Globally 264 million people suffer from depression. Depression and anxiety cost $1 trillion each year in lost productivity. Here we are proposing the prediction of onset of depression using commercial medical claims data. We are using 2297 medical indicators for the 2 years of observation window and predicting depression risk probability for one year of prediction window. We developed a CNN + BiLSTM + Attention neural network architecture and trained it on 7.2 million patients' claims. This and another previous study suggests that medical claims data holds information to predict depression. Using the medical claims dataset predicts depression is an inexpensive method and a good representative of the population due to its ease of gathering and ubiquity. 

The structure of the talk would be as follows: 

  • Domain background
  • Data used and how it is architected for modeling
  • Network architecture; Sparse multidimensional time-series for binary class classification
  • Evaluation techniques
  • Conclusion; Future work and other application of the work

Instructor's Bio

Behlool Sabir, Lead Data Scientist at Fidelity Investments

Behlool is a data scientist at Fidelity Investments in AI CoE of the healthcare group. He has more than 7 years of industry experience in machine learning with a mix of academic research, where he co-authored on predictive healthcare work, and experience on business implementation of ML systems built on large datasets in the domains of healthcare, BFSI & aviation. He is currently focusing on predictive analytics in the healthcare space leveraging medical claims information. He holds a master's degree in Physics from IISER Pune with research experience in Nonlinear Dynamics and Complex Systems.


  • 01

    Depression Onset Prediction on a Large Medical Claims Data Using Deep Learning

    • Depression Onset Prediction on a Large Medical Claims Data Using Deep Learning