Response to Letter by Thacker
We appreciate the interest by Dr Thacker in our study exploring the relation between stroke and poststroke dementia (PSD) and the impact of prestroke level of cognitive function on this relationship.1
In our study incident stroke was associated with a more than doubled risk of subsequent dementia, and this association was independent on prestroke cognitive function and other risk factors for cognitive decline. In Kaplan–Meier analyses (the Figure), the cumulative proportion of survivors in persons without incident stroke was 97.6%. Among this group (n=6011), dementia-free survival was 96.0% in persons with an MMSE score of <26 at study entry, and 98.8% in persons with an MMSE score of ≥26, corresponding with an age, sex, education and APOE genotype adjusted hazard ratio (95% CI) for dementia of 1.9 (1.53 to 2.34) for people with a low versus a high MMSE score at study entry. In individuals experiencing a stroke during follow-up (n=713), the cumulative proportion of survivors without dementia at the end of the follow-up period was 87.8% in the group with a MMSE score of <26 at last follow-up before incident stroke, and 92.5% in the group with a MMSE score of ≥26 at last follow-up before stroke, corresponding to an age, sex, education and APOE genotype adjusted hazard ratio for dementia of 1.5 (1.02; 1.96) for people with a low versus a high MMSE score before stroke.
In Cox models relating the individual effect of incident stroke as a time-varying exposure with the risk of poststroke dementia, persons with incident stroke had a 2-fold increased risk of subsequent dementia compared to persons remaining free of stroke during follow-up. This risk did not differ between strata of MMSE performance before stroke (sex, age, education, and apoEε4 genotype adjusted hazard ratio for high prestroke MMSE score 2.1 (95% CI, 1.53 to 2.94) and for low prestroke MMSE score 2.1 (95% CI, 1.51 to 2.90).
We fully agree that a mere dichotomization of probability values at the completely arbitrary cut-off level of 0.05 is uninformative, which is why we presented the absolute probability values. Indeed, effect estimates in combination with the confidence intervals of those effect estimates better reflect the information in the data than probability values. When we explored the impact of prestroke cognitive function on the association between incident stroke and subsequent dementia by adding an interaction term to the model that contained variables for incident stroke (yes/no) and prestroke cognitive performance, the estimate of the interaction term of incident stroke×prestroke cognitive performance was close to 1.0 (hazard ratio 1.01, 95% CI, 0.68 to 1.79, P=0.7) strongly suggesting that the effect of stroke on cognitive function is truly independent of prestroke cognitive function.