I often have an argument or discussion with my spouse about facts versus opinions. Although both concepts represent information, my brain is mostly concerned with facts as a scientist and a problem solver. My spouse, having spent years in the hospitality, banking, and real estate business, is “all about the customer,” with success rooted largely in recognition of and alignment to their opinions. In my current role as CMO at ASCP where we have “concierge customer service” as one of our principles, I have adapted to listening, understanding, and operationalizing the opinions of a diverse group of individuals. However, when it comes to science, I feel that it is important to remain with facts until the point where science runs out of answers, and we have to guess about something. Where I run into trouble is that I have the opinion that people who do not understand the facts should not necessarily espouse their opinions. Someone once said an expert is someone who knows everything about a topic as well as everything that is wrong about a topic—referring to common misunderstandings that flow through our common knowledge. Opinions are like belly buttons—everyone has one, but some are cleaner than others.
When it comes to journalism, I am unfortunately a purest. I just want the facts. I was, therefore, a bit taken aback by a New York Times article discussing an innovative technology in neurosurgery for intraoperative consultations for brain tumors. My visceral negative response emerged from the surgeon discussed in the article owning part of the company that built the device for the study. My secondary concerns stem from the article discussing intraoperative consultations done by pathologists and, yet, not a single pathologist was interviewed for the article. But the biggest concern I had which made me delve deeper into the topic was the enormous number of inaccurate facts or complete untruths presented in the article. It is hard to say it wasn’t poor journalism but, as a scientist, I had to go to the source.
The scientific article in question was “Near real-time intraoperative brain tumor diagnosis using stimulated Raman histology and deep neural networks” published on Jan 6, 2020. The New York Times article, “A.I. Comes to the Operating Room” was published the same day. I read the article myself, and when it didn’t quite pass my sniff test, asked three of my colleagues who are experts in this area to also read the paper and the news report. Jane Brock, a breast pathologist who is truly an innovative thinker and dreams of the day when pathologists can study tissue immediately with confocal laser imaging and rapid molecular testing—part of her research—said of the technology, “this is a great paper and a great microscope.” She further mentioned how brain is ideal because it is homogenous, easily flattened, and amenable to artificial intelligence (AI) review because of the limited parameters needed to be evaluated for clinical decision making intraoperatively. “This is definitely the future of pathology—getting rid of frozen sections in favor of fresh tissue imaging,” Dr. Brock said. “It also means you can take tissue for research/molecular diagnostics, image it, and not waste it just by imagining it [on frozen section]. The time savings are huge.” It was exciting to hear Dr. Brock’s enthusiasm for the technology conceptually and how it could be a boon for pathologists’ ability to consult during surgery.
Dr. Rebecca Folkerth, a surgical neuropathologist for more than 20 years before she became the neuropathologist for the Office of the Medical Examiner of New York, had some concerns about the science in the paper and the maturity of the technology to replace a consultation with a pathologist today. “The questions asked [in the paper] were of necessity extremely basic, and hardly represent the real world as encountered in the operating room and neuropathology laboratory,” said Dr. Folkerth. She was concerned in the news article with the statement “Final pathologic diagnosis is increasingly driven by molecular rather than morphologic criteria.” “That is actually true for a minority of nervous system lesions, such that the ‘gold standard’ for diagnosis remains ‘macroscopic (gross) pathology’ [essentially imaging] and ‘cytologic and histoarchitectural features’ as well as clinical and laboratory findings. In other words,” concluded Dr. Folkerth, “the training and experience of a physician [pathologist] is what allows the synthesis of all data points to arrive at a comprehensive interpretation.” With regard to the immediate application to real world practice, Dr. Folkerth said, “It is telling that no ‘gliosis/treatment effect’ cases were analyzed [because] distinction of post-treatment changes from a neoplastic process in the brain is one of the most difficult encountered in clinical neuropathology.” Other cellular processes that occur in the brain were also not in the study. “A glaring omission of this paper,” Dr. Folkerth concluded, “were the consequences of the errors in ‘predicting diagnosis’. Were these ‘class A’ [leading to radical changes]?” In the article, the surgeon states that neuropathologists “hate frozen sections” which Dr. Folkerth says is simply not true. Perhaps the author should have interviewed an actual neuropathologist for this piece. Both Dr. Brock and Dr. Folkerth had concerns about the practicality of the technology where Dr. Brock felt it was currently “too expensive” (relative to current practice which provides more information across the spectrum of neuropathology) while Dr Folkerth was not clear who this process could work in smaller or decentralizes or underserved surgical settings (as the new article suggests) without a lower cost. Dr. Folkerth agreed with Dr. Brock, however, that this technology “may well represent a revolution in intraoperative decision-making and outcome.”
Lastly, I spoke with Dr. Jason Hornick, who in addition to be an internationally renowned surgical pathologist, has been in charge of quality of intraoperative consultations at one of the premiere hospitals in the US. The “frozen section [procedure] does not often take longer than 30 minutes, and is not often ‘far less accurate’ than in the study,” Dr. Hornick began. “The rates of significant discordance between intraoperative diagnosis and the final diagnosis in published surveys (for all of surgical pathology) are generally less that 1.5%.” Dr. Hornick (and I) agree that the surgeon quoted in the article is bashing the practice of pathology without any accuracy to his statements. He is not quoting facts and he is not quoting opinions. He is simply saying things that are wrong and not supported by data or years of experience. Dr. Hornick also mentioned that the cost of frozen sections are trivial compared with the cost of this technology and, as pointed out by Dr. Folkerth, the pathologists’ intraoperative consultation is much broader than just reviewing a slide for signs of tumor. Dr. Hornick said it best: “The expert consultation provided intraoperatively by the pathologist to the surgeon is not restricted to making an accurate diagnosis; pathologists are uniquely suited to integrating the patient’s clinical history, imaging, prior pathology, and surgical findings to assist the surgeon in making surgical decisions.” He concluded, “The intraoperative consultation is not a laboratory test; it is a consultative opinion by an expert physician who often understands the patient’s disease better than the surgeon.”
With these three experts’ views including their clarification of the facts and their opinions, it seems pretty clear, in my opinion, that this news article is presenting an inappropriate picture of the practice of pathology and making claims about this technology which are not, in fact, accurate or fair. What struck me, however, was the comment by the surgeon that when he was working with pathologists, he may only ask two questions because of the very long time for the frozen section but with this technology he can ask six or seven questions and get the answers really quickly. If this surgeon needed to know the answers to those question, that is, if they were mission critical to patient care, why didn’t the surgeon work with the pathology team and demand higher quality and faster turnaround time rather than investing time and money in a novel, expensive technology (from which he profits) which can only provide a fraction of the answers that a pathologist can provide? Unless the technology completely replaces a pathologist—which it doesn’t seem to be able to do—the pathologists and frozen section labs still have to be available. Thus, costs are increased, not decreased. The only parameter for increased value for the patient would appear to be time savings; however, most clinicians and pathologists would agree that the value to a patient of a two-minute AI read versus less than 20-minute intraoperative consultation is minimal compared with the cost difference.
But outside of all of this, as the entire field of pathology faces pressure from technologies—largely driven by non-pathologists—we have to realize that pathologists’ consultations, whether intraoperative or on permanents, are a tool of quality directly for the patient. If a surgeon is able to perform a surgery and discard tissue, use a system from which he/she profits, or depends solely on a computer algorithm informed by a couple of hundred cases, where is the quality check for the patient? How do we know that was tumor that was removed? Pathologists are paid to perform intraoperative consultations and ASCP works very hard to ensure that pathologists are not only fairly paid for their work but that their involvement occurs whenever it can improve quality and care for the patient. Dr. Folkerth alone has seen thousands of cases on intraoperative consultation for neuropathology and the collective knowledge of currently living neuropathologists would be millions of cases. If such knowledge were captured by an AI across the full spectrum of neuropathology, the technology would truly be remarkable. But you can’t have such integration of knowledge without involving pathologists.
-Dan Milner, MD, MSc, spent 10 years at Harvard where he taught pathology, microbiology, and infectious disease. He began working in Africa in 1997 as a medical student and has built an international reputation as an expert in cerebral malaria. In his current role as Chief Medical officer of ASCP, he leads all PEPFAR activities as well as the Partners for Cancer Diagnosis and Treatment in Africa Initiative.