Samyukta Mullangi, Medical Oncologist at Tennessee Oncology and VP of Clinical Strategy at OpenEvidence, shared on LinkedIn:
“Earlier this month at Association for Value-Based Cancer Care, I was on a panel with Lavi Kwiatkowsky, who derailed our conversation about ePROs in the most useful way.
Speaking about the value based care movement’s love for predictive modeling, he asked, ‘what is the point of all this prediction when AI agents can check in on a patient daily?’ Essentially, Lavi was asking whether cheap universal support weakens one rationale for prediction – resource allocation, or deciding which patient requires extra attention from a scarce resource – in this case, the nursing pool.
Another example – Lavi discounted the value of prognostic models that estimated 3 or 6 month mortality. ‘Short of a 100% certainty,’ he said, ‘I’ve never seen these influence a decision to start or stop therapy.’Here Lavi is speaking to another important concept – the value of information. A prediction has no intrinsic clinical value if it doesn’t change an action.
We may have overestimated the long-term importance of predictive models because they were developed in an era when attention was expensive. AI agents may make attention abundant, which means prediction will become narrower in focus, and move downstream. When observation is cheap enough, monitoring can substitute for prediction. A provocative thought!”
Other articles featuring Samyukta Mullangi on OncoDaily.