Statistics from Altmetric.com
Editor,—The recent study by Hutton, Colver, and Mackie1 is in some respects a useful addition to our knowledge of survival in cerebral palsy. Unfortunately there are substantial problems with the paper; we note three of them below.
1. In figure 1A it appears that in the most seriously affected group, who had a Lifestyle Assessment Score (LAS) ⩾70%, there is100% survival to age 9. This scarcely seems plausible when, as the graph indicates, 20% of these survivors die in the next 9 years. The explanation is that the most severely disabled children, with LAS 70% or more, have to survive to age 5 to be assessed by LAS. Thus the severely disabled children who die before 5 have no LAS, and are excluded (actually, it appears from the graph that some children are evaluated even later than age 5).
The resulting bias could have serious consequences. For example, in a lawsuit involving a neurologically devastated 2 year old child a plaintiff may cite Hutton et al to argue for 100% survival over the next seven years.
2. Hutton et al's results show that, as is well known, low IQ and/or poor mobility correlate with reduced life expectancy. In his commentary, Dr Rosenbloom usefully asks whether extreme immobility or mental impairment would give an even greater reduction. The answer is clearly yes, as indicated by our own work2 3 and is also acknowledged by other workers in the area.4 5 Indeed it must be so because an extensive literature shows a much shortened survival in the persistent vegetative state,6 which may be regarded as the extreme case of disability.
In neither HCM nor in Dr Hutton's previous study of the Merseyside area7 is this possibility mentioned. As a result, the latter study has frequently lead plaintiffs to overestimate survival of children with the most severe disabilities.
3. In their table 5, Hutton et alstate that in our California study2 the odds ratios for various hazards were lower than in several other studies. For example, in table 2 of our article we gave a hazard ratio of 3.8 for two year olds who were tube fed, compared with those who were not. Huttonet al then proceeded to speculate at some length on what accounts for the transatlantic difference.
The real reason is simple: the California database has many more variables other databases, so the marginal effect of any one of them—that is, when the others are held constant—is smaller. For example, our table 1 showed that if no other factors are taken into account the hazard ratio for tube feeding (compared with children who could self feed) was 23.6—a much larger ratio than the above 3.8, and in fact about as large as any in Hutton et al's table 5. In addition, the definitions of mobility etc in the various studies are very different.
We thank David Strauss for his interest in our work but he fails to substantiate his claims that there are “substantial problems” with it.
First, we wish to correct an error in our article on page 470, column 2, line 11: “dying before” should read “surviving until”.
With regard to Strauss's remarks on LAS, both the abstract and the results section include the phrase “survived to age 5”, so Strauss has not explained our result, but merely repeated this information. Even the brief precis of the LAS in the paper makes it clear that it would be difficult, if not impossible, as well as unwise, to attempt to complete it for a 2 year old. That someone might misquote our work is true, but not our responsibility.
With regard to mobility and mental ability, we have reported exactly what is measured, and have referenced other work which includes measures different from ours. It appears that Strauss wishes we had speculated about information we do not have. Note that IQ was constructed to have a mean of 100, and standard deviation of 15. On this scale, fewer than 4 in 10 000 people would have an IQ less than 50, our definition of severe cognitive disability. Fewer than 5 in 10 million people would have an IQ of less than 20, the number mentioned by Dr Rosenbloom.
It seems obvious that persistent vegetative state or indeed leukaemia or need for ventilation twice a year are clearly additional, and largely independent risk factors for death which any court would take in to account.
With regard to hazard or odds ratios, our discussion does comment on multivariate versus univariate models. Whether univariate or multivariate, one still has a relative statistic, so that any lack of similarity in the baseline categories will be relevant. Strauss claims that “the real reason is simpl(y)” the difference between multivariate and univariate analyses: he gives hazard ratios for tube feeding, and states that the univariate ratio of 23.6 is “about as large as any in Hutton et al's table 5”. However, Strauss fails to quote a factor—hand use—which, unlike tube feeding, can be easily compared with our work, and that of South East Thames. In correspondence with Hutton, Strauss stated “Re hand use: our multivariate OR was 1.52 and our univariate OR was 5.69”: a value which is substantially less than 23.6, and than our results. Further, Strauss states that this lower ratio is for a more disabled group: “But our definitions are very different from yours. Our ‘bad’ group is ‘no functional use of hand’...while your ‘severe manual disability’ group is much more inclusive'...”. Thus the “reason” Strauss gives fails to explain the difference in results for manual disability.