Ethics, error, and initial trials of efficacy

A normative framework for artificial intelligence as a sociotechnical system in healthcare

Artificial intelligence (AI) tools are of great interest to healthcare organizations for their potential to improve patient care, yet their translation into clinical settings remains inconsistent. One of the reasons for this gap is that good technical performance does not inevitably result in patient benefit. We advocate for a conceptual shift wherein AI tools are seen as components of an intervention ensemble. The intervention ensemble describes the constellation of practices that, together, bring about benefit to patients or health systems. Shifting from a narrow focus on the tool itself toward the intervention ensemble prioritizes a "sociotechnical" vision for translation of AI that values all components of use that support beneficial patient outcomes. The intervention ensemble approach can be used for regulation, institutional oversight, and for AI adopters to responsibly and ethically appraise, evaluate, and use AI tools.

Inefficiencies and Patient Burdens in the Development of the Targeted Cancer Drug Sorafenib: A Systematic Review

Failure in cancer drug development exacts heavy burdens on patients and research systems. To investigate inefficiencies and burdens in targeted drug development in cancer, we conducted a systematic review of all prelicensure trials for the anticancer drug, sorafenib (Bayer/Onyx Pharmaceuticals). We searched Embase and MEDLINE databases on October 14, 2014, for prelicensure clinical trials testing sorafenib against cancers. We measured risk by serious adverse event rates, benefit by objective response rates and survival, and trial success by prespecified primary endpoint attainment with acceptable toxicity. The first two clinically useful applications of sorafenib were discovered in the first 2 efficacy trials, after five drug-related deaths (4.6% of 108 total) and 93 total patient-years of involvement (2.4% of 3,928 total). Thereafter, sorafenib was tested in 26 indications and 67 drug combinations, leading to one additional licensure. Drug developers tested 5 indications in over 5 trials each, comprising 56 drug-related deaths (51.8% of 108 total) and 1,155 patient-years (29.4% of 3,928 total) of burden in unsuccessful attempts to discover utility against these malignancies. Overall, 32 Phase II trials (26% of Phase II activity) were duplicative, lacked appropriate follow-up, or were uninformative because of accrual failure, constituting 1,773 patients (15.6% of 11,355 total) participating in prelicensure sorafenib trials. The clinical utility of sorafenib was established early in development, with low burden on patients and resources. However, these early successes were followed by rapid and exhaustive testing against various malignancies and combination regimens, leading to excess patient burden. Our evaluation of sorafenib development suggests many opportunities for reducing costs and unnecessary patient burden in cancer drug development.

Ethics and epistemology of accurate prediction in clinical research

All major research ethics policies assert that the ethical review of clinical trial protocols should include a systematic assessment of risks and benefits. But despite this policy, protocols do not typically contain explicit probability statements about the likely risks or benefits involved in the proposed research. In this essay, I articulate a range of ethical and epistemic advantages that explicit forecasting would offer to the health research enterprise. I then consider how some particular confidence levels may come into conflict with the principles of ethical research.