How Big Data Could Mitigate Cancer Drug Side-Effects
Updated: Apr 26
22 August 2013
As an oncologist, when I sit with patients to discuss starting a new chemotherapy regimen, their first questions are often ‘How will it make me feel?’ and ‘How did patients like me feel with this treatment?’ Regrettably, this information is generally missing from U.S. drug labels and from published reports of clinical trials — the two information sources most commonly available to people trying to understand the clinical effects of cancer drugs. -Ethan Basch
It’s no secret that the side effects of anti-cancer drugs are often considerable. The dawn of genomics-driven therapies with fewer molecular targets means fewer side effects as well… in theory. And in theory, there is no difference between theory and practice. Even the newest, most directed therapies carry adverse side effects.
About a week ago I attended a clinical trials panel discussion at the UCSD Medical School featuring Robert Abraham of Pfizer Oncology and Elizabeth Barrett-Connor, professor of epidemiology at UCSD. Dr. Barrett-Connor expressed her frustration at both the relatively modest increases in lifespan of advanced cancer patients with most new targeted therapies (many among the order of magnitude of months) and the staggering cost, some around $10,000 per month.
However, Dr. Barrett-Connor was most upset by the extreme side effects of cancer therapeutics, and was adamant about actually weighing the benefit of lifespan against the cost and benefit to the patient. The FDA is most concerned with progression-free survival (PFS) and overall survival (OS) of cancer patients when comparing the efficacy of new drugs in clinical trials. Side effects are considered in the prevalence of (very) adverse side effects, and in subtracting PFS from OS there is some discerning quality of life, but this is quite certainly not enough.
Could there be more metrics used for quality of life? Could these be incorporated into the FDA’s guidelines for approval? Would this cause a shift in how drugs are made?
Ethan Basch produced a fantastic perspective article in the New England Journal of Medicine (1) available for free.
Basch also discussed the daily use of smart devices to prompt patients for temporally and structurally consistent quick surveys of their well being during trials, and ways to incorporate this data into easily accessible, more transparent forms.
Perhaps this could be taken a step further and incorporated into data (even if supplemental) for published clinical trials? Perhaps this might make oncologists more likely to prescribe drug A vs. drug B if drug A’s side effects were more available? Would this give the company producing drug A an upper hand in their market?
Personalized medicine aims to match patients with the best possible therapies based on factors unique to the patient like their endogenous genes and genes of diseased tissue. “Best possible” thus far has been severely biased toward PFS and OS. We must do a better job including side effects into these equations, and with a slightly different approach this can be a reality. As I touched upon last month, the technology exists to make big data based clinical trials.
Regardless of success or failure, the results from cancer clinical trials would be published (…) for future analysis to find trends not otherwise comprehensible without such a macro view. Items investigated could include: 1) What genetic profiles predict response to therapy or non-response?? 2) Are there unforeseen similarities between cancer types?? 3) Do mechanisms of resistance correlate between types of cancer and types of drugs? 4) Do these insights correlate with other defined risk factors?? 5) Questions or correlations that have yet to be considered(!)
To that list I should add:
How do patient genetic profiles correlate with side effects to new drugs?
This information could be used to better tailor specific therapeutic regimens, and could allow oncologists more informed therapeutic recommendations for their patients.
The big data approach also allows accumulated data to become more valuable over time, as the sheer magnitude allows for macro view of trends and correlations not normally visible with smaller sample sizes. Data to discern adverse effects from drugs could go from sample sizes of dozens to tens of thousands. It also allows for meta-analyses for queries not yet considered as well. For example, it could identify a rare genetic group that has no side effects to drug A, and allow for prescription of more cycles of therapy, or fewer additional drugs.
1. Basch E. Toward patient-centered drug development in oncology. The New England journal of medicine. 2013 Aug 1;369(5):397-400. PubMed PMID: 23822654. Epub 2013/07/05. eng.