The five P's, and what AI actually contributes

Part 2 — The five P's, and what AI actually contributes Leroy Hood proposed P4 Medicine — Predictive, Preventive, Personalized, Participatory — more than a decade ago. A fifth P, Precision, has since been added, because the molecular resolution it demands is now routinely achievable. These are not a menu you pick from. They depend on each other in order. Predictive. Risk can be estimated before disease exists. Whole-genome sequencing establishes what you were born with. Polygenic risk scores turn that into calibrated probabilities. Biomarkers and AI models add the acquired layer on top, and epigenetic or organ-specific aging clocks report biological rather than chronological state. A concrete example: in the PATHFINDER 2 study presented at ESMO in 2025, adding a methylation-based blood test to standard breast, cervical, colorectal and lung screening produced a more than seven-fold increase in cancers detected, at 99.6% specificity. Roughly three-quarters of the cancers it found were types with no recommended screening test at all. Preventive. Prediction without action is a horoscope. This P turns a risk estimate into something you can do — lifestyle change where risk is behavioral, closer surveillance where it's genetic, pre-emptive treatment where the trajectory is known. Personalized. Treatment matched to the individual: drug selection informed by your genotype, dosing adjusted to how you metabolize it, predictable adverse reactions avoided. Clinically the most mature P, and the least implemented — mostly because the data sits in silos no clinician has time to reconcile. Participatory. The patient as an active party rather than a recipient. Wearables, patient-reported outcomes, genuinely shared decisions. Easy to reduce to a consumer-app gesture; done properly, it supplies the continuous signal that episodic clinic visits can't generate. Precision. Diagnosis and treatment at molecular resolution. Genomics, transcriptomics, proteomics and metabolomics combined to identify the mechanism of disease in one specific person, assign them to a correct subtype rather than a broad label, and match them to a therapy aimed at that mechanism. Personalized and Precision sound like the same thing. They aren't. Personalized describes an intention — fit the treatment to the person. Precision describes a standard of evidence — the fit has to be demonstrable at the level of molecules and data. Where AI sits AI is not a sixth P. It's the layer underneath all five: risk modeling and patient stratification, multi-omics integration that condenses high-dimensional data into usable evidence, knowledge graphs and digital twins that make a patient's state queryable, and generative design of the therapy itself. Remove that layer and the five P's revert to what they were for most of the last decade: a description of a desirable future. One thing worth being clear about. None of this replaces the physician. The output of all five P's is evidence, and evidence goes to the clinician who makes the decision. The goal is better evidence, sooner, for more patients — not fewer doctors.