Tuesday, 11 March 2014

Day 214: Individuation.

At first it was a leit motif, but now it is a drum-beat.

The tobacco defence in the Blais-Létourneau trial is hitting hard on the very idea of wronged smokers as a collectivity. Today's witness insisted that even the calculation of a common level of harmful exposure to smoke is inappropriate.

Statistician Mr. Laurentius Marais made clear that he thinks it is impossible to use epidemiology to decide whether a smoker with cancer can be considered more likely than not to have become ill as a result of smoking, no matter how much he or she smoked.

Are they trying to hang their hat on a Howard Engle break-up-the-class hook?

Backing away from a population-wide approach

Mr. Marais was engaged by JTI-Macdonald to "review and comment on" the analysis prepared for the plaintiffs by Montreal epidemiologist, Jack Siemiatycki.

Mr. Siemiatycki had estimated that over 100,000 Quebecers suffered from the four diseases involved in the Blais lawsuit as a result of smoking. From the amount of illness attributed to smoking, he calculated how much additional risk a smoker would have to face before one could say that their illness was more likely than not to have been caused by smoking (a doubled risk). Combining studies on risk,, he calculated that this level of risk happened if someone smoked a pack of cigarettes a day for 3-5 years, depending on which illness was involved.

Mr. Siemiatycki's report was novel in that it translated the well established link between smoking and disease into workable numbers that could be applied in a mass tort-class action.

Now it is the industry witnesses' turn to try to reverse that work. By trying to smash Justice Riordan's confidence in Mr. Siemiatycki's calculations, Mr Marais seems to be the advance scout in a campaign to unravel the logic of collective recovery. Decertification by the back door?

Death by a thousand cuts

Mr. Marais seemed to have an endless pile of criticisms against Mr. Siemiatycki's work, and a long list of unrecoverable fatal flaws that he felt were made by one of Canada's more renowned epidemiologists.

"With the greatest of respect for Dr. Siemiatycki' he would begin, in a somewhat mannered delivery, only to follow up with a harsh, if diplomatically phrased, attack.

Dr. Siemiatycki's methods were wrong. His assumptions were wrong. His model was wrong. His slopes were wrong. There were mistakes in his calculations. The results are unreplicable. The conclusions are unusable.

So many criticisms - large and small - seemed intended to overwhelm. And why not?  Epidemiology and statistics are, let's face it, a little mysterious. And with that mystery, perhaps, a little mischief.

Mr. Marais certainly explained some things very clearly, and at length. But more than once he moved smoothly from the over-explained to the under-explained, ending up with a conclusion that directly opposed Mr. Siemiatycki's, without an overly clear trail of how he got there.

Is this a technique to make the juror or judge think that they are to blame for not following or understanding?

Wool over Eyes?

Even with my limited grasp of statistics, I heard explanations that seemed designed to obscure more than enlighten.

Yesterday, Justice Riordan had asked a question the answer to which, I thought, would have been an explanation of standard deviation. ("If you do a study and the average height is 5'9", plus or minus 4" then could I not be comfortable in assuming, at 95% confidence that any individual who came in would be at least 5'5"?)

The answer he was given by Mr. Marais, amplified this morning, did not provide him with this statistical tool. Instead it directed him to think that nothing could be derived from the mean.

Another questionable illustration today came in the form of a visualization of how "random effects model" (don't ask!) was a false way to make inferences in meta-analyses.

He presented a graph suggesting that such an approach would falsely predict Quebec self-reported obesity rates, and would lead one to think they are in the middle of the pack of North American levels, instead of at a much lower value. [His illustration, with the Quebec value shown as a red dot, is shown below].

Justice Riordan seemed quite taken with this -- "I find that very interesting."

Say again?  Creating comparative groups by measuring against number of jurisdictions, not number of people?? Selecting the fattest global population (USA) as your frame of reference?

To my eyes, it seemed to illustrate the problems with bad data and fuzzy thinking, not techniques within the meta-analysis machine. But, Judge Riordan found it very interesting, and he was the audience.

The bottom line: no threshold value to attribute legal responsibility. 

Mr. Marais made clear that he felt that Mr. Siemiatycki had consistently biased his results to create a larger number of potential claimants. His slope for he dose-response was too high, and it was anchored at the wrong point. A curvilinear dose-response would produce lower values, etc.

These, together with his elaborated concerns with confidence levels, lead him to conclude that it was not possible to find any "statistically reliable" threshold level of smoking that could be used to attribute responsibility to cigarettes. "Not 5 pack years, not 50 pack years, not 150 pack years!"

So where does that leave the judge?

Mr. Marais seems to have been under instruction to offer only negative criticism of the plaintiffs' estimates of the harms done by smoking, and not to propose an alternative way of measuring how many people have become ill through smoking.

Close to the end of Mr. Pratte's questions, Justice Riordan addressed the lawyer and suggested that Quebec law did not require the perfect to be the enemy of the good.

"I hear everything that Dr. Marais is saying – it is very interesting testimony. This man is a scientist and is very competent. But I have a different role than a peer reviewer. My level of satisfaction is dictated to me by the Civil Code - Article 2804;--- Evidence is sufficient if it renders the existence of a fact more probable than its non-existence -- That is the test that I have to apply.

This testimony is on point and well founded within the confines of a scientific debate – But it is only fair to let you know that I think there is something else I have to look at." 

A final volley

Mr. Pratte acknowledged the unusual intervention, but did not seem to deviate from his prepared questions.

He asked his witness to again beat on the drum of individuation. "The probability of causation ... is not a concept that evaluates or assesses or describes any individual person. It is a descriptive measure on average of an entire population." 

Dr. Siemiatycki's approach, even if it were not replete with problems, would result in "over-counting and overcompensation" of those who are close to the threshold value of a relative risk of 2. "Only 50 out of 100 cases would have been caused by the exposure .. but the court would be accepting 100 out of 100." 

(He did not draw attention to the under-counting and under-exposure on the other side of the threshold, for whom the court would be rejecting 100 out of 100, even if 50 had become ill as a result of exposure).

Nor could a higher level of threshold risk be set. There was "simply no place that the meta analysis lead you to as the appropriate critical amount."  There was no scientific consensus in favour of a "balance of probabilities" approach, which was  a "hugely controversial point" among scholars.

Mr. Pratte's last questions asked him to confirm the importance of variables that made Quebec smokers different from one another  -- when they started smoking, whether they had quit, and how long ago, how heavily they smoked. 

The court adjourned at 3:00 p.m. Tomorrow - the second anniversary of this trial - the plaintiffs will cross-examine Mr. Marais.