When will individuals meet their personalized probabilities? A philosophical note on risk prediction
- PDF / 552,743 Bytes
- 7 Pages / 595.276 x 790.866 pts Page_size
- 7 Downloads / 148 Views
ESSAY
When will individuals meet their personalized probabilities? A philosophical note on risk prediction Olaf M. Dekkers1 · Jesse M. Mulder2 Received: 13 October 2020 / Accepted: 13 November 2020 © Springer Nature B.V. 2020
Abstract Risk prediction is one of the central goals of medicine. However, ultimate prediction–perfectly predicting whether individuals will actually get a disease–is still out of reach for virtually all conditions. One crucial assumption of ultimate personalized prediction is that individual risks in the relevant sense exist. In the present paper we argue that perfect prediction at the individual level will fail–and we will do so by providing pragmatic, epistemic, conceptual, and ontological arguments. Keywords Prediction · Individual risk · Reference class · Philosophy · Ontology
Joe Meet Joe. Joe, a 60-year-old male, with hypertension, overweight and hypercholesterolemia, wants to know his cardiovascular risk. Based on the Framingham risk score, you calculate his predicted 10-year risk for myocardial infarction to be 10%. Joe, not completely satisfied, asks whether a more precise estimation would be possible, maybe even a truly ‘personalized risk estimation’. You decide for a more personalized approach, a risk prediction model including measured circulating proteins [1], and provide Joe with a 5.8% 5-year predicted risk. Joe tries to reformulate this risk: ‘Suppose I have seventeen identical copies, one of us will get a myocardial infarction in the next 5 years.’ But realizing that his predicted risk, although more personalized now, seems still a population average, Joe further challenges the risk prediction: ‘Now I know I’m one of those seventeen identical copies. But why is it not possible to tell me which one of those I am? Why can’t you provide me with a truly individualized risk?’ Joe challenges you to come up with a risk score of either 0 or 1 (for the given time-frame). (Mind that Joe explicitly adopts a frequentist’s approach to risk. * Olaf M. Dekkers [email protected] 1
Department Clinical Epidemiology LUMC Leiden (OMD), Leiden University Medical Center, Albinusdreef 2, 2333 Leiden, ZA, The Netherlands
Department of Philosophy and Religious Studies, Utrecht University (JMM), Utrecht, The Netherlands
2
Although such an approach is often displayed in epidemiology [2], other approaches to risks are possible [3]). Risk classification for different conditions is an old enterprise; and the Framingham study for instance, with its developed risk scores, has clearly contributed to cardiovascular risk management [4]. However, Joe is right in challenging the risk prediction, for the ultimate prediction–perfectly predicting whether individuals will actually get a disease–is still out of reach for virtually all conditions (ignoring prediction over exceedingly short time frames, the only plausible exceptions are monogenetically caused diseases). The optimists may hasten to add: ‘not yet, but in the future, we will be able to have perfect predictions’, an optimism echoed in the
Data Loading...
