User Tools

Site Tools


We use machine learning, text-mining, and prior knowledge in medical ontologies to enable the learning health system. Our research group is part of the Center for Biomedical Informatics Research at Stanford. We analyze longitudinal EHR data, including unstructured data, to answer clinical questions, generate insights, and build predictive models. See our work in the news or watch the short video below.

Group information



Answering clinical questions:

Insights from data:

Predictive Modeling

  • Predicting Diagnoses of Depression: We developed a model that uses electronic medical record (EMR) data for predicting the diagnosis of depression up to 12 months before the diagnosis of depression.
  • Rapid identification of slow healing wounds: We demonstrate that it is possible to build a model for identifying delayed healing wounds with an Area Under the Curve (AUC) of 0.842 that works across all wound types.
  • Implications of non-stationarity on predictive modeling using EHRs: Under the non-stationarity in the underlying dataset, the performance advantage of complex methods such as stacking relative to the best simple classifier disappears. Ignoring non-stationarity can thus lead to sub-optimal model selection in predictive modeling tasks.
  • Predicting patient ‘cost blooms’: We develop models that identify new entrants to the upper decile of per capita healthcare expenditures in the next year.

Selected Talks

Building a Machine Learning Healthcare System, at the Parc Forum, Summer 2016.

Performing an Informatics Consult, Grand rounds in Medicine at Stanford, Feb 1 2017.

Other talks

start.txt · Last modified: 2018/01/22 10:08 by acallaha