1 mortality database approximately 12% of deaths have been assigned - - PDF document

1

Allowing for garbage codes in mortality analyses matters! The evidence from Central and Eastern Europe Agnieszka Fihel1, France Meslé2

1University of Warsaw, Warsaw (Poland); 2Institut National d’études Démographiques, Paris

Corresponding author: Agnieszka Fihel, a.fihel@uw.edu.pl Paper submitted to the 2017 International Population Conference, IUSSP, Cape Town, South Africa, 29 October to 4 November 2017 Extended abstract

• Introduction. Mortality data by detailed causes of death allow for the analysis of

epidemiological trends and impacts of health policies. Comparing mortality trends in an international perspective constitutes an important methodological challenge and in order to avoid arriving at false conclusions, a very detailed insight into the cause-of-death data is

• required. Inter alia, a procedure concerning ill-defined and unknown causes of death needs to

allow for country-specific registration rules and practices. This concerns in particular countries, where the assignment of deaths to ill-defined and unknown causes is relatively

• frequent. We use the example of Poland to show the importance of accurate analytical

techniques when studying the trends in mortality by single causes of death. Indeed, according to the WHO Mortality Database (WHO, 2015b) mortality due to acute myocardial infarction (AMI) or cerebrovascular diseases in Poland was recently at a similar level than that registered in some West European countries (Austria, Germany or the United Kingdom). Other studies, however, show that the level of risk factors related to cardiovascular diseases is still significantly higher in Poland than in Western Europe (Bandosz et al., 2012; Bobak and Marmot, 1996; Tykarski et al., 2005), whereas the advancement of treatment methods is significantly lower (EOHCS and WHO, 1999; McKee and Nolte, 2004). The inconsistency between on the one hand, relatively low mortality due to AMI and cerebrovascular diseases in Poland and on the other hand, relatively high level of risk factors and less advanced treatment methods might stem from the Polish coding practices concerning the so-called garbage codes (GCs). It is highly possible, in point of fact, that GCs take over a vast part of mortality that in other countries is registered as due to well-defined medical conditions. Garbage codes is a term introduced by Murray and Lopez (1996) to designate all causes of death that are not useful in the analyses of public health and mortality, that is: (1) codes that cannot or should not be considered as underlying causes of death (i.e. symptoms, signs and ill-defined conditions), (2) that constitute intermediate causes of death (i.e. heart failure), or (3) that remain unspecified within larger groups of causes (i.e. malignant neoplasm of other and ill-defined sites). Garbage codes are used with a different frequency and in a different context from one country to another. For the 105 countries associated within the World Health Organization (WHO)

2

mortality database approximately 12% of deaths have been assigned to the GCs1 since 1990 (Mathers et al., 2005). In 20 countries, of which 5 in Europe2, this percentage was relatively high, that is equal to 20% or higher. In Poland more than 6% of deaths are registered due to unknown and ill defined causes (the 18th chapter of the 10th revision of the International Statistical Classification of Diseases and Related Health Problems, ICD-10) each year, and further 19% due to other GCs. The highest number of deaths attributed to GCs in Poland concerned cardiovascular diseases, among which the most frequent remained cardiac arrest (code I46 according to the ICD-10), heart failure (I50) and generalized and unspecified atherosclerosis (I70.9). These causes of death together with atherosclerotic cardiovascular or heart disease (ICD-10 code: I25.0, .1), another GC, constituted in Poland more than a half (54%) of all deaths assigned to cardiovascular diseases in 2013. Cardiovascular GCs have particularly low informative value and therefore, they remain useless in advanced analyses of epidemiologic situation, assessment of health policy efficiency and international comparisons (Naghavi et al., 2010).

• Objective. In this study, we aim at presenting and testing an accurate technique for

redistributing cardiovascular GCs across well-defined causes of deaths, which might improve comparability of mortality data in the international perspective. We focus here on Poland where the frequency of GCs registration is very high and spatial differences in registration of cause-specific mortality remain important (i.e. Wojtyniak et al., 2012). The common and widely accepted practice in using cause-specific mortality data so far has consisted of redistributing age-, sex- and cause-specific death counts (or seldom: death rates) assigned to GCs proportionally across all other (well-defined) causes (WHO, 2013, 2015a). This method imposes a strong, unrealistic assumption that in case of each disease and each external cause

• f death the difficulty in recognizing the appropriate diagnose is the same. Also, this practice

favors causes of death that are already large and important, even though pathophysiologic or statistical links between those causes and GCs may not exist at all. Finally, for some countries such a proportional redistribution seems inappropriate3. The method that we propose considers selected GCs and selected well-defined causes of death only, in their country- specific context.

• Data. The data concern death counts by age, sex and specific causes of death in Poland

derived from the World Health Organization mortality database (WHO, 2015b), national statistical offices and the Human Cause-of-Death database (causesofdeath.org) – in the period covered by the 10th revision of ICD, that is since 1997, – due to four cardiovascular GCs4: atherosclerotic cardiovascular or heart disease (ICD-10 code: I25.0, .1), cardiac arrest (I46), heart failure (I50), atherosclerosis (I70).

1 The following causes were included: ill defined (ICD-10 codes starting with R), ill-defined

cardiovascular diseases (I47.2, I49.0, I46, I50, I51.4, I51.5, I51.6, I51.9, I70.9), neoplasms of unspecified sites (C76, C80, C97).

2 Cyprus, Greece, Poland, Portugal and San Marino. 3 For instance, in case of Russia or Ukraine ill-defined causes (constituting a part of GCs) are assigned

more often instead of cardiovascular diseases (Meslé and Vallin, 2003, 2012).

4 These causes of death were chosen because of their low informative value and high frequency in

national registration.

3

• Method. The method consists of two steps. In the first step we search for geographical

relationships in assignment of GCs in Poland. Following the method proposed by S. Ledermann (1955), we analyze proportions of deaths assigned to ill-defined causes and selected well-defined causes of death at regional level. Ledermann showed that in French and Italian regions with high proportion of deaths due to ill-defined causes, proportion of deaths due to pulmonary tuberculosis was low, and vice versa, which proves a certain exchangeability between those causes of death. In a later study for France similar method was adopted to deal with deaths assigned to three large groups of ill-defined causes (Vallin and Meslé, 1988). We aim at finding similar reverse relationships in shares of death due to cardiovascular GCs and selected well-defined causes (respiratory diseases, well-defined cardiovascular diseases, falls or complications stemming from medical interventions) and groups of causes (whole chapters of the ICD-10) registered at the regional level. We calculate statistical correlations in Stata. For statistically significant correlations we perform, in a second step, a redistribution of cardiovascular GCs across well-defined causes of death proportionately to the slope of

• correlation. To each well-defined cause of death we attribute the proportion of GC deaths

according to the slope in linear regression between the GCs time series and each well-defined cause of death time series. If for a given GC the total of slopes exceeds -1, we rescale proportionately the slopes so that their sum is equal to -1. Preliminary results. We performed an analysis for 380 regions at the NUTS-4 level and found negative, statistical correlations between shares of deaths assigned to the cardiovascular GCs and shares of deaths assigned to some ICD chapters. On the basis of slopes of linear regressions between those time series we redistributed deaths assigned to GCs across and inside selected ICD chapters. As a result, the level of mortality from well-defined cardiovascular diseases increased in Poland significantly, by 45%, from neoplasms by 18% and from respiratory diseases by 11%. Mortality due to AMI (ICD code: I21) and stroke (I64) became considerably higher than in some West European countries (such as for example England and Wales or Germany) and comparable to that registered in other countries of Central Europe (such as the Czech Republic, Figures 1-4). Obviously, when dealing with GCs there is no single general solution for all countries and all causes of death. This is why we seek to develop and test the most detailed method, adjusted to each country’s specificity concerning coding practices and epidemiologic reality.

4

Figure 1. Standardized death rate (per million) due to acute myocardial infarction (I21) for selected countries, 1997–2013 Source: Authors’ elaboration on the basis of WHO (2015b). Figure 2. Standardized death rate (per million) due to acute myocardial infarction (I21) for selected countries, adjusted and non-adjusted for Poland, 1997–2013 Source: Authors’ elaboration on the basis of WHO (2015b).

200 400 600 800 1000 1200 1400 1600 1800 2000 1997 1999 2001 2003 2005 2007 2009 2011 2013 SDR PER MILLION CZECH GERMANY JAPAN POLAND ROMANIA ENGLAND 200 400 600 800 1000 1200 1400 1600 1800 2000 1997 1999 2001 2003 2005 2007 2009 2011 2013 SDR PER MILLION CZECH GERMANY JAPAN POLAND ROMANIA ENGLAND

5

Figure 3. Standardized death rate (per million) due to stroke (I64) for selected countries, 1997–2013 Source: Authors’ elaboration on the basis of WHO (2015b). Figure 4. Standardized death rate (per million) due to stroke (I64) for selected countries, 1997–2013, adjusted and non-adjusted for Poland, 1997–2013 Source: Authors’ elaboration on the basis of WHO (2015b).

500 1000 1500 2000 2500 3000 1997 1999 2001 2003 2005 2007 2009 2011 2013 SDR PER MILLION CZECH GERMANY JAPAN POLAND ROMANIA ENGLAND 500 1000 1500 2000 2500 3000 1997 1999 2001 2003 2005 2007 2009 2011 2013 SDR PER MILLION CZECH GERMANY JAPAN POLAND ROMANIA ENGLAND

6

Literature Bandosz, P., O’Flaherty, M., Drygas, W., Rutkowski, M., Koziarek, J., Wyrzykowski, B., Bennett, K., Zdrojewski, T., Capewell, S., 2012. Decline in mortality from coronary heart disease in Poland after socioeconomic transformation: modelling study. British Medical Journal 344. doi:http://dx.doi.org/10.1136/bmj.d8136 Bobak, M., Marmot, M., 1996. East-West mortality divide and its potential explanations: proposed research agenda. British Medical Journal 312, 421–425. EOHCS, WHO, 1999. Health Care Systems in Transition - Poland. WHO Regional Office for Europe, Copenhagen. Fihel, A., Muszyńska, M., Wróblewska, W., 2014. Umieralność z przyczyn nieznanych i niedokładnie określonych oraz jej trwałe zróżnicowanie terytorialne w Polsce. Studia Demograficzne 165, 83–102. Ledermann, S., 1955. La répartition de decés de causa indeterminée. Review of the International Statistical Institute 23, 47–57. Mathers, C., Fat, D.M., Inoue, M., Rao, C., Lopez, A.D., 2005. Counting the dead and what they died from: an assessment of the global status of cause of death data. Bulletin of the World Health Organization 83, 171–180. McKee, M., Nolte, E., 2004. Health sector reforms in central and eastern Europe: how well are health services responding to changing patterns of health? Demographic Research Special collection 2, 163–182. Meslé, F., Vallin, J., 2012. Mortality and causes of death in 20th-century Ukraine, Demographic Research Monographs. Springer, Dordrecht. Meslé, F., Vallin, J., 2003. Mortalité et causes de décès en Ukraine au XXe siècle, Les cahiers de l’INED. INED, Paris. Murray, C., Lopez, A.D., 1996. The global burden of disease: a comprehensive assessment of mortality and disability from diseases, injuries, and risk factors in 1990 and projected to 2020. Harvard University Press, Cambridge. Naghavi, M., Makela, S., Foreman, K., O’Brien, J., Pourmalek, F., Lozano, R., 2010. Algorithms for enhancing public health utility of national causes-of-death data. Population Health Metrics 8, 9. Tykarski, A., Posadzy-Małaczyńska, A., Wyrzykowski, B., Kwaśniewska, M., Pająk, A., Kozakiewicz, K., Rywik, S., Broda, G., 2005. Rozpowszechnienie nadciśnienia tętniczego oraz skuteczność jego leczenia u dorosłych mieszkańców naszego kraju. Wyniki programu WOBASZ. Kardiologia Polska (Polish Heart Journal) 63, 1–6. Vallin, J., Meslé, F., 1988. Les causes de décès en France de 1925 à 1978, Travaux et

• Documents. INED, Paris.

WHO, 2015b. WHO Mortality Database. WHO, 2015a. World Health Statistics 2015. WHO, Geneva. WHO, 2013. WHO methods and data sources for global causes of death 2000-2011. WHO, Geneva. Wojtyniak, B., Rabczenko, D., Pokarowski, P., Poznańska, A., Stokwiszewski, J., 2012. Atlas umieralności ludności Polski w latach 1999–2001 i 2008–2010 [Mortality atlas for the population of Poland, 1999-2001 and 2008-2010].