Forty-one studies were included. There were 26 studies on HRT and cognition: 4 RCTs (n=8,603), 5 longitudinal observational studies (n=21,933) and 17 cross-sectional studies (n=23,649). There were 18 studies on HRT and risk of dementia: one RCT (n=7,428), 11 cross-sectional studies (n=3,302) and 6 longitudinal studies (n=8,250).
Verbal memory (22 studies): effect sizes were calculable for 17 studies (50 comparisons). RCTs (n=3) and longitudinal studies (n=3) found no consistent evidence of a benefit related to HRT use. There was some evidence of a benefit related to HRT use in cross-sectional studies (n=11), with seven reporting positive results, but there was significant heterogeneity (p=0.000).
Visual memory (10 studies): effect sizes were calculable for 6 studies (one RCT and 5 cross-sectional studies; 15 comparisons). The results were inconsistent.
Cognitive speed (8 studies): the pooling of 2 RCTs showed no significant effect; the findings of the other studies (one longitudinal and 5 cross-sectional) were inconsistent.
Executive performance/concept formation (22 studies): effect sizes were calculable for 16 studies. The findings of RCTs (n=3) and longitudinal studies (n=3) were inconsistent. There was some evidence of a benefit associated with HRT in cross-sectional studies (n=10), with nine reporting positive results, but there was significant heterogeneity (p=0.000).
Cognitive test screening (16 studies): effect sizes were calculable for 11 studies (3 RCTs, 4 longitudinal studies and 5 cross-sectional studies examined). Findings were inconsistent, with a wide range of effect sizes. One RCT (n=2947) reported a small but statistically significant negative effect in the HRT group (MD -2.63, 95% CI: -2.69, -2.56). Overall, there appeared to be no positive effect related to HRT and possibly a small negative effect.
Risk of dementia (18 studies): the findings were inconsistent. The only RCT (n=7,428) found a significantly increased risk in the HRT group (RR 1.38, 95% CI: 1.02, 1.88). Four of the 6 longitudinal studies and two of the 11 cross-sectional studies reported a significantly reduced risk of dementia associated with HRT.
Heterogeneity: effect sizes and directions of effect varied widely across the forest plots, and nearly all analyses had statistically significant heterogeneity which precluded pooling of the data. Meta-regression analyses found no significant associations between effect size and either age or duration of therapy, and subgroup analyses did not in most cases explain the heterogeneity. There was no statistically significant evidence of publication bias.