Using sensitivity analysis to evaluate the fracture benefit of parathyroidectomy

Studies suggesting that parathyroidectomy prevents more fractures than nonoperative management are all observational and could be biased by unmeasured confounding. The Robustness of Inferences to Replacement is a novel method for sensitivity analysis that assesses the robustness of study findings against these concerns. We aimed to inform discussions about the strength of available observational data indicating that parathyroidectomy reduces fracture risk versus nonoperative management for patients with primary hyperparathyroidism.

Studies that compared incident fractures after parathyroidectomy versus nonoperative management were identified via a comprehensive literature review. The Robustness of Inferences to Replacement was calculated for each study finding to determine how many patients would have to be replaced with patients for whom there was no benefit of parathyroidectomy for the results to be overturned. The Fraction of Effect Above the Statistical Significance Threshold, which determined the extent to which the observed treatment effect exceeded the threshold for statistical significance, was also calculated.

Six retrospective cohort studies involving 217,343 patients were identified. Parathyroidectomy was associated with a 31%-59% reduction in fractures at any site and a 16%-62% reduction in hip fractures. The Fraction of Effect Above the Statistical Significance Threshold and Robustness of Inferences to Replacement indicated that up to 90% of the observed differences would have to be attributed to unmeasured confounding to change the study result.

The observational data showing that parathyroidectomy reduces fractures are likely robust to unmeasured confounding and probably represent a true benefit of surgery. The findings demonstrate how the Robustness of Inferences to Replacement can be used to inform conversations about whether observational studies are sufficiently robust to support clinical decision-making.