Chapter 92
Cancer — 19
There’s far-reaching potential for the use of digital twins to determine optimal cancer treatment at the individual level in the future. That requires setting up a huge dataset with hundreds of thousands, if not millions, of patients who have been diagnosed and treated for cancer. It would include their electronic health records, labs, medical scans, pathology whole slide images, genomic data, liquid (micro) biopsy results, treatments, and outcome. This multimodal data serves up the potential for a newly diagnosed patient, by nearest neighbor AI analysis, to have their digital twin(s) identified and so have confidently predicted what treatment will work because it has previously worked in the twin. Such a strategy would be complementary to randomized clinical trials. Of course, no twin is perfect. Every cancer has unique features at the molecular level, and every individual is unique.
But in comparison to the heterogeneity of participants in cancer trials and their variable responses to treatment, a digital twin is a great potential improvement. There is no such resource built to date, except for very small ones for a single type (by organ) of cancer. We’ll only be able to know the effectiveness of digital twins when they are fully built at scale and we can track outcomes of patients whose treatments were informed by this strategy. Barriers for building such a data resource include concerns over privacy and security of the data, along with the funding requirements for algorithmic development, validation, and data acquisition with automated longitudinal participant data inputs. Eventually, we’re likely to see the infrastructure for digital twins develop, but, unfortunately, the near-term prospects remain remote.
PREVENTION
Testing current strategies to prevent cancer are yet another dimension of the need to reboot our approach. Take the example of low-dose aspirin to prevent cardiovascular disease and colon cancer, once widely recommended, and still taken by at least twenty-nine million Americans when last assessed. In over nineteen thousand participants in a randomized trial of low-dose aspirin (100 mg) or placebo with five-year follow-up, there was a significant excess of all-cause deaths (14% more), deaths from cancer (31% higher), and major bleeding events (38% higher) in the aspirin group (fig. 6.5).
Furthermore, there was no reduction in cardiovascular disease or disability-free survival. While that trial predominantly enrolled people aged seventy and older, another even larger randomized trial in men aged fifty-five and older, and women sixty years and older, found the same lack of protection against cardiovascular disease and significant excess of hemorrhagic events. In 2022, four years after these and other trials with compelling evidence were published, the US Preventive Services Task Force stopped recommending aspirin to prevent colon cancer or cardiovascular events. In their guidelines, the hedge for cardiovascular events was that for people aged forty to fifty-nine years, “clinicians should decide for themselves.” How fuzzy is that, and how much better could we do if we precisely defined risk before testing an intervention, and recommending it at mass scale, based on just one factor—age?