Fluoxetine and fractures after stroke: an individual patient data meta-analysis of three large randomised controlled trials of fluoxetine for stroke recovery.

Background: Observational studies have shown that selective serotonin reuptake inhibitors are associated with an increased risk of bone fractures, but the association can be confounded by indication and other sources of systematic bias that can be minimised in randomised controlled trials (RCTs).

Aim: Our aim was to report the rate, site, context, and predictors of fractures after stroke, and whether the fractures modified the effect of fluoxetine on modified Rankin score (mRS) at six months in an individual patient data meta-analysis of 5907 patients enrolled in three RCTs of fluoxetine (20mg for six months) for stroke recovery.

Methods: We classified fractures by treatment allocation, site (and thus likelihood of osteoporosis) and context, then performed multivariable analyses to explore independent predictors of fractures. We explored whether the trend towards a poorer mRS at 6 months was explained by a fracture excess. Risk of bias was assessed using GRADE.

Results: Among 5907 patients randomised at a mean of 6.6 days (SD3.6) post-stroke onset and followed for six months, the number with fractures at 6 months was 93 (3.15%) in the fluoxetine group vs 41 (1.39%) in the control group (difference 1.76%, 95% CI 0.10 to 2.51%). 128 patients with fractures were suitable for further analyses. Of these 102 (80%) were in sites typically affected by osteoporosis; 115 (90%) were associated with falls and one (1%) with a seizure. Independent fracture risk factors were female sex (hazard ratio (HR) 1.96; 95% CI 1.37 to 2.81, p=0.0002), age>70 years (HR 2.30, 95% CI 1.52 to 3.49, p<0.001), previous fractures (HR 0.63 for no previous fractures, 95% CI 0.42 to 0.94, p=0.0227), and randomised treatment (fluoxetine) (HR 2.39; 95% CI 1.64 to 3.49, p<0.001). The common odds ratio for the effect of fluoxetine on mRS at 6 months was unchanged after excluding fracture patients. Risk of bias was high for imprecision.

Conclusion: Fractures were more common in the fluoxetine group but the absolute risk of fractures was small and risk estimates were imprecise. Most fractures occurred with a fall, and in osteoporotic locations. Fractures did not modify the effect of fluoxetine on functional outcome.