Hello everyone! I'm Narges Razavian. I'm a postdoc at Clinical Machine Learning lab lead by Dr. David Sontag. My background is machine learning for bioinformatics and medical informatics.
Our most recent publication is a large study, where we built 42,000 health features (including temporal features) from administrative claims data, to learn a model for predicting Type 2 diabetes onset. We trained and independently tested our model on data from an initial cohort of 4.1 million individuals over 8 years.
You can read it here:
Population-Level Prediction of Type 2 Diabetes From Claims Data and Analysis of Risk Factors
Narges Razavian, Saul Blecker, Ann Marie Schmidt, Aaron Smith-McLallen, Somesh Nigam, and David Sontag, Big Data. January 2016.
We are converting our full cohort into common data format developed by OHDSI, and plan to release our models (not just for diabetes, for all diseases) compatible with this standard data format, so everyone could plug them into their data and get predictions. I will present this paper on the meeting of Jan 27th, 2016, and look forward to collaborating on the next steps!
We have also built a more advanced machine learning model for detecting (3 months in advance) multiple (270) diseases, using temporal convolution networks and lab values of the individual in the past 3 years. I can discuss that one further into the future! This work is still under review: http://arxiv.org/abs/1511.07938