Assessment of grouping variable should have been blind in trial of dementiaBMJ 2001; 322 doi: https://doi.org/10.1136/bmj.322.7300.1491/b (Published 16 June 2001) Cite this as: BMJ 2001;322:1491
EDITOR—Huusko et al studied the effect of intensive geriatric rehabilitation on dementia patients with hip fracture.1 The reported study was probably the first randomised study with predetermined subgroup analysis according to the degree of dementia in this growing patient group. The degree of dementia was classified by the score on the minimental state examination.2 The median length of hospital stay of patients with hip fracture who had moderate dementia was 47 days in the intervention group and 147 days in the control group (P=0.04). The corresponding values for patients with mild dementia were 29 days in the intervention group and 46.5 days in the control group (P=0.002).
Huusko et al concluded that patients with hip fracture and mild or moderate dementia can often return to the community if they are provided with active geriatric rehabilitation. Pioneering work is never easy. Even though the study seemed to be well conducted, the patients in the intervention group were stated to have a highly significant deeper level of dementia compared with the control group (P<0.001). This was considered as coincidental, even though such a difference in outcome would be considered to be definitive proof of treatment effect in any medical trial. The degree of dementia was used as grouping variable when testing the effect of intervention on the outcome and thus probably had a fundamental effect on the obtained results.
The problem is that the degree of dementia was stated about one week after admission to the geriatric ward of the central hospital or the local hospital, 10 days after surgery and randomisation. The person scoring the minimental examination knew the treatment group, and this person was probably different in the control and in the intervention group. This may have caused biased classification of dementia and may explain the observed difference in minimental state examination between the study groups.3 Thereafter, recovery of patients in the intervention group without true dementia may have been compared with patients in the control group with mild dementia, and so on, which might explain an appreciable part of the observed results in the study.
Conducting this sort of randomised study is very demanding, but it is recommended to try to blind assessment of the main grouping factor.