PERFORMANCE OF A MACHINE-LEARNING GENE RISK SCORE BIOMARKER ON PREDICTING RESPONSE TO SEMAGLUTIDE: A PROSPECTIVELY FOLLOWED MULTI-CENTER BIOBANK AND OUTCOMES REGISTRY

Background: Our phenotype-guided approach to antiobesity medications is associated with greater weight loss compared to the obesity medicine standard of care. For instance, the abnormal post-prandial satiety phenotype (“hungry gut”-HG) is associated with greater weight loss in response to GLP-1 receptor agonists (GLP-1RA) liraglutide and exenatide that are known to prolong post-prandial satiety. We developed a machine-learning gene risk score (ML-GRS) to predict HG and validated it in patients taking liraglutide against placebo in a randomized controlled trial. The ML-GRS was developed by exploring the associations between variables related to food intake regulation and single nucleotide polymorphisms within an expert-guided set of 40 genes. Our model predicted the response to liraglutide with an AUC of 0.69 (DOI:10.1016/S0016-5085(23)01311-2). However, the effect of this ML-GRS in predicting weight loss response to semaglutide, a more effective GLP-1 RA, is still unknown.

Methods: We established a multi-center biobank and outcomes registry of adults undergoing weight loss interventions at Mayo Clinic (IRB: 21-011737). The registry collects information on demographics, anthropometrics, and weight loss interventions with their outcomes. For this report, we enrolled individuals with obesity prescribed semaglutide 0.25-2.4mg. Using saliva or blood samples, we performed genetic studies and generated a ML-GRS, a continuous variable varying from 0 to 1. Our analysis was divided as following: 1) ML-GRS < 0.50 (“Hungry Gut Positive” [HG+]), and 2) ML-GRS≥ 0.50 (“Hungry Gut Negative” [HG-]). Our endpoints were the total body weight loss (TBWL)% at 3, 6, 9 and 12 months and the probability of the ML-GRS to predict response to semaglutide, defined as TBWL ≥5% at 12 months. Continuous data were analyzed using paired t-test and categorical data using Fisher’s exact test, with p-value <0.05 considered statistically significant. Results are presented as mean±standard deviation.

Results: We included 84 participants: age 47.6±10.9, BMI 38.8±6.9 kg/m² (Table 1). Compared to HG-, HG+ had superior TBWL% at 9 months (-14.4± 6.6% vs. -10.3 ± 7.0%, p=0.045) and at 12 months (-19.5 ± 11.4% vs. -10.0 ± 9.3%, p=0.01) but not at 3 and 6 months (Figure 1, A and B). When used to predict response, the ML-GRS had an AUC of 0.76 (95% CI [0.57-0.94], p=0.04), with a PPV equal to 0.95.

Conclusion: Our HG ML-GRS could serve biomarker predicts response to semaglutide. Our test could be employed in clinical practice to select responders to semaglutide, thereby reducing obesity heterogeneity. Further prospective studies are needed to assess the validity of our biomarker in a more diverse population and with different weight loss interventions.