EDITOR: We would like to congratulate Albertin and colleagues on their study into the effects of adding nitrous oxide on the MACBAR of sevoflurane given with two target-controlled concentrations of remifentanil in women [1]. They used a modified up-down sequential allocation technique to determine the MACBAR of sevoflurane, described initially by Jung and Choi [2]. This involved two stages. The first used a standard up-down sequence using predetermined equally spaced test levels until three reversals were observed. For the second stage, the testing interval was reduced, restarting the up-down sequence at the nearest level to the average and continuing for four reversals. The logic is that by reducing the testing interval, the value of the final estimator is more precise and the doses used in the second stage are closer to the true median dose. A number of issues follow. The methods, as described in this paper, do not appear to have been followed consistently in that authors appear to have broken their own rule of achieving three reversals prior to moving on to the second stage of the sequential allocation technique in Figures 2 and 4. Figures 1 and 3, although more in keeping with the stated design, show considerable variability and limited stability in the first stage. They also present estimates of ED95 (should be EC95 for sevoflurane concentration), which are extrapolated well beyond the limits of the data as presented. Such extrapolations are better suited to discussion rather than results. They also report similar estimates of precision for EC50 and EC95, where the latter is usually more imprecise. Also, incidentally Figure 4 appears to have an error in the correct plotting of the MACBAR and error bar. The main issue, however, is how to best present and analyse these data given that the method described by the authors removes up to approximately 60% of the subjects from the analysis. The question is whether we can simply ‘ignore’ the initial data? Intuitively the answer is ‘yes’ for Figures 2 and 4, but perhaps ‘no’ for Figures 1 and 3. Also whilst the simulations by Jung [2] show the expected reductions in mean square error by reducing the testing interval, it is not clear what should be done about the variability of data with larger testing intervals where this is trending or when the reversals are unbalanced as in this study. (The schematic in the Jung paper [2] showed a balanced example). A logical approach is then to consider all the data together. Whilst this will decrease the precision of the estimates, it implies that we are not discarding the variability of otherwise valid observations. The authors present an up-down analysis and correctly use logistic regression as a back-up analysis. In this instance, Table 1 shows the logistic regression results reported in the study and an additional analysis of all the data using a loglogistic maximum likelihood approach. They confirm the highly significant (P = 0.001) effects of nitrous oxide and alfentanil on MACBAR of sevoflurane. However, the fiducial 95% confidence intervals (CI) for the estimates of ED95 are relatively imprecise compared to the original report.Table 1: Original results with 95% CI for EC95.In conclusion, again the authors are to be commended on using a dose-varying methodology. Care must be taken when using regression methods to extrapolate point estimates beyond the data as collected and should not be assigned the same weight as median estimates. M. O. Columb P. Tandon Acute Intensive Care Unit, South Manchester University Hospital, Wythenshawe, Manchester, UK
Read full abstract