First off, I am happy that we both seem to agree on a qualitative fact: there is indeed a tradeoff between what statisticians call type I and type II errors. At one extreme, you can let everything through, advance technology rapidly, and suffer some side effects (type I bias). Or you can block everything, stop technology, and suffer no side effects (type II bias). If we agree on this qualitative point, the key is whether we are currently at a Pareto optimum. Is our current system optimizing the type I vs. type II tradeoff? I have a numerical scenario below which you can critique, but first to your points.
It's faster to run experiments on animals than people.
I'm not gainsaying the utility of animal models. I just think the goal needs to be to get to humans as soon as the safety data is in,
because people are dying.
I think you are vastly overestimating what society has to
gain by deregulating medicine. You'll get a one-time gain
of 10 years of progress at the cost of an unknown number of
lives.
Well, the reason sulfalinamide/thalidomide were heavily covered in 1938/1962 respectively was that those were relatively
rare events. So I would somewhat disagree that the number of lives would be unknown. But, ok, let's take as a given that some would die. On the other side of the ledger, we both agree that tens of millions of people each year are dying from cancer and heart disease. So let's consider two scenarios for a cure for condition X, which kills 1 million people per year.
In scenario I, we do it status quo and safe, with no deaths. Very generously, let us grant that a cure appears in 10 years. This is generous because a regulated market may never iterate upon the cure if it is radical/different (e.g. Barry Marshall and H. pylori).
In scenario II, we accelerate the cure in a deregulated market. The R&D phase takes 1 year and costs us 100 deaths from test pilots / early adopters; the scaling phase takes 2 years and costs us another 900 deaths from volunteers. These numbers are vastly in excess of any reasonable safety testing paradigm in a deregulated space (no one died in Banting & Best's experiments) and I cite them as extremely conservative upper bounds.
Ok. Then in scenario I, the status quo, you had
- 0 die from testing
- cure appears at end of 10 years
- 10 million people die over those 10 years
- 10 million deaths
In scenario II, you had
- 1000 die from testing over 3 years
- 3 million die from disease over those yeers
- cure appears in year 3
- no further deaths
- 3 million + 1000 total deaths
So scenario II saves ~7 million lives. Feel free to play with the numbers, but that's the kind of calculus I think we need to engage in, one that explicitly reckons with the cost of delay. In reality, the number of deaths attributable to R&D won't be close to 1000, though it won't be zero. But there is no reasonable scenario in which R&D actually consumes anything close to as many lives as the disease itself.