One of the great disappointments of my life is the manner in which the hallowed calling of science has become infested with the plague of ideology. This unfortunately extends to health research. The latest example is a supposed study in Science magazine claiming that a widely used risk-scoring algorithm is racially biased. (Science Article) The algorithm takes inputs and identifies patients who are at highest risk for health service need, and consequently most likely to have higher spending. The output is used for determining which patients to focus on in care management programs. The setting for the authors analysis was one, that’s right, one, hospital and they looked at all patients covered under risk-based contracts from 2013 to 2015. 70% of the patients were covered by commercial insurance and 30% by Medicare. The analysis is very twisted and you have to understand what these people did to be able to come to their conclusion. This algorithm typically uses large sets of insurance claims data as its input and its typical ranking output is actually a measure of cost. As the authors found, the algorithm makes no distinction between black and white patients, that is, it accurately ranks them regardless of race by expected cost, based on illness severity, utilization and spending and other factors found in the insurance claims data set. But now the authors claim, using other data not typically available in insurance claims, that the black patients have a higher illness burden and therefore should be ranked higher for intervention. They say the algorithm is therefore racially biased. You should be able to immediately spot the logical flaw. The algorithm uses race neutral input data and has race neutral outputs. How can an algorithm which displays no bias in the output based on race be biased? It isn’t.
The authors have identified a separate problem, which is whether the algorithm has an adequate, or the best, set of data on which to do its work. They illustrate this problem by later positing alternate algorithms with different ranking methods, that result in more or less black patients being ranked higher. So the issue isn’t the algorithm displaying any racial bias, because it clearly doesn’t, but whether the data used or the factors weighted most heavily in the output are maximized. They give us no criteria for determining which approach would be best. They seem solely focussed on which approach ranks black patients the highest, which may or may not be indicative of who most needs care management attention. The authors then go on to give us a lecture on how disparities in care arise based on race, which might explain the gap between illness burden and spending. Among other things, they ignore the possibility that white patients may get too much unneeded care and just assume that black patients aren’t getting enough care, resulting in lower spending. There is no support in the research for this notion. And there is no showing that using the algorithm as currently configured results in worse actual health care or health outcomes for black patients. Remember, these are insured patients, so cost isn’t likely to play as much of a role in access to care. And different care seeking behaviors could account for any difference in care levels. It is clear, however, from reading the study, that these authors started with a belief that the system is racist and the algorithm must therefore be racist. There is absolutely no logical support for this attitude, but of course, logic has nothing to do with it.
Science magazine and Nature, two of the most prominent research magazines, exhibit rampant ideological bias. You read their editorials and you know that they are promoting particular viewpoints, not fact-based research. This is demonstrated in their commentary on this piece of research, which they described as “automating racism”. That is an absurd conclusion. If you wonder why people feel like they can’t trust anything anymore, it is because of the blatant intrusion of ideology into every area of our lives. When it occurs in science, it is a tragedy. We need to know that research is based on neutral fact-finding, not a determination to advance a particular political view of the world.