Biotech Torts - Winter 2002

Epidemiologic Proof

I.          Basic questions:

What is the probability that D's product causes the type of injury in question?

If D's product does cause the type of injury suffered by P, what is the probability that it caused P's injury?

II.          P Values

P stands for the probability that D's product causes injuries of the type that P suffers.

P values are usually stated in terms of the probability that the results of the study could be due to chance, i.e., that the detected harm was not caused by D's product.

P is a continuum 0 --------------------------> 1

P = .90/.10 90% sure that the result was not due to change - 9 times in 10

P = .5/.5 50% sure - a coin toss (no info about causation)

P = .10/.90 90% sure that the result was due to chance (i.e., that there is no causation)

General accepted level of certainty is .95/.05, the results are due to chance only 1 in 20 times.

For Bendectin, there are 2 issues: Are we 95% certain that Bendectin causes any birth defects.

Then, what is the relative risk, i.e., what is the chance that Bendectin caused the P's birth defect.

The P's baby had a birth defect and there was no other obvious cause: how can it be that Bendectin did not cause the birth defect?

III.          Background Level of Injury

How common are birth defects?

____% of babies have a birth defect of any type?

____% of babies have a serious birth defect

____% of pregnancies end in miscarriage

Why miscarriages?

5,000,000 births a year

30,000,000 women took Bendectin, ____ babies with serious birth defects are expected.

Animal studies are not predictive of human birth defects ‑ Thalidomide.

IV.          The nature of the defect

Asbestos litigation

Lung cancer

Lung fibrosis

Mesothelioma

What was Thalidomide and why was it given?

What did it cause?

Phocomelia is a rare defect, , but seen in a bunch of Thalidomide cases.

What is DES and why was it given?

DES causes clear cell adenoma

How long does it take?

How common is it?

How did we ever discover the linkage?

Low background + unusual injury + good reporting

Contrast with Agent Orange

Exposed population?

_______________ men and women, mostly men.

Nature of the injury?

Expected background level of birth defects?

_____% background rate X ____________ kids = ?

Cancers ‑ ____% expected X __ persons at risk = ?

How many cases of Agent Orange injuries would there need to be for them to be discoverable?

Is there credible evidence that Agent Orange causes injuries?

Why is the VA paying?

V.          Relative Risk and Confidence Intervals

The key value is the relative risk that D's product caused P's injuries.

Relative risk is the % increase in the number of injuries (over background) due to D's product.

A relative risk of 1 (100%) means that the product does not cause the injury.

A relative risk of 2 (200%) means that the product doubles the number of expected injuries.

Some courts equated a relative risk of 2 with the negligence standard of preponderance of the evidence.

What does a relative risk of .5 imply?

What should be the relative risk of clear cell adenoma from DES?

Relative risk of Thalidomide induced injuries?

Assume that a new anti‑nausea drug, Vomital, causes a generalized increase in serious birth defects (but no unusual birth defects) of 100 per 1,000,000 births.

What is the relative risk of a Vomital induced birth defect?

Some courts use 2 as the minimum for caustion because of more probable than not standard

How many birth defects would Vomital have to cause to get to RR 2?

How sure are we of the relative risk?

Confidence Intervals for Vomital (graph)

95% =6-2.7 / 90% =9-2.0 / 80%=1.6-1.8 / 50%=1.66

VI.          Study Design

Key factors: relative risk and size of study group

Why is Thalidomide ideal?

What about DES ‑ latency is a problem

What about Vomital? 

How many women would you have to study to establish the causation to P = .05?

Study design is critical.

Must carefully identify the population at risk

Must carefully identify the control population

Must be matched

Must identify the condition of interest

Retrospective analysis of data is a problem

Confounding factors

Coffee studies

Causes cancer

Confounded by correlation with smoking

De caf causes cancer

Confounded by people with health problems off high lead

Wine drinking reduces heart disease?

Confounding factors?

Trolling for correlations

Lab tests are 95% confidence level

In 20 tests you would expect 1 false positive.

At 95%, 2 of you could be murders.

Criminal v. Civil Standards

VII.          Dose-response Curves

Step function versus smooth, versus hyperbolic

Graph with long tail buried in background

Step function with threshold above routine exposure values.