Mortality data from Eurostat

For this study, we accessed cause-of-death statistics from the European Statistical Office (Eurostat) for the data-year 2014. As at the end of March 2017 this included reported data from all the 28 European Union member countries (EU28) [14]. The EU28 includes these member countries as at 1 January 2013: Austria, Belgium, Bulgaria, Croatia, Cyprus, the Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, the Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden and the United Kingdom.

The mortality statistics published by Eurostat, like other national and international mortality statistics, are based on one single underlying cause of death. If two or more causes are recorded on the form of medical death certificate, one cause should be selected and used for primary tabulation. The cause used to compile official cause-specific mortality statistics should be designated the underlying cause of death, defined as “(a) the disease or injury which initiated the train of morbid events leading directly to death, or (b) the circumstances of the accident or violence which produced the fatal injury” [15] (page 31). This definition was introduced in the sixth revision of the International Statistical Classification of Diseases (ICD-6), with the purpose of obtaining essential information for prevention: “From the standpoint of prevention of death, it is necessary to break the chain of events or to cure at some point. The most effective public health objective is to prevent the precipitating cause from operating” [15] (page 31).

Definition of rankable causes

The cause-of-death statistics from Eurostat were grouped according to the extended mortality shortlist of 2012 [14]. This list contains in total 86 cause-categories (including subgroups and groups of residuals) instead that the 65 ones included in the previous version. We used these 86 categories to select two different lists of mutually exclusive causes for this study ( Table 1 ). In our first list, we grouped three main categories, which contained subcategories: malignant neoplasms, diseases of the circulatory system, and accidents (unintentional injuries). In the second ranking list, we split these three categories. If a cause category contained subcategories, the most detailed cause-category was selected (except for three main cause-of-death categories in the first list). In the first list, we identified 28 causes, and in the second, 51 causes ( Table 1 ). All cause-of-death categories starting with “other”, “symptoms”, “ill-defined”, and “event of undetermined intent” were excluded from the selection of cause categories. The ranking of selected causes was based solely on the number of deaths, and not rates. Crude rates would have given the same result, in contrast to standardized rates because the selection of standard population may impact on the ranking. We mostly followed the procedures used by the US National Center for Health Statistics (NCHS) for ranking causes of death [9,10], with one important exception. We selected causes from the list of 86 causes, the NCHS use a list of 113 causes [9]. We included one residual category, which identifies the percentage of all deaths not covered by the lists of leading causes. In addition to the causes that were ranked, the category “Symptoms, signs and ill-defined conditions”, excluding “Sudden infant death syndrome - SIDS” (ICD-10 codes R00-R99, excl. R95) [16], is presented separately. The proportion of deaths classified to this set of causes, as the underlying cause of death, is one internationally used indicator of the quality of mortality data [12,17,18].

Global Health Estimates 2015

In addition to data from Eurostat, we have also used summary estimates of leading causes of death for the year 2015, ranked by the World Health Organization (WHO) [19]. The last available estimates contain figures for the four years 2000, 2005, 2010 and 2015 (2000-2015) [19,20]. The data sources behind these estimates are a mix of information from vital registration systems, periodic and non-national-level samples, and disease surveillance systems. Based on all this information, statistical models were used to adjust for completeness, underreporting of specific causes of death, redistribution of unknown information on age, sex and causes of death, and to interpolate and extrapolate where national data was missing for some years. The WHO Global Health Estimates (GHE) 2015 on mortality used a list of, in total, 202 cause-of-death categories [19,20], instead of the 163 covered in the previous GHE for 2014 [21,22]. The estimates of WHO for the years 2000-2015 were based on latest available national information on levels of mortality and cause distributions as at the end of October 2016 together with available information from WHO programs [19]. Because of substantial revisions of the method and data sources behind the global and regional estimates, the WHO estimates for the years 2000-2015 [20] are not directly comparable with previous WHO estimates for 2000-2012 [21] or earlier versions. We present WHO`s global ranking of leading causes of mortality, and for the WHO European Region for the year 2015. The WHO European Region includes 53 countries [20]. In addition to the 28 European Union countries (EU28) listed below, these countries are defined into the region: Albania, Andorra, Armenia, Azerbaijan, Belarus, Bosnia and Herzegovina, Georgia, Iceland, Israel, Kazakhstan, Kyrgyzstan, Monaco, Montenegro, Norway, Republic of Moldova, Russian Federation, San Marino, Serbia, Switzerland, Tajikistan, The former Yugoslav Republic of Macedonia, Turkey, Turkmenistan, Ukraine, and Uzbekistan. It must be added, the WHO regional grouping as of 2015 excluded analyzing information from WHO Member States with a population of less than 90 000 population in 2015 [20]. Regarding the European region, Andorra, Monaco, and San Marino were not included in GHE 2015 [20], but included in the GHE 2014 analysis [21].

Ethics

This study was solely based on aggregate cause of death statistics for Europe and estimates from the World Health Organization, globally and the WHO European Region. All the data we used was available to the public on the internet, which means that the information was anonymous without any conditions relating to privacy considerations. Therefore, no ethical or governmental permissions were required for this study.

Main findings in context

We have shown how two different methods gave different leading underlying causes of mortality in Europe, and how this affected the rank order of suicide. Based on the Eurostat’s short list of causes of death, we selected rankable causes and used different levels of details for three cause categories: circulatory system diseases, malignant neoplasms, and accidents. These three causes ranked high when using aggregated groupings, all among the top five leading causes ( Table 2 ). The importance of circulatory diseases (accounted for 38% of all deaths in EU-28 in 2014), and cancers (26% of all deaths), was shown also when these disease groups were split into more specific diagnoses (6 out of the 10 leading causes were subgroups of circulatory diseases, and cancers, Table 3 ). When we disaggregated unintentional injury deaths, no subcategory of accident mechanism (eg, transport, falls and poisoning) was among the 10 leading causes.

These results were in accordance with previous studies from Europe: Griffiths et al. [12], demonstrated how the leading causes in lists of the top ten in England and Wales, in 2003, varied as a consequence of using grouped or more detailed categories of malignant neoplasms and accidents. In the England and Wales study, the authors grouped suicide together with the majority of injury and poisoning deaths of undetermined intent because most deaths coded to undetermined intent in England and Wales are thought to be deaths from self-harm for which the coroner returns an open verdict (insufficient information that deceased intended to die) [12]. Despite this broader grouping, suicide was not among the top ten leading causes in England and Wales, for males or females. Another study from Becker et al [7], showed how the list with leading causes of death changed when using the broad category for all cancers, and when cancers were disaggregated by site, based on data for western Europe, 2001. Suicide ranked as the 14^th leading causes of death in the western part of Europe in 2001, and accounted that year for 1.7% of all deaths (for males and females suicide was ranked as the 9^th and 26^th, respectively) [7].

In the United States, in 2015 (the most recent year of available death data), suicide was ranked as the 10^th leading cause of death [23,24], and this rank has been unchanged since 2008 [25]. In 2015 suicide accounted for 1.6% of all deaths in the United States (number of suicide deaths was 44 193, and the crude rate was 13.7/100 000 population) [24,26]. In the last 40 years, since 1975, suicide has been among the top 12 leading causes of death in the US [24].

Suicide globally

Regarding suicide globally, the WHO publication on global suicide prevention [8], stated suicide was the 15^th leading cause of mortality globally in 2012. GHE 2014 [22] was used as data source in this WHO report, and necessarily corresponding to what we have reported from the same source. However, the uncertainty of the global cause of death estimates is great. One example: According to estimates in GHE 2015 [19] the magnitude of suicide deaths globally in 2000 was over-estimated by 135 000 deaths in GHE 2014 [22]; 748 000 and 883 000 suicide deaths, respectively. This relatively large difference in the estimated number of suicides in 2000 only had a minor effect of the rank (suicide moved down from 14^th to 15^th rank in GHE 2014 and GHE 2015 in the year 2000, respectively). A newer report on mental health globally states that suicide rates are increasing globally [27]. However, and as we presented, this is not in line with the two most recent estimates of the GHE [19,22], which both indicate that global suicide rates have decreased. We may add that Bertolote and De Leo [28] have also shown decreasing trends in suicide globally, when assessing trends from 1992 to 2009 using available mortality data in WHO’s data bank from 62 countries. WHO has previously used a similar method to shown that suicide rates were quite stable during a period from 1970 to 1996, for the 39 countries for which complete data was available [29].

Leading causes of mortality (and morbidity) can give useful information supporting the advance of public health, but are only a starting point for supplemental analysis [7] eg, on cause specific mortality rates, trends over time, and causes by age and sex. Without a clear understanding of what the ranking means and its inherent limitations [7,9,10], it is easy to cross the line between good use and misuse of statistics [30,31]. There are many ways to misuse statistics in science, and one of the most common forms of the misrepresentation of data are to exaggerate the significance of results [30]. By saying that a specific cause is “among the leading causes of mortality”, without presenting the numbers and the percentage of all deaths, this may result in misinterpretation and undermine the public`s trust in public health science.

Limitations

This study has limitations. First, the rankings do not necessarily present the causes of mortality of greatest public health importance. Some causes of public health importance may not be among the ranked causes because they either are incorporated into broader categories (eg, road traffic accidents), or they appears infrequently in official mortality statistics. This is the case for conditions that to some extent should be used as the underlying cause, when more than one condition is reported on the death certificate (eg, alcohol conditions, diabetes, and mental disorders). It is a common misunderstanding, eg, in medical research, that mortality statistics mirror the prevalence of specific conditions reported on the death certificate, and judge cases as “misclassified” if official mortality statistics do not match the number of cases identified on the death certificate [32]. This means that many studies assessing quality of mortality statistics are flawed [10,32]. We must always keep in mind that the rankings present underlying causes of death [15], among those cause categories available to be ranked [7,9], which implies that only one cause is used when two or more diseases, injuries or other conditions are recorded on the death certificate. Further, quality of mortality statistics may affect the ranking (and all other mortality analysis). The debate regarding quality [17], is often a mixed debate regarding completeness, and misclassification, and also the terms used in ICD. For example, Rockett and co-authors [33] questioned the dichotomy between suicide and unintentional (accidents) deaths, and argued the need for statistics that better reflect the burden of deaths due to deliberate, self-destructive behavior (self-injury deaths). This category group all suicides together with a major proportion of accidental - and undetermined drug intoxication deaths [33]. The self-injury category would constitute the eighth leading cause of death in US [33], and the subject is part of an ongoing debate regarding classification of suicide and drug intoxication deaths in US [34]. In addition, the number of deaths allocated to symptoms, ill-defined and unknown causes of death is of special concern for the quality of mortality data [12,17]. When it comes to injury deaths, lack of information on the external cause of injuries and poisonings may overestimate number of accidents and underestimate suicide [12,17,35,36]. Third, a change in the ranking of a specific cause over time may occur even though the proportion of deaths classified to that cause did not change [9]. Similarly, two populations with different risk levels (rates) for a specific cause (eg, suicide), may have the same rank for this condition [9]. Introduction of new revisions of the classification system and coding rules is another factor that can substantially affect the ranking. Moving to ICD-10 in 1999 in US affected the ranking of five out of the ten leading causes: diabetes, influenza and pneumonia, Alzheimer`s disease, suicide, and chronic liver disease and cirrhosis [37].

Finally, we did not assess leading causes in different age groups. In general, the pattern differs in the way that injury deaths (accidents, suicide, homicide) account for relatively more deaths in younger age groups, in contrast to older age groups [9,12].

CONCLUSIONS

To ensure that ranking causes of death is not driven by political motives and funding considerations, standard methods and official tabulation lists should be used. A ranking of causes of death should always present the proportion of all deaths and number of deaths allocated to the ranked causes, similarly for the residuals not covered, and a data quality indicator.