301 LETTERS to the EDITOR Human insulin and hypoglycaemia SIR,-Dr Colagiuri and colleagues’ (June 13, p 1432) contribution to the conflicting evidence about lack of awareness of hypoglycaemia and human insulin should not lead to an end of the debate and thus human insulin being cleared. The discrepant findings in respect of differences in patients studied and methodologies used need to be examined. The randomisation Colagiuri et al used resulted in very unequal groups of between 6 and 13 patients. The probability of this happening by chance alone is low (XI=4-24, p=004). This imbalance led to an unequal distribution of treatment periods. Patients were significantly more likely to be treated with porcine insulin during the first two months and more likely to be treated with human insulin during the second two months (=3-92, p = 0-05). In a clinical study the patient receives more attention and this often leads to an improvement of glycaemic control during the investigation. Tighter control during the second half of Colagiuri’s study (when more patients were treated with human than animal insulin) could have increased the number of hypoglycaemic episodes with human insulin. Glycosylated haemoglobin A1 (HbA1) and fructosamine, measured at study entry, crossover, and study end, could confirm or refute such a trend in glycaemic control. In our double-blind crossover trial,’ in which hypoglycaemia associated with human insulin showed a symptom pattern that could impair the patient’s ability to recognise hypoglycaemia, stable glycaemic control was documented in this way. Colagiuri et al state that HbA1 was measured for most patients, but "that interpretation of the results was difficult because of the number of laboratories used and the lack of a quality control mechanism for data comparison". This argument may not be valid since the outcome interest is in the difference between HbA1 concentrations measured at different time points in the same patient, not between-patient differences. Hypoglycaemia awareness per se is difficult to assess. The number of blood glucose measurements done at the patient’s initiative that result in the detection of hypoglycaemia are of little value as the main outcome measure in the investigation of this condition. Hypoglycaemia awareness is highly dependent on glycaemic control and does not assess differences in symptom patterns that may well be relevant for recognition of hypoglycaemia. The blood glucose concentrations at which the patient is prompted to do a measurement may be more meaningful. Altman and Doré2 have identified three important aspects of standards for the reporting of clinical trials. First, how the sample size was calculated on the basis of previous knowledge should be discussed. Second, the method of randomisation should be adequately described. Third, baseline characteristics of the different treatment groups should be presented and, if necessary, statistical analysis should explore the effect of any discrepancies. Colagiuri and colleauges’ report fails on all three counts. This is especially worrying with respect to possible imbalances in baseline characteristics, in view of the fact that the two possible statistical FREQUENCY OF HYPOGLYCAEMIC EPISODES ACCORDING TO SEVERITY (CALCULATED FROM DATA OF COLAGIURI ET AL) tests of balanced randomisation that we could do suggested a breakdown of their randomisation process. In the analysis, we are presented with the mean number of hypoglycaemic episodes recorded per patient per day, which ignores the fact that every hypoglycaemic event poses a threat to life and data should therefore be analysed both by episode and by patient.’ When an analysis by episodes is done for the frequency of hypoglycaemic events according to severity as assessed by blood glucose concentrations, results are obtained that sit uneasily with Colagiuri’s conclusions. There is a significant trend (X2 for trend = 3-79, p = 0-05) for hypoglycaemia being recognised at lower glucose concentrations with human than with porcine insulin (table). A similar trend emerges when the analysis is done by patient, but it cannot be tested statistically with the information provided. The way Colagiuri et al present data makes it difficult for the reader to recognise this trend, which, contrary to these researchers’ assertion, supports the hypothesis of diminished awareness of hypoglycaemia during treatment with human insulin. The debate about human insulin should rely on careful examination of scientific data rather than unfounded allegations. 3,4 As we have demonstrated, Colagiuri’s study cannot be regarded as showing conclusively that there is no difference in hypoglycaemia awareness when patients are treated with human rather than porcine insulin. Indeed, the outcome measure that may have been least affected by the weaknesses of the study supports the hypothesis of less reliable recognition of hypoglycaemia during human insulin treatment. Since there has long been a consensus among diabetes specialists that human insulin offers no real advantages over highly purified animal insulins5,6 and that insulin therapy should not be altered in patients who are satisfactorily controlled, it is surely the responsiblity of those proposing a therapeutic change to demonstrate the complete safety of what they propose before it is implemented. This did not happen with human insulin, and now there is clearly an ethical obligation for the manufacturers of this insulin to fund an adequately sized randomised controlled trial with severe hypoglycaemia as the outcome measure. There is, unfortunately, little indication that they are willing to do this.7 Department of Epidemiology and Public Health, University College London, London WC1 E 6EA, UK MATTHIAS EGGER Department of Public Health Medicine, University of Glasgow GEORGE DAVEY SMITH 1. Egger M, Davey Smith G, Teuscher AU, Teuscher A. Influence of human insulin on symptoms and awareness of hypoglycaemia: a randomised double blind crossover trial. BMJ 1991; 303: 622-26. 2. Altman DG, Doré CJ. Randomisation and baseline comparisons in clinical trials. Lancet 1990; 335: 149-53. 3. Barfoed N Human insulin on trial. Nova Nordisk Magazine 1991; December: 3. 4 Smith R. Human insulin. Novo Nordisk Magazine 1992; June: 27. 5. Egger M, Davey Smith G, Teuscher A. For debate: Human insulin and hypoglycaemia: the need for a large randomised trial. BMJ (in press). 6. Wolff SP Trying times for human insulin. Nature 1992; 356: 375-76. 7. No funding for definitive insulin study. Scrip 1992; 1690 (Feb 7): 22. SIR,--The Lancet has certainly played an important part in the debate about human insulin,l Dr Colagiuri and colleagues’ report being the latest contribution in your columns. With a colleague, I described 3 patients who had severe hypoglycaemia after transfer to human insulin and described the results of a retrospective survey among 176 insulin-treated patients that was prompted by these case histories.2 Two years later, the first randomised double-blind trial showing a different hypoglycaemic symptom pattern during treatment with human and porcine insulin appeared.3 Results in