The number of alcohol-related deaths was estimated by identifying deaths that included any alcohol-related cause as the underlying or the contributing cause of death. The inclusion of a cause of death on the death certificate, either as the underlying or a contributing cause, indicates that the cause was considered as part of the causal chain leading to death by the physician responsible for writing the death certificate. Furthermore, in ICD-10 there is a set of exceptions to the rules regulating when a cause should be reported as underlying and contributing, affecting about 20% of all deaths (The National Board of Health and Welfare
2010). Consequently, all deaths including at least one alcohol-related cause of death are classified as alcohol-related deaths in this study. Over the study years, the underlying cause of death was alcohol-related on average in 43% of all the alcohol-related deaths. ICD-9 was used in 1991–1996, and ICD-10 was used in 1997–2008. The following causes were classified as alcohol-related in ICD-9: alcohol-related psychosis and mental disorders (291A-F, 291W, 291X), alcohol dependence (303), alcohol abuse (305A), alcohol-related nerve damage (357F), alcoholic myopathy (425F), alcoholic gastritis (535D) and alcoholic liver disease including liver cirrhosis (571A–D). In ICD-10, the following causes were classified as alcohol-related: alcohol-related psychosis and mental disorders (F10.0–9), alcohol-related nerve damage (G31.2), alcohol-induced epilepsy (G40.5), alcoholic myopathy (G72.1), alcoholic gastritis (K29.2), and alcoholic liver disease including liver cirrhosis (K70.0–4, K70.9), alcoholic pancreatitis (ICD-10: K85.2, K86.0) and accidental alcohol poisoning (ICD-10: X45). The proportion of alcohol-related deaths in 1996 and 1997 was similar (5.1 and 5.3%, respectively), indicating that the shift between ICD-9 and ICD-10 did not substantially influence the likelihood of classifying a death as alcohol-related. The number of smoking-related deaths was estimated using the indirect method developed by Preston et al. (
2010). The method is based on data from 21 high income countries for the period 1950–2007 and uses age- and sex-specific lung cancer death rates as indicators of the population level health damage from smoking. The method develops a regression model that uses lung cancer mortality to predict smoking-related mortality in other causes of death. The coefficients from this regression model and information on expected lung cancer death rates among non-smokers as well as the observed lung cancer mortality rates in the Swedish population are used to estimate the overall fraction of deaths attributable to smoking. The original method is restricted to ages 50 and above, ages at which most lung cancer deaths occur, and we rely on an extension of this method presented in Martikainen et al. (
2014) to also cover ages between 30 and 49 years. For a more detailed account, see Preston et al. (
2010). In our analyses, lung cancer death was defined as any death having lung cancer as the underlying or a contributing cause. Lung cancer deaths were identified as malignant neoplasms of either lung, trachea, or bronchus [162A, 162C-E, 162W, or 162X in ICD9 (1991–1996) and as C33, C34.0–3, or C34.8–9 in ICD 10 (1997–2008)].
The estimated number of alcohol- and smoking-related deaths and all-cause deaths were used to calculate mortality rates and life expectancies. Age standardized mortality rates (ASMR) were calculated to assess trends in alcohol and smoking-related mortality over the time period by education and sex. We also estimated temporary life expectancies (Chiang
1984; Preston et al.
2000) between the age 30 and 74. The temporary life expectancy reflects the average number of years lived within a set age bracket (Arriaga
1984; Chiang
1984), in this case 30–74.
To assess the contribution of alcohol and smoking to life expectancy, three sets of life expectancies were calculated by sex, period and education. First, observed life expectancy was calculated using the observed death rates. Second, life expectancy was re-calculated after excluding all alcohol-related deaths from the calculation of death rates. Third, life expectancy was re-calculated after excluding all smoking-related deaths. When calculating life expectancy without alcohol or smoking, deaths classified as alcohol and smoking-related, respectively, were right-censored. Thus, the time at risk remained the same in the different expectancy calculations, only the number of deaths differed.