Mortality due to traumatic spinal cord injuries in Europe: a cross-sectional and pooled analysis of population-wide data from 22 countries

Study design, data sources and bias

A population-based cross-sectional epidemiological study was conducted using 2012 data from 22 European countries: Austria; Belgium; Croatia; Cyprus; Czech Republic; Denmark; Estonia; Finland; Germany; Ireland; Italy; Latvia; Lithuania; Netherlands; Portugal; Romania; Slovakia; Slovenia; Sweden; Switzerland; Turkey; and the United Kingdom. The inclusion of countries and the year of study was determined by data availability – the most recent year with the largest number of countries for which data were available was chosen.

All analyses are based on data obtained from Eurostat. The provided dataset contained a record for each death registered in the respective country, along with the cause of death that was derived from the death certificate. Eurostat routinely disseminates the aggregated data on causes of death classified using the European Shortlist for Causes of deaths which includes 86 causes [11, 12]. However, for the purpose of the study, both information on the external cause (E-code) and on the nature of injury (N-code) were provided for each death using the coding system of the 10th revision of the International Classification of Diseases and Related Health Problems (ICD-10) [13]. All data were provided as absolute numbers of deaths, broken down by sex and 5-year age groups.

Eurostat collects data on deaths from countries of the 28 member states of the EU, the former Yugoslav Republic of Macedonia, Albania, Iceland, Norway, Liechtenstein, and Switzerland under the framework of an EU regulation (No 328/2011) [14] which defines the scope, variables, and helps to ensure that the procedures of collecting and coding are harmonized and standardized to produce comparable results.

However, it must be noted that differences may exist in applying the recommended WHO’s updates in individual countries, or in the coverage of residents dying abroad or non-residents dying in the reporting country and these could - to a certain extent - affect the between-country comparability of the data. Furthermore, procedures on the level of responsible national authorities, or on the level of individuals may differ from country to country. All the above factors could result in selection bias. The authors of this study had no opportunity to control for such bias post-hoc.

Case definition, variables and statistical methods

Rate ratios with 95% CI were used to quantify the relative differences in mortality rates between age groups and sexes. Proportions of TSCI-related deaths out of the overall number of injury-related deaths were calculated - for this purpose, deaths with unspecific causes (e.g. “Other and unspecified effects of external causes”), deaths caused by exposure to heat, frost, intoxications (T15-T65), and cases with other generalized causes (T66-T78; T80-T88) were excluded from the pool of injury related deaths.

Pooled mortality rates were estimated — to model possible heterogeneity of rates in the different countries, the random effects model was applied by the DerSimonian and Laird method. In order to assess the heterogeneity of pooled estimations, 95% prediction intervals and I² were used. In order to allow for better comparability of our findings, pooled estimates were calculated based on crude rates and age-standardized rates. To estimate the total number of deaths attributable to TSCI in Europe, the crude pooled mortality rates were extrapolated to the population of the 28 member states of the EU and to the whole Europe (48 countries, as defined by the UN) [16].

The R statistical language and environment was used for all analyses in this study [17].

Additional data and results are presented in an online appendix – tables and figures in the appendix are denoted with A (e.g. Additional file 1: Table S1 and Additional file 1: Figure S1 etc.).

Main findings

We conducted a population-based cross-sectional study using data obtained from Eurostat in order to calculate population-based mortality rates attributable to TSCI in 22 European countries in 2012. We identified a total of 1840 TSCI-related deaths were identified, of which 1084 (59%) were males. We found a pooled age-standardized TSCI-related mortality rate of 6.7 per million (95% CI: 5.2 to 8.2) overall, 9.4 (95% CI: 7.3 to 11.5) for males, and 4.5 (95% CI: 3.4 to 5.6) for females. Extrapolating these results, 3152 (95% CI: 2441 to 3915) deaths would occur in 2012 in the EU-28 and 4570 (95% CI: 3538 to 5675) deaths in the whole Europe. TSCI-related deaths contributed by 2% (95% CI: 1.8% to 2.2%) to the overall injury related mortality. 61% of fatal TSCI were located in the cervical spine area. To our knowledge, this is the largest study that reports TSCI-related population-based mortalities to date which brings valuable information that can inform further research or prevention strategies.

Comparison to other studies and interpretation

Despite the relatively large number of published studies dealing with epidemiology of TSCI, studies reporting population-based mortality rates are rare. A study from the US state of South Carolina reported a crude mortality rate of 27.4 per million between 1981 and 1998 [18], whereas a study from Austria reported an average crude mortality rate of 4.1 and an age-standardized mortality rate of 2.38 (95% CI: 1.61 to 6.63) per million for the period of 2002–2012 [19]. The relatively large difference between the results in these two studies may stem from the different study periods, but could also suggest varying methodological approaches – our pooled estimates fall in-between them. Most of the other published studies focus on incidence or prevalence, and report mortality rates or deaths as standardized mortality ratios [5, 20–23], percent out of total numbers of patients admitted to a hospital after TSCI [21, 24], or life expectancy and survival of patients years after injury [4, 5, 8, 23, 25–27], and thus they are not directly comparable to these results.

One of the key findings of this paper is a relatively large variation of mortality rates that is observed, despite analyzing data for the same time period which were collected under unified standards for definitions and procedures. These could reflect true differences between the analyzed countries due to various levels of implementation of preventive measures, varying level of health care, or in general a varying level of risk of TSCI between the populations of these countries. However, there are other factors which might be driving these differences and they must be considered when interpreting these findings.

First, there might be various coding practices within countries or even within institutions, which could influence the overall picture of causes of deaths in the respective country and in Europe. An example of the influence of such practices might be the differing proportions of deaths coded as multiple injuries. Countries reporting large proportions of deaths caused by multiple injury tend to have lower TSCI-mortality rates, indicating that some deaths that were actually caused by TSCI could be ‘hidden’ under these categories. For example, the lowest TSCI related mortality in our study was observed in Turkey which reported 49% of injury related deaths being due to a multiple injury, whereas Finland with the highest TSCI mortality attributed only 5% of deaths to multiple injuries (on average, the 22 countries reported 22% of injury related deaths as caused by multiple injury).

Second, it has been previously reported that nations with similar economies tend to have similar patterns of TSCI epidemiology [7]. Such a difference was found also in this study (pooled crude rate of 5.3 per million in countries with lower GDP, and 8.4 in countries with higher economy level). Although more detailed investigation is needed to provide stronger evidence, it is suggested that some of the between country variability of mortality rates could be explained by varying GDP of the respective countries.

The observed between-country heterogeneity of our findings is in line with the heterogeneity reported for incidence or prevalence rates globally [3–5, 7, 9]. These findings together could suggest that both the risk of attaining a TSCI, and dying from it differs quite substantially between the countries; this could be a result of demographic differences, but could also reflect the level of attention that is devoted to prevention in the respective country.

Bias and limitations

There are limitations to this study that we acknowledge. Different epidemiological studies use different definitions for case ascertainment. Therefore, a potential limitation of this study is that the sensitivity of the ICD-10 codes used to identify cases and deaths was not analyzed, nor compared to other definitions. However, it has been shown that using ICD-10 coding to define TSCI is superior to previous versions of the ICD [28].

This study analyzed data from 22 countries and it is possible that differences in practices of coding the causes of death between them introduced a certain level of selection bias. We are aware of this limitation, but were not able to control for it, as we had no access to the death certificates themselves. However, the data were submitted by each country to Eurostat under specific guidelines included in an EU-regulation and we assume that this setup helped achieving the highest possible degree of harmonization that was possible under the circumstances.

In addition, we were not able to calculate incidence rates for the included countries, as data that would allow this were not available. As a consequence, we were not able to put the reported mortality rates in context with the overall occurrence of TSCI in the respective countries. Doing so would help interpreting our results and the heterogeneity that was observed between the mortality rates in our study.

Certain aspects of our findings, such as the age distribution of deaths may be to a certain extent biased by the varying period between the injury itself and the time of death. Our results should be interpreted with this in mind.

There is substantial heterogeneity in our pooled estimates (as documented by high I²) which should be taken into account when interpreting them.