I am a clinician working everyday with patients in an Oncology ward, then I would be very pleased to find a tool to improve the management of clinical decision and the diagnostic process with my patients.
Artificial Intelligence (AI) is the new concept in town, and Institutions and Medical Schools are suffering a real challenge trying to stablish a practical role for this tool.
Searching the Ocean: AI and PubMed
When you want to look for information in Medicine, you go to PubMed. If you type (Artificial Intelligence) AND (Clinical Practice) AND (Systematic Review)”, you are going to obtain 301 results right now(October 2023), the majority published in the last three years. Many articles are dedicated to specific task in relation with several medical specialties such as Dermatology, Surgery, Ophtalmology, Rehabilitation, Pathology, but only a few have been published concerning the application to clinical diagnosis or clinical reasoning in general in the real world of a busy clinician.
Managers and Department Directors are very interested in this tool based on the capacity to get fast and complete information about “Big Data” and the relationship with several variables that could be modified to increase “business” in general. But, here, we faced one of the dangers of this important tool, the “kidnapping of the Secret” for Corporations and the Industry in general, without a direct and clear improvement in the clinical practice and in the attention to patients in particular.
Comment to a Paper
We would like to comment here a paper published by Yin J, Ngiam Ky, Teo HH in 2021 titled “Role of Artificial Intelligence Applications in Real-Life Clinical Practice: Systematic Review” (1), because the authors tried to disentangle the real application of AI to clinical medicine. They conducted a literature search between 2010 and 2020, crossing the words “artificial intelligence”, “machine learning” with other associated with clinical implications (“clinical”, “healthcare”, “medical”, etc). Initially, they searched 17.945 journal articles, but only 51 remained after applying specific inclusion criteria.
Only a few studies evaluated AI in the clinical practice, had a direct application to patients, and only 13 of them were Randomized Clinical Trials. Most of the studies were trying to validate previous retrospective data sets. Perhaps, the most clear application until now, from the point of view of diagnosis, had been the ability of AI to detect diabetic retinopathy in initial stages
The conclusion of this paper is that to much work must be done, with a more direct implication to the clinical environment, moving from theory to practice.
The intrinsic value of AI in Medicine
There is no doubt that AI is going to have a big impact in clinical medicine, but their definitive role is going to be directly linked to the vision of utility for patients and providers. We have had in the last years some examples of the implication of the computation development in the field of prognosis and therapeutic in cancer, and now we are able to detect a mutation in a cancer cell from a patient and, after searching in a big database, to select a specific drug associated with the mutation detected. Also, the implications about prognosis are clear, widening the landscape of several diseases grouped inside the same name, as we know now in breast cancer.
But, what can we say about the application of AI in the clinical diagnosis in the daily practice? To give a proper answer to this question we have to expose two important characteristics of the diagnostic process: the question of the cognitive biases and the uncertainty environment where a clinician works.
We really don´t know if the AI will have a positive impact on the reduction of cognitive biases, but probably there is not going to be an absolute reduction of them, because in a way the role of the person searching for information and the discrimination and decision after considering the available information is going to be under a human with his or her knowledge, emotions and personality.
Another situation is the question of uncertainty. This “quality” of the diagnostic process is associated to the amount of the diseases classified until now in Medicine, more than a 10.000, and also because entities different in etiology and prognosis can share the same signs and symptoms. Here, the Artificial Intelligence can have and important role, helping providers to increase the differential diagnosis, and even to disentangle particular data of every disease, not only in the way they appeared in the clinic, but also detecting small and definitive differences in the laboratory tests and even in the pathology assessment. But, as we say at the beginning, the search of data , and the application must be accessible, easy and generalized in the real clinical practice.
Artificial Intelligence, in different applications, is present already in the clinical world, incorporated into the radiology software, rules of decision in cancer, patient safety in surgery, but their role is going to be expanded very fast. I think two main areas in Medicine should be prioritized, the Emergency Department and the General Medicine Centers, two basic pillars today of the activity of the modern medicine.
Author: Dr. Lorenzo Alonso Carrión