RLMS is probably the largest population study from post-transition Russia to date, and provides a unique opportunity to study the determinants of health at a time of major social change and profound mortality crisis. The study sample was representative of the national population in terms of age, sex and geography, and the findings are therefore likely to be generalisable across Russia. The findings are supported further by the generally high response rate and the relatively low proportion of respondents with missing data (less than 0.5% for education and area of residence, less than 2% for smoking and alcohol consumption, and 6% for household income).

The study had several potential limitations. First, deaths were reported by the relatives and could have therefore been under-reported. However, mortality rates were similar to the national figures 27, and the association between education and mortality 410, and smoking and mortality 9 were in line with other studies suggesting the absence of any major bias in mortality ascertainment.

Second, twenty-five percent of individuals left RLMS without explanation, and these were more likely to be young, less educated, urban residents with higher incomes. People living alone, who were excluded because of the risks of under-reporting of mortality, were older and less affluent than other respondents. The study sample may thus not be fully representative for the national population but the similarity of the associations between education, smoking and mortality here and in other studies 4910 indicates good internal validity of the findings.

Third, difficulties in measuring income may have led to underestimation, as a result of inflation, wage arrears 28, and widespread non-monetary earnings (e.g. bartering or exchanging of favours). However, more detailed analyses elsewhere showed that material (non-monetary) factors did not predict mortality strongly 2425.

Fourth, alcohol consumption was not measured by individual type, and some alcohols, such as surrogate alcohols which are widely consumed, appear to predict mortality particularly strongly 7.

Finally, variations in factors between years could have influenced the results, since baseline data was taken from the year of entry into RLMS, and subjects entered in different years, although financial measures were adjusted for inflation. Two observations suggest that this effect is not likely to have been important, though. First, the prevalence of poor or very poor self-rated health in the study population varied very little between years (from 17% in 1994 and 1996 to 20% in 1995 and 1998); and second, adjusting the analyses for year of entry did not greatly affect the final results (data not shown).

Finally, using self-rated health a dichotomous measure, rather than utilising the full 5-point scale, could have led to the loss of some information. However, both measures were strongly and significantly associated with mortality, which suggests that any information loss was likely to have been small.

In RLMS, more than half the respondents reported "average" self-rated health, similar to the 1996 New Russia Barometer survey (45% in men and 49% in women) 29, and Nicholson et al (50% of both sexes) 30. However, there were some differences between the Russian studies: fewer respondents in RLMS reported "poor" or "very poor" health than in either the NRB (17% of men and 29% of women) 29, or in Nicholson et al (30% of men and 40% of women) 30. The prevalence of fair and poor subjective health in RLMS were almost identical to 2 rounds of the World Values Survey before the transition (1981 and 1990) 31. Self-rated health all the Russian studies was far worse than in English speaking countries where it is typically "good", although somewhat better than the "poor" health 14 typically reported in one Lithuanian study.

The strong, independent association between self-rated health and mortality in RLMS is consistent with studies in other countries 14, and the progressive relationship between worsening self-rated health and increased risks of mortality in men was similar to findings elsewhere 15. The relationship was less evenly graduated in women, but this could be due to their low prevalence of good and very good health.

Considering these findings in the light of the Russia's rapid fluctuations in mortality as well as theoretical mechanisms proposed for the association between self-rated health and mortality 14, may cast further light on these findings.

A key theory is that that self-rated health influences mortality because it is an "inclusive" measure that summarises objective health status 1432. The strong association between self-rated health and mortality in this study and the similarity of the relationships between education and both outcomes support this notion The high prevalence of poor subjective health in Russia may simply indicate high rates of disease, reflected in the low national life expectancy 3.

However, three findings suggest that serious illness is not the only influence on subjective health, and that other factors limit the explanation of mortality by health. First, international differences in the prevalence of self-rated health do not always follow national mortality rates. Self rated health was worse in Lithuania 14 than in this and other Russian studies 2930, despite Lithuania's higher life expectancy 22. Similarly, within India, the regions with the worst self-rated health do not experience the highest mortality 33. Cultural differences in reporting may offer a partial explanation, but are unlikely to account for the second, and related, finding that the prevalence of subjective health in Russia has not changed during the transition 31, despite major fluctuations in life expectancy 23. This could perhaps reflect the high frequency of sudden deaths in Russia 23 where the two commonest causes of premature mortality, cardiovascular disease (which includes sudden cardiac deaths) and external causes, may lead to unexpected deaths in apparently fit individuals who do not perceive themselves as ill 210.

Third, important individual level differences in the predictors of self-rated health and mortality also suggest that there were other factors at play. For example, the association between satisfaction and self-rated health, but not mortality, may indicate that psychological factors influence subjective ratings of health. The same was true of income, and although the connection between material measures and health is less clear, some positive emotion may possibly be associated with both. Although this range of variables was limited, such findings are consistent with another study where perceived stress predicted subjective symptoms of angina, but not the objective electrocardiographic signs of heart disease 34.

In contrast, although smoking and frequent alcohol consumption predicted mortality as expected 911, neither led to worse subjective health. In fact, somewhat paradoxically, frequent drinkers reported better health than more moderate consumers. This is the first time that better health has been demonstrated amongst Russians with risky health lifestyles individuals in a single, individual-level study. These findings are consistent with gender differences in health in Russia, where men, whose life expectancy 2 and health behaviours 1935 are worse than women, have better 17 or similar self-rated health 30.

There are at least two plausible explanations for these findings. First, Russians are known to drink to "help them forget everyday cares and difficulties" 36, and whilst they may forget their health concerns, the adverse physical consequences of heavy alcohol consumption remain. Second, some smoking and alcohol related deaths may not have been preceded by ill-health. As discussed above, rapid onset deaths, notably from cardiovascular disease and external causes 21023, are common, and alcohol has been implicated in sudden death 23. Although the role of smoking is less clear, it could act as a background risk factor for cardiovascular disease.

Abstainers from alcohol, as well as frequent drinkers, were at higher risk of death, consistent with the J-shaped curve seen elsewhere 37, but in contrast non-drinkers also experienced worse self-rated health than moderate consumers. Although the reasons are not clear, as elsewhere, some may previously have been heavy users of alcohol who were now too ill to drink 38.

Two further theoretical mechanisms linking self-rated health to mortality were considered, but these were not supported by our findings. The first was that poor health leads to more risky behaviours, which themselves lead to premature death 14. However, as noted above, self-rated health and health behaviours were not closely associated here. The second mechanism was that self-rated health acts as a marker of socioeconomic circumstances, and the latter influence mortality 14. Again, however, these data did not support this mechanism, since subjective health predicted mortality independently of income and education, and it did not interact significantly with either variable here or in another study 39.

In summary, although self-rated health predicted mortality as expected 14, differences in the associations with different predictor variables indicate that subjective evaluations of health may incorporate psychological and other influences in addition to subsequently fatal illness, whereas some health-damaging factors such as smoking and drinking are not considered. Frequent sudden deaths amongst apparently healthy individuals, some of whom had significant behavioural and other risk factors,, may partly explain these differences, and could also explain why self-rated health remained unchanged during Russia's "mortality crisis". These data offered only limited support for several theoretical mechanisms for the influence of self-rated health on mortality.

Further research should study individual causes of death, and a range of health outcomes, for example chronic disease 1516. International comparisons would be help to identify any systematic variations the associations shown in this study, for example in relation to rates of sudden death. In the short-term, however, our results indicate that, at least in Russia, extrapolating the predictors of self-rated health to those of mortality should be done with caution.