An automated computer assistant specialized in the diagnosis of transplant rejection

- FR- EN
Human kidney section magnified 400 times by a polychromatic immunofluorescence m
Human kidney section magnified 400 times by a polychromatic immunofluorescence microscope. Inserm/Oriol, Rafael

In a recent study published in the journal Nature Medicine , a research team from Université Paris Cité, Inserm and AP-HP, led by Professor Alexandre Loupy at the Centre d’expertise de la transplantation multi-organ de Paris, has created an automated computer assistant that can correct 40% of misdiagnoses of allograft rejection in humans and better guide patient management.

Rejection is the main cause of graft failure after renal transplantation. It is therefore a major public health problem in view of the worldwide shortage of organs.

The diagnosis of rejection is based on an international classification that has become considerably more complex over the last 30 years. It is now necessary for physicians to analyze and integrate complex and extremely diverse data - histological, immunological, or transcriptomic data - to make a correct diagnosis, which will guide the therapeutic management of patients.

This complexity in diagnosing rejection, initially necessary to better understand and define its type and severity, has become a daily problem for physicians, who are faced with situations where it can be difficult to make a correct diagnosis.

Therefore, in view of the increasing number of diagnostic errors that are continuously documented in the scientific literature, the international transplant societies have called on researchers worldwide to react and find a solution to simplify and make the diagnosis of rejection more reliable.

The research team had a novel idea to solve this problem.

" Precision medicine needs diagnostic tools that are reliable, robust, accurate, widely validated and demonstrate a real and measurable benefit for patients. It is also essential that these digital systems are ethical and benefit from complete transparency in their construction as well as in the interpretation and reporting of results. In our study, we were able to demonstrate that an automated computerized assistant allowed doctors to make more accurate diagnoses," says Professor Alexandre Loupy.

"Our main challenge was to establish a consortium of experts in transplantation, nephrology, pathology, data science, epidemiology, biostatistics, computer programming and artificial intelligence to develop this computer system and recruit patients from around the world to test whether it could correctly diagnose rejection. The results are clear: more than 40% of diagnoses are corrected and reclassified by the machine. This tool will allow better treatment of patients and also improve clinical trials and the development of immunosuppressive treatments.

The consortium first conducted a systematic review of the scientific literature to collect all diagnostic rules for the classification of rejection published over the last 30 years. Physicians and pathologists then worked with data scientists, developers and computer programmers to translate these diagnostic rules into a computer algorithm covering all possible rejection scenarios. They then created an easy-to-use, online automated computer assistant that instantly interprets the complex medical data provided by the physicians with the algorithm to provide patients with a diagnosis by strictly applying the rules of the international classification.

" It’s sort of a conversational agent that specializes in rejections," says Daniel Yoo, data scientist and co-first author of the study. "We have developed a system that is intelligent and very easy to use. Doctors can get a correct diagnosis for their patients with a few clicks. The computer assistant also provides them with a report of the analysis as well as a decision tree that explains the algorithm’s reasoning."

The second part of the study was to demonstrate the clinical utility of this computerized assistant, i.e. its ability to correctly identify rejection. To do this, the researchers recruited more than 4,000 kidney transplant patients from 20 European and North American transplantation reference centers. For each patient, they had the initial diagnoses of the physicians, as well as all the data necessary for the automated system to make its own diagnosis. This allowed them to compare the human with the machine and determine which one was making the most relevant diagnosis.

" One of the strengths of this computer assistant is that it can also handle large databases and improve clinical trials," emphasizes Valentin Goutaudier, nephrologist and epidemiologist, first author of the study, " The computer system allowed us to reclassify more than 40% of erroneous diagnoses of rejection among the patients we had recruited, and to make more accurate diagnoses. As a result, these results allow us to better guide patient management."

The results of this study were published on May 4, 2023 in the journal Nature Medicine and are accompanied by an editorial. This is a major advance towards precision medicine supported by automated computer systems, and the first study in any medical specialty to show that a computer assistant can help doctors make a better diagnosis.

This computerized assistant, which improves the diagnostic performance of rejection phenomena, has been validated by all international transplantation societies and will soon be used by transplantation teams worldwide to improve patient management. It will also allow standardization of rejection diagnoses in next-generation clinical trials to facilitate the development of new immunosuppressive treatments.

Transplantation is not the only medical specialty facing increasingly complex data. Other specialties, such as oncology and immunology, where varied and complex data are increasingly used, may look to automating disease classifications to improve patient management.