Beyond Prediction: Drawing Causal Inference from Big Data S. Reza - - PowerPoint PPT Presentation

▶
beyond prediction drawing causal inference from big data
SMART_READER_LITE
LIVE PREVIEW

Beyond Prediction: Drawing Causal Inference from Big Data S. Reza - - PowerPoint PPT Presentation

Sep 08, 2022 433 likes •589 views

Beyond Prediction: Drawing Causal Inference from Big Data S. Reza Jafarzadeh, DVM, MPVM, PhD Section of Rheumatology Boston University School of Medicine 2 Disclosure Boston University Slideshow Title Goes Here Funding NIH R03, R21

slide-1
SLIDE 1

Boston University School of Medicine

Beyond Prediction: Drawing Causal Inference from Big Data

  • S. Reza Jafarzadeh, DVM, MPVM, PhD

Section of Rheumatology

slide-2
SLIDE 2

Boston University Slideshow Title Goes Here Boston University School of Medicine

Disclosure

  • Funding
  • NIH
  • R03, R21
  • Rheumatology Research Foundation
  • Innovative Research Award
  • Pfizer
  • Global Awards for Advancing Chronic Pain Research

2

slide-3
SLIDE 3

Boston University Slideshow Title Goes Here Boston University School of Medicine

Beyond Prediction

  • To identify a disease (an outcome/exposure) in a big

data

  • Machine learning
  • ICD-10 codes
  • Questionnaire
  • “Have you ever been told by a doctor or other health

professional that you have some form of arthritis, rheumatoid arthritis, gout, lupus, or fibromyalgia?”

3

slide-4
SLIDE 4

Boston University Slideshow Title Goes Here Boston University School of Medicine

Prediction; Problem

  • Uncertainty
  • Sensitivity: Pr(T+|D+)
  • Specificity: Pr(T-|D-)
  • Probability Laws

Pr 1

  • Pr(A or B) = Pr(A) + Pr(B)
  • If A and B are independent

4

slide-5
SLIDE 5

Boston University Slideshow Title Goes Here Boston University School of Medicine

Prediction with Uncertainty

  • We want perfect prediction (i.e., T+ = D+)
  • Our prediction could be True or False
  • Pr(True or False) = Pr(True) + Pr(False)
  • Pr(True Positive) = Prevalence * Se
  • Pr(False Positive) = (1 – Prevalence)*(1 – Sp)
  • Pr(T+) = Prevalence * Se + (1 – Prevalence)*(1 – Sp)
  • Quantitative bias analysis
  • Incorporate uncertainty

5

slide-6
SLIDE 6

Boston University Slideshow Title Goes Here Boston University School of Medicine

Inference with Uncertainty

  • “Have you ever been told by a doctor or other health professional that you have

some form of arthritis, rheumatoid arthritis, gout, lupus, or fibromyalgia?”

  • 54.4 million adults (22.7% of US population) reported by CDC*
  • 91.2 million adults (36.8% of US population)

* Barbour et al 2017 6

slide-7
SLIDE 7

Boston University Slideshow Title Goes Here Boston University School of Medicine

Causal Inference; Big Data

  • Data is random
  • Get a distinct realization by repeating an experiment
  • Data distribution can be learned
  • By repeating an experiment
  • Need to represent data-generating distribution
  • That corresponds to observed data
  • By a statistical model

7

slide-8
SLIDE 8

Boston University Slideshow Title Goes Here Boston University School of Medicine

Causal Inference; Target

  • Define a target: What we wish to learn from data
  • Average treatment effect (ATE)
  • Model coefficients
  • I want to exponentiate model coefficients to get odds ratios
  • Estimate target parameter
  • Parametric modeling
  • Machine learning
  • Incorporate uncertainty in estimating target
  • Use statistical theory

8

slide-9
SLIDE 9

Boston University Slideshow Title Goes Here Boston University School of Medicine

Randomized Longitudinal Study

  • Causal treatment effect can be validly estimated
  • Treatment assignment randomized at each visit
  • A: treatment (or exposure)
  • L: measured confounders
  • U: unmeasured/unknown confounders
  • Y: outcome
  • Randomized treatment assignment depends only on
  • Prior treatment history

9

slide-10
SLIDE 10

Boston University Slideshow Title Goes Here Boston University School of Medicine

Randomized Longitudinal Study

  • Causal treatment effect can be validly estimated
  • Treatment assignment randomized at each visit
  • A: NSAID, physical activity
  • L: age, sex, BMI, pain
  • U: genetics, physically-demanding job
  • Y: radiographically-measured joint space width
  • Randomized treatment assignment depends only on
  • Prior treatment history

10

slide-11
SLIDE 11

Boston University Slideshow Title Goes Here Boston University School of Medicine

Observational Study

  • Why different persons received different treatments?
  • Why a person received a different treatment at a

different visit?

  • A: NSAID, physical activity
  • L: age, sex, BMI, pain
  • U: genetics, physically-demanding job
  • Y: radiographically-measured joint space width
  • Probability of receiving treatment depends on
  • Prior treatment history
  • Measured confounders history

11

slide-12
SLIDE 12

Boston University Slideshow Title Goes Here Boston University School of Medicine

Causal Inference; Framework

  • Causal framework to formalize reasoning
  • Association
  • Seeing
  • Intervention
  • Doing
  • Causation
  • Imagining
  • Targeted learning
  • Unifies estimation (machine learning) and causal inference

12

slide-13
SLIDE 13

Boston University Slideshow Title Goes Here Boston University School of Medicine

Targeted Learning

13

slide-14
SLIDE 14

Boston University Slideshow Title Goes Here Boston University School of Medicine

Thank You!

14