Recent advances in machine learning technology may help doctors to decide on better treatments related to Acute severe ulcerative colitis (ASUC). Ulcerative colitis (UC) is a disease condition that is characterized by relapsing inflammation in the colon. About 15%-25% of UC patients have at least 1 episode of ASUC, which can be life-threatening and may develop into serious complications that require colectomy (removing the affected part of the colon).
In a new study by scientists from INSERN, France, researchers have collected information from blood works and expression levels of biomarkers from colonic biopsies from 47 ASUC patients when they were admitted to the hospital and their responds to steroid treatment. These information were fed into a neural network-based algorithm to identify parameters and biomarkers that are able to predict whether the patient responds to steroid treatment. Nine biomarkers and 5 blood parameters that are related to status of inflammation [C-reactive protein (CRP) and albumin] and anemia/blood loss (hemoglobin, hematocrit and transferrin) are shown to be important in deciding whether the patients will respond to steroid treatment. Then, scientists fed a new set of information from a separate group of patients (29 patients) to the algorithm and found that the algorithm predicts which patient responds to steroid treatment with 84% accuracy.
The biomarkers identified in this study are expression of microRNAs (miRNAs) in colonic biopsies. miRNAs are nucleic acid molecules in cells that mainly work to reduce gene expression. Up til now, many biomarkers for UC have been proposed mainly for diagnostic purpose; however, the miRNAs proposed in this study have prognostic values in helping doctors to decide the optimal treatment for patients.
Machine learning is slowly making its way to disease diagnosis, prognosis and predictions to treatment outcome. Machine learning is good for diagnosing diseases which are evolving and have a spectrum of symptoms, such as inflammatory bowel diseases (UC is under this umbrella). Furthermore, in some diseases, such as UC, patients may not respond uniformly to certain treatment regimens. Given the right laboratory tests and record of the symptoms, machine learning offers an unbiased way of diagnosis, and allow physicians to devise optimal treatment at the early stage of the disease.