One major discussion at NASH-TAG this year was about the inconsistency in ballooned hepatocyte identification and how this inconsistency inflates screen fail rates and possibly placebo response across studies. This conversation is part of a thorough exploration of this issue.

This conversation starts with Roger Green asking Stephen Harrison how clinical trial analysis might change for the better pending implementation of what researchers learned in this study. Stephen suggests that the area in greatest need of improvement is efficacy analysis at the back end of clinical trials. Specifically, Stephen notes that variability in placebo response rates is the single largest factor determining which Phase 2b and Phase 3 trials are deemed success or failure. He asks whether there is a way to correlate the AI assessment of ballooned hepatocyte improvement to changes in fibrosis and, separately, whether we should be looking at more tissue. Stephen notes that the present approach has led us to kill good drugs due to analytical error. Louise Campbell agrees and makes a different point, which is that over time, exhaustion leads to consistent changes in the ways experts evaluate data. Quentin takes Louise’s point as an interesting question about controlling for intra-rater variability. After some interplay, Stephen discusses how he was trained to read slides (by Dr. Brunt, lead author on this paper). He was taught first to get an overall feel for whether the slide architecture looks like NASH before scouring for ballooned hepatocytes. Quentin notes that the AI methodology (qBallooning2) incorporates some assessment of fibrosis into the identification of balloon cells.

The episode and this conversation are sponsored by HistoIndex. Conversation 14.5 is a discussion of how artificial intelligence-driven assistive technology can improve the consistency of ballooned hepatocyte scoring in advanced fibrosis and support the development of robust outcomes for fibrosis studies.