Where conflict's medical consequences remain unchanged.

www.thelancet.com Vol 381 March 16, 2013 901 Submissions should be made via our electronic submission system at http://ees.elsevier.com/ thelancet/ northwest of England. We followed up almost 15 000 patients over an 8-year period using a district-wide diabetes register encompassing both primary and secondary care. Our fi ndings indicate a falling mortality rate in type 2 diabetes. Simmons and colleagues recruited a large number of individuals in the screening and control practices, but a relatively small number of patients with diabetes was followed up. The intraclass correlation coeffi cient, which can only be indirectly estimated owing to the confounding eff ects in the two paired clusters, is unreliable. It is also possible to overestimate the stability of a trial of less than 40 clusters owing to the probability of large standard errors. Furthermore, the validity of permutation tests becomes unreliable since there are imbalances in the cohort cluster size and consequently distortion in the paired matching results. The assumptions of the Cox proportional hazard modelling, which tests the mortality risks in the screened and controlled cohorts, is normally based on error variations in the cohort groups and not within matched pairs. Finally, Simmons and colleagues have not revealed whether there were adjustments for other covariates in the all-cause mortality in both cohorts. We suggest that Simmons and colleagues present their hazard ratios alongside a measure of time such as median time to death in comparison with the screened and controlled cohorts.