Predicting risk after aneurysm surgery – a reply

I thank Drs Brown and Price for their comments. I agree that I would have overestimated mortality hazard had I combined hazard ratios associating single variables with death in the general population. I limited that problem by using conservative estimates of hazard ratios from multivariate analysis, not univariate analyses. I am sorry I did not make that clear. For instance, studies of fitness and survival have reported hazard ratios of 1.67-2.82 for heart failure, independent of peak oxygen consumption, for which the independent hazard ratio was 0.96-0.98 for each ml.kg−1.min−1 1-9. I used a lower hazard ratio of 1.5 for the diagnosis of heart failure because I also incorporated the ventilatory equivalent for carbon dioxide in my calculator, for which I used a hazard ratio of 1.05, rather than 1.07 10. I was fully aware how I would disadvantage the performance of a calculator by importing survival curves and hazard ratios from completely different populations. This was intentional. I wanted to determine whether I could calculate survival using data from the general population, i.e. not surgical populations, and definitely not the population of 1000 or so patients we observed. My main aim is to develop a single generic survival calculator that could be used for any surgical or non-surgical population that uses survival factors common to all populations, and that can also accommodate factors specific to different populations. My aim is not to generate a model restricted to patients undergoing abdominal aortic aneurysm (AAA), although that is the population I used for this paper. I think that the calculator works well for AAA populations, in that the expected survival curves match reported survival curves for all randomised controlled trials (11, Supporting Information) (Fig. 1). As Brown and Price observe, uncertainty is much greater for the individual than for a population. The expected survival curves will generally match the observed curves for cohorts, because only a minority of a cohort – even AAA populations – have one of the co-morbidities in the calculator, so the curve follows the background population curve. Survival curves for individuals with a major co-morbidity will be much more sensitive to the hazard ratio chosen for that co-morbidity. In response to the correspondents' comments I have developed a version of the calculator, ‘AAA with variable hazard ratios’, for which users can choose hazard ratios different to those I have used (this can be accessed at https://www.dropbox.com/sh/i57xhrtyop0pq31/AACEvMTOPRzdMEQnhLbCL2G8a?dl=0 (my thanks to Dr Nick Wisely for the new input page)).