Commentary: problems in Finnish or British data—or a true difference?BMJ 1998; 316 doi: https://doi.org/10.1136/bmj.316.7125.105 (Published 10 January 1998) Cite this as: BMJ 1998;316:105
- Seppo V P Koskinen, assistant chief physician ()a
- a National Public Health Institute, Department of Health and Disability, Mannerheimintie 166, FIN-00300 Helsinki, Finland
Chaturvedi et al report a clear class gradient in mortality among diabetic people. In two separate data sets the death rate in the lowest socioeconomic group was roughly twice that in the highest group. In our recent study, covering the total Finnish population, we only found a non-significant 7% excess mortality among diabetic women in the lowest class compared with the highest class.1 Among diabetic men the corresponding excess was significant but still small (25%). The discrepancy between our results and those of Chaturvedi et al may reflect deficiencies in the Finnish or British data sets, or both, or reveal a true difference between the countries.
Because of our large data set, including 11 215 deaths among diabetic people, random variation could have had only a minor effect on our findings. The validity of the results can, however, be questioned. As the mortality follow up was complete, we are left with the possibility that the method of case ascertainment produced biased results. In the Finnish study diabetic people were identified on the basis of entitlement to free medicines for the treatment of diabetes, while the British study also included people with diabetes treated by diet alone.
If diet treated diabetes, with better than average prognosis, is particularly common in the higher social classes, the exclusion of people with diet treated diabetes might have biased the Finnish results.2 Secondly, in the higher social classes a particularly large proportion of drug treated diabetic people may have insulin dependent diabetes,2 which tends to increase the risk of death more than non-insulin dependent diabetes.3 Thus, analysing both types of diabetes together might have artificially diluted the socioeconomic differentials in mortality. Thirdly, if a particularly large proportion of white collar employees with mild diabetes did not bother to obtain entitlement to free drugs, our results would also be biased.
All three potential sources of bias would be expected to have only a minor impact on younger diabetic people, most of whom suffer from insulin-dependent diabetes. As we found no socioeconomic gradient in mortality among women aged 30–49 and only a slight gradient among men of the same age, it seems unlikely that these potential sources of error played a decisive role in our results. Moreover, the national drug register data and analyses of representative population samples gave very similar estimates of the prevalence of drug treated diabetes.4
The second main question is whether the British data are reliable. Could the observed socioeconomic gradient in mortality be the result of random variation or of problems in the representativeness of the British datasets?
Random variation might have played a role in the British study. The wide confidence intervals around the mortality ratios include the corresponding values found in the Finnish study. However, the similarity of the findings from two separate datasets increases confidence in the British results.
Civil servants in London and patients of diabetes clinics in London may not be representative of the total British population. On the other hand, there is no obvious reason why the results, particularly in the Whitehall study, would be very different from those in the total population. Tables 1 and 2 in Chaturvedi et al's paper do, however, raise questions about health related selection. Among people with diabetes, heart disease seems to have been much more common in the lower than in the higher social classes. This difference is far greater than that in the non-diabetic population, and, paradoxically, diabetes seems actually to decrease the probability of heart disease in the higher social classes. Could this finding be partly explained by either early retirement or a decline in socioeconomic position among those diabetic members of the higher classes who develop heart disease? Could such selection influence the results more in Britain than in Finland, where pensioners were included in the analysis and were classified according to their former occupation?
The final main alternative is that there really is a true difference between the countries. In the British data there are clear socioeconomic differences in most risk factors, including smoking and blood pressure. The Finnish data do not include direct information on risk factors, but indirect evidence, based on mortality from specific causes, suggests that among people with diabetes there are only small class differences in smoking.
Further research should be carried out to assess whether (and why) the situation truly differs between countries. We need results from Finnish datasets that cover persons with diabetes treated by diet and include information on the disease (type, treatment, duration) and risk factors. British analyses of (pooled?) population based datasets with a large number of deaths would help in assessing the validity of the findings of Chaturvedi et al. The situation in other countries should also be studied.