You are here

  • Home
  • Archive
  • Volume 69, Issue 6
  • A closer look at the role of healthcare in the recent mortality decline in the Netherlands: results of a record linkage study

Article Text

Article menu

Download PDFPDF

Research report
A closer look at the role of healthcare in the recent mortality decline in the Netherlands: results of a record linkage study
  1. F Peters,
  2. W J Nusselder,
  3. J P Mackenbach
  1. Department of Public Health, Erasmus MC, Rotterdam, The Netherlands
  1. Correspondence to F Peters, Department of Public Health, Erasmus MC, P.O. Box 2040, Rotterdam 3000 CA, The Netherlands; f.peters{at}erasmusmc.nl

Abstract

Background Since 2002, Dutch mortality rates decreased rapidly after decades of stagnation. On the basis of indirect evidence, previous research has suggested that this decline was due to a sudden expansion of healthcare. We tested two corollaries of this hypothesis—first, that the decline was concentrated among those with ill-health and second, that the decline can be statistically accounted for by increases in healthcare utilisation.

Methods We linked the Dutch health interview survey to the mortality register and constructed two cohorts, consisting of 7691 persons interviewed in 2001/2002 and 8362 persons interviewed in 2007/2008, each with a 5-year mortality follow-up (659 deaths in total). The change in mortality between both cohorts was computed using Cox proportional hazard models. We estimated the change in mortality by severity of chronic conditions and with respect to the inclusion of indicators of healthcare utilisation.

Results Between the two study cohorts, mortality declined by 15% (95% CI 2% to 29%), and mortality reduction was greatest for those suffering from fatal and non-fatal conditions with a decline of 58% (95% CI 35% to 78%). Even after adjustment for health status and risk factors, most indicators of healthcare utilisation were associated with higher instead of lower mortality and changes in healthcare utilisation did not explain the decline in mortality.

Conclusions Our results only partly confirm the hypothesis that an expansion of healthcare explains the recent mortality decline in the Netherlands. Owing to confounding by health status, it is difficult to reproduce the mortality-lowering effects of healthcare utilisation of individual level studies in the open population.

  • MORTALITY
  • HEALTH SERVICES
  • HEALTH POLICY

https://doi.org/10.1136/jech-2014-204905

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    Introduction

    Mortality rates at advanced ages improved dramatically during the past decades in most high-income countries.1 A notable exception is the Netherlands, where mortality rates at older ages stagnated and partly even increased during the 1980s and 1990s.2–4 Since about 2002, however, Dutch life expectancy has improved rapidly and is particularly caused by a mortality decline at age 65 years and older.5

    Coinciding with this unexpected mortality decline, a major healthcare reform was implemented in the Netherlands, in 2001, in which the budget caps on hospital expenditures that had been introduced in 1983 were abolished6 ,7 and a fee-for-service financing system was introduced.8 Although the reform primarily aimed at reducing the waiting time for elective surgeries, more money was spent in virtually any area of the healthcare system, including other elements of hospital care and pharmaceutical care.9 ,10 While Dutch healthcare expenditure, expressed as a percentage of gross domestic product (GDP), had remained roughly constant at around 8% during the 1980s and 1990s, after the reform it suddenly started to increase and reached a level of 11.8% in 2011—the second highest level worldwide.11

    On the basis of an analysis of routinely collected data at the aggregate level, Mackenbach et al5 recently argued that the expansion of healthcare was the only plausible explanation for the sudden improvement in Dutch life expectancy. While there were no clear changes in the prevalence of risk factors for mortality, utilisation of healthcare services rose disproportionally strongly in the elderly and innovative and effective treatments, such as percutaneous transluminal coronary angioplasty, cardiac catheterisations and lipid lowering drugs, were performed more often and at higher ages due to changes in guidelines abolishing age limits.5 Furthermore, end-of-life decisions involving the withdrawal of further treatments for seriously ill patients were performed less often in these years,12 possibly indicating a change in attitude towards life-saving treatments among the elderly.

    The simultaneous trend break in the time series of mortality and healthcare expenditure suggests an association between the two but does not prove a causal relationship. To test the hypothesis that ‘more healthcare for the elderly’5 explains the sudden improvement in Dutch life expectancy after 2001, we examine two specific corollaries of that hypothesis. First, we expect that the mortality decline since 2001 is concentrated among those with more severe chronic conditions, since this group is most sensitive to a lack of provision of medical care and likewise will benefit most from more and better treatment. Second, we expect that in a multivariate analysis, increases in healthcare utilisation statistically explain the decline of mortality.

    We tested these two predictions using individual-level data, obtained by linking the national health interview survey to the mortality registry. Detailed information on chronic conditions allowed us to stratify the survey respondents by severity of disease, while the stability of survey and sampling methodology over time enabled us to estimate changes in mortality and the contribution of changes in healthcare utilisation to those changes in mortality, after adjustment for sociodemographic factors, behavioural risk factors and various health indicators.

    Data and methods

    Data

    We used data from the ongoing national Dutch health interview survey (Permanent Onderzoek Leefsituatie, POLS), and a random sample of the non-institutionalised Dutch population, with response rates at about 60% and an annual sample size of about 10 000 persons. Primarily, we utilised information of the separate health module of the questionnaire with a response rate of 60% and 58% in 2001 and 2002, and 64% and 64% in 2007 and 2008. Using a unique data key provided by Statistics Netherlands, we linked individual records of this survey to mortality registers allowing a complete mortality follow-up over the four consecutive years. We limited our analysis to adults older than 35 years of age. As upper limit, we used the age of 85 years at interview to avoid selectivity associated with a higher proportion of people in nursing homes above that threshold. Further, we excluded persons with missing information regarding the study variables (in men 25% and 18% in 2001/2002 and 2007/2008, and in women 24% and 17% in 2001/2002 and 2007/2008). Sampling weights were used to adjust for selective non-response and to resemble the Dutch non-institutional population.13

    To assess changes in mortality over time, we pooled the information from the first two surveys that contained the revised health module for 2001 and 2002 (men: N=3697 men; women: N=3994), and from the past two surveys in 2007 and 2008 (men: N=3996; women: N=4366). Each survey year was linked to four consecutive years of mortality follow-up.

    Measures of non-fatal and fatal chronic conditions

    To assess the first prediction of our study hypothesis, we identified persons without chronic conditions, only non-fatal conditions, only fatal chronic conditions and those suffering non-fatal and fatal conditions. As non-fatal conditions, we defined reporting (within the previous 12 months) at least one of the following: diabetes, skin disease, eczema, bowel disease, urinary incontinence, arthrosis, rheumatism, back pain, serious head pain and disease of the neck or arm. As fatal conditions we defined reporting at least one of the following: to ever have had cancer, stroke, myocardial infarction, serious heart problems or reporting to have had either peripheral artery disease or serious lung disease during the previous 12 months. A detailed overview about the classification of conditions and details on the included diseases is provided in online table A1 in the web supplement.

    Independent variables

    To assess the second prediction of our study hypothesis, we used various indicators for the volume of healthcare utilisation, measured by the number of visits within the year before the interview to the general practitioner (GP) (0, 1 and more), to the medical specialist (0, 1 and more) and to the hospital (0, 1 and more without surgery, 1 and more with surgeries). Further, we assessed the number of prescribed medicines consumed during the recent two weeks (0, 1 and more but no medicine for heart problems, lipid level and hypertension; 1 and more including medicine for heart problems or lipid level or hypertension).

    Confounders

    Sociodemographic confounders comprised of education (low: primary education only; middle: vocational training or high school; high: university degree) and marital status (married, divorced/widowed and never married). Behavioural risk factors included smoking status (never, ex-smoker and current smoker) and body mass index (BMI; underweight for BMI<18.5, normal/overweight for BMI 18.5–30, obese for BMI>30). Health status was measured by the presence of disability (no, yes=major difficulties or only with help at least in one of six items: conversation, reading, visual impairment, carrying, walking and bending) and self-rated health (not bad, bad).

    Data analysis

    We modelled the survival time until death by using a Cox proportional hazards model. We tested the assumption of proportional hazards by including an interaction of all covariates and process time, and by estimating Schoenfeld residuals. The process time in all models was age. We estimated pooled models with gender as strata, and separate models for men and women. To measure differences over time, we coded year at interview: 0=interviewed in 2001/2001 and 1=interviewed in 2007/2008. We further included an interaction term for this dummy and the index for chronic conditions (0=no chronic condition; 1=at least one non-fatal condition and no fatal condition; 2=at least one fatal condition but no non-fatal conditions; 3=at least one non-fatal and at least one fatal condition). To study the effect of the healthcare utilisation, we first entered sociodemographic and behavioural risk factors and health status indicators.

    Results

    Changes in sociodemographic characteristics, risk factors, health status and healthcare utilisation

    Differences between the two study cohorts are displayed in table 1. Several favourable changes occurred between the survey years 2001/2002 and 2007/2008. In the second period, there were more people with higher education (+0.4% in men and +4.1% in women) and more never-smokers (+4.2% in men and +3.3% in women). However, there were some unfavourable changes as well, since the share of people with obesity increased (+2.1% in men and +2.2% in women) and the fraction of those without any chronic conditions decreased (−2.8% in men and −3.9% in women).

    View this table:
    Table 1

    Sample characteristics

    Overall, the changes in primary and secondary healthcare utilisation were modest, while larger changes occurred in the use of pharmaceutical drugs (table 1). Men more often reported a visit to the GP (+0.8%) and to the hospital (+1.0%), and the use of prescribed medicines (+6.1%). Women more often reported a visit to a medical specialist (+0.2%) and to the hospital (+0.8%), and like men more often used prescribed medicines (+3.4%). Medication use increased, particularly for drugs given for the treatment of heart problems, and the lipid lowering and blood pressure lowering drugs (+5% in men and +6.2% in women).

    Changes in mortality stratified by severity of chronic conditions

    Overall, we found a significant mortality decline between both cohorts, which was more pronounced among those with more severe chronic conditions (table 2). In the cohort 2007/2008, the HR was 0.85 (95% CI=0.72 to 0.99) as compared to the cohort 2001/2002, with a larger decrease among men (HR=0.79, 95% CI=0.65 to 0.79) than among women (HR=0.95, 95% CI=0.73 to 1.22). However, the 95% CIs of the estimates for men and women overlap to a large degree.

    View this table:
    Table 2

    Reduction in mortality between cohort 2007/2008 and cohort 2001/2002

    When we conditioned on the presence of chronic conditions, we did not detect significant declines in mortality in the first three categories, but did find mortality declines in the most severe category (table 2). Among those without chronic conditions or with non-fatal conditions or fatal conditions only, there was no statistically significant decrease in mortality. By contrast, there was a statistically significant decrease of mortality among those suffering one or more non-fatal and one or more fatal conditions simultaneously (change in HR=−0.58, 95% CI=−0.78 to −0.38). This finding is replicated in the separate estimation for men (change in HR=−0.62, 95% CI=−0.83 to −0.41) but not in that for women (change in HR=−0.32, 95% CI=−1.06 to 0.41). Again the CIs between the results for men and women overlap to a large degree so that inferences about gender differentials cannot reliably be made.

    Contribution of changes in healthcare utilisation to decline in mortality

    The simultaneous inclusion of all variables of healthcare utilisation in our model could not detect an association with mortality. Despite adjustments for several risk factors and various health indicators, most HRs were above one indicating an elevated mortality risk with higher utilisation of healthcare (table 3). In the separate model for women, hospital visits with surgery involved were significantly associated with a higher mortality risk (HR=1.63, 95% CI=1.09 to 2.44).

    View this table:
    Table 3

    Effect of health care utilisation on mortality (cohort 2001/2002 and 2007/2008 pooled)

    Changes in healthcare utilisation did not contribute to the improvement in survival between the cohorts 2001/2002 and 2007/2008. Shown in table 4 are the estimated crude changes in mortality between both cohorts (left panel), the changes in mortality after adjustment for sociodemographic characteristics, risk factors and health status (centre panel) and the changes in mortality after additional adjustment for healthcare utilisation (right panel). While adjusting for confounders did not explain the decrease in mortality in the full sample, it increased the HR in the separate model for men from 0.79 (95% CI=0.65 to 0.97) to 0.82 (95% CI=0.67 to 1.01). Adjusting for healthcare utilisation had the opposite effect and made the decline in mortality larger by a 1% point, both in the full sample and among men only. In the separate model for women, the decrease in mortality was not statistically significant in all three models.

    View this table:
    Table 4

    Crude and adjusted HRs for mortality of the cohort 2007/2008 relative to the cohort 2001/2002 (Reference group, HR=1)

    Discussion

    We exploited a rich nationally representative health survey with register-based mortality follow-up to study the recent decline in Dutch mortality since 2001. Our results only partly confirmed the hypothesis that the expansion of healthcare explains the mortality decline. While the greatest mortality improvement was indeed found in the subgroup with the most severe chronic conditions, changes in healthcare utilisation could not statistically account for the mortality decline in our study population.

    Strengths and limitations

    Our study analysed the association between healthcare utilisation and mortality at the individual level. This opens up the black box of the ecological association reported between trends in life expectancy and healthcare expenditures at the country level.5 While ecological studies often control only for a few and quite crude confounders, such as GDP and smoking,14 we were able to include detailed information on sociodemographic and behavioural risk factors and different health status indicators. A broad range of chronic conditions allowed stratified analyses by severity of chronic illness.

    Although we have used a large nationally representative survey, selection bias due to the non-response in POLS (about 40%) and exclusion of the institutionalised population might be an issue. We further excluded about 20% of the sample due to missing information for at least one of the variables in the model. In general, those not responding in a survey and people living in institutions have a less favourable health status.15 ,16 In our study, the survey response rate was about 4% points higher in the cohort 2007/2008 than in the cohort 2001/2002, and about 7% points fewer cases were excluded due to missing information, while the proportion of the institutionalised population was lower.17 ,18 In totality, these changes may have led to a slightly more healthy sample composition in the cohort 2007/2008. Even if this would have affected the estimated mortality decline for the full sample, it is unlikely to affect the results that the more chronically ill had the greatest improvement in survival. The latter finding was also found in another Dutch sample that included people in institutions.17

    Our study builds on self-reported information on health status and chronic conditions. Previous research concluded that self-reported information on chronic conditions was fairly accurate.19 Self-rated health is generally considered a reasonable proxy for the objective health status.20 Inconsistencies in self-reported health over time have been reported, but this referred to a much longer time span than in our study and was found to be relevant mainly among younger people.21 Nevertheless, if non-differential misclassification of health conditions has occurred, our estimates of the effect of health conditions on mortality will be downwardly biased, and control for health status in our analysis of the effect of healthcare utilisation on mortality may have been incomplete (see below).

    This study did not really capture the time before the introduction of healthcare reform, since the latter started in 2001 and we used data from 2001 onwards. Hence, we may have missed some of the effects of the expansion of healthcare. However, healthcare spending and utilisation did continue to rise, at least until 2011,11 ,22 so that our analysis still compares a period with more utilisation to a period with less utilisation. In a sensitivity analysis, we compared our study cohort 2007/2008 with a cohort constructed from two earlier POLS survey years, 1997/1998, in which identical information had been collected for healthcare utilisation, but not for all of the other variables. The results of this analysis confirmed our second main finding: there is a significant positive association between healthcare use and mortality, and changes in healthcare use did not contribute to the explanation of mortality decline (see online supplementary table A2 and A3 in appendix). This was also true after including each indicator of healthcare utilisation separately in the models and after including an interaction of study cohort and healthcare utilisation (see online supplementary table A4 and A5 in the appendix).

    Interpretation

    We found that mortality decline was concentrated among those with chronic conditions, as predicted, but could not establish a beneficial effect of healthcare utilisation on mortality. In fact, we rather found the opposite: a higher risk to die among those who had used more care. Empirical findings suggesting that healthcare seem to do more harm than good have been reported before and have been described either as an ‘anomaly’23 or as ‘the paradox of health care’.24–26 These paradoxical results have usually been ascribed to imperfect control of confounding23 ,27 and we believe that this also applies to our analysis. Although we controlled for sociodemographic characteristics, risk factors and various aspects of health status, we have probably not sufficiently adjusted for the nature of the health conditions and the severity of illness, partly because of the self-reported nature of our data. Likewise, the indicators of healthcare utilisation in our analysis are quite crude and more detailed information on hospital treatments or pharmaceutical drugs would have been preferable. Moreover, our analysis contained data on healthcare utilisation in the years prior to the interview only, which could lead to reverse causation if health status as reported during the interview is partly determined by prior treatment. Additionally, since the institutionalised population was not included, we missed the most severely ill persons in our analysis.28 Owing to all these issues, we believe that our results cannot be interpreted as a refutation of the idea that healthcare saves lives, or for that matter, of the hypothesis that increases in healthcare utilisation have contributed to the recent declines in mortality in the Netherlands.

    While we found that mortality decline was concentrated among those with chronic conditions, we also found an increase in the prevalence of chronic conditions between the 2001/2002 and 2007/2008 cohorts. This increase in prevalence is likely to be the result of earlier and better detection of diseases, perhaps as a consequence of the same changes in healthcare that contributed to the decline of mortality. The change in hospital financing from fixed budgets to a fee-for-service scheme in 2001 created incentives to admit more persons for milder and less specific symptoms.10 Diagnostic procedures involving brain CT and MRI were applied more often after 2001, which presumably led to an increase in the incidence of conditions like ischaemic stroke.29 This also suggests that in our analysis the category of fatal and non-fatal chronic conditions may have contained milder cases on an average in 2007/2008 than in 2001/2002. We, therefore, reran our analysis of mortality decline among those with and without chronic conditions (as presented in table 2) with additional controls for self-perceived health and OECD disability (results not shown). However, the results were robust to this adjustment for health status.

    In our study sample, the biggest changes in healthcare utilisation occurred in pharmaceutical care, particularly because more people took drugs for heart problems, lipid levels and hypertension. This is in line with other studies arguing that the recent improvement in Dutch survival was mainly due to better (pharmaceutical) care for cardiovascular diseases.5 ,30 Between 2000 and 2003, the utilisation of statins increased by 70% in the Netherlands.31 During this time there were several guideline changes in the prescription of these lipid-lowering agents, effectively increasing the prescription rates for older and sicker patients.32 Unlike previous reports, we find only modest changes in hospital visits.5 We believe that this is due to the fact that our sample includes only persons who were able to participate in the survey and has missed out persons with more severe diseases who are more dependent on hospital care. The spending and activity growth of the hospital sector since 2001 was predominantly driven by a more intensive treatment of those already involved in specialist or hospital care.10 Diagnostic categories for which hospital admissions increased most strongly were mainly related to diseases at older ages, for example, pneumonia, transient ischaemic attack and gonarthrosis.10 Hence, the changes in Dutch healthcare mainly affected older and sicker individuals, who were not fully included in the sample of the Dutch health interview survey. Also the indicators of healthcare utilisation may not have been sensitive enough to fully capture the intensification of treatment.

    Implications

    The jury on the explanation of the recent mortality decline in the Netherlands is still out. Our study confirms that this decline cannot be explained by changes in sociodemographic characteristics, risk factors or health status, as has been argued before,5 but has not been able to demonstrate that it can be explained by the increase of healthcare utilisation. Further studies are needed, using more detailed information on the health status (allowing better control for confounding) and on healthcare utilisation (allowing a better distinction between life-saving and non-life-saving interventions). Recent examples of linkages of survey and/or mortality data with GPs and hospitals, and health insurance data show that such studies are now becoming feasible.33–35

    What is already known on this subject

    • Since 2002, Dutch mortality rates decreased rapidly after decades of stagnation.

    • On the basis of indirect evidence, previous research has suggested that this decline was due to a sudden expansion of healthcare.

    What this study adds

    • Our study confirms that the decline cannot be explained by changes in sociodemographic characteristics, behavioural risk factors or changes in health status.

    • While the greatest mortality improvement was indeed found in the subgroup with the most severe chronic conditions, changes in healthcare utilisation could not statistically account for the mortality decline in our study population.

    References

    1. ↵
      1. Rau R,
      2. Soroko E,
      3. Jasilionis D, et al
      . Continued reductions in mortality at advanced ages. Population Devel Rev 2008;34:747–68. doi:10.1111/j.1728-4457.2008.00249.x
      OpenUrlCrossRef
    2. ↵
      1. Nusselder WJ,
      2. Mackenbach JP
      . Lack of improvement of life expectancy at advanced ages in the Netherlands. Int J Epidemiol 2000;29:140. doi:10.1093/ije/29.1.140
      OpenUrlAbstract/FREE Full Text
    3. ↵
      1. Nusselder WJ,
      2. Mackenbach JP
      . Rectangularization of the survival curve in the Netherlands, 1950-1992. Gerontologist 1996;36:773–82. doi:10.1093/geront/36.6.773
      OpenUrlAbstract/FREE Full Text
    4. ↵
      1. Janssen F,
      2. Nusselder WJ,
      3. Looman CWN, et al
      . Stagnation in mortality decline among elders in the Netherlands. Gerontologist 2003;43:722–34. doi:10.1093/geront/43.5.722
      OpenUrlAbstract/FREE Full Text
    5. ↵
      1. Mackenbach JP,
      2. Slobbe L,
      3. Looman CWN, et al
      . Sharp upturn in life expectancy in the Netherlands: effect of more health care for the elderly? Eur J Epidemiol 2011;26:903–14. doi:10.1007/s10654-011-9633-y
      OpenUrlCrossRefPubMed
    6. ↵
      1. Schut FT,
      2. Varkevisser M
      . Tackling hospital waiting times: the impact of past and current policies in the Netherlands. Health Policy 2013;113:127–33. doi:10.1016/j.healthpol.2013.05.003
      OpenUrlCrossRefPubMedWeb of Science
    7. ↵
      1. Casparie AF,
      2. Hoogendoorn D
      . Effects of budgeting on health care services in Dutch hospitals. Am J Public Health 1991;81:1442–7. doi:10.2105/AJPH.81.11.1442
      OpenUrlCrossRefPubMedWeb of Science
    8. ↵
      1. Schut FT,
      2. Van de Ven WP
      . Rationing and competition in the Dutch health-care system. Health Econ 2005;14(Suppl 1):S59–74. doi:10.1002/hec.1036
      OpenUrlCrossRefPubMedWeb of Science
    9. ↵
      1. de Meijer C,
      2. O'Donnell O,
      3. Koopmanschap M, et al
      . Health expenditure growth: looking beyond the average through decomposition of the full distribution. J. Health Econ 2013;32:88–105. doi:10.1016/j.jhealeco.2012.10.009
      OpenUrlCrossRefPubMedWeb of Science
    10. ↵
      1. van de Vijsel AR,
      2. Engelfriet PM,
      3. Westert GP
      . Rendering hospital budgets volume based and open ended to reduce waiting lists: does it work? Health Policy 2011;100:60–70. doi:10.1016/j.healthpol.2010.11.014
      OpenUrlCrossRefPubMedWeb of Science
    11. ↵
      OECD. OECD Health Statistics 2014. 2014.
    12. ↵
      1. van der Heide A,
      2. Onwuteaka-Philipsen BD,
      3. Rurup ML, et al
      . End-of-life practices in the Netherlands under the Euthanasia Act. N Engl J Med 2007; 356:1957–65. doi:10.1056/NEJMsa071143
      OpenUrlCrossRefPubMedWeb of Science
    13. ↵
      1. Stam S,
      2. Knoops K
      . Lange tijdreeksen gezonde levensverwachting. Beschikbaarheid van enquøetedata gezondheidsindicatoren. Den Haag/Heerlen, The Netherland: Statistics Netherlands, 2009.
    14. ↵
      1. Nixon J,
      2. Ulmann P
      . The relationship between health care expenditure and health outcomes. Eur J Health Econ 2006;7:7–18. doi:10.1007/s10198-005-0336-8
      OpenUrlCrossRefPubMed
    15. ↵
      1. Klijs B,
      2. Nusselder WJ,
      3. Looman CW, et al
      . Contribution of chronic disease to the burden of disability. PLoS ONE 2011;6:e25325. doi:10.1371/journal.pone.0025325
      OpenUrlCrossRefPubMed
    16. ↵
      1. Kulhanova I,
      2. Hoffmann R,
      3. Eikemo TA, et al
      . Educational inequalities in mortality by cause of death: first national data for the Netherlands. Int J Public Health 2014;158:A8188.
      OpenUrl
    17. ↵
      1. Deeg DJH,
      2. van Vliet MJG,
      3. Kardaun JWPF, et al
      . Understanding the mortality decline at older ages improved life course or improved present period? Annu Rev Gerontol Geriatr 2013;33:259–91. doi:10.1891/0198-8794.33.261
      OpenUrlCrossRef
    18. ↵
      The Netherlands Institute for Social Research. Rapportage ouderen 2006. Veranderingen in de leefsituatie en levensloop. The Hague, The Netherlands: The Netherlands Institute for Social Research, 2006.
    19. ↵
      1. Kriegsman DM,
      2. Penninx BW,
      3. van Eijk JT, et al
      . Self-reports and general practitioner information on the presence of chronic diseases in community dwelling elderly. A study on the accuracy of patients’ self-reports and on determinants of inaccuracy. J Clin Epidemiol 1996;49:1407–17. doi:10.1016/S0895-4356(96)00274-0
      OpenUrlCrossRefPubMedWeb of Science
    20. ↵
      1. Idler EL,
      2. Benyamini Y
      . Self-rated health and mortality: a review of twenty-seven community studies. J Health Soc Behav 1997;38:21–37. doi:10.2307/2955359
      OpenUrlCrossRefPubMedWeb of Science
    21. ↵
      1. Salomon JA,
      2. Nordhagen S,
      3. Oza S, et al
      . Are Americans feeling less healthy? The puzzle of trends in self-rated health. Am. J. Epidemiol. 2009;170:343–51. doi:10.1093/aje/kwp144
      OpenUrlAbstract/FREE Full Text
    22. ↵
      Statistics Netherlands. Ziekenhuisopnamen [hospital admissions]. Secondary Ziekenhuisopnamen [hospital admissions] 2014. http://www.cbs.nl/nl-NL/menu/themas/gezondheid-welzijn/cijfers/extra/2010-ziekenhuisopname.htm
    23. ↵
      1. Leger SS
      . The anomaly that finally went away? J Epidemiol Community Health 2001;55:79. doi:10.1136/jech.55.2.79
      OpenUrlFREE Full Text
    24. ↵
      1. Hofmann B
      . The paradox of health care. Health Care Anal 2001;9:369–86. doi:10.1023/A:1013854030699
      OpenUrlCrossRefPubMedWeb of Science
    25. ↵
      1. Ricketts TC,
      2. Holmes GM
      . Mortality and physician supply: does region hold the key to the paradox? Health Serv Res 2007;42(6 Pt 1):2233–51. doi:10.1111/j.1475-6773.2007.00728.x
      OpenUrlCrossRefPubMedWeb of Science
    26. ↵
      1. Cochrane AL,
      2. St Leger AS,
      3. Moore F
      . Health service ‘input’ and mortality ‘output’ in developed countries. J Epidemiol Community Health 1978;32:200–5. doi:10.1136/jech.32.3.200
      OpenUrlAbstract/FREE Full Text
    27. ↵
      1. Young FW
      . An explanation of the persistent doctor-mortality association. J Epidemiol Community Health 2001;55:80–4. doi:10.1136/jech.55.2.80
      OpenUrlAbstract/FREE Full Text
    28. ↵
      1. Polder JJ,
      2. Barendregt JJ,
      3. van Oers H
      . Health care costs in the last year of life—the Dutch experience. Soc Sci Med 2006;63:1720–31. doi:10.1016/j.socscimed.2006.04.018
      OpenUrlCrossRefPubMedWeb of Science
    29. ↵
      1. Vaartjes I,
      2. O'Flaherty M,
      3. Capewell S, et al
      . Remarkable decline in ischemic stroke mortality is not matched by changes in incidence. Stroke 2013;44:591–7. doi:10.1161/STROKEAHA.112.677724
      OpenUrlAbstract/FREE Full Text
    30. ↵
      1. Bonneux L
      . Success has many fathers, failure remains an orphan. Eur J Epidemiol 2011;26:897–8. doi:10.1007/s10654-011-9640-z
      OpenUrlCrossRefPubMed
    31. ↵
      1. Walley T,
      2. Folino-Gallo P,
      3. Stephens P, et al
      . Trends in prescribing and utilization of statins and other lipid lowering drugs across Europe 1997–2003. Br J Clin Pharmacol 2005;60:543–51. doi:10.1111/j.1365-2125.2005.02478.x
      OpenUrlCrossRefPubMedWeb of Science
    32. ↵
      1. Teeling M,
      2. Bennett K,
      3. Feely J
      . The influence of guidelines on the use of statins: analysis of prescribing trends 1998–2002. Br J Clin Pharmacol 2005;59: 227–32. doi:10.1111/j.1365-2125.2004.02256.x
      OpenUrlCrossRefPubMedWeb of Science
    33. ↵
      1. Snijders BE,
      2. van der Hoek W,
      3. Stirbu I, et al
      . General practitioners’ contribution to the management of community acquired pneumonia in the Netherlands: a retrospective analysis of primary care, hospital, and national mortality databases with individual data linkage. Prim Care Respir J 2013;22:400–5. doi:10.4104/pcrj.2013.00085
      OpenUrlCrossRefPubMed
    34. ↵
      1. Koek H,
      2. Soedamah-Muthu S,
      3. Kardaun J, et al
      . Short-and long-term mortality after acute myocardial infarction: comparison of patients with and without diabetes mellitus. Eur J Epidemiol 2007;22:883–8. doi:10.1007/s10654-007-9191-5
      OpenUrlCrossRefPubMedWeb of Science
    35. ↵
      1. Smits FT,
      2. Brouwer HJ,
      3. Zwinderman AH, et al
      . Morbidity and doctor characteristics only partly explain the substantial healthcare expenditures of frequent attenders: a record linkage study between patient data and reimbursements data. BMC Fam Pract 2013;14:138. doi:10.1186/1471-2296-14-138
      OpenUrlCrossRefPubMed

    Supplementary materials

    • Supplementary Data

      This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

      Files in this Data Supplement:

    Footnotes

    • Contributors FP and JPM developed the concept of the study. FP drafted the manuscript, and developed and performed the statistical analysis. JPM and WJN critically reviewed the different versions of the manuscript. All the authors have read and approved the final version to be published.

    • Funding This work was supported by Netspar and is part of the project “Causes and consequences of rising life expectancy in the Netherlands”. The funding organisation did not participate in and did not influence the design and conduct of the study, collection, management, analysis or interpretation of the data, preparation, reviewing or approval of the manuscript.

    • Competing interests None.

    • Provenance and peer review Not commissioned; externally peer reviewed.

    • Data sharing statement We performed a record-linkage study, where information from health interview surveys was linked to mortality registry. For this purpose, the analysis of the data took place in the closed environment of Statistics Netherlands to preserve anonymity of the study participants.