Social media platforms significantly influence users’ outlook, shaping and molding their likes, dislikes, and thought processes, and even affecting mood and well-being. The content posted by social media users can potentially reveal signs of their emotions, indicating underlying mental and emotional health concerns. Given the rise in mental health disorders and the lack of access to effective healthcare, there is a growing need to explore research avenues that could benefit users experiencing anxiety-related disorders and depression, two of the most common conditions. While these conditions are explicitly studied, the interconnection between mental health and other stress-related chronic diseases like heart ailments and diabetes is also on the rise. In this article, an approach that leverages semantic representation techniques and deep neural models to analyze social media data, particularly community-level tweets, for predicting Atherosclerotic Heart Disease (AHD) mortality rates in vulnerable communities is presented. Transformer-based Masked Language Models (MLMs) were utilized to represent tweet data contextually, and Transfer learning methods were adopted for further optimization. Experiments revealed that tweets correlate with community-level AHD mortality rates, and finetuning experiments on open benchmark datasets highlighted the efficacy of finetuned MLM models, which outperformed all other language models considered for the study.