Summary of suggested changes and improvements in the conduct of clinical trial endoscopy
| Suggested improvement or change (in order of presentation in the paper) | Importance | Ease of implementation |
| Colonoscopy only for UC trials. | ++ | ++ |
| Require split dosing for colonoscopy preps. | +++ | +++ |
| Avoid early morning colonoscopy for trial participants. | ++ | ++ |
| Standardise bowel prep to polyethylene glycol 3350. | +++ | +++ |
| Require vendors to present videos to central readers at the same resolution as recorded (no downsampling). | +++ | +++ |
| Capture metrics for colonoscopy acquisition times (site reader) and viewing times (central reader) and set minimum standards. | ++ | ++ |
| Involve site endoscopists as readers. | ++ | + |
| Central reading training programmes by GI societies. | +++ | + |
| Better training and collaborative use of ancillary personnel. | +++ | + |
| Design new scoring systems (endoscopic outcome instruments), especially for UC, that better reflect inflammatory burden and are validated for their context of use, possibly using machine learning. | ++++ | + |
| Harmonise central reader qualification processes with clinical credentialing requirements. | ++ | ++ |
| Insist on more transparency regarding vendor central reader training programmes and harmonisation (see also above ‘Central reading training programmes by GI societies’). | +++ | ++ |
| Embrace ML to inform development of new scoring systems. | +++ | + |
| Read algorithms (aggregation of the input of more than one reader per video into the final score): choose statistical over non-statistical data aggregation methods. | ++++ | +++ |
| Create prespecified thresholds for acceptable versus unacceptable bowel preps, possible implementation with ML algorithms prior to presentation to central readers. | +++ | ++ |
GI, gastrointestinal; ML, machine learning; UC, ulcerative colitis.