User login
PHILADELPHIA – A novel score accurately predicted the size and presence of esophageal varices in a noninvasive manner, which may help clinicians avoid unnecessary esophagogastroduodenoscopy in patients with cirrhosis, according to a recent award-winning presentation at the annual meeting of the American College of Gastroenterology.
Although there are two validated scores for predicting esophageal varices (EV), platelet count to spleen diameter ratio and Baveno VI criteria, they have drawbacks, Tien Dong, MD, from the University of California, Los Angeles said.
“The limitations to these existing scores and criteria are both of them require imaging to do, so because of that, they oftentimes are limited in national clinical use,” Dr. Dong said in his presentation of his team’s abstract, which won the ACG Governors Award for Excellence in Clinical Research. “The other thing is that, even though it’s recommended, sometimes spleen diameter on a normal ultrasound is not reported. Furthermore, elastography – even though it’s becoming more and more common – is still not yet readily available across the country.”
Dr. Dong and his colleagues sought to identify noninvasive clinical predictors of EV to create a predictive score to identify EV to overcome these drawbacks. They gathered endoscopy data from the Olive View–UCLA and West Los Angeles Veterans Affairs Hospital to create a discovery cohort (165 patients) and tested the score on patients from Ronald Reagan UCLA Medical Center in a validation cohort (73 patients).
They used a random forest classifier machine learning algorithm “to create a forest of decision trees so that it can tell us which variables it believes to be the most predictive of our outcomes of interest,” Dr. Dong said.
The algorithm identified several variables that appeared to be predictive of EV presence, such as international normalized ratio, aminotransferase, platelet mean, hemoglobin, albumin and blood urea nitrogen less than or equal to 3, which the researchers used to create an EV-endoscopy (EV-Endo) score.
In the discovery cohort, area under the curve (AUC) for the presence of EV was 0.81, compared with an AUC of 0.82 in the validation cohort, while there was an AUC of 0.77 in the discovery cohort and an AUC of 0.79 for small/absent vs. medium/large EV. Patients with Child-Pugh class A cirrhosis had an AUC of 0.81 for EV presence and an AUC of 0.77 for EV size. The researchers then created a cutoff score of 3.48 or less and 7.70 or more for EV presence, which had a sensitivity and specificity of 93.9% and 97.5%, respectively. EV-Endo score EV size cutoff scores were also 3.48 or less and 7.70 or more, with a sensitivity of 95.8% and specificity of 95.0%.
Dr. Dong reports no conflicts of interest.
SOURCE: Hauer M et al. ACG 2018. Presentation 31.
PHILADELPHIA – A novel score accurately predicted the size and presence of esophageal varices in a noninvasive manner, which may help clinicians avoid unnecessary esophagogastroduodenoscopy in patients with cirrhosis, according to a recent award-winning presentation at the annual meeting of the American College of Gastroenterology.
Although there are two validated scores for predicting esophageal varices (EV), platelet count to spleen diameter ratio and Baveno VI criteria, they have drawbacks, Tien Dong, MD, from the University of California, Los Angeles said.
“The limitations to these existing scores and criteria are both of them require imaging to do, so because of that, they oftentimes are limited in national clinical use,” Dr. Dong said in his presentation of his team’s abstract, which won the ACG Governors Award for Excellence in Clinical Research. “The other thing is that, even though it’s recommended, sometimes spleen diameter on a normal ultrasound is not reported. Furthermore, elastography – even though it’s becoming more and more common – is still not yet readily available across the country.”
Dr. Dong and his colleagues sought to identify noninvasive clinical predictors of EV to create a predictive score to identify EV to overcome these drawbacks. They gathered endoscopy data from the Olive View–UCLA and West Los Angeles Veterans Affairs Hospital to create a discovery cohort (165 patients) and tested the score on patients from Ronald Reagan UCLA Medical Center in a validation cohort (73 patients).
They used a random forest classifier machine learning algorithm “to create a forest of decision trees so that it can tell us which variables it believes to be the most predictive of our outcomes of interest,” Dr. Dong said.
The algorithm identified several variables that appeared to be predictive of EV presence, such as international normalized ratio, aminotransferase, platelet mean, hemoglobin, albumin and blood urea nitrogen less than or equal to 3, which the researchers used to create an EV-endoscopy (EV-Endo) score.
In the discovery cohort, area under the curve (AUC) for the presence of EV was 0.81, compared with an AUC of 0.82 in the validation cohort, while there was an AUC of 0.77 in the discovery cohort and an AUC of 0.79 for small/absent vs. medium/large EV. Patients with Child-Pugh class A cirrhosis had an AUC of 0.81 for EV presence and an AUC of 0.77 for EV size. The researchers then created a cutoff score of 3.48 or less and 7.70 or more for EV presence, which had a sensitivity and specificity of 93.9% and 97.5%, respectively. EV-Endo score EV size cutoff scores were also 3.48 or less and 7.70 or more, with a sensitivity of 95.8% and specificity of 95.0%.
Dr. Dong reports no conflicts of interest.
SOURCE: Hauer M et al. ACG 2018. Presentation 31.
PHILADELPHIA – A novel score accurately predicted the size and presence of esophageal varices in a noninvasive manner, which may help clinicians avoid unnecessary esophagogastroduodenoscopy in patients with cirrhosis, according to a recent award-winning presentation at the annual meeting of the American College of Gastroenterology.
Although there are two validated scores for predicting esophageal varices (EV), platelet count to spleen diameter ratio and Baveno VI criteria, they have drawbacks, Tien Dong, MD, from the University of California, Los Angeles said.
“The limitations to these existing scores and criteria are both of them require imaging to do, so because of that, they oftentimes are limited in national clinical use,” Dr. Dong said in his presentation of his team’s abstract, which won the ACG Governors Award for Excellence in Clinical Research. “The other thing is that, even though it’s recommended, sometimes spleen diameter on a normal ultrasound is not reported. Furthermore, elastography – even though it’s becoming more and more common – is still not yet readily available across the country.”
Dr. Dong and his colleagues sought to identify noninvasive clinical predictors of EV to create a predictive score to identify EV to overcome these drawbacks. They gathered endoscopy data from the Olive View–UCLA and West Los Angeles Veterans Affairs Hospital to create a discovery cohort (165 patients) and tested the score on patients from Ronald Reagan UCLA Medical Center in a validation cohort (73 patients).
They used a random forest classifier machine learning algorithm “to create a forest of decision trees so that it can tell us which variables it believes to be the most predictive of our outcomes of interest,” Dr. Dong said.
The algorithm identified several variables that appeared to be predictive of EV presence, such as international normalized ratio, aminotransferase, platelet mean, hemoglobin, albumin and blood urea nitrogen less than or equal to 3, which the researchers used to create an EV-endoscopy (EV-Endo) score.
In the discovery cohort, area under the curve (AUC) for the presence of EV was 0.81, compared with an AUC of 0.82 in the validation cohort, while there was an AUC of 0.77 in the discovery cohort and an AUC of 0.79 for small/absent vs. medium/large EV. Patients with Child-Pugh class A cirrhosis had an AUC of 0.81 for EV presence and an AUC of 0.77 for EV size. The researchers then created a cutoff score of 3.48 or less and 7.70 or more for EV presence, which had a sensitivity and specificity of 93.9% and 97.5%, respectively. EV-Endo score EV size cutoff scores were also 3.48 or less and 7.70 or more, with a sensitivity of 95.8% and specificity of 95.0%.
Dr. Dong reports no conflicts of interest.
SOURCE: Hauer M et al. ACG 2018. Presentation 31.
REPORTING FROM ACG 2018