AI tools ‘effective’ at predicting remission after Cushing’s surgery
Review: Standardized studies needed before they can be widely adopted
Written by |
Artificial intelligence (AI) tools may help predict which people with Cushing’s disease are likely to achieve remission after surgery, according to a new systematic review and meta-analysis.
To evaluate how accurately AI can forecast remission outcomes, a team of researchers from the U.S., Iran, and Australia analyzed results from five previously published studies involving nearly 2,000 patients. The studies used diverse AI-based models — tools that detect patterns in large data sets — to estimate the likelihood of remission after surgery based on patients’ clinical characteristics, hormone levels, and imaging findings.
Across the studies, the AI-based models were generally “highly effective” at predicting remission, correctly identifying most patients who improved after surgery while showing moderate ability to identify those less likely to do so, researchers reported.
“These results underscore the significant potential of AI algorithms in enhancing clinical decision-making and improving the prediction of remission in [Cushing’s disease],” researchers wrote, while noting that larger, standardized studies are still needed before such tools can be widely adopted due to differences in study methods and lack of external validation.
The study, “Prediction of remission in cushing’s disease using artificial intelligence: A systematic review and meta-analysis,” was published in Neurosurgical Review.
Surgery not always successful for bringing patients into remission
Cushing’s disease is caused by long-term exposure to high levels of the hormone cortisol due to a tumor in the brain’s pituitary gland, which results in hallmark Cushing’s symptoms.
The main first-line treatment for Cushing’s disease is a surgical procedure, called transsphenoidal surgery, to remove the tumor through the nasal cavity. While surgery is usually successful for bringing patients into remission, meaning cortisol levels return to normal and symptoms disappear, it isn’t effective for everyone.
Doctors typically assess remission by measuring cortisol levels shortly after surgery, but this approach does not always capture long-term outcomes. Some patients who do not meet early remission criteria may improve later, an occurrence known as delayed remission, while others may experience persistent or recurring disease years after treatment.
Predicting remission after surgery “is crucial for improving treatment strategies and patient outcomes,” the researchers wrote.
AI-based models performed well in predicting remission after surgery
Better approaches to predict remission could help identify patients likely to experience delayed remission and avoid unnecessary additional therapy, while flagging those at risk of persistent or recurrent disease who may need closer monitoring or earlier intervention.
As interest in using AI-based prediction tools grows, the team of researchers set out to evaluate how accurately AI-based models can predict remission after surgery and how these tools might support clinical decision-making by analyzing data from previously published studies.
They searched five medical databases for relevant studies published through December 2024. Out of 1,571 records initially identified, five studies involving 1,938 patients were included in the final analysis; four were conducted in China and one in Italy.
Across these studies, the researchers tested a wide range of AI-based models built using 17 different machine-learning algorithms, which are sets of mathematical rules that help computers learn from data to make predictions. The models were developed using combinations of clinical information, hormone measurements, and imaging data — mostly MRI scans — to estimate the likelihood of remission after transsphenoidal surgery. Most models were tested only within their original study populations rather than in independent patient groups.
Our findings demonstrate that AI-based models are highly effective in predicting remission outcomes, offering excellent sensitivity, specificity, and overall diagnostic performance.
Overall, AI-based models performed well in predicting remission after surgery. Across the included studies, the models correctly identified most patients who achieved remission, with a pooled sensitivity of 93%, and showed moderate ability to identify those less likely to improve, with a specificity of 78%.
However, the researchers noted that results varied across studies, which may partly reflect differences in the machine-learning algorithms used and may restrict “the generalizability and reproducibility of the findings.”
Overall predictive performance was measured using the area under the receiver operating characteristic curve (AUC), a standard metric that reflects how well a model distinguishes between different outcomes. Scores range from 0 to 1, with higher values indicating better performance. The combined AI models achieved an AUC of 0.91, indicating strong predictive performance.
“Our findings demonstrate that AI-based models are highly effective in predicting remission outcomes, offering excellent sensitivity, specificity, and overall diagnostic performance,” the researchers wrote. “However, further research with standardized methodologies and external validation is essential to enhance these models’ reliability and clinical applicability in diverse patient populations.”
