(102) Drug target Mendelian Randomisation vs regression methods: estimating causal effects of sclerostin inhibition on bone mineral density and fracture risk

Background: Sclerostin is a protein produced by bone cells whose inhibition significantly increases bone mineral density (BMD) and decreases fracture risk in trials.

Objectives: This study aims to estimate effects of serum sclerostin on BMD and fracture risk using observational data via two analytic methods: traditional regression approach versus Mendelian randomization (MR).

Methods: We performed 4 analyses: (1) linear regression of sclerostin levels on heel BMD, and cox regression on fracture risk, adjusted for age and sex; (2) additionally adjusting for comorbidities; (3) one-sample MR using two-stage least squares regression; and (4) two-sample MR using inverse-variance weighted method. For analysis of (1), (2), and (3), we used individual-level UK Biobank data only. For the two-sample MR analysis, 3 meta-analyzed GWAS of sclerostin levels, and publicly available GWAS for heel BMD and fracture risk were used. Instrumental variables for both MR methods were extracted from previous knowledge. Results from analysis (1) and (2) are reported in standard deviation (SD) change and hazard ratio (HR) per SD decrease in sclerostin levels. Results from analysis (3) and (4) are reported in SD change and odds ratio (OR) per SD decrease in genetically predicted sclerostin levels. All estimates were benchmarked to the causal effect in previous trials, suggesting sclerostin decrease BMD and increase fracture risk.

Results: Of 51,444 UK Biobank participants with measured sclerostin, ~28K were eligible for analyses (1) and (2) when assessing BMD, and ~50K when studying fracture risk. ~17K and ~29K passed the further quality control and were included in analysis (3). Sclerostin meta-analyzed GWAS used in analysis (4) included ~50K participants, BMD GWAS ~400K and hip fracture GWAS ~700K. Results from analyses (1) and (2) suggested a negative association between sclerostin inhibition and heel BMD (Beta = -0.22 (-0.25, -0.19) in both), and a positive association with fracture risk (HR=1.29 (1.2, 1.38)). One sample and two sample MR showed a positive association between genetically predicted decrease in sclerostin levels (Beta=1.50 (0.52,2.48), and Beta=1.00 (0.92,1.08)), and a negative association between genetically predicted decrease in sclerostin levels and fracture risk (OR=0.84 (0.69,1.00), and OR=0.16(0.08,0.30)).

Conclusions: The two analytic methods give conflicting effect estimates of serum sclerostin, with MR results aligning with, whereas regression results contrasting, the anticipated direction. Our study shows that whilst MR approximated the known causal effects, regression analyses remained biased despite adjustment for available confounders. Further research is required to probe the sources of biases inherent in the regression method in this scenario.