Take, for instance, the hypothetical case of a man who comes into the ER complaining of intermittent left-side chest pain that occasionally comes when he walks up the stairs and that lasts from five minutes to three hours. His chest exam, heart exam, and ECG are normal, and his systolic blood pressure is 165, meaning it doesn’t qualify as an urgent factor. But he’s in his sixties. He’s a hard-charging exectuive. He’s under constant pressure. He smokes. He doesn’t exercise. He’s had high blood pressure for years. He’s overweight. He had heart surgery two years ago. He’s sweating. It certainly seems like he ought to be admitted to the coronary care unit right away. But the algorithm says he shouldn’t be. All those extra factors certainly matter in the long term. The patient’s condition and diet and lifestyle put him at serious risk of developing heart disease over the next few years. It may even be that those factors play a very subtle and complex role in increasing the odds of something happening to him in the next seventy-two hours. What Goldman’s algorithm indicates, though, is that the role of these other factors is so small in determining what is happening to the man right now that an accurate diagnosis can be made without them. In fact […] that extra information is more than useless. It’s harmful. It confuses the issues. What screws up doctors when they are trying to predict heart attacks is that they take too much information into account. p. 137
There are lots of other reasons to collect a full history on patients admitted to ER for chest pains but arriving at a diagnosis of whether to route them to a costly coronary care unit is not one of them. Another reason why it is often not a good idea to collect too much patient information is because it tends to increase the confidence of the doctor in their diagnosis when there is in fact no grounds for such increased confidence (i.e., they are attending to information that, for the most part, is irrelevant and possibly misleading for the present purposes but which falsely increases their confidence in their diagnosis because they think they are getting the “full picture”). There are important lessons here to be considered when designing web applications to support decision making: giving users more information to incorporate into their decision making is not a good idea if that information is causally impotent or irrelevant for the decisions the system is designed to support. In fact, the mere presence of such information might negitively impact decision making by falsely implying the relevance of such information to the decision at hand.
