Limits to forecasting in personalized medicine: An overview

Biomedical research is generating massive amounts of information about potential prognostic factors for health and disease. However, few prognostic factors or systems are robustly validated, and still fewer have made a convincing difference in health outcomes or in prolonging life expectancy. For most diseases and outcomes, a considerable component of the prognostic variance remains unknown, and may remain so for the foreseeable future. I discuss here some of the main problems in medical forecasting that pose obstacles to personalized medicine. Their recognition may help identify solutions to improve personalized prognosis, or at least understand and cope with the component of the future that we cannot predict. Much prognostic research is stuck at generating “publishable units”, without any interest in conclusively proving their worth, let alone moving them into real life applications. Information is reported selectively and reporting is deficient. The replication record of prognostic claims is poor. Even among replicated prognostic effects, few are convincingly shown to add much information besides what is already known through more simple, traditional measurements. There are few efforts to systematize prognostic knowledge. Most prognostic effects are subtle when traced to the molecular level, where most current research operates. Many researchers, clinicians, and the public are not appropriately educated to interpret prognostic information. We still have not even agreed on what the important health outcomes are that we want to predict and intervene for, and some subjectivity may be unavoidable. Finally, without concomitant effective, affordable, and non-harmful interventions, prognosis alone is of questionable value, and wrong prognosis or a wrong interpretation thereof can be harmful. The identification of these problems also suggests a roadmap on what could be done to amend them. Solutions include a systematic approach to the design, conduct, reporting, replication, and clinical translation of prognostic research; as well as the education of researchers, clinicians, and the general public. Finally, we need to recognize that perfect individualized health forecasting is not a realistic target in the foreseeable future, and we have to live with considerable residual uncertainty.