Consistency of Cause-of-Death Mortality Data across Subnational Entities in the USA, France, Germany and Russia
Pavel Grigoriev, Max Planck Institute for Demographic Research
Vladimir M. Shkolnikov, MPIDR
Jacques Vallin, Institut National d''Etudes Démographiques
France Meslé, Institut National d''Etudes Démographiques
Inna Danilova, National Research University Higher School of Economics
Magali Barbieri, Institut National d''Etudes Démographiques
Dmitri A. Jdanov, National Research University Higher School of Economics
In the current study, we use an indirect technique to analyze the consistency of cause-specific mortality statistics at the subnational level in four countries, which have rather different systems on producing information on causes of death - France, Germany, the USA, and Russia. Based on the cause-specific mortality data, we estimate the prevalence of major groups of causes of death at the subnational level (regions) and compare it with the inter-regional average prevalence. We visualize deviations from the inter-regional average through the heatmap to identify suspicious outliers that are unlikely to be solely explained by real differences in morbidity and mortality prevalence.
Using the proposed approach, we found that the French system of producing information on causes of death results in a high level of comparability across subnational entities. In Germany, the USA, and Russia, we found cases of significant deviation from an inter-regional average level. That potentially indicates specific peculiarities in certifying causes of death and/or selecting the underlying cause of death.
The causes of death, which are easy to be recognized as underlying, showed high consistency across subnational entities in all the four countries. On the contrary, causes that do not have very strict diagnostic criteria and/or often accompanied by comorbidities tended to have higher variability.
Though ICD-manualsprovide detailed instructions on the coding process, specifying the sequence ofcauses of death and on selecting the underlying cause, there is still a scopefor actual coding practices to vary [2-6]. The process of selecting theunderlying cause highly depends on the information reported in the deathcertificate and on the interpretation of the ICD-rules by the coder. Comparabilityissues arise not only in international comparison. Certifying and codingpractices may vary significantly within countries as well [1,8-10].
Countries choosedifferent ways of producing information on causes of death. In this paper, wecompare four countries (France, Germany, the USA, and Russia) which have ratherdifferent systems of producing cause-of-death mortality data. We analyze howthese differences affect the within-country consistency of cause-specificmortality statistics.
Data and Methods
We used mortality datafor areas of France, Germany, the USA and Russia on the first level ofadministrative division (hereinafter regions). The data was taken for 5-yearperiods (2005-2009 for France, Germany, and Russia; 2008-2012 for the USA) andfor 67 major groups of causes of death. To eliminate the potential bias causedby a small number of events only those regions with a population over 1mln.were included in the analysis.
To identify potentialinconsistencies in certifying and coding practices across subnational entities weestimate the prevalence of the major groups of causes of death in the regional mortalitystructures. The cause-specific share of the all-cause age-standardized deathrate was used as an indicator of cause-specific mortality prevalence:
Sr,c= SDRr,c / SDRr ,
where SDRr,c is theage-standardized death rate for cause c in region r, SDRr isthe all-cause age-standardized death rate in region r.
For each possiblecombination region/cause we have calculated the deviation from thecross-regional mean:
Vr,c = ∑│Sr,c- ̅S,c│/ ̅S,c *100%,
with ̅S,c asthe mean of regional Sr,c.
The heatmap is a tablewith colored cells which is used for a visual presentation of the matrices V.Each row of the heatmap corresponds to a particular cause; each column representsa specific region. Cells are colored based on the values of Vr,c.The examples of heatmaps for France and Russia are presented in theAppendix.
Cells (region/causeintersections) with relatively high deviation from the inter-regional averagelevel are clearly visible on the heatmaps. In some cases, the level ofdeviation is too high to be explained by real differences in epidemiologicalpatterns. Such suspicious cases might be caused by the differences in approachto certifying causes of death. We identify most problematic areas analyzing thetotal number of suspicious cells for region or cause.
The comparativeanalysis showed that France has the highest consistency of cause-specificmortality data at subnational level. In Germany, the USA, and Russia, thecolor dispersion is higher, i.e. we observe more outliers on the heatmap.That potentially indicates inconsistency at regional level in certifying causesof death and/or choosing the underlying cause. Among the three countries, Russiahas the highest level of within-country inconsistency.
Thecauses of death, which are easy to be recognized and adjudicated as underlyingshowed high consistency across subnational entities in all the four countries. Lessspecific causes, which do not have very strict diagnostic criteria and/or oftenaccompanied by comorbidities, tend to show higher variability. This possibly indicatesthat medical professionals may have a different view on whether information onthese causes should be reported in the death certificate and whether thesecauses should be indicated as underlying by manual coding.
1. FrançaE, de AbreuDX,RaoC, LopezAD. Evaluationof cause-of-death statistics for Brazil,2002-2004. IntlJEpidemiol.2008;doi:10.1093/ije/dyn121.
2. JouglaE, PavillonG, RossollinF,De SmedtM, BonteJ. Improvement of the quality and comparability ofcauses-of-death statistics inside the European community.EUROSTAT task force oncauses of death statistics.Rev Epidemiol Sante Publique.1998;46(6):447-456.
3. MurrayCJ, DiasRH, KulkarniSC, LozanoR, StevensGA, EzzatiM. Improving the comparability of diabetes mortalitystatistics in the United States and Mexico.Diabetes Care.2008;doi:10.2337/dc07-1370.
4. PercyC, MuirC. The internationalcomparability of cancer mortality data.AmJEpidemiol.1989;129(5):934-46.
5. LuTH, WalkerS, JohanssonLA, HuangCN.An international comparison study indicated physicians habits in reportingdiabetes in part I of death certificate affected reported national diabetesmortality. J Clin Epidemiol.2005;58:11507;
6. LuTH, WalkerS, AndersonRN,McKenzieK, BjorkenstamC, HouWH. Proportion of injury deaths with unspecifiedexternal cause codes:a comparison of Australia, Sweden, Taiwan and theUS.Injury Prevention.2007;13(4):276-281.
7. Ylijoki-SørensenS, SajantilaA,LaluK, BøggildH, BoldsenJL, BoelLWT. Coding ill-defined and unknown cause ofdeath is 13 times more frequent in Denmark than in Finland.Forensic SciInt.2001;115(1-2):15-32.
8. DanilovaI, ShkolnikovVM, JdanovDA,MesléF., VallinJ. Identifying potential differences in cause-of-death coding practicesacross Russian regions.Population Health Metrics.2016;doi:10.1186/s12963-016-0078-0
9. LuTH, LeeMC, ChouMC. Accuracy ofcause-of-death coding in Taiwan:types of miscoding and effects on mortalitystatistics.Int.J.Epidemiol.2000;doi:10.1093/ije/29.2.336
10. HartelohP, de BruinK., KardaunJ.Thereliability of cause-of-death coding in The Netherlands. European Journal ofEpidemiology.2010;doi:10.1007/s10654-010-9445-5
Presented in Session 1197: Mortality and Longevity