The prediction problem

A major problem in all preference or utility assessment, particularly holistic assessments that require the visualization of a health state in its entirety, is that people must often be asked to assess their preferences for health states that they have not yet experienced. That is, they must predict how they will feel about future health states that may arise as a result of their decisions.

Unfortunately, people are generally poor predictors of future experiences for several reasons.

Predictions about acute experiences
For acute states, people will experience the state for a short period, but they will experience the memory of the state for the rest of their lives, and it is the memory of the state that will be used in making future decisions. This would pose no problem if memory was always veridical, but striking studies suggest systematic biases in the memory of unpleasant experiences that can affect decision making.

In one study, patients undergoing colonoscopy or lithotripsy provided minute-by-minute ratings of the discomfort of the procedure as they were undergoing it. After the procedure, and a month later, they provided an overall rating of discomfort. Procedures varied considerably in length (from 4-69 minutes), but overall ratings had no relation to the length of the procedure. Instead, ratings were related to two “snapshots” in memory: the peak discomfort the patient experienced during the procedure and the discomfort the patient experienced at the end of the procedure (Redelmeier and Kahneman 1996).

This “peak and end rule” implies that it is possible to improve such experiences as colonoscopy by extending them — for example, by leaving the scope inserted in the rectum “unnecessarily” for a few minutes at the end of the procedure. In fact, Redelmeier, et al. (2003) tested exactly that intervention in a randomized controlled trial, and found that it resulted in better evaluations of the procedure and increased likelihood of repeat screening.

Would you agree to have an artificially lengthy colonoscopy if you know that you would remember it better?

Predictions about chronic states
Two powerful forces make it difficult to predict how we’ll feel about chronic states: contrast effects and adaptation effects.

Contrast effects refer to the impact of our previous state on our current state. Someone who is very happy yesterday and sad today will experience the sadness more acutely than someone who was not particularly happy the day before, due to the contrast between the states. Lottery winners find that subsequent events in their lives, judged in comparison to the experience of winning a jackpot, lack frisson. Paraplegics, on the other hand, find that subsequent events in their lives, judged in comparison to losing their ability to walk, represent significant victories. (These two groups were studied by Brickman et al., 1978).

Adaptation is the other great force in ongoing experiences and chronic states, and people regularly underestimate the human capacity for adaptation. A regularly reported finding in large-scale studies of utilities for commonly-understood disabilities is that respondents without a disability rate the utility of life with the disability significantly lower than respondents who actually live with the disability; blindness, for example, appears much worse to the sighted or newly blind than to those who have lived without sight for some time. (G. Ardine De Wit 2000)

Several recent studies illustrate the power of adaptation. Smith et al. (2006) found that people with colostomies gave higher time-tradeoff utilities to life with a colostomy than people without colostomies. Of particular interest were a large group of subjects who had previously had colostomies which had then been reversed. These subjects gave essentially identical utilities as those who had never had colostomies. This suggests that it is not merely experience with a health state that impacts valuation, but current experience — that is, having adapted to the health state and not to another.

Surrogate predictions
An often-tempting alternative to requiring patients to accurately predict their evaluation of future health states is to have an agent who has more experience with the health states provide surrogate judgments on a patient’s behalf. It seems infeasible for most patients to identify friends or relatives who happen to be living in the set of health states that the patient needs to evaluate. On the other hand, clinicians who have cared for patients in all of the relevant health states are usually available. Might clinicians be capable of taking their patients’ perspectives and providing surrogate measures of utility for them?

Although little research has investigated this question, the results of at least one investigation are not promising. Elstein et al. (2004) studied 120 prostate cancer patients and their physicians. Patients provided time tradeoff utilities for three hypothetical future states and for their own current health; physicians provided surrogate time tradeoff utilities for the hypothetical states and the patient’s current health. As a group, physicians provided higher utility assessments than those actually provided by patients — a result that might seem promising, particularly if patients underestimate adaptation and physicians expect it. Unfortunately, however, although physicians were reasonably accurate in predicting the preference order in which patients would place the four health states, there was no correlation between the utilities provided by the patient and his or her physician within each health state.


  1. Oh wait. Yes, I have. I’m sorry, but I just don’t have it in me right now to type it all out again. Besides, it was just ramblings anyway. You didn’t want to hear me go on and on about this, right?

  2. john kolp

    boy george

    that is a tterriable answear

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>