DR. TED Deep learning Recommendation of Treatment from Electronic - - PowerPoint PPT Presentation

• DR. TED

Deep learning Recommendation of Treatment from Electronic Data

David Ledbetter Melissa Aczon Randall Wetzel, M.D.

Children’s Hospital Los Angeles (CHLA) Virtual Pediatric ICU (VPICU)

GTC April 7th 2016

1

Outline

• Problem
• Data
• Models
• Results
• Summary

2

3

Problem

There’s actually a patient there For an individual patient, can we recommend the most effective treatment?

• Doctors generate implicit models

○ Requires significant training ○ Combination of academic, clinical experience, and medical research

• Use model to generate treatment

strategies for new patients

○ Limited time (other patients, rapid deteriorations) ○ Limited capacity* to ingest data

*Miller. The Magical Number Seven, Plus or Minus Two. Psychological Review, 63 (2): 81-97, 1956

4

Traditional Approach

• Generate explicit model from clinical data to predict

which treatments will give best patient outcomes

○ Leverage 10+ years of electronic health records (EHR)

■ ~12,000 patient encounters from CHLA PICU ■ (patient, treatment, outcome) triples

• Learn the most important relationships utilizing state-of-the-

art information extraction techniques

5

Moving Forward

My CPU is a neural-network processor; a learning computer. The more CHLA PICU data I have, the more I learn.

6

Moving Forward

7

DRTED

Input Model Output

8

DRTED

Input Model Output

• Convert non-uniformly sampled time-series data into image representation
• Image representation enables exploitation via advanced computer vision

algorithms

9

Data Structure - Overview

• 161 measurements

○ 53 labs/vitals ○ 108 drugs/interventions

• 12 hours of data

○ Sampled every 5 minutes ○ (144 samples)

10

Data Structure - Patient Snapshots

Labs Vitals Drugs Inter. Time

11

DRTED

Input Model Output

• Utilizes patient snapshot as input ‘image’
• VGG-style architecture

○ Heavily exploit temporal relationships with 1-D convolutions

• Generates mortality prediction given

fixed time window

• NVIDIA GTX Titan X used for training

12

v1: Convolutional Neural Network

• Basic structure is a feedback loop

○ At each time, t, a vector X is input ○ An output is generated and fed back into the network

• Advantages:

○ Native comprehension of the temporal dimension ■ Including non-uniform samples ○ Increased temporal memory ○ Formal feedback mechanism ○ Generate predictions for all vitals

13

v2: Recurrent Neural Network

Mortality Prediction Physiology forecasting Patient Vitals Patient Treatments

Kernel XV XT ym yp

14

DRTED

Input Model Output

• Results for holdout set of 3372

patients with encounter length

• f at least 12 hours

○ DRTED AUC - 90.3% ○ PIM 2* AUC - 83.0%** Notes: *Pediatric Index of Mortality **Published PIM 2 AUC

15

Assessment - Mortality

• Predictions generated for 5

key vitals + Mortality ○ Heart Rate ○ Diastolic Blood Pressure ○ Systolic Blood Pressure ○ Respiratory Rate ○ Pulse Oximetry

• Accurate prediction of vitals

and mortality enable prediction of treatment effects

16

Predictions

Respiratory Rate Pulse Oximetry

Probability of Survival

Heart Rate Systolic Pressure Diastolic Pressure Time (hours) Time (hours) Time (hours)

Utilize machinery to predict effect of each treatment on patient

17

Predicted Treatment Effect

Patient diagnosed with: Cardiac Arrest Cardiomyopathy Epileptic Seizures Pneumothorax Eventually treated with Piperacillin Vancomycin Epinephrine Phenylephrine

• Applied deep learning methods on 10+ years of

Pediatric ICU data

○ Able to generate state-of-the-art mortality predictions ○ Able to generate physiology predictions ○ Able to generate predictions of treatment/therapy effects

• Framework/machinery is being extended to provide

additional Decision Support Services to clinicians

18

Summary

Contact:

ledbetdr@gmail.com macs.aczon@gmail.com

Interested?

Machine Learning in Healthcare Conference

August 19th, 20th Children’s Hospital Los Angeles http://www.mucmd.org/

19

Machine Learning in Healthcare

20

Backups

21

Augment Data - Example

Original 0-8 Hours

22

Patient Snapshots - High Survive Survivors

Labs Vitals Drugs Inter.

Sample surviving patients with high predicted probability of survival

23

Patient Snapshots - Low Survive Died

Labs Vitals Drugs Inter.

Sample non-surviving patients with low predicted probability of survival

24

Patient Snapshots - Low Survive Survived

Labs Vitals Drugs Inter.

Sample surviving patients with low predicted probability of survival

25

Patient Snapshots - High Survive Died

Labs Vitals Drugs Inter.

Sample non-surviving patients with high predicted probability of survival Patient encounter lasts for 4 days but no data during first 72 hours

• Minimize: ||xt+1

V - ht V||α + β|y - ht y|

• First term represents ability to predict future vital readings

from current information

• Second term represents ability to predict outcome from

current information

• Alpha term represents vector cost weight for vital prediction
• Beta term represents cost weight of mortality prediction

26

Optimization

• Each patient is represented as a sequence of nt (m+1)-length vectors

○ m+1 → # of measurements + Δt ○ nt → # of discrete time steps

• Vitals receive forward fill + median imputation
• Intermittent exogenous inputs are delta functions

○ Continuous drugs/interventions are propagated

• Δt informs the algorithm how far into the future it needs to predict

○ training is allowed to ‘cheat’ - knows when next measure is ○ But that’s OK, we just want to learn the relationships ○ At test time, we specify Δt to predict precise point in time

27

Implementation

28

LSTM

http://colah.github.io/posts/2015-08-Understanding-LSTMs/

• Intrinsic difficulty of assessment is ambiguity of truth
• Best metric would be A/B test

○ Average outcome of patients whose doctors have access to decision aid vs. ○ Average outcome of patient whose doctors do not have access to decision aid ○ Not practical for initial development or iteration

• Instead, develop intuitive quantification

○ Provide adequate feedback for iteration ○ Base on simple assumption: ■ Maximizing frequency of recommendations of actual treatments used in successful cases is good

29

Assessment - Treatment Response (cont.)

• Compress interventions and drugs into treatment response

y = ΔHealthIndex * treatment where: ΔHealthIndex = HealthIndexi+1 - HealthIndexi HealthIndexi is the expected survival at time ti as computed by survival model treatment is a vector: [t1, t2, …, tn], with ti ∈ {0, 1} indicating presence of treatment categories

• Elements of y contain:

○ positive values for treatments that contributed to improvement ○ negative values for treatments detrimental to patient condition ○ 0 for treatments not utilized

30

Assessment - Treatment Response