User login
Endometrial cancer is the most frequently occurring uterine cancer. Early-stage patients have about a 95% 5-year survival, but distant recurrence is associated with very poor survival, according to Sarah Fremond, MSc, an author of the research (Abstract 5695), which she presented at the annual meeting of the American Association for Cancer Research.
“Most patients with endometrial cancer have a good prognosis and would not require any adjuvant treatment, but there is a proportion that will develop distant recurrence. For those you want to recommend adjuvant chemotherapy, because currently in the adjuvant setting, that’s the only treatment that is known to lower the risk of distant recurrence. But that also causes morbidity. Therefore, our clinical question was how to accurately identify patients at low and high risk of distant recurrence to reduce under- and overtreatment,” said Ms. Fremond, a PhD candidate at Leiden (the Netherlands) University Medical Center.
Pathologists can attempt such predictions, but Ms. Fremond noted that there are challenges. “There is a lot of variability between pathologists, and we don’t even use the entire visual information present in the H&E [hematoxylin and eosin] tumor slide. When it comes to molecular testing, it is hampered by cost, turnaround time, and sometimes interpretation. It’s quite complex to combine those data to specifically target risk of distant recurrence for patients with endometrial cancer.”
In her presentation, Ms. Fremond described how she and her colleagues used digitized histopathological slides in their research. She and her coauthors developed the AI model as part of a collaboration that included the AIRMEC Consortium, Leiden University Medical Center, the TransPORTEC Consortium, and the University of Zürich.
The researchers used long-term follow-up data from 1,408 patients drawn from three clinical cohorts and participants in the PORTEC-1, PORTEC-2, and PORTEC-3 studies, which tested radiotherapy and adjuvant therapy outcomes in endometrial cancer. Patients who had received prior adjuvant chemotherapy were excluded. In the model development phase, the system analyzed a single representative histopathological slide image from each patient and compared it with the known time to distant recurrence to identify patterns.
Once the system had been trained, the researchers applied it to a novel group of 353 patients. It ranked 89 patients as having a low risk of recurrence, 175 at intermediate risk, and 89 at high risk of recurrence. The system performed well: 3.37% of low-risk patients experienced a distant recurrence, as did 15.43% of the intermediate-risk group and 36% of the high-risk group.
The researchers also employed an external validation group with 152 patients and three slides per patient, with a 2.8-year follow-up. The model performed with a C index of 0.805 (±0.0136) when a random slide was selected for each patient, and the median predicted risk score per patient was associated with differences in distant recurrence-free survival between the three risk groups with a C index of 0.816 (P < .0001).
Questions about research and their answers
Session moderator Kristin Swanson, PhD, asked if the AI could be used with the pathology slide’s visible features to learn more about the underlying biology and pathophysiology of tumors.
“Overlying the HECTOR on to the tissue seems like a logical opportunity to go and then explore the biology and what’s attributed as a high-risk region,” said Dr. Swanson, who is director of the Mathematical NeuroOncology Lab and codirector of the Precision NeuroTherapeutics Innovation Program at Mayo Clinic Arizona, Phoenix.
Ms. Fremond agreed that the AI has the potential to be used that way.”
During the Q&A, an audience member asked how likely the model is to perform in populations that differ significantly from the populations used in her study.
Ms. Fremond responded that the populations used to develop and test the models were in or close to the Netherlands, and little information was available regarding patient ethnicity. “There is a possibility that perhaps we would have a different performance on a population that includes more minorities. That needs to be checked,” said Ms. Fremond.
The study is limited by its retrospective nature.
Ms. Fremond and Dr. Swanson have no relevant financial disclosures.
Endometrial cancer is the most frequently occurring uterine cancer. Early-stage patients have about a 95% 5-year survival, but distant recurrence is associated with very poor survival, according to Sarah Fremond, MSc, an author of the research (Abstract 5695), which she presented at the annual meeting of the American Association for Cancer Research.
“Most patients with endometrial cancer have a good prognosis and would not require any adjuvant treatment, but there is a proportion that will develop distant recurrence. For those you want to recommend adjuvant chemotherapy, because currently in the adjuvant setting, that’s the only treatment that is known to lower the risk of distant recurrence. But that also causes morbidity. Therefore, our clinical question was how to accurately identify patients at low and high risk of distant recurrence to reduce under- and overtreatment,” said Ms. Fremond, a PhD candidate at Leiden (the Netherlands) University Medical Center.
Pathologists can attempt such predictions, but Ms. Fremond noted that there are challenges. “There is a lot of variability between pathologists, and we don’t even use the entire visual information present in the H&E [hematoxylin and eosin] tumor slide. When it comes to molecular testing, it is hampered by cost, turnaround time, and sometimes interpretation. It’s quite complex to combine those data to specifically target risk of distant recurrence for patients with endometrial cancer.”
In her presentation, Ms. Fremond described how she and her colleagues used digitized histopathological slides in their research. She and her coauthors developed the AI model as part of a collaboration that included the AIRMEC Consortium, Leiden University Medical Center, the TransPORTEC Consortium, and the University of Zürich.
The researchers used long-term follow-up data from 1,408 patients drawn from three clinical cohorts and participants in the PORTEC-1, PORTEC-2, and PORTEC-3 studies, which tested radiotherapy and adjuvant therapy outcomes in endometrial cancer. Patients who had received prior adjuvant chemotherapy were excluded. In the model development phase, the system analyzed a single representative histopathological slide image from each patient and compared it with the known time to distant recurrence to identify patterns.
Once the system had been trained, the researchers applied it to a novel group of 353 patients. It ranked 89 patients as having a low risk of recurrence, 175 at intermediate risk, and 89 at high risk of recurrence. The system performed well: 3.37% of low-risk patients experienced a distant recurrence, as did 15.43% of the intermediate-risk group and 36% of the high-risk group.
The researchers also employed an external validation group with 152 patients and three slides per patient, with a 2.8-year follow-up. The model performed with a C index of 0.805 (±0.0136) when a random slide was selected for each patient, and the median predicted risk score per patient was associated with differences in distant recurrence-free survival between the three risk groups with a C index of 0.816 (P < .0001).
Questions about research and their answers
Session moderator Kristin Swanson, PhD, asked if the AI could be used with the pathology slide’s visible features to learn more about the underlying biology and pathophysiology of tumors.
“Overlying the HECTOR on to the tissue seems like a logical opportunity to go and then explore the biology and what’s attributed as a high-risk region,” said Dr. Swanson, who is director of the Mathematical NeuroOncology Lab and codirector of the Precision NeuroTherapeutics Innovation Program at Mayo Clinic Arizona, Phoenix.
Ms. Fremond agreed that the AI has the potential to be used that way.”
During the Q&A, an audience member asked how likely the model is to perform in populations that differ significantly from the populations used in her study.
Ms. Fremond responded that the populations used to develop and test the models were in or close to the Netherlands, and little information was available regarding patient ethnicity. “There is a possibility that perhaps we would have a different performance on a population that includes more minorities. That needs to be checked,” said Ms. Fremond.
The study is limited by its retrospective nature.
Ms. Fremond and Dr. Swanson have no relevant financial disclosures.
Endometrial cancer is the most frequently occurring uterine cancer. Early-stage patients have about a 95% 5-year survival, but distant recurrence is associated with very poor survival, according to Sarah Fremond, MSc, an author of the research (Abstract 5695), which she presented at the annual meeting of the American Association for Cancer Research.
“Most patients with endometrial cancer have a good prognosis and would not require any adjuvant treatment, but there is a proportion that will develop distant recurrence. For those you want to recommend adjuvant chemotherapy, because currently in the adjuvant setting, that’s the only treatment that is known to lower the risk of distant recurrence. But that also causes morbidity. Therefore, our clinical question was how to accurately identify patients at low and high risk of distant recurrence to reduce under- and overtreatment,” said Ms. Fremond, a PhD candidate at Leiden (the Netherlands) University Medical Center.
Pathologists can attempt such predictions, but Ms. Fremond noted that there are challenges. “There is a lot of variability between pathologists, and we don’t even use the entire visual information present in the H&E [hematoxylin and eosin] tumor slide. When it comes to molecular testing, it is hampered by cost, turnaround time, and sometimes interpretation. It’s quite complex to combine those data to specifically target risk of distant recurrence for patients with endometrial cancer.”
In her presentation, Ms. Fremond described how she and her colleagues used digitized histopathological slides in their research. She and her coauthors developed the AI model as part of a collaboration that included the AIRMEC Consortium, Leiden University Medical Center, the TransPORTEC Consortium, and the University of Zürich.
The researchers used long-term follow-up data from 1,408 patients drawn from three clinical cohorts and participants in the PORTEC-1, PORTEC-2, and PORTEC-3 studies, which tested radiotherapy and adjuvant therapy outcomes in endometrial cancer. Patients who had received prior adjuvant chemotherapy were excluded. In the model development phase, the system analyzed a single representative histopathological slide image from each patient and compared it with the known time to distant recurrence to identify patterns.
Once the system had been trained, the researchers applied it to a novel group of 353 patients. It ranked 89 patients as having a low risk of recurrence, 175 at intermediate risk, and 89 at high risk of recurrence. The system performed well: 3.37% of low-risk patients experienced a distant recurrence, as did 15.43% of the intermediate-risk group and 36% of the high-risk group.
The researchers also employed an external validation group with 152 patients and three slides per patient, with a 2.8-year follow-up. The model performed with a C index of 0.805 (±0.0136) when a random slide was selected for each patient, and the median predicted risk score per patient was associated with differences in distant recurrence-free survival between the three risk groups with a C index of 0.816 (P < .0001).
Questions about research and their answers
Session moderator Kristin Swanson, PhD, asked if the AI could be used with the pathology slide’s visible features to learn more about the underlying biology and pathophysiology of tumors.
“Overlying the HECTOR on to the tissue seems like a logical opportunity to go and then explore the biology and what’s attributed as a high-risk region,” said Dr. Swanson, who is director of the Mathematical NeuroOncology Lab and codirector of the Precision NeuroTherapeutics Innovation Program at Mayo Clinic Arizona, Phoenix.
Ms. Fremond agreed that the AI has the potential to be used that way.”
During the Q&A, an audience member asked how likely the model is to perform in populations that differ significantly from the populations used in her study.
Ms. Fremond responded that the populations used to develop and test the models were in or close to the Netherlands, and little information was available regarding patient ethnicity. “There is a possibility that perhaps we would have a different performance on a population that includes more minorities. That needs to be checked,” said Ms. Fremond.
The study is limited by its retrospective nature.
Ms. Fremond and Dr. Swanson have no relevant financial disclosures.
FROM AACR 2023