Clinical prediction models are built per disease, trained on many features, and do not transfer between conditions. This paper presents the Cross-Disease Biomarker Levelling Framework (CDBLF): a single formula — Meff = M × (1 − D) / (1 + D) — that predicts patient outcomes across five diseases in four organ systems using two measurements per patient and zero fitted parameters. Results across seven public datasets: AUC 0.808 in NSCLC (n=159), outperforming PD-L1 CPS (AUC 0.64), senior oncologist judgment (AUC 0.72), and the SCORPIO machine learning system trained on 9,745 patients (AUC 0.76). AUC 0.889 in melanoma (n=19). AUC 0.933 at the molecular level in basal cell carcinoma (n=11). AUC 0.853 in type 2 diabetes (n=367). AUC 0.763 in heart failure (n=299). All patient-level data is included. The formula requires no training, no machine learning, no software licence. Two measurements. One formula.
Raimo van der Klein (Sun,) studied this question.