Phenotype and disease subclassification are among some of the fundamental challenges facing precision medicines. The task of deep disease phenotyping is to observe and define a set of clinical features that represent the clinical expression of a disease by looking at a patient’s medical history, anthropometric measurements, recent events and activities for common patterns. If a diverse group of patients can be successfully classified into appropriate subtypes based on the way the disease is behaving in their bodies and their own common features, more personalised and, therefore, effective and preventative treatment may be possible.
Disease phenotyping is also a powerful tool to help translate cancer research into potential cancer treatments. Classifying cancer patients based on differences in the disease’s expression in their bodies (the disease phenotype) can lead to better patient selection for clinical trials and inclusion in other research studies. For example, it may identify why some patients survive or fare better than others, controlling for patient care performance at the hospital or the hospital’s revenue. However, many key phenotypic variables in cancer, such as specific tumour behaviours (e.g., metastasis), laboratory findings (e.g., gene amplification), tumour morphology (e.g., histopathologic features) and response to treatment (e.g., the effect of the chemotherapy on the size of a tumour) are only available in clinical notes and are difficult to map together.
Funded by the European Convergence Programme, Professor Shang-Ming Zhou and his team and partners are using AI technologies including machine learning, deep learning and natural language processing to extract clinically meaningful information from patients’ notes and electronic health records. They will then identify patients with colorectal cancer and use deep phenotyping to predict the development of cancer stages, including the potential return of cancer after treatment and associated multimorbidity. Such deep phenotyping uses advanced AI algorithms to integrate the wealth of health data and reveal connections and interactions between phenotypic factors. Ultimately, we will seek to explore how and why cancer affects people differently and suggest how treatment and prevention could be individualised for sufferers.
