AI Tool Predicts Infection Risks for Stem Cell Transplant Patients with Oral Mucositis

A new tool can predict the infection risks oral mucositis poses to stem cell transplant patients

University at Buffalo researchers and collaborators have completed a series of studies that reveal how much painful mouth sores known as oral mucositis increase infection risks in stem cell transplant patients and how artificial intelligence can be used to more accurately predict those risks. Their paper, published in the journal Cancers, revealed that patients undergoing hematopoietic stem cell transplants (HSCT) for blood cancers who develop oral mucositis are at nearly four times the risk of developing a severe infection compared to those without the condition. This is the first time that risk has been quantified. The paper is the most comprehensive synthesis to date of recent findings on individual risk factors for oral mucositis, whether the transplant involves a patient's own stem cells or donor cells. Risk factors are identified as specific drugs, such as methotrexate, high-dose chemotherapy, female gender, younger age, kidney issues, and reactivation of the herpes simplex virus. A significant portal for infections "Oral mucositis is not simply a source of discomfort; it serves as a significant portal for infections in immunocompromised patients," says Satheeshkumar Poolakkad Sankaran, DDS, corresponding author on the paper and research scientist in the Division of Hematology/Oncology in the Department of Medicine at the Jacobs School of Medicine and Biomedical Sciences at UB. "All my patients with oral mucositis experience poorer outcomes, adversely impacting their quality of life." For this reason, he says, screening every cancer patient for oral mucositis risk ahead of time makes sense because the condition is so common -- it occurs in up to 80% of HSCT patients. "Knowing risk factors can help doctors spot patients at high risk early," he says. "This can allow for preventive steps, like oral hygiene or cryotherapy, where extremely cold temperatures are used to reduce inflammation, thus improving outcomes and quality of life." To better assess who is at risk, Poolakkad Sankaran and colleagues published a paper in July in Support Cancer Care describing a nomogram tool they developed to predict which patients are more likely to develop oral mucositis. A nomogram is a statistical instrument that is used to model relationships among variables. The researchers used age, gender, race, total body irradiation, and fluid/electrolyte disorders to estimate risks of developing ulcerative mucositis, a severe form of oral mucositis. "This nomogram simplifies complex data for clinicians, enabling targeted oral care before HSCT," explains Joel Epstein, DMD, co-author at the City of Hope Comprehensive Cancer Center. Explainable AI better predicts adverse events At the Multinational Association of Supportive Care in Cancer 2025 meeting in June, Poolakkad Sankaran presented additional related findings on a nomogram-based model that can better predict adverse events. He explains that this model was evaluated against a new framework that uses explainable AI, employing machine learning algorithms to assess intricate clinical and demographic facts. Explainable AI is designed to provide the rationale behind the output of an AI system. "The AI model exhibited enhanced predictive accuracy, recognizing patterns linked to toxicities that conventional nomograms failed to detect," he adds. "By synthesizing demographic and clinical data, the system can predict adverse events, facilitating individualized therapy modifications to reduce toxicities." Sankaran is validating the model with other cancer adverse events such as immune-related adverse events in a larger cohort, working with Roberto Pili, MD, a co-author and associate dean for cancer research and integrative oncology in the Jacobs School. Their ultimate goal is widespread clinical adoption of the model in assessing cancer patients. "These interconnected studies underscore the oral-systemic connection in cancer therapy, urging multidisciplinary collaboration among oncologists, dentists, and AI specialists," Poolakkad Sankaran says. "As cancer management such as HSCT and immunotherapy grows -- particularly for older patients -- these tools promise reduced complications, shorter hospitalizations, and lower costs."

Researchers reveal the infection risks oral mucositis poses to stem cell transplant patients, introduce AI tool to predict them

BUFFALO, N.Y. - University at Buffalo researchers and collaborators have completed a series of studies that reveal how much painful mouth sores known as oral mucositis increase infection risks in stem cell transplant patients and how artificial intelligence can be used to more accurately predict those risks. Their paper, published Aug. 14 in the journal Cancers, revealed that patients undergoing hematopoietic stem cell transplants (HSCT) for blood cancers who develop oral mucositis are at nearly four times the risk of developing a severe infection compared to those without the condition. This is the first time that risk has been quantified. The paper is the most comprehensive synthesis to date of recent findings on individual risk factors for oral mucositis, whether the transplant involves a patient's own stem cells or donor cells. Risk factors are identified as specific drugs, such as methotrexate, high-dose chemotherapy, female gender, younger age, kidney issues, and reactivation of the herpes simplex virus. A significant portal for infections "Oral mucositis is not simply a source of discomfort; it serves as a significant portal for infections in immunocompromised patients," says Satheeshkumar Poolakkad Sankaran, DDS, corresponding author on the paper and research scientist in the Division of Hematology/Oncology in the Department of Medicine at the Jacobs School of Medicine and Biomedical Sciences at UB. "All my patients with oral mucositis experience poorer outcomes, adversely impacting their quality of life." For this reason, he says, screening every cancer patient for oral mucositis risk ahead of time makes sense because the condition is so common -- it occurs in up to 80% of HSCT patients. "Knowing risk factors can help doctors spot patients at high risk early," he says. "This can allow for preventive steps, like oral hygiene or cryotherapy, where extremely cold temperatures are used to reduce inflammation, thus improving outcomes and quality of life." To better assess who is at risk, Poolakkad Sankaran and colleagues published a paper in July in Support Cancer Care describing a nomogram tool they developed to predict which patients are more likely to develop oral mucositis. A nomogram is a statistical instrument that is used to model relationships among variables. The researchers used age, gender, race, total body irradiation, and fluid/electrolyte disorders to estimate risks of developing ulcerative mucositis, a severe form of oral mucositis. "This nomogram simplifies complex data for clinicians, enabling targeted oral care before HSCT," explains Joel Epstein, DMD, co-author at the City of Hope Comprehensive Cancer Center. Explainable AI better predicts adverse events At the Multinational Association of Supportive Care in Cancer 2025 meeting in June, Poolakkad Sankaran presented additional related findings on a nomogram-based model that can better predict adverse events. He explains that this model was evaluated against a new framework that uses explainable AI, employing machine learning algorithms to assess intricate clinical and demographic facts. Explainable AI is designed to provide the rationale behind the output of an AI system. "The AI model exhibited enhanced predictive accuracy, recognizing patterns linked to toxicities that conventional nomograms failed to detect," he adds. "By synthesizing demographic and clinical data, the system can predict adverse events, facilitating individualized therapy modifications to reduce toxicities." Poolakkad Sankaran is validating the model with other cancer adverse events such as immune-related adverse events in a larger cohort, working with Roberto Pili, MD, a co-author and associate dean for cancer research and integrative oncology in the Jacobs School. Their ultimate goal is widespread clinical adoption of the model in assessing cancer patients. "These interconnected studies underscore the oral-systemic connection in cancer therapy, urging multidisciplinary collaboration among oncologists, dentists, and AI specialists," Poolakkad Sankaran says. "As cancer management such as HSCT and immunotherapy grows -- particularly for older patients -- these tools promise reduced complications, shorter hospitalizations, and lower costs." Other UB co-authors are Parakshit Padhi, MD, Lauryn Rudin, MD, and Deepshika Kewlani, an MD candidate, all of the Jacobs School; and Ridham Varsani of the UB School of Dental Medicine. Other co-authors are from Boston Medical Center, the University of Connecticut Health Center, and the City of Hope National Cancer Center. The authors are part of the Worldwide Extension of Buffalo's Research Innovation Group in Hem/Onc Talent (WE-BRIGHT) Network, based in the Jacobs School, which is inspiring transformation through mentoring and instruction in oncology to produce next-generation cancer care scientists. Some of the work was partially funded by the Kaleida Health Foundation.