healthcare spending, while the example of the USA shows that substantial investments in healthcare spending do not necessarily lead to large improvements in life expectancy [38, 39]. This highlights the relevance of contextual factors, such as a universal coverage of healthcare services or more general cultural aspects. Clearly, investments in healthcare systems have a larger impact on life expectancy in countries where costly but also effective medical treatments were held back.

Our results attributed a considerable part of the increase in Dutch life expectancy to changes in Dutch healthcare expenditures, i.e., about 47 % for Dutch women and about 30 % for Dutch men in 2000–2009. Studies that combined knowledge on the effect of new medical treatments with cause-specific mortality data and disease prevalence, estimated that innovations in healthcare have contributed to at least 50 % of the gains in life expectancy during recent decades, which is in line with our estimates [40].

A major way in which variations in Dutch healthcare expenditures could have affected mortality, is the budgeting of hospital care. During the early 1980s, at the time the improvements in Dutch life expectancy started to slow down, policymakers introduced fixed hospital budgets resulting in a considerable reduction in the admission of new patients, of employed personnel and even in the closure of some hospitals [41, 42]. To cope with the budgeting of their resources hospitals invested less in new medical technologies and limited the volume of treatment for their patients [42]. Together with the legalization of euthanasia in 1985 and an increasing incidence of end-of-life decisions (e.g., the withdrawal of artificial nutrition), this reflected a general attitude towards a less aggressive treatment of older and terminally-ill patients [7]. Consequently, the dramatic improvements in survival due to innovations in the treatment of cardiovascular diseases observed in many Western countries particularly at older ages, where probably not fully realized in the Netherlands where mortality rates even increased for some age groups [3, 4].

At the end of the 1990s, complaints about excessive waiting times for elective surgeries piled up, so that policymakers abolished the fixed hospital budgets at the end of 2000 and replaced them with activity-based funding [43]. As a result hospital admissions, treatments, and pharmaceutical prescriptions increased rapidly, particularly among the elderly [8]. Thereby, effective life-saving treatments, e.g., lipid- and blood pressure lowering drugs, were applied more often and prescribed to older patients than before the reform [8, 44]. This more active medical treatment potentially explains the reversal in mortality trends from diabetes, stroke, pneumonia and symptoms and ill-defined conditions [8]. That the change in healthcare expenditures was not merely driven by higher prices of care was also confirmed by earlier analyses that ascribed about two-thirds of the increase to an extension of the volume of care [45].

The precise causal mechanisms how more spending translated into better survival in the Netherlands need to be understood in more detail. Recent studies revealed that particularly the survival of persons with more severe chronic conditions improved during the trend reversal of Dutch life expectancy, which indirectly confirms a possible impact of improved healthcare [46, 47]. However, so far the decline in mortality could not be linked to healthcare utilization at the individual level [46].

Although the impact of smoking did not explain the trend reversal in Dutch mortality it did affect trends in life expectancy. During the study period, the lower gains in life expectancy for Dutch women, as compared to Dutch men, were to a large extent caused by smoking; this is in line with the general theory of the smoking transition through which women progress with a delay of several decades as compared to men [48]. It has been reported that, after the exclusion of smoking-related causes of death, the stagnation in mortality decline occurred at the same time for men and women [7]. Our study results show that also the resumption of Dutch mortality decline after 2001 was similar for men and women, when the differential impact of smoking was accounted for. This concordance of sex-specific life expectancy trends before and after the Dutch trend reversal calls for a more general explanation that affected all Dutch inhabitants in the same way; we believe we have identified this in the form of financing of healthcare. No other plausible alternative explanation, that could have affected men and women in the same way, is available.