Trottier Symposium Lecture
Lorne Trottier- engineering student graduate from McGill and philanthropist
Topic: Is that a fact? Making sense of the facts (in various media)
Timothy Carfield and Jimmy Carlinus-guest speakers; Moderator Joe Swartz
Facts: observations that stand the test of time: observations have to be repeatable. Facts may
turn out to be wrong. There are some things that don’t change. Theory to explain things ex.
Fact: Apple, when attached to a tree, will not fall. Theory: gravity.
BPA-molecular structure is a fact. Whether or not it leaches out of baby bottles and affects their
health is not a fact. It still needs to be explored. FDA removed to investigate, not because it
causes harm necessarily.
Flushed with Pride book- parody- Thomas Crapper created the toilet- FALSE!! He was a real
person though. He was a sanitary engineer and did do improvements in plumbing.
Vitamin D- not preventing osteoporosis? Helping for diabetes?
Moops- massive online courses. McGill is getting into this- first free course. Take a peek at the
promo. (January 2014 McGillX)
Sherlock Holmes is Swartz’s favorite literary character.
Johny Ionnas- from Standford. over 100 publications. Paper where he observed scientific
studies (peer reviewed) were flawed.
Most scientists are under a lot of pressure to get results (“very nice papers”-avg6-7pgs).
Analogy using poem- piece of papyrus found in desert translated word for word.- like science:
trying to find a very nice results but not that accurate.
Diet causes cancer- believes that this is true. Randomly picked 50 ingredients from a cookbook
and looked at peer reviewed papers in pubmed and see if any are related. 40/50 were found-
10 missing ex. Ingredient vanilla not there, but vanillin is. With relative risk of 2, serving will
double cancer risk (?). 5-10% of correct estimate- above 10% not accurate study. Most are
larger studies- .99 per servings fruits and vegetables relative risk reduction. Studies goes both
ways (good and bad). Z scores : very little action of studies with no significant results.
->Empirical studies suggest that most of the claimed statistically significant effects in traditional
medical research are false positives or substantially exaggerated.-> know this by replication.
Genotyping helped figure this out with very large scale studies. Average replication rate 1.2%.
Outside of genetics- nongenetic biomarkers- can personalize medicine if id this. Larger studies
found some effects, but much smaller effects (1.1-1.4) than measured using smaller scale
studies. If use more complex models/multivariant model (genetic, clinical, etc.), found model less
effective 2 time than 1 . 80% of literature, no one has tried to apply, replicate, do anything with
it. Usually same author that use the paper again.
Who would pay for replication studies? Government wants “new”, so not really. Looked at
research of oncology drug targets- only 10-25% replicatable projects. 50% of research
according to one source not replicable- needed to progress in science. This may be worse in
other countries. US studies tend to have more exaggerated results probably since a lot of
pressure to get more funding.
Why some results not credible? Not due to fraudulence. Bias, random error (see multiple
comparisons)- usually a lot of both. “soft sciences”(pharmacology, med, psychology) have more
Need to learn to work with small values- can mean a lot (ex. 1% cancer risk reduction for fruits
and vegetables). VERY rare that there will be a significant impact.
Solution: work together- ie large scale collaborations. Ex. Smoking- 159 determinants when
individual studies- with large scale collaboration 7 genes that regulate smoking that weren’t in
159 and are strongly reproducible (unlike 159 genes). Can use this info and make a difference
when info is used.
Solution 2: Present results differently. Ex. Equator- improve transparency of info presented
Solution 3: registration- instead of being secretive about research- this is a study we are doing-
this is our plan- success in randomized trials (for 8 yrs now-1300 trials registered).
Different levels: 0=no registration (ex. Done out of blue); level 1: registration of dataset; level 2:
etc….; level 5: public streaming of entire plan-