Mesmerized by clinical prediction models

I am not a surgical oncologist but I came across with this study:  A deep survival interpretable radiomics model of hepatocellular carcinoma patients. Authors conclude:

In summary, novel deep radiomic analysis provides improved performance for risk assessment of HCC prognosis compared with Cox survival models and may facilitate stratification of HCC patients and personalization of their treatment strategies.

The main results are shown in a Table.

I am surprised to see such low values for the c-index. Hence it is hard to see any value in their model. Then I wondered whether prediction of survival in HCC is truly so difficult. With a quick search I managed to find this study which reported a c-index between 0.70-0.72 using a flexible parametric survival regression. Another study reported a c-index of 0.78 using a Cox model. So it seems that better c-indices are reported with more traditional approaches using different baseline variables. At times it seems we are mesmerized by available methods when old tricks would work just fine.

Leave a Reply

Your email address will not be published. Required fields are marked *