Comparison of relative risks obtained in ecological and individual studies: some methodological considerations

This paper is concerned with the problem of estimating relative risks from ecological correlation studies. In the first part, some of the biases encountered when analysing aggregated data are discussed and in particular attention is focused on the shape of the dose-response relationship obtained from aggregated and non-aggregated data, on the need for extrapolation and on the scale of aggregation. In the second part some empirical observations are made on these points by means of four examples concerning the relative risk between smoking and different pathologies. The estimates of relative risks derived from French geographical data and from case control or cohort studies are compared. The performance of ecological studies is discussed with respect to the strength of the risk factor considered, the geographical distribution of counfounding factors and the adjustment of different models.