Effect of antihypertensive treatment on cognitive function of older patientsBMJ 1996; 313 doi: https://doi.org/10.1136/bmj.313.7050.166a (Published 20 July 1996) Cite this as: BMJ 1996;313:166
- C D Irvine,
- S H Irvine
- Research registrar Vascular Research Unit, Bristol Royal Infirmary, Bristol BS2 8HW
- Professor of psychology Human Assessment Laboratory, University of Plymouth, Plymouth PL4 8AA
Effect is not proved
EDITOR,—On the basis of the results of two cognitive tests Martin J Prince and colleagues conclude that hypertensive drug regimens in elderly people impair cognition no more than does a placebo.1 This may lead to the changes suggested by the key message “Concerns about damaging cognition should not deter doctors from treating hypertension in older patients.” We have concerns about the study's design: the nature of the outcome measures used, the lack of any run in period before treatment to minimise practice effects, and the use of slopes as measures of individual learning rates.
Any battery of cognitive tests for elderly people should test four fundamental cognitive factors: perceptual speed, working memory capacity, access to long term memory, and spatial orientation.2 In this study the authors analysed the variance associated with the paired associate learning task and the Reitan trail making test. Both tasks have methodological constraints that make testing of cognitive factors imprecise. Paired associate learning tasks are poor predictors of learning in real life situations.3 Because they operate under severely controlled time conditions the outcome in Prince and colleagues' elderly population was an extremely restricted range (mean 17 (SD 1.7)). Consequently, estimates of the internal consistency of the paired associate learning tasks4 are close to zero.
The Reitan trail making test is assumed to assess attention, concentration, and psychomotor skills. In the authors' study the average time to complete it over five occasions fell, which suggests that the test scores were subject to practice or learning effects, or both, which are likely to mask treatment effects. A validated battery of cognitive tests exists with limitless numbers of parallel forms, which ameliorate practice effects and negate memory effects.2 5
For any standardised slope (correlation coefficient) based on five individual measures the limits of the 95% confidence interval are 0.75 and −0.75. This means that the result of correlating five scores with five time intervals must show a value of 0.75 or −0.75 in order to be treated as a non-random value for a single hypertensive patient. The authors admit that many scores were identical from point to point in time. Whenever this is so the correlation (slope) will fall within random limits, making any possible treatment effect redundant.
While we agree that there are no significant differences among the means in the treatment and placebo groups, there may have been confounding factors in this study's design and execution. We therefore advocate the more conservative conclusion of “not proved.”