Breast cancer patients often discontinue their long-term treatments, such as hormone therapy, increasing the risk of cancer recurrence. These discontinuations may be caused by adverse patient-centered outcomes (PCOs) due to hormonal drug side effects or other factors. PCOs are not detectable through laboratory tests, and are sparsely documented in electronic health records. Thus, there is a need to explore complementary sources of information for PCOs associated with breast cancer treatments. Social media is a promising resource, but extracting true PCOs from it first requires the accurate detection of real breast cancer patients. We describe a natural language processing (NLP) pipeline for automatically detecting breast cancer patients from Twitter based on their self-reports. The pipeline uses breast cancer-related keywords to collect streaming data from Twitter, applies NLP patterns to filter out noisy posts, and then employs a machine learning classifier trained using manually-annotated data (n = 5,019) for distinguishing firsthand self-reports of breast cancer from other tweets. A classifier based on bidirectional encoder representations from transformers (BERT) showed human-like performance and achieved Fscore of 0.857 (inter-annotator agreement: 0.845; Cohen’s kappa) for the positive class, considerably outperforming the next best classifier—a recurrent neural network with bidirectional long short-term memory (Fscore: 0.670). Qualitative analyses of posts from automatically-detected users revealed discussions about side effects, non-adherence and mental health conditions, illustrating the feasibility of our social media-based approach for studying breast cancer related PCOs from a large population.