Related Experiment Video
Updated: Jul 8, 2026
10:27
Testing Targeted Therapies in Cancer using Structural DNA Alteration Analysis and Patient-Derived Xenografts
Published on: July 25, 2020
When survival models fail: An interpretable anomaly-detection approach for high-risk phenotypes in resected solid
Claudio Ricci1, Laura Alberici2, Vincenzo D'Ambra1
1Department of Internal Medicine and Surgery (DIMEC), Alma Mater Studiorum, University of Bologna, Bologna, Italy.
Surgery
|July 6, 2026
Summary
This study identifies a high-risk phenotype in pancreatic solid pseudopapillary tumors using anomaly detection. Key factors include lymph node ratio, hospital type, and male sex, improving risk stratification for this rare cancer.
Area of Science:
- Oncology
- Medical Informatics
- Surgical Pathology
Background:
- Solid pseudopapillary tumors of the pancreas (SPTP) are rare neoplasms with generally favorable prognoses post-resection.
- Traditional survival models are often unreliable for SPTP due to the scarcity of cancer-specific deaths.
- This research introduces an interpretable anomaly-detection framework to identify high-risk phenotypes within SPTP.
Purpose of the Study:
- To identify clinicopathologic patterns associated with a "high-risk phenotype" in patients with resected pancreatic solid pseudopapillary tumors.
- To develop a more reliable method for risk stratification in SPTP, overcoming limitations of conventional survival models.
- To leverage an interpretable anomaly-detection framework for uncovering subtle indicators of poor prognosis.
Main Methods:
- Utilized the Surveillance, Epidemiology, and End Results (SEER) registry (2000-2021) to identify patients who underwent pancreatic resection for SPTP.
- Modeled cancer-specific death as an ultrarare event using unsupervised anomaly detection algorithms: Isolation Forest and Local Outlier Factor.
- Employed Mann-Whitney U tests and SHapley Additive exPlanations (SHAP) for assessing group differences and feature relevance, respectively. External validation was performed on a systematically reviewed cohort.
Main Results:
- Patients who died from SPTP exhibited significantly higher anomaly scores compared to survivors, indicating distinct patterns.
- Key contributors to the high-risk anomaly pattern included elevated lymph node ratio, non-metropolitan hospital type, male sex, atypical resection, and Stage IV disease.
- The anomaly detection model demonstrated strong performance with an in-sample AUC of 0.892 and external validation AUC of 0.975.
Conclusions:
- An interpretable anomaly-detection approach effectively identifies high-risk phenotypes in diseases with seemingly indolent behavior.
- For resected pancreatic solid pseudopapillary tumors, an elevated lymph node ratio, treatment in non-metropolitan settings, and male sex collectively define the "high-risk phenotype."
- This framework offers a novel method for improved risk stratification and personalized management of SPTP.
