Comparison of simplified phases of evaluation of a drug versus artificial intelligence (AI) in healthcare
| Study phases | Drug | AI in healthcare44–46 |
| Phase 0 preclinical/discovery | Compound/drug target development Preclinical/lab studies | Proof-of-concept studies (usually on a static/retrospective dataset) Algorithm development and performance metrics evaluation |
| Phase I safety | Safety assessment Evaluating metabolism and optimal therapeutic dosage Adverse effects | Feasibility to implement into an existing workflow ‘Real world’ evaluation of algorithm performance Safety evaluation |
| Phase II efficacy and safety | Prospective efficacy and safety evaluation/clinical trial (in a larger study group, >100 patients with controls) | Prospective efficacy and safety evaluation/ clinical trial (in a larger study group, potentially multi-departmental or hospital-wide) |
| Phase III therapeutic efficacy | Efficacy and safety clinical trial (>1000 patients with controls) Medium to long term adverse event monitoring | Efficacy and safety clinical trial (potentially hospitalwide or multitrust, with controls) Medium-term to long-term performance evaluation compared with control/existing non-AI workflows |
| Phase IV safety and effectiveness | Postmarket surveillance | Postdeployment surveillance |