Making Risk-Benefit Assessments of Medical Research Protocols

Respectively, The Declaration of Helsinki, Section B, Item 18; The Belmont Report, Part C, Section 2 (1979); and 45 CFR § 46.111.a.2 (2001). I should note that the authors of The Belmont Report admit that talk of “balancing” and “risk-benefit ratios” is metaphorical, but offer no replacement for the metaphor.

45 CFR § 46.111.a.2 (2001).

Here I assume that the expected value of not running the protocol will always be zero, since it represents no change in the status quo.

The authors of The Belmont Report speak against crude expected value calculation and in favor of more subtle weighting. In fact, they suggest that there is a side-constraint against certain kinds of risks: … assessment of the justifiability of research should reflect at least the following considerations … Brutal or inhumane treatment of human subjects is never morally justified. (Part C, section 2.)

Still, the existence of side-constraints like these would not affect the principal argument of this section. If someone proposed that expected value calculation should take place only if certain side-constraints are satisfied, then they are still proposing a method which requires the calculations that I think are often impossible to make.

Improving Informed Consent for Research Radiation Studies , National Institutes of Health Radiation Safety Committee (October 17, 2001).

See, e.g., “1990 Recommendations of the International Commission on Radiological Protection,” ICRP Publication 60 (New York: Pergamon Press, 1991) (especially chapter 3).

I say “usually” because some protocols produce such spectacular results that one could judge them greatly efficacious even if there were no other medical knowledge about the same disease or procedure. The first uses of penicillin on syphilis were like this. The antibiotics cured the disease in almost everyone they were given to.

For an overview of the problems with such scales, see Nord E. , Cost-Value Analysis in Health Care: Making Sense out of Qalys(Cambridge: Cambridge University Press, 1999). Cf. Broome J. , “Qalys,” Journal of Public Economics 50 (1993): 149–167.

Someone might say that my large numbers are obfuscatory; that this problem can be reduced to a much more simple one:

Imagine that there is a one in two chance that a treatment will reduce a single patient's nausea a little bit. Imagine that the treatment also had a one in 150 chance of reducing the patient's lifespan by four months. Question: Is it rational for the patient to opt for the treatment or not?.

Here some simple division has been done. A one in 10,000 chance of helping 5,000 people has been reduced to a one in two chance of helping one person; and a one in 15,000 chance of hurting 100 subjects has been reduced to a one in 150 chance of helping that same, single person.

This reduction is suspect. One reason is that the original problem involves different groups of people. That fact may (or may not) be morally relevant, so it cannot simply be eliminated. Another reason is that the reduction ignores the possibility that morality requires us to be (the impersonal equivalent of) risk-averse — i.e., to discount the value of outcomes based upon their likelihoods.

Let me explain. There is an intuitive difference between (a) a one in two chance of getting two units of benefit, and (b) a one in 10,000 chance of getting 10,000 units of benefit. Choosing the first can seem rational even if the latter does not. The reason is that a person might be risk-averse: He might count the value of 10,000 units of benefit, in a risky situation, as less than five-thousand times the value of two units of benefit, when the latter might be gained in a less risky situation.

In light of this, there seem to be some situations in which it's illegitimate to reduce problems of decision-making, at least without the further assumption that the decision-maker isn't, or shouldn't be, risk-averse.

Of course, that is a point about rationality, not morality. And one might think this difference is important. After all, I am asking an explicitly utilitarian question: Whether doing something has a greater expected value than something else. In other words, I have asked a question of maximization. And one might think that this sort of “reduction” is legitimate if we only want to know how to maximize goodness.

Yes and no, I think. On one hand, it's natural to interpret the question “does doing (a) instead of (b) maximize expected value?” as one in which there is no risk-aversion. That is, it's natural to interpret it as one in which reduction is legitimate. But on the other hand, that is not the only interpretation of the question. One need not assume that a moral person should be risk-neutral with respect to benefits, just as one need not assume that a rational person will be risk-neutral. Morality might require us to play it safe with the interests and lives of others.

Of course, someone might think that a risk-averse decision-maker could not claim to be “maximizing potential benefit.” As I've admitted, in a way that's right — the most natural interpretation of “maximize potential benefit” is one in which there is no risk-aversion. But it seems to me there is another way in which a person can still be said to maximizing potential benefit, because after the benefits are discounted for risk-aversiveness, he then goes on to maximize benefit.

When we ask ourselves whether this calculation can be made despite the vagueness about probabilities and such, we must not confuse that question with another: Can we know that the protocol is ethically acceptable despite the vagueness about probabilities and such? The two questions are easy to confuse, and yet here the only question is whether the calculation can be made on the intuitive level.

Later on I will argue that we can judge the morality of a protocol without making calculations, and thus that we might be able to judge a protocol's morality despite terrible vagueness about the surrounding scientific facts.

See, e.g., Emanuel E. Wendler D. Grady C. , “What Makes Clinical Research Ethical?” JAMA 283 (2000): 2701–2717.

Another principle I won't consider is this: A protocol is moral only if it constitutes the best use of the resources that would be devoted to it. This may or may not be a principle governing the distribution of funds for medical research. But whether it is or not, my focus is on the principles that IRBs should use when assessing protocols, and IRBs are simply not equipped to make the comparative judgements required by this principle. They cannot be expected to think through all possible uses of research funds.

This definition of competence is drawn from Buchanan A. Brock D. , Deciding for Others: The Ethics of Surrogate Decision Making (New York: Cambridge University Press, 1989): At 23ff. Buchanan and Brock's definition also includes a requirement that the subject understand the nature of the options presented, but since I am treating informed consent as a separate requirement, it is not included here. In addition, because there is such a large literature on competence, I will say relatively little about the two conditions that are included in my definition.

It is worth nothing one other interaction between the agreement principle and the other standards for ethical research. The agreement principle says that a protocol is moral just in case some competent decision-makers would enter into the protocol, but it does not say how many decision-makers must be willing to do so. In particular, it does not require that there be enough competent decision-makers to adequately power the study. However, I take that requirement to be covered by another standard for ethical research — the one that says that in order for a study to be ethical, it must be scientifically valid.

See, e.g., Feinberg J. , The Moral Limits of the Criminal Law, Volume Three: Harm to Self (Oxford: Oxford University Press, 1986): 27–97; Raz J. , The Morality of Freedom (Oxford: Oxford University Press, 1986).

See Mill J. S. , On Liberty (Indianapolis, IN: Hackett, 1978).

This example brings out a complexity in liberalism that was not discussed earlier. I have said that if an offer wouldn't be accepted by an informed and competent decision-maker, then we should ban it. But another commonly-accepted liberal principle is that the government should not bother banning things that no one is going offer, or that no one would do anyway. On those grounds we might not bother banning the hand-feeding of the lions, even though such a ban would be justified by liberal principles.

One of those subtle points is that a limited voluntarist principle would normally include the requirement that acceptable decision-makers make voluntary and uncoerced choices. I have left those factors aside because they are not central to my discussion.

Consumer-safety legislation might have other rationales as well. It might be justified by an appeal to the unfair bargaining position between consumers and manufacturers. It also might be justified by appeal to utility.

The objection under consideration might be put forward by people of at least three different outlooks: (1) those who were consequentialists about morals in general; (2) those who felt that researchers who receive federal funds (and hence are regulated by IRBs) are bound, like the government itself, to work for the public good; and (3) those who felt that federal research money was allocated for the specific purpose of promoting the general good, and hence that researchers who receive federal funds are obligated to fulfill that purpose.

It is worth nothing that the requirement not to treat money as a benefit is distinct from the improvement principle as well. The improvement principle merely says that a protocol has acceptable risks and benefits only if it will do more good than harm, and that claim says nothing about what counts as a benefit.

For an extended discussion of these points, see Feinberg J. , The Moral Limits of the Criminal Law, Volume Four. Harmless Wrongdoing (Oxford: Oxford University Press, 1988): 176–276; A. Wertheimer, Exploitation (Princeton: Princeton University Press, 1996): 278–309.