Cut-off Point Selection for Biomarkers from a Personalized Medicine Perspective: A Practical Approach

Objectives: Despite its drawbacks, converting continuous biomarkers into either positive or negative is common in medicine; the reason is that a physician usually needs to choose between two options i.e. to classify someone as healthy or unhealthy. According to personalized medicine, to allow for incorporating preferences into decision-making, cut-points should be selected based on individualized costs and benefits of treatment/decision i.e. based on a cost jointly agreed upon by physician and/or patient. Traditional measures such as Youden s Index fail to do dichotomization at the patient level and modern ones such as net-benefit-fraction are not comprehensible to physicians. We went deep into these measures to help select individualized cut-points. Study Design and Setting: Dichotomization methods were outlined focusing on cost-sensitive measures including Misclassification-cost-term (MCT), generalized-Youden (GY), and net-benefit-fraction (NBF) and their consistency in selecting the optimal cut-off point was proved mathematically. Convertible terms representing preferences such as cost, harm-to-benefit ratio and threshold probability for treatment were clarified. In an experiment, sex-specific cut-points were identified to dichotomize fasting blood sugar (FBS) for pre-diabetes definition in a cohort of 1212 men and 1758 women, aged 20-60 years with 12-year follow-up. Results: The area under the curve (AUC) as a measure to show the discrimination power of FBS for incidence of Type-2 diabetes was 0.77 (95% CI: 0.73-0.81) and 0.79 (95% CI: 0.76-0.82) for males and females, respectively. The Youden s index resulted in the cut-off points of 94 for males and 95 for females. As a practice, the two threshold probabilities of %10 and %20 for developing diabetes resulted in FBS cutoffs of 89 and 95 respectively in females. For males, cutoff selection resulted in 94 at both probabilities. These cut-offs could be helpful to decide for prevention strategies regarding physician/patient s preferences.Conclusion: The probability threshold for treatment is suggested as a tangible cost index for patients/physicians and Net Benefit Fraction as an understandable and practical tool for cut-off selection. The probability threshold for treatment is an understandable index which is defined as the specific probability of disease at which the clinician prefers to make intervention (i.e. treatment or diagnostic test) for each individual patient. It could easily be specified by formal methods (such as clinical trials, clinical decision analysis) or subjective estimate (the domain expert and/or the patient opt for). The NBF is an understandable unitless measure which is defined as the fraction of the incidence rate that could be predicted and prevented appropriately regarding harm-to-benefit of treatment.

کلیدواژه ها:

Marker ، Dichotomization ، Cut-off ، Harm ، Benefit ، Decision Making