Controversy surrounding the Hippocratic oath is not new. Taken by doctors around the world before starting to practice medicine, the oath, which is more than 2400 years old, establishes a number of ethical principles in medicine. Yet what appears in principle to be a straightforward matter is often at the heart of heated debate about the relevance and adequacy of the oath in modern times. Many parts of the ancient oath text are indeed controversial by modern standards, including beginning the oath by swearing by a number of ancient healing gods. Also, the oath does not, naturally, mention the importance of respecting patient privacy, particularly when dealing with social media, and this may help build a case for inadequacy. But what attracted my attention more than these and similar arguments is a statement made by Vinod Khosla, one of the primary founders of Sun Microsystems, during a keynote speech at the 2018 Digital Medicine Conference.
The question Mr. Khosla asked was in fact a bit shocking: Has the Hippocratic oath become hypocritical? To put the question in context, the keynote speech was about advancing medicine in the digital era, and the main thesis was that it has become almost impossible, at least statistically speaking, to do good without doing some harm. So insisting on “doing no harm”—a core part of the Hippocratic oath—consequently means losing the ability to do great good. This argument comes to mind repeatedly whenever I discuss with my research students ways to optimize the performance of a classifier used in medicine with regard to the cost of false positives vs. the cost of false negatives. It is worth mentioning here that the “do no harm” imperative, while being an integral part of the oath for many medical institutions, is actually not present in the original version of the Hippocratic Oath. Rather, this phrase appears in another work by Hippocrates called Of the Epidemics. Still, this does not change the main argument of Mr. Khosla nor the main message in this article.
Modern medicine in this digital era is poised to benefit from the availability of big data, advanced machine learning algorithms, and decreased cost of computational power and computer storage. But whether the task is early detection of a disease, diagnosing a detected lesion as malignant or benign, or timely prediction of an epileptic seizure, one of the main challenges is that there is a fundamental tradeoff between decreasing the possibility of false positive events (asserting that there is a problem when this is not the case) vs. lowering the possibility of false negative events (missing the occurrence of a true problem). This fundamental tradeoff dictates that even at the best possible performance point some harm may be done to some patients as a side effect.
Ultimately, the use of technology in medicine, even when all possible efforts are made to optimize that use, will entail some unavoidable collateral harm. While the upside could be better access to higher quality or more cost-effective health care, the downside is the potential harm done to some patients. Consequently, swearing to do no harm—whether by ancient gods or any other value-based system—while using technologies that offer a positive upside but a potential side effect may indeed be described as a hypocritical act. However, the oath text of many medical schools uses an extra word that may resolve this issue while starting a new debate: intentionally. Do no harm intentionally. Would accepting the possibility of a small percentage of false negative events qualify as intentionally doing harm?
This new debate becomes even more complicated if we consider, for instance, how a classifier system applied to a mammogram (using computer aided diagnosis output or a diagnosis generated manually by a radiologist) is optimized. The cost of a false negative is considered higher than the cost of a false positive. In the first case, an opportunity to identify a malignant tumor at an early stage is lost, which puts the patient at a huge risk associated with late detection of the tumor. In the latter case, the patient will go through the pain and cost of doing a biopsy only to find out that the lesion is benign. The cost of a false negative is set much higher than the cost of false positive—in fact, much, much higher—resulting in a state where more than 80% of biopsies turn out to be negative. This means that many more patients go through the agony and horror of the period between ordering a biopsy and actually getting the negative results (days, not hours) than what should happen in an optimal situation. Whether the reason is being legally more defensive or being overly conservative for the interest of the patient is an argument for another day. But going through such practice while claiming to adhere to the Hippocratic oath may also be a hypocritical act in itself. Just like in most other cases, it’s difficult to judge the intentions involved in making these decisions.
The probabilistic nature of the use of modern technologies in medicine such as machine learning or advanced pharmaceuticals will always keep the door open for some harm to be done as a small cost for gaining a greater good. Judging how to best manage this probabilistic nature in a complicated world of conflicting financial incentives and legal complications may affect how we eventually try to minimize the inevitable harm. This may make it even more difficult to answer the provoking question posed by Mr. Khosla as to whether or not the Hippocratic oath has now become hypocritical.