Thirty-three publications were included. Some publications evaluated more than one drug. These publications provided 54 antihypertensive drug/control ‘studies’ (the review defined a study as exposure to an antihypertensive medication): 40 case-control studies and 14 cohort studies.
Diuretics (38 studies included in analyses, n=384,533): there was no significant difference between diuretics overall and control (RR 0.90, 95% CI: 0.79, 1.03), based on 38 studies. Thiazide diuretics were associated with a statistically significant reduction in the risk of any fracture compared with control (RR 0.86, 95% CI: 0.81, 0.92), based on 25 studies (including 16 high-quality studies); significant heterogeneity was found (p=0.01). There was no statistically significant difference between nonthiazide diuretics and control (RR 1.19, 95% CI: 0.91, 1.57), based on 11 studies; significant heterogeneity was found (p<0.001). Moderate to large heterogeneity was found for almost all analyses and subgroup analyses of diuretics (p<0.19 to p<0.001); heterogeneity remained after stratification by age, gender and study quality.
Beta-blockers (8 studies, n=224,407) were associated with a statistically significant reduction in the risk of any fracture compared with control (RR 0.86, 95% CI: 0.76, 0.98). Significant heterogeneity was found (p<0.01); this subsided when the analysis was restricted to hip fractures.
Angiotensin-converting enzyme inhibitors were associated with a statistically significant reduction in the risk of any fracture compared with control (RR 0.81, 95% CI: 0.73, 0.89), based on one study (n=151,420).
Calcium-channel blockers were associated with a statistically significant increase in the risk of femoral neck fracture compared with control (RR 1.96, 95% CI: 1.16, 3.30), based on one study (n=484).
There was no statistically significant difference in the risk of any fracture between α-blockers and control (RR 1.12, 95% CI: 0.42, 3.02), based on 2 studies (n>9,142).
The funnel plot for thiazide diuretics was asymmetrical, suggesting the possibility of publication bias, but no evidence of publication bias was found using Egger’s test.