website_banner_vfinal.png
MIDS Capstone Project Spring 2022

ASROM: Automated Systematic Review of Medical Literature

ASROM empowers medical professionals to make the right decisions at the right time. It is a non-commercial web tool designed to provide medical practitioners evidence based answers to clinical questions. The ASROM tool works using a number of Machine Learning (AI) models trained on the PubMed database. It searches through articles in PubMed and PubMed Central to find the most relevant articles to the clinical question that is asked.

Our Mission

Empowering medical professionals to make the right decisions at the right time.

The Problem

The Medical Profession should be guided by evidence based medicine when it comes to diagnosis, treatment and prognosis of medical conditions. In the past, medical knowledge was just anecdotal practice based on an attending physician or professors’ thoughts with local variations in medical practice on what the best way to treat a condition was.

Best practices and information was very slowly disseminated via journals and conferences. However, with the advent of the internet and electronic health records which improved storage and accessibility of health information through to distributing medical knowledge and research, there has been a move towards medical practice based on evidence and data.This has led to exponential growth of clinical information available to the practitioner which posed a new question of how to make the information usable by practitioner in making decisions. To address this, systematic reviews and meta-analysis are conducted in finding the right evidence and studies.

Systematic reviews are considered the gold standard of evidence. However, they are time-consuming and expensive, very narrowly focused, delayed and often outdated, and they can be biased.

Impact

The impact of ASROM will be to:

  • Decrease the barrier to access to high quality evidence and information for health care professionals for their clinical questions
  • This will encourage an evidence based practice which will improve patient care and outcome

How It Works

ASROM will take the clinical question and use a machine learning model to break it down into relevant PIO (Population, Intervention and Outcome) format and use these terms to search through the PubMed database. The PubMed Database would exclude non-human search and is up to date to 19 February 2022.

  1. Using these MeSH terms we search through the PubMed MeSH terms, Abstracts and the Title Fields of PubMed to find the most relevant articles within the PubMed and Pubmed Central database. If no Mesh Terms exist related to the search term, a vector based nearest Mesh Term will be used (Word2Vec that was trained on the whole PubMed and PubMed Central database is used). A semantic search is also used based on the BM25 search algorithm.
  2. From here, articles will be ranked according to relevance. This is currently being done by the level of evidence which is based on specific search criteria. Order of importance of ranking is:
    • Systematic and Meta-analysis review
    • Randomized Control Study / Trials
    • Remainder
  3. A Machine Learning Named Entity Recognition tool will help to extract the most important information from the study to help the medical professional not have to read through the whole study:
    • Population of the study
    • Intervention that were used in the study
    • Outcomes from the study
    • Conclusion of the study
    • If an appropriate confidence interval is available in the text that is relevant to the clinical question asked, this will also be extracted and placed into a forest plot.

Disclaimers

This tool is intended for use by medical professionals to get better access to information and does consist of specific medical advice

This tool is not a replacement for clinical research, clinical judgment or medical guidelines.

The current tool uses information from PubMed Central and Pubmed up to the date of 19 February 2022. The creators cannot guarantee the accuracy, completeness or usefulness of any information contained or returned.

Given the nature of machine learning and the complexity of the dataset that is being used, not all relevant articles or studies may be shown.

screen_shot_2022-04-11_at_11.30.36_pm.png

Video

ASROM Demo Video

ASROM Demo Video

If you require video captions for accessibility and this video does not have captions, click here to request video captioning.

Last updated:

May 19, 2022