User login
BOSTON – A machine-learning model based on standard clinical and laboratory values is able to predict nonalcoholic steatohepatitis (NASH) with a sensitivity of 72%-81%, an investigator reported at the annual meeting of the American Association for the Study of Liver Diseases.
The tool, dubbed NASHmap, could serve as an initial screening tool to reveal more potential undiagnosed patients with NASH, according to Jörn M. Schattenberg, MD, with the metabolic liver research program in the department of medicine at University Medical Centre Mainz (Germany).
While not intended to replace current scoring systems, NASHmap potentially could be applied as a clinical decision support tool within electronic medical record systems, allowing greater numbers of patients with suspected NASH to be evaluated and referred to specialists for further testing, according to Dr. Schattenberg.
“To me, this is an at-risk population,” Dr. Schattenberg said in an interview. “There’s a lot of talk of increasing numbers of end-stage liver disease, and these are the cases that are waiting to happen. I think if we identify them, we can manage them better.”
“I’m not saying they should all get drugs – I’m saying they have to be informed about their condition because they may not have a clue they have liver disease,” he continued.
The machine-learning approach described here by Dr. Schattenberg included an exploratory analysis based on 704 patients with NASH or non-NASH nonalcoholic fatty liver disease (NAFLD) in the NAFLD Adult Database from the National Institute of Diabetes, Digestive, and Kidney Diseases.
The best-performing model they identified included 14 variables. Ranked by contribution to predictive power, those variables included hemoglobin A1c, aspartate aminotransferase, alanine aminotransferase, total protein, AST/ALT ratio, body mass index, triglycerides, height, platelets, white blood cells, hematocrit, albumin, hypertension, and sex.
That 14-variable model had good performance when tested on the Optum Humedica electronic medical record database, according to Dr. Schattenberg and colleagues.
In the analysis for final evaluation, including 1,016 patients with histologically confirmed NASH, the area under the curve was 0.76, they reported.
A simplified, five-variable model including just hemoglobin A1c, AST, ALT, total protein, and AST/ALT ratio also had good performance, with an area under the curve of 0.74, they said in their report.
Dr. Schattenberg provided disclosures related to AbbVie, Novartis, MSD, Pfizer, Boehringer Ingelheim, BMS, Intercept Pharmaceuticals, Genfit, Gilead, and Echosens. Several study coauthors reported employment with Novartis and stock ownership.
BOSTON – A machine-learning model based on standard clinical and laboratory values is able to predict nonalcoholic steatohepatitis (NASH) with a sensitivity of 72%-81%, an investigator reported at the annual meeting of the American Association for the Study of Liver Diseases.
The tool, dubbed NASHmap, could serve as an initial screening tool to reveal more potential undiagnosed patients with NASH, according to Jörn M. Schattenberg, MD, with the metabolic liver research program in the department of medicine at University Medical Centre Mainz (Germany).
While not intended to replace current scoring systems, NASHmap potentially could be applied as a clinical decision support tool within electronic medical record systems, allowing greater numbers of patients with suspected NASH to be evaluated and referred to specialists for further testing, according to Dr. Schattenberg.
“To me, this is an at-risk population,” Dr. Schattenberg said in an interview. “There’s a lot of talk of increasing numbers of end-stage liver disease, and these are the cases that are waiting to happen. I think if we identify them, we can manage them better.”
“I’m not saying they should all get drugs – I’m saying they have to be informed about their condition because they may not have a clue they have liver disease,” he continued.
The machine-learning approach described here by Dr. Schattenberg included an exploratory analysis based on 704 patients with NASH or non-NASH nonalcoholic fatty liver disease (NAFLD) in the NAFLD Adult Database from the National Institute of Diabetes, Digestive, and Kidney Diseases.
The best-performing model they identified included 14 variables. Ranked by contribution to predictive power, those variables included hemoglobin A1c, aspartate aminotransferase, alanine aminotransferase, total protein, AST/ALT ratio, body mass index, triglycerides, height, platelets, white blood cells, hematocrit, albumin, hypertension, and sex.
That 14-variable model had good performance when tested on the Optum Humedica electronic medical record database, according to Dr. Schattenberg and colleagues.
In the analysis for final evaluation, including 1,016 patients with histologically confirmed NASH, the area under the curve was 0.76, they reported.
A simplified, five-variable model including just hemoglobin A1c, AST, ALT, total protein, and AST/ALT ratio also had good performance, with an area under the curve of 0.74, they said in their report.
Dr. Schattenberg provided disclosures related to AbbVie, Novartis, MSD, Pfizer, Boehringer Ingelheim, BMS, Intercept Pharmaceuticals, Genfit, Gilead, and Echosens. Several study coauthors reported employment with Novartis and stock ownership.
BOSTON – A machine-learning model based on standard clinical and laboratory values is able to predict nonalcoholic steatohepatitis (NASH) with a sensitivity of 72%-81%, an investigator reported at the annual meeting of the American Association for the Study of Liver Diseases.
The tool, dubbed NASHmap, could serve as an initial screening tool to reveal more potential undiagnosed patients with NASH, according to Jörn M. Schattenberg, MD, with the metabolic liver research program in the department of medicine at University Medical Centre Mainz (Germany).
While not intended to replace current scoring systems, NASHmap potentially could be applied as a clinical decision support tool within electronic medical record systems, allowing greater numbers of patients with suspected NASH to be evaluated and referred to specialists for further testing, according to Dr. Schattenberg.
“To me, this is an at-risk population,” Dr. Schattenberg said in an interview. “There’s a lot of talk of increasing numbers of end-stage liver disease, and these are the cases that are waiting to happen. I think if we identify them, we can manage them better.”
“I’m not saying they should all get drugs – I’m saying they have to be informed about their condition because they may not have a clue they have liver disease,” he continued.
The machine-learning approach described here by Dr. Schattenberg included an exploratory analysis based on 704 patients with NASH or non-NASH nonalcoholic fatty liver disease (NAFLD) in the NAFLD Adult Database from the National Institute of Diabetes, Digestive, and Kidney Diseases.
The best-performing model they identified included 14 variables. Ranked by contribution to predictive power, those variables included hemoglobin A1c, aspartate aminotransferase, alanine aminotransferase, total protein, AST/ALT ratio, body mass index, triglycerides, height, platelets, white blood cells, hematocrit, albumin, hypertension, and sex.
That 14-variable model had good performance when tested on the Optum Humedica electronic medical record database, according to Dr. Schattenberg and colleagues.
In the analysis for final evaluation, including 1,016 patients with histologically confirmed NASH, the area under the curve was 0.76, they reported.
A simplified, five-variable model including just hemoglobin A1c, AST, ALT, total protein, and AST/ALT ratio also had good performance, with an area under the curve of 0.74, they said in their report.
Dr. Schattenberg provided disclosures related to AbbVie, Novartis, MSD, Pfizer, Boehringer Ingelheim, BMS, Intercept Pharmaceuticals, Genfit, Gilead, and Echosens. Several study coauthors reported employment with Novartis and stock ownership.
REPORTING FROM THE LIVER MEETING 2019