http://news.mit.edu/2018/doctors-rely-gut-feelings-decision-making-0720
One of the most prominent topics in medical and telemedicine conferences, journals and meetings is “the role of technology and digital health in clinical medicine.” As the Chief Medical Officer of an up-and-coming telemedicine company that utilizes sophisticated predictive software and artificial intelligence platforms, one might assume that I would side in favor of “the machines” and eschew traditional clinical dogma. Nothing could be farther from the truth, and I recently found support in a recently published study from the MIT computer science department, a study which may represent a cardinal study in the evolving discourse of the integration of technology into clinical medicine. For physicians everywhere, it validates the outspoken declaration issued with firm conviction (but perhaps lacking in hard evidence) that human insight is critical for optimal patient care. Clinicians who steadfastly believe in the sacred value and importance of “the art of medicine” finally have a leg on which to stand. Allow me to explain given recent personal events that transpired in my clinical ICU practice.
I admitted a patient to the ICU with a very simple and solvable dilemma, a mild to moderate bleed subsequent to a kidney biopsy. Although her blood counts decreased slightly (Hb concentration from 9 to 8 g/dL) and she had a high heart rate (120 bpm), an urgent CT showed that the bleed was small and had been contained in the retro-peritoneum. My surgical colleague reassured me that these bleeds are rarely fatal and that, after administering fluids and one unit of packed red blood cells to the patient, she should improve and recover quickly. When she arrived in the ICU we found that her blood counts had stabilized and had even returned to their original levels. Although her heart rate was still a little bit high, by all other objective measures she was improving. But she just didn’t look right. I hovered over her silently begging for a piece of objective data to validate my clinical sense; the science of medicine to support the art of medicine, but hard evidence of impending disaster eluded me. An ultrasound of her abdomen didn’t show worsening of the bleed, and an echocardiogram was normal, save for a small, non – intrusive pericardial effusion. Repeated blood test showed similarly stable results, and I retired to the call room for a few moments of rest. However sleep evaded me, knowing, as I did, that “something was awry.” I checked on her repeatedly and cycled blood tests as frequently as medically plausible, but everything continued to be stable. Simply put – there was nothing that I could fix. And yet I knew that we were in trouble. The morning quickly came, and I signed out with the ominous warning to “please check on her because I don’t have a good feeling.” My colleague rushed to evaluate her, and was reassured by the relative calm that he observed. But still somehow, within two hours the patient was dead of unknown causes (blood counts immediately before her arrest were stable). Sometimes, it would appear, intuition and a “clinical hunch” can provide valuable information where scoring mechanisms and algorithms fail. After all, the SOFA and APACHE scores didn’t provide early warning alerts; her bedside clinician’s gut feeling did.
I suppose that heroic stories of following one’s gut instinct in the absence – or even in the face of contradicting – objective data is ubiquitous among my physician readers. I still remember the incredulous look I received from my junior resident when I ordered an echocardiogram on a patient who I was convinced had diastolic heart failure, even though two days earlier one had been performed and interpreted as normal. I playfully punched him on the arm when the repeat test demonstrated a speckled pattern consistent with amyloidosis and the patient was whisked away to receive therapy for her multiple myeloma. In another instance, I broadcasted to all who would listen that I saved a patient who had a met criteria for a high stroke score from dangerous thrombolytic therapy, because I noticed “something off.” A completely normal MRI confirmed my suspicion of a conversion disorder and she was transferred to the psychiatry unit.
In the spirit of full transparency, I concede that my intuition has no doubt also fallen short at times and led me astray. But that is beside the point!
In “When Breath Becomes Air,” Paul Kalanithi remarks that there is a component of wisdom learned in medical training that finds expression in every decision that a physician undertakes. He is spot on. The ability to walk into a room and almost immediately discern and understand, to reach clinical decisions from a brief moment of insight, is not teachable. It is learned through experience, being alert and continuously attentive, and paying heed to patients’ subtle cues. But it is often far from objective. I don’t think that a machine would have been alerted or picked up on the strangely distant manner in which the aforementioned patient described crippling neurological symptoms (la belle indifference), but I did. As a pulmonologist I wonder how a machine could identify Kussmaul respirations or “guppy breaths,” or differentiate tachypnea due to true air hunger from simple anxiety (not that I can with complete certainty either).
Simply put: there is no replacement for clinical intuition.
At the same time, there is no replacement for high-quality integrated objective data.
The study cited above from MIT evaluated physician’s clinical decisions as compared with choices determined by an artificial intelligence algorithm. With the power of high-speed computer processing behind it, one would think that the deep learning of the computer would reign supreme. However, in many instances, it was the emotional reaction of the physician, the subjective assessment, the counter transference to the patient, which informed the clinician’s decisions and guided the doctor’s actions.
There are innumerable studies demonstrating the importance and benefit of algorithms to clinical care. Almost the entirety of Atul Gawande’s “The Checklist Manifesto” is dedicated to exploring the phenomenon. We have shown that predictive analytics can stave off disastrous situations, and that deep learning can synthesize data at a rate far superior to that of humans and with access to many more data points. For the management of diabetes, for example, artificial intelligence will soon be the standard of care, as companies such as DreamMed are proving. I was once rounding with my father, a New Jersey Best Doctor in Endocrinology, when I quipped that using artificial intelligence for insulin management was “like having you in their pocket all the time.” “No,” he replied, “it’s even better than having me.” The superiority of advanced computing is manifest in the synthesis of big data and the instantaneous performance of complicated calculations and predictions thereof. But, absent objective data, when a situation requires management guided by wisdom, intuition, and experience, there is simply no replacement for a well-trained and wise clinician.
As my fellow physicians and I wade our way through the choppy waters of digital health integration into our daily practice, it is imperative for us to appreciate the utility of the incredible computing power that we have at our fingertips. At the same time, policymakers, engineers, programmers and administrators must recognize that medicine is much more than just a science. It is a beautiful and majestic art form as well.