Correspondence www.thelancet.com Vol 375 April 24, 2010 1435 model, the adjusted hazard ratio for the lowest HbA 1c decile versus the optimum decile was 1·48 (95% CI 1·28–1·71). Eishu Nango and Takeo Saio raise the possibility that duration of diabetes is a confounding factor. Although significant in our models (for each year duration, adjusted hazard ratio 1·01, 95% CI 1·01–1·02; p=0·0002), we originally excluded diabetes duration empirically on the basis of its interaction with baseline morbidity. Furthermore, the inclusion of diabetes duration in the models did not meaningfully change any of the key relations. For example, in model 1 (table 3 in original paper), the adjusted hazard ratios when duration of diabetes was added to the base-case model were: insulin versus oral hypoglycaemic agents, 1·47 (1·38–1·58); HbA 1c decile 1 versus decile 4, 1·51 (1·30–1·74); and HbA 1c decile 10 versus decile 4, 1·80 (1·55–2·08). Hans DeVries challenges the notion of an increased mortality risk with insulin. We believe that there is a case to answer for insulin. For example, we did a further sensitivity analysis on a subgroup of patients who had the very lowest morbidity, and at baseline had no left-ventricular dysfunction, no renal disease, a Charlson score of less than 4, and a baseline HbA 1c of 9·0% or less. In this healthy group, the adjusted hazard ratio for insulin-based treatment versus oral combination treatment was 2·08 (1·75–2·47). This pattern of increased mortality with insulin persisted under all analytical conditions. Kennedy Cruickshank highlights weight gain as a confounding factor. We agree that weight-gain is an issue, although its effect is likely to present a linear pattern across the HbA 1c range; thus it probably did not have an effect here. We discussed the availability of cause of death in our report. To reiterate, data were available for a proportion of deaths (about 25%) but we concluded that this was insufficient to make an inference. Several correspondents discuss hypoglycaemia as a direct or indirect cause of death in those with low HbA 1c concentration. Although we speculated that this could be an issue in the discussion section of our report, we did not have any specific data to support this, although very recent data allude to this being a concern. 1 Within our data, an HbA 1c of about 7·5% was without doubt the value with the lowest likelihood of adverse outcome. The survival pattern between the two cohorts differed in that it was not as pronounced across the HbA 1c range in the oral combi- nation group as in the insulin group. Although some of the specific decile values did not achieve significance at the conventional level, the pattern of association across the range was clear. Although we discuss the limitations of the study in our report, most of the comments that we have received relate to limitations typical of this study design. Although we acknowledge that our findings disagree with convention, they are nonetheless consistent with evidence from randomised controlled trials. We remain convinced our data are robust and warrant further investigation. CJC has received research grants from various health-related organisations, including Astellas, Diabetes UK, the European Association for the Study of Diabetes, the Engineering and Physical Sciences Research Council, Ferring, GlaxoSmithKline, Lilly, Medtronic, the Medical Research Council, Pfizer, Sanofi-Aventis, the National Health Service, and Wyeth, and consults for Amylin, Aryx, Astellas, Boeringher Ingelheim, Bristol-Myers Squibb, Diabetes UK, Eisel, Ferring, GlaxoSmithKline, Ipsen, Lilly, Medtronic, Merck, Pfizer, Sanofi-Aventis, Takeda, and Wyeth. CDP consults for Astellas, Ferring, Lilly, Medtronic, Sanofi-Aventis, and Wyeth (Pfizer). *Craig J Currie, Chris D Poole currie@cardiff.ac.uk Department of Medicine, School of Medicine, Cardiff University, Pharma Research Centre, Medicentre, Cardiff CF12 4UJ (CJC); and Department of Epidemiology, Pharmatelligence, Cardiff, UK (CDP) 1 Kubiak T, Wittig A, Koll C, et al. Continuous glucose monitoring reveals associations of glucose levels with QT interval length. Diabetes Technol Ther 2010; 12: 283–86. Method of delivery and pregnancy outcomes In their survey of the risks of caesarean section without medical indication, Pisake Lumbiganon and colleagues (Feb 6, p 490) 1 make serious errors in their analysis of the data, which completely change the conclusions of the paper. First, women cannot elect to have a spontaneous vaginal delivery, only to plan for one. Thus the data for spontaneous vaginal delivery need to be combined with those of operative vaginal delivery and emergency intrapartum caesarean section with indications to get a realistic picture. The same is true for elective caesarean section without medical indication, which, assuming that the section is planned for 39 weeks, should include women that will go into labour before that date (and become emergency caesarean section without medical indication) according to the intention-to-treat principle. Second, although adjustment for confounding factors is important academically, women themselves do not have this luxury. These factors are part of their reality, and are an inseparable part of their overall risk. For a woman presenting antenatally, therefore, the crude data give the most accurate assessment of their overall risk. And the crude data clearly show that those undergoing elective caesarean section without indication have the lowest risk for both mother and baby of any group. The difference is even more pronounced when grouped according to intention, as above. Thus, the rates of maternal and perinatal mortality and mor- bidity for planned spontaneous vaginal delivery become 3·48% and 3·22%, respectively, compared with 2·37% and 0·53%, respectively, for those undergoing elective caesarean section without medical indication (odds ratios 0·67, 95% CI 0·51–0·89, for maternal and perinatal mortality, Getty Images