You'd be surprised at the number of cancer treatment studies that are deeply flawed:
- Positive effects may have a low confidence due to small sample size, the joke is that if you can fit the laser pointer between the lines it's considered a success. Cancer is a very tough disease and sometimes positive results are due to noise in the dataset.
- Some studies don't consider overall survival (important because you might not die of cancer but you might die sooner from a side effect like Parkinson's caused by the treatment). See mammograms and colonoscopies for treatments that look like they are almost entirely ineffective.
- Don't compare against the standard of care (its easier to show positive results if you aren't using the best treatments available)
- Allow for self selection (the treatment isn't blind or double blind and people drop out of the control group, skewing the results)
Imo he's an excellent source of the latest data driven results related to cancer and other treatments.