Data may not have been summarised appropriatelyBMJ 1996; 313 doi: https://doi.org/10.1136/bmj.313.7053.361b (Published 10 August 1996) Cite this as: BMJ 1996;313:361
EDITOR,—June M C McArdle and colleagues end their paper by saying that “the failure to reduce morbidity in the combined approach [routine care plus support from a breast care nurse and a voluntary organisation] is difficult to explain.”1 This logic is not supported by the data presented.
An overconcentration on P values at the expense of descriptive trends seems to have led the authors to conclude that “scores were consistently lower in patients offered support from [a] breast care nurse alone compared with the other groups, which were similar to each other.” My interpretation of their data is that the poor showing of the combined approach is explained simply by two opposing effects, one larger than the other. Take, for example, their results for anxiety at 12 months. The means were 4.8 (routine care), 4.4 (routine care plus nurse only), 6.3 (routine care plus voluntary organisation only), and 5.8 (routine care plus nurse and voluntary organisation combined). The effect of the nurse can be measured as 4.4 - 4.8 = −0.4 and as 5.8 - 6.3 = −0.5. The effect of the voluntary organisation can be measured as 6.3 - 4.8 = 1.5 and as 5.8 - 4.4 = 1.4. These effects combine to give more morbidity than if routine care alone is given. Similar conclusions can be drawn for most other measures reported at 12 months or averaged over four postoperative visits.
I recognise the dilemma in using parametric methods to summarise skewed distributions, with the resulting mismatch between presentation and results of statistical testing. Consequently, readers are presented with an inferior description of the trends. But how inferior? Is my manipulation of means justified? I have merely used my skills to interpret differently those means that the authors themselves regard as less than appropriate. My gut answer to my own questions is that means of skewed distributions do often characterise the appropriate message, but I would qualify this by saying that a more appropriate description should also have been given. In my experience, the main difficulty lies not in being able to summarise data more appropriately but in being able to do so within the confines of the space allowed.
A final point relates to inequality of loss to follow up in the study. At 12 months the loss was 24% for the routine care group compared with 6% (nurse only), 8% (voluntary organisation only), and 16% (combined). The psychological morbidity of those lost to follow up and of those who remained could have been different. This point should have been discussed in relation to the results.