Then there are biases about the biases. Highly cited expositions of biases in clinical care, such as those of the insightful emergency physician Pat Croskerry (Academic Medicine, 2003), among many others) very often surmise the presence of biases in clinical care, without the kind of empirical evidence that established the biases in the first place. Sometimes, new and probably useful biases are proposed (such as "search satisfycing"), without any empirical evidence, at all in any domain, for their existence. They are merely postulates. (Granted, empirical evidence is very difficult to generate, this the reason I don't do this kind of research anymore.) Finally, the descriptions of the biases applied to medicine are often strained, or just plain wrong. My favorite is the bastardization of "anchoring and adjustment" into a description of any time a physician seizes upon a diagnosis and discounts disconfirming evidence or fails to consider alternatives. This is not anchoring and adjustment. Anchoring refers to a numerical anchor, and failure to adjust away from it when providing numerical estimates. Here is a summary of the original descriptions, from the wikipedia entry on anchoring and adjustment:
The anchoring and adjustment heuristic was first theorized by Amos Tversky and Daniel Kahneman. In one of their first studies, participants were asked to compute, within 5 seconds, the product of the numbers one through eight, either as or reversed as . Because participants did not have enough time to calculate the full answer, they had to make an estimate after their first few multiplications. When these first multiplications gave a small answer – because the sequence started with small numbers – the median estimate was 512; when the sequence started with the larger numbers, the median estimate was 2,250. (The correct answer was 40,320.) In another study by Tversky and Kahneman, participants observed a roulette wheel that was predetermined to stop on either 10 or 65. Participants were then asked to guess the percentage of the United Nations that were African nations. Participants whose wheel stopped on 10 guessed lower values (25% on average) than participants whose wheel stopped at 65 (45% on average). The pattern has held in other experiments for a wide variety of different subjects of estimation.While it is certainly possible that you can anchor to and fail to adjust away from a non-numerical thing like a diagnosis, this is conceptually strained and I am not aware of any empirical support for it. When something is as carefully and clearly defined as anchoring and adjustment, I do not think we should play loose with it.
Sunk cost bias (my erstwhile mentor and co-author Hal Arkes, pictured above, provided one of the earliest descriptions) refers to situations where decision makers base decisions about expenditures not on future expected benefits, but rather on irretrievable costs that have already been incurred. It is beyond the scope of this post to explain the proposed underlying mechanisms, but sunk cost bias has well-recognized real world examples. Wall Street traders recognize it easily and call it "throwing good money after bad." A proverbial description is "failure to cut bait." Presidents Johnson and Nixon, according to the recent PBS documentary, failed to end the Viet Nam war sooner, because doing so, they feared, would represent wasting the lives that had already been lost, while apparently failing to recognize that overall casualties would simply be greater.
I have long looked for sunk cost bias in medicine, and the only recurring example I have seen is when something is ordered but later information suggests it should not be administered - say, 2 units of blood for a hemoglobin level of 4.5, which on repeat testing is 7.5. Should we give the blood just so it is not wasted? This is a mild form of sunk cost bias. We have incurred costs, which are sunk, crossmatching and thawing it for this patient, but the patient is not expected to benefit, and may be harmed from the blood. We give it not for expected future benefit, but rather because we have sunk costs into the course of action by the time the repeat hemoglobin is received.
This year, the eureka! moment came. The first example was a kidney-pancreas transplant recipient who was desperately ill, whose kidney had failed and had required dialysis for several weeks. There were multiple infections with multi-drug-resistant bacteria, as well as poor wound healing from multiple surgical and decubitus wounds. The patient was receiving insulin. The patient was also receiving immunosuppression for the remote possibility that the transplants would recover. The main focus of recovery in these dire circumstances is for the patient, not her transplanted organs. But those organs represent massive sunk costs, especially to sub-communities. The primary team pushed to discontinue them, appropriately weighing the threats from infection and poor wound healing to be greater than the benefits of possible transplanted organ recovery. The team was told to "wean" the immunosuppressives, but elected instead to stop them cold turkey. Months later, the patient continues to suffer from poor wound healing and MDR bacterial infections. The patient's life and well-being was threatened by sunk kidney bias.
At nearly the same time, another patient had failure of a kidney transplant, was started on hemodialysis, immunosuppression was continued to preserve the pancreas and the possibility of renal recovery, and the patient died from invasive pulmonary mucormycosis. A month after starting dialysis, the patient died, essentially a result of immunosuppression ("good money") to potentially save a sunk kidney (and an afloat pancreas). The sunk kidney was "bad money" because it was likely irretrievable. Maybe it was not, you say, it may have recovered! Thus is one essence of sunk cost bias - overestimation of the probability that the good money will save the bad.
While the correct decision regarding immunosuppression in both of these cases cannot be known, and notwithstanding other caveats of retrospective analysis, these cases highlight the dangers of sunk cost, and of failure to avoid biases it may engender. Whenever sunk cost, or any other sacrosanct goal leads you to fail to accurately weigh costs and benefits of all courses of action, the results can be lethal.