Authors' replyBMJ 1996; 313 doi: https://doi.org/10.1136/bmj.313.7055.495 (Published 24 August 1996) Cite this as: BMJ 1996;313:495
- Charles R Gillis, Director,
- David J Hole, Principal epidemiologist
EDITOR,—We wish to respond to several points raised in the previously published letters1 and the above letters about our paper.2 None of Ian Campbell's points1 change our results or their interpretation. We considered selective bias by analysing our data on the basis of hospital catchment areas. Specialists and non-specialists worked in the same hospital serving the same catchment areas. There was a 4.6% advantage in five year survival for those catchment areas where the specialists saw 44% of patients compared with those where the specialists saw only 9% of patients. As both analyses showed an advantage for specialist care, selection bias is an unlikely explanation of the differences seen.
We relied heavily on tumour size as the main prognostic factor because of our stated concerns about the accuracy of nodal assessment. Using actual tumour size, or six rather than three categories, as a covariate in the survival analysis leaves the hazard ratio for specialists relative to non-specialists unaltered at 0.83.
Our data fit the proportional hazards model well. Analysis of the data in two separate periods of follow up (0-5 and 5-10 years) produced ratios of 0.81 (95% confidence interval 0.70 to 0.93) and 0.89 (0.71 to 1.11) respectively for specialists compared with non-specialists.
To say that our study was biased because we did not take all prognostic factors into account is to say that all non-randomised studies of disease are biased. While randomised controlled trials give an indication of efficacy in a selected population, observational studies show how effective findings are in the general population. Also, when there are difficulties in carrying out a randomised trial, observational studies are the only means of proceeding.
We agree with Michael Baum1 that the maximum difference that could be attributed to the application of optimal adjuvant treatment is 4%, compared with the 11% advantage that we observed for specialists in patients aged 50-64. We present no evidence to explain this observation. We believe that the only way to test the validity of our findings is to extend the period of the study to the present, with the cooperation of the specialists. This is now being done.
We agree with Adrian Ball, in the letter above, that the number of patients treated should never be the sole criterion for designating a service as a specialist service. The ultimate assessment of survival will come from prospective audit. This will give precise meaning to the term “specialist.”
Log in using your username and password
Log in through your institution
Sign up for a free trial