Frailty, an age-related syndrome, is associated with poor post-operative outcomes. The impact of community-level social determinants of health (SDoH) on pre-operative frailty has not been investigated yet. We developed a machine learning model to predict pre-operative frailty using an institutional dataset and applied it to a more geographically diverse population from the OneFlorida+ Clinical Research Consortium. Computable phenotyping for SDoH stratification using unsupervised learning was employed to identify distinct patient profiles based on zip code-level SDoH characteristics. We applied multivariate logistic regression to examine the association between SDoH profiles and pre-operative frailty risk. Adverse community-level SDoH profiles are independently associated with higher pre-operative frailty risk; patients from the disadvantaged SDoH profile had 1.21 times higher odds (95% CI 1.16–1.26) of being frail compared to the advantaged SDoH cluster after adjusting for potential confounders. Considering patients’ social context could improve pre-operative care and surgical outcomes, informing clinical practice and policies.