MediBo

An AI Medical Chabot to prevent adverse drug reactions (ADRs)

Abstract

With ground-breaking research and development in the data science industries like machine learning, artificial intelligence and natural language processing, creation of intelligent and responsive systems such as artificially intelligent chat-bots has become possible. Powered with capabilities of processing natural language, these chatbots can not just have meaningful conversations with users, but also go beyond and use the power of machine learning and data science to offer easily accessible and customizable services which are otherwise highly-priced or currently unavailable in mass markets. Indeed, the introduction of chatbots has brought much more than the promise of creation of a new array of services. A very relevant example is the use of chatbots in the health-care industry where introduction of chatbots can actually save human lives - the value of which is incomparable. In this project, we carefully study state-of-the-art techniques about chatbot techniques and using our training at the Data Science Institute to create a contextually-aware chat-bot that can help its users (primarily patients) for providing after-care services specially related to drugs and dosage.

Introduction

Intelligent chatbots conduct either textual or auditory exchange to sufficiently simulate a conversation with a human. For consumers, chatbots are used to provide information using natural language processing systems, keyword scans, and similar wording patterns. Chatbots also offer 24 hour access and are able to alert and monitor interactions with users and can provide assistive systems. In healthcare, chatbots offer real-time symptom to disease assistance. These symptom support systems are in wide use with medical chatbots being utilized by patients, clinicians, nurses, pharmacists and others in the healthcare system. While the market is flooded with symptom-checking chatbots, there are gaps within the healthcare system that could be supported by chatbots, including providing real-time support in determining adverse drug reactions or side effects. In the United States alone, more than 2 million people suffer adverse side effects from medication prescribed by their doctor, with more than 100,000 of those people dying each year. In order to solve this problem, a team of student researchers at Columbia University’s Data Science Institute, with support from corporate affiliate, Synergic Partners, are working to create a chatbot that engages with patients who have been prescribed medication by a clinician.

MediBo - a contextually aware medical chatbot — In its full form, the chatbot system will accept the patient’s medication as an input. Next, the chatbot will engage with thepatient to determine whether an adverse side effect is beingexperienced by the patient. Finally, the chatbot will advise the patient on the steps the patient should take depending on the severity of the side effect. From a technical standpoint, the chatbot will be focused on converting word patterns or similar words into a standardized symptom list. For example, the patient may say they “feel queasy” and the chatbot will convert that into a standardized symptom, in this case, “nausea”. The chatbot will then access the data for the particular prescription and determine the level of severity of that symptom and suggest the next steps the patient should take.