On Fβ -score for medical diagnostics tests of binary diseases: proposing new measures of accuracy.

Abstract

Accurate differentiation between health states - diseased or non-diseased - is essential in clinical diagnostics. Optimal cut-off points, or thresholds used to classify test results, are crucial for precise diagnoses. This work introduces the Harmonic Mean of F-score and inverse F-score (HF), a novel metric for a balanced assessment of diagnostic accuracy. HF integrates Specificity (Sp) and Negative Predictive Value (NPV) into the Negative F-score (NF_γ), ensuring a comprehensive evaluation of true negatives and negative test reliability. Prioritizing both true positives and true negatives, HF was used in optimal cut-off point estimation under binary disease classification. Simulation results revealed that the HF measure performed well, often surpassing established methods in specific settings. The HF measure and cut-off point selection criterion were applied to real-life data, showcasing its ability to provide a balanced evaluation of diagnostic accuracy. The HF measure frequently outperformed traditional metrics. The HF metric's flexibility, allowing parameter adjustments to accommodate diverse scenarios, enables researchers and clinicians to tailor its emphasis on specific aspects of diagnostic performance depending on the context.