Drug Efficacy in the Wild Tim Vaughan 17 June 2011

Proprietary & Confidential 1

Drug Efficacy in the WildTim Vaughan17 June 2011

Proprietary & Confidential 2

Contents

PatientsLikeMe What can MikeFromFinland teach us, and vice versa? Lithium delays progression of ALS?! PatientsLikeMe’s observational study Finding patients like me Results Predictive modeling / What is my outcome? Concluding remarks

Proprietary & Confidential 3

PatientsLikeMe web site

Proprietary & Confidential 4

PatientsLikeMe background – Three brothers

Proprietary & Confidential 5

Stephen Heywood (alsking101)

Proprietary & Confidential 6

What can Mike teach us, and vice versa?

Lithium

Proprietary & Confidential 7

Lithium delays progression of ALS?!

Fornai et al., PNAS 105:2052-2057 (2008)

Proprietary & Confidential 9

Timeline

Proprietary & Confidential 10

Patients track their progress

Proprietary & Confidential 11

The “kitchen sink” plot

Proprietary & Confidential 12

Random control may not be a “patient like me”

Proprietary & Confidential 13

Demographics – age

Proprietary & Confidential 14

Demographics – onset site

Proprietary & Confidential 15

Demographics – sex

Proprietary & Confidential 16

Matching algorithm

Proprietary & Confidential 17

Matching across the entire sample

Proprietary & Confidential 18

Pre-treatment progression bias reduced

Proprietary & Confidential 19

Results of lithium treatment

Proprietary & Confidential 20

Kaplan-Meier for patients & data

Proprietary & Confidential 21

Biases and other stuff that worried us

Self-selection for treatment “Recruitment bias” Data reported (vs. data opportunity) Outliers (e.g. PMA and PLS) “Optimism bias” at treatment start

Proprietary & Confidential 22

What Mike (and PatientsLikeMe) can learn

Proprietary & Confidential 23

Conclusions

Structured, self-reported patient data, despite being subject to bias (like all patient data!), has value

Think about bias, and then think about bias again (Repeat) “Pair programming” for statistics