6.S897/HST.S53: Machine Learning for Healthcare

Instructor: Professor David Sontag (OH: TBD)
Teaching Assistant: Maggie Makar (OH: Wednesdays 5-7. Location 56-162)
Graduate level; Units 2-0-4 (counts as an AAGS subject)
Time: Wednesdays 2:30-4pm
Location: 56-154
Contact: 6.s897hst.s53[at]gmail.com or through Piazza

Course description

Explores machine learning methods for clinical and healthcare applications. Covers concepts of algorithmic fairness, interpretability, and causality. Discusses application of time-series analysis, graphical models, deep learning and transfer learning methods to solving problems in healthcare. Considers how newly emerging machine learning techniques will shape healthcare policy and personalized medicine.

Schedule

Schedule is tentative and subject to change

Lecture 1: What makes healthcare unique?

Problem set 0 Out

[Slides] Lecture 1

Feb 15, 2017

Lecture 2: Supervised learning for risk stratification

Problem set 0 due. Problem set 1 Out.
[Slides] Lecture 2
[required] Big Data In Health Care: Using Analytics To Identify And Manage High-Risk And High-Cost Patients.
[optional] Population-Level Prediction of Type 2 Diabetes using Claims Data and Analysis of Risk Factors.

Feb 22, 2017

Lecture 3: Predicting the outcome of interventions: causal inference from observational data

[Slides] Lecture 3
[required] A targeted real-time early warning score (TREWScore) for septic shock.
[optional] Survival Models.
[optional] Causality references

Lecture 4: Ethics of machine learning to prioritize care: bias, fairness and accountability

L4 Reflection questions due (at noon).
[Slides] Lecture 4
[required] Machine bias and accompanying statistical analyses
[required]One of the following three readings
[required] Learning Fair Representations
[referenced in class] Learning Non-Discriminatory Predictors
[referenced in class] The variational fair autoencoder
[referenced in class] Fairness in Learning: Classic and Contextual Bandits

March 8, 2017

Lecture 5: Machine learning in radiology (Guest lecture. Amir Tahmasebi and Christine Swisher, Philips Healthcare)

[Slides] Lecture 5 part 1 (Amir Tahmasebi), part 2 (Christine Swisher)
[background] Oxford machine learning class material (Lectures 8-10 and 12)

March 15, 2017

Lecture 6: AI for computational pathology (Guest lecture: Andrew Beck, MD, PhD, BIDMC and PathAI)

L6 Reflection questions due (at noon).
[Slides] Lecture 6
[required] Deep learning-based assessment of tumor-associated stroma for diagnosing breast cancer in histopathology images.
[required] Systematic analysis of breast cancer morphology uncovers stromal features associated with survival.
[required] Dermatologist-level classification of skin cancer with deep neural networks.
[optional] Deep Learning for Identifying Metastatic Breast Cancer.
[optional] Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs.

March 22, 2017

Lecture 7: Physiological and laboratory time-series

L7 Reflection questions due (at noon).
[Slides] Lecture 7
[required] Identifying and mitigating biases in EHR laboratory tests.
[required] Discriminative Switching Linear Dynamical Systems applied to Physiological Condition Monitoring.
[optional] Recurrent Neural Networks for Multivariate Time Series with Missing Values.

March 29, 2017

No lecture. Spring break.

Lecture 8: Preventing high-cost care (Guest lecture: Nigam Shah, MBBS, PhD, Stanford School of Medicine)

No reflection questions for this lecture.
[Slides] Lecture 8
[required] Implications of non-stationarity on predictive modeling using EHRs.
[optional] Predicting patient 'cost blooms' in Denmark: a longitudinal population-based study.

April 12, 2017

Lecture 9: Clinical text & natural language processing

L9 Reflection questions due (at noon).
[Slides] Lecture 9
[required] Challenges in clinical natural language processing for automated disorder normalization. (2015)
[required] Electronic medical record phenotyping using the anchor and learn framework. (2016)
[optional] Aspiring to Unintended Consequences of Natural Language Processing: A Review of Recent Developments in Clinical and Consumer-Generated Text Processing. (2016)
[optional] Natural language processing: an introduction. (2011)
[optional] Medical Language Processing: Applications to Patient Data Representation and Automatic Encoding. (1995)

April 19, 2017

Lecture 10: Interpretability of machine learning models

L10 Reflection questions due (at noon).
[Slides] Lecture 10
[required] Intelligible Models for HealthCare: Predicting Pneumonia Risk and Hospital 30-day Readmission. (2015)
[required] Rationalizing Neural Predictions. (2016)
[required] The Mythos of Model Interpretability. (2017)
[optional] Algorithms for Interpretable Machine Learning. (2014; video)

April 26, 2017

Lecture 11: Clustering for subtype discovery

L11 Reflection questions due (next Monday at noon).
[Slides] Lecture 11
[required] Prognostic Physiology: Modeling Patient Severity in Intensive Care Units Using Radial Domain Folding. (2012)
[required] Comorbidity Clusters in Autism Spectrum Disorders: An Electronic Health Record Time-Series Analysis. (2014)
[optional] Cluster Analysis and Clinical Asthma Phenotypes. (2008)

Lecture 12: Disease progression modeling

No reflection questions for this lecture.

[Slides] Lecture 12
[optional] Unsupervised Learning of Disease Progression Models. (2014)
[optional] Temporal disease trajectories condensed from population-wide registry data covering 6.2 million patients. (2014)
[optional] Single-Cell Trajectory Detection Uncovers Progression and Regulatory Coordination in Human B Cell Development. (2014)

May 10, 2017

Lecture 13: Finding optimal treatment policies

L13 Reflection questions due (at noon).
[Slides] Lecture 13
[required] Informing sequential clinical decision-making through reinforcement learning: an empirical study. (2011)
[required] A Reinforcement Learning Approach to Weaning of Mechanical Ventilation in Intensive Care Units. (2017)
[optional] Statistical Methods for Dynamic Treatment Regimes. (2013, book; MIT link)
[optional] Optimal medication dosing from suboptimal clinical examples: A deep reinforcement learning approach. (2016)
[optional] Learning Cost-Effective and Interpretable Treatment Regimes. (2017)
[optional] What-If Reasoning with Counterfactual Gaussian Processes. (2017)

May 17, 2017

Project presentations

Problem sets

Problem set 0 [Deadline: Wednesday 02/15/2017 at midnight]

For the first problem set, please complete the steps required to get access to MIMIC data outlined here One of the steps will require you to fill out a Data Use Agreement (DUA) where you will be asked for:

A reference name. Write David Sontag
The general research area for which the data will be used: Write 6.S897/HST.S53

Problem set 1 [Deadline:March 7th 11:59pm EST through Gradescope]
Problem set 2 [Deadline:April 25th 11:59pm EST through Gradescope]

Grading

25% homework (2 problem sets)
50% course project (You must discuss with the instructors in advance if you are coregistered with 6.862 and wish to use the same project for both classes. Because of the larger number of credits, we expect your project to be more ambitious.)
25% participation (weekly responses to readings)

Projects

Project proposals (one per group): April 7th at 5pm.
Project presentations: May 17th in class.
Project report (one per group): May 18th at 11:59pm.

Collaboration:We strongly prefer 3-5 registered student per group, but we do permit groups with fewer students. Doing something related to your research is fine, but your class project should be distinct and you should be able to isolate your contributions to the project from those of any collaborators outside of the class.

Relationship to other classes: You must ask instructors for permission before submitting project proposal if you wish to use the same project for our class and another class (it should also be stated clearly in the proposal itself). If it is one project for two classes you must:

Produce a project that is twice as large in depth and content as would have been required for either class individually
Obtain permission from the instructor of the other class

Moreover, all students in the project should be enrolled in both classes.

Meeting with instructors to discuss projects:

David's office hours: 3-4:30pm on 3/28 and 4/4 in E25-545
Maggie's office hours: Wednesdays 5-7 in 56-162

15% proposal
15% presentation
70% report

Proposal: At most 3 pages, one per group. Send as a PDF to 6.s897hst.s53[at]gmail.com. Clearly state the following:

Problem you wish to tackle
Description of data you plan to use
Proposed approach and methods
Evaluation plan
Timeline
What each student in the group will do.

We understand that much of this would be preliminary at this stage, but these details are important for us to ensure that you are on the right track.

Detailed presentation and proposal write up guidelines can be found here