Using Machine Learning Techniques to Predict a List of Prescription Medications in the Obstetrics and Gynecology Service

A medical prescription is the result of a cognitive process where the doctor, from the acquired knowledge, listens to the patient’s symptoms, and with the learned expertise, performs a physical examination in search of signs, confronts with the acquired data and decides on an action. If the action is therapeutic, the medical prescription is issued. In particular, errors in prescriptions and treatments during pregnancy lead to damage, in some cases irreversible to the patient and the fetus, which is why it is essential to have tools that help us to reduce such events. Several studies have been conducted to address the problem from different scopes, studying the impact on the reduction of errors through the digitization of medical records, identifying the requirements for the automation of medical care, as well as using machine learning by means of classification methods through diagnosis by related groups. In this context, different machine learning techniques were implemented and evaluated in this study, where fully connected neural networks have so far been the best for prediction of prescriptions in the obstetrics and gynecology service. The databases associated with the analysis have 13 numerical characteristics and two in text, where the latter by their nature were subjected to natural language processing (NLP) techniques. Currently, 4 models have been built for 4 diagnoses and thus predict the variability based on the diagnosis of up to 7 types of drugs, reaching accuracy even in tests from 0.5 to 0.84.