Medication adherence improvement on Rheumatoid Arthritis patients based on past medical records

Current drug design and medical treatment guidelines are defined by generic procedures on which each patient that suffers from a disease is treated equally and without any distinction based on any previous genetic/epigenetic study. Most of the time, this generic treatment approach is prone to failure, due to each individual's distinctive genetic characteristics that block the activation pathway of a treatment drug. As a consequence, general treatment guides increases a patient's disease remission time, reducing their quality of life and increasing the financial treatment costs for all parties involved. Until now, the effectiveness of a drug on a specific patient is evaluated based on trial and error over a sequence of periodical medical evaluations that include several clinical tests that are logged on each patient's medical record; this information is used to select suitable drug guidelines until the most effective one is found. This process occurs for every patient that suffers a disease, specially on rheumatic diseases like Arthritis. To tackle this problem, we propose a medical information system (TiMed) that takes as input a collection of medical records and generalizes the temporal sequence effective variables that determine the most suitable drugs for Arthritis patients, using a Recurrent Neural Network (RNN) with Long Short Term Memory (LSTM) units, achieving a 97.4% accuracy over a 150 real data input dataset.