The results of our study agree with the findings of a EU report in 2006 , including data from 21 different European countries, in which differences in LE at birth between the lowest and highest socio-economic groups are estimated to be between 4-6 years among men and 2-4 years among women. They also agree with German studies conducted with other datasets and confirm that in Germany there is a socio-economic gap in life expectancy. For example, a recent German study  conducted with administrative records from the German Public Health Pension System found that LE rises almost linearly with lifetime earnings - a proxy for socio-economic status. An analysis conducted with data of the LE-Survey  showed that while 45-year-old men, who work as "Beamte" (public officers), have a remaining LE of about 32 years, manual workers have a remaining LE of about 26 years. Among women, the LE gap between 45-year-old "Beamte" and simple employees is estimated to be about 5 years. The data of the socio-economic panel (SOEP) have provided the basis for different analyses which all showed the existence of a socio-economic gap in LE. Lauterbach et al.  calculated that the probability to reach retirement age for the men belonging to a lower income group (< 1,500 Euro) compared to those belonging to a higher income group (> 4,500 Euro) is 79,1% and 90,0%, respectively. Lampert et al.  estimated that the difference in "healthy" LE among men and women earning less than 60% of the median of the income in Germany and those earning more than 150% of the median is about 14 healthy years for the men and 10 for the women.
Possible reasons that would explain the differences in LE by socio-economic status have been amply discussed in the literature . They include explanations related to different life styles among lower and higher socio-economic groups. This has also been investigated by another study using the German MONICA/KORA data , which showed that men and women with a higher educational level have a lower consumption of tobacco and alcohol, are more likely to engage in leisure-time physical activity, have a lower Body Mass Index, and have less job strain.
Our results also show that if German men and women with lower income have diabetes or myocardial infarction then their LE is reduced more by the disease than the LE of richer men. Adjustment for diabetes and myocardial infarction for people with myocardial infarction and diabetes, respectively, yields only very limited changes of the LE estimates. This means that the influence of the interaction of these diseases on the reduction of LE is negligible, and that the reduction in LE lies mostly in the social differences in our regression model. This indirectly confirms the results of a German study, also conducted in the region of Augsburg, which showed that the number of deceased persons with a first time myocardial infarction was about 60% higher among those belonging to a lower socio-economic group compared to those belonging to a higher socio-economic group . Another German study demonstrated that persons with diabetes mellitus belonging to a lower socio-economic group are more susceptible to diabetes-specific complications, such as micro and macrovascular complications, compared to those who belong to a higher socio-economic group . Clearly, a higher frequency of diabetes-specific complications could contribute to a shorter life expectancy.
Income and education seem to have similar effects on the LE of both men and women. Computation of LE based on separate analyses of income and education yielded similar estimations (79.06 vs. 79.71 and 81.83 vs. 81.96 for the men in the lower and higher groups, respectively, and 85.57 vs. 85.95 and 88.57 vs. 88.37 for the women). These figures also show that the social group with both the highest income and education is the group with the best LE in our analysis.
The group of those men and women who have an income corresponding to the lowest tercile and an education corresponding to a qualification for university entrance/completion of undergraduate studies has the worst LE in our analysis. This group would also probably show the most significant reduction in LE when affected by a disease . In our dataset, the limited number of cases did not make an estimation of LE within the group of patients with diabetes or myocardial infarction possible. This remains an important topic for future research.
The value of the estimated mean life expectancy is higher than the mean LE reported by the 2002/2004 life table for Germany (men: 75.89; women: 81.55) and by the Bavarian life table, whose values (men: 76.47; women: 81.92) are the second highest among the German states. Our estimation remains slightly higher, even after correction of the left data truncation. However, this hardly affects the study results, whose major purpose is to show a gradient among different socio-economic groups and not to calculate a precise estimate of the LE of a newborn baby, if the current death rates continue to apply throughout his or her life. A possible reason for the difference resides in the small number of old and deceased people in the available dataset. This is particularly true in the case of women where the number of deceased participants (522) is almost half compared with men (1032), implying that the survival data of women is more heavily censored. Another reason could also lie in an over-representation of health conscious persons in the dataset, as health conscious persons are, presumably, more inclined to participate in a study dealing with health issues, and having a healthier lifestyle, they might live longer.
Most 95% confidence limits of the LE value overlap with other groups. This is largely due to the relatively small number of deceased people in the available data. However, it should be noted that the degree of overlap between the confidence intervals does not directly quantify the statistical significance regarding the existence of a difference between the life expectancies of the different groups.
The use of a parametric method was appropriate for the present analysis. Parametric methods are particularly useful to investigate small changes induced in the distribution function by certain data variables (e.g. sex, socio-economic status in the case of life expectancy). Also, they are more powerful than non-parametric methods for sparse data, provided the assumed shape for the underlying distribution function is a reasonable description of the data. The principal advantage of non-parametric methods is that they are free of any assumptions regarding the underlying distribution function, which protects against potential biases in the results if a parametric method is used with a distribution function that is actually a poor match for the data. However, the larger freedom regarding possible distribution functions explored by non-parametric methods means that datasets need to be very large to obtain accurate results with this method. Also, it is often difficult to extrapolate the estimated non-parametric distribution function into regimes that are not well sampled by the data. This is for example the case in building a life table with the MONICA/KORA data; the tail of the age distribution is in fact not well determined by non-parametric methods due to the small number of deceased people in the available data.
The small sample was clearly a limitation of this analysis and biased the estimates of LE. The inclusion in the survey of only people aged between 25-74 years and the absence of people living in institutions also limited a precise quantification of life expectancy. The survey is also limited by the regional data collection and possible recruitment bias.
While these limitations did not allow a robust estimation of absolute life expectancy, the results of the study confirmed, as have other German and international studies, the existence of a socio-economic gap in life expectancy. The study also provides important new information by addressing a public health topic rarely discussed to date, i.e. the socio-economic differences in the relative reduction of LE of people who have diabetes mellitus or myocardial infarction.