Biomarkers identified by machine learning could assist in diagnosing liver diseases
Key takeaways:
- Machine learning models identifying significant biomarkers were highly accurate at predicting MASLD and hepatic fibrosis.
- More research is needed assessing the models in a larger population.
Machine learning may help identify biomarkers and develop predictive models that could assist in the diagnosis of metabolic dysfunction-associated steatotic liver disease and hepatic fibrosis, researchers reported.
“The current reliance on liver biopsy for diagnosis and a lack of validated biomarkers are major factors contributing to the overall burden of metabolic dysfunction-associated steatotic liver disease (MASLD),” Mohammad Alfrad Nobel Bhuiyan, PhD, assistant professor, director of biostatistics and computational biology and co-director of medical student research in the departments of medicine, pathology and translational pathobiology, and biochemistry and molecular biology at Louisiana State University Health Sciences Center, told Healio. “This study investigates the association between biomarkers and hepatic steatosis and stiffness measurements measured by FibroScan.”
In a study published in The Journal of Clinical Endocrinology & Metabolism, researchers obtained National Health and Nutrition Examination Survey data from 12,471 adults from 2017 to March 2020 (mean age, 38.3 years; 51.7% women). Based on results from liver ultrasound transient elastography, adults with a controlled attenuation parameter (CAP) score of at least 238 dB/m were defined as having MASLD, and those with a median liver stiffness of 7 kPa or greater were considered to have hepatic fibrosis. Machine learning was performed to identify the most important variables for diagnosing MASLD and hepatic fibrosis and develop predictive models.
Of the study group, 5,266 were identified as having MASLD and 1,343 were defined as having hepatic fibrosis.
Biomarkers tied to MASLD
After propensity score matching, all variables examined in the study except for iron and bilirubin were associated with MASLD. Adults with MASLD had a higher HbA1c (6.23% vs. 5.45%; P < .001), plasma fasting glucose (124.09 mg/dL vs. 100.26 mg/dL; P < .001) and BMI (31.98 kg/m2 vs. 24.81 kg/m2; P < .001) than those without MASLD. Of lipid variables examined, the most statistically significant P value was with triglycerides, (MASLD, 123.92 mg/dL; no MASLD, 74.37 mg/dL; P < .001). Of biochemistry profile variables, the strongest association was observed in uric acid, in which adults with MASLD had a concentration of 5.7 mg/dL vs. 4.98 mg/dL for adults without MASLD (P < .001).
In logistic regression analyses, the likelihood of having MASLD increased with older age (adjusted OR = 1.024; 95% CI, 1.018-1.03), higher BMI (aOR = 1.17; 95% CI, 1.146-1.195), increased plasma fasting glucose (aOR = 1.007; 95% CI, 1-1.013), higher insulin (aOR = 1.033; 95% CI, 1.02-1.047) and higher alanine aminotransferase (aOR = 1.029; 95% CI, 1.016-1.043). MASLD odds decreased with higher blood urea nitrogen (aOR = 0.953; 95% CI, 0.934-0.972). Black adults were less likely to have MASLD than Mexican American adults (aOR = 0.466; 95% CI, 0.334-0.649).
Variables tied to hepatic fibrosis
All examined variables were different between adults with hepatic fibrosis and those without hepatic fibrosis except for blood urea nitrogen, total bilirubin and total protein. The hepatic fibrosis group had a higher HbA1c (6.34% vs. 5.8%; P < .001), plasma fasting glucose (128.12 mg/dL vs. 111.15 mg/dL; P < .001) and BMI (35.68 kg/m2 vs. 28.6 kg/m2; P < .001). In lipid profile variables, the largest difference between adults with hepatic fibrosis and adults without the disease was with HDL cholesterol (hepatic fibrosis, 48.3 mmol/L; no hepatic fibrosis, 54.18 mmol/L; P < .001). Of biochemical variables, alanine aminotransferase was the most significantly different between adults with hepatic fibrosis and those without hepatic fibrosis (27.11 U/L vs. 20.48 U/L; P < .001).
Hepatic fibrosis odds increased with older age (aOR = 1.008; 95% CI, 1.001-1.015), higher BMI (aOR = 1.106; 95% CI, 1.089-1.123), higher insulin (aOR = 1.01; 95% CI, 1.004-1.017), higher aspartate transferase (aOR = 1.017; 95% CI, 1.004-1.03), blood urea nitrogen (aOR = 1.024; 95% CI, 1.004-1.044) and gamma-glutamyl transferase (aOR = 1.006; 95% CI, 1.002-1.009). Women were less likely to have hepatic fibrosis than men (aOR = 0.651; 95% CI, 0.511-0.828).
High accuracy for both models
The logistic regression model for predicting MASLD had an overall accuracy of 79.59%, with a true positive detection rate of 72.14% and a true negative detection rate of 84.65%. The model for predicting hepatic fibrosis had an accuracy of 86.07%, with a true positive detection rate of 98.01% and a true negative detection rate of 20.92%.
“These findings indicate that assessing a variety of biomarkers, across demographic, metabolic, lipid and standard biochemistry categories, may provide valuable initial insights for diagnosing patients for MASLD and hepatic fibrosis,” Bhuiyan said.
Bhuiyan added that more research is needed using a large population size.
For more information:
Mohammad Alfrad Nobel Bhuiyan, PhD, can be reached at nobel.bhuiyan@lsuhs.edu; X (Twitter): @alf_nobel.
Perspective
Back to TopThis study by Boullion and colleagues represents a significant advancement in the early diagnosis of MASLD using machine learning to identify key biomarkers. The authors used NHANES data to link biomarkers of MASLD and liver fibrosis to create logistic regression models with hepatic steatosis and stiffness data. By employing sophisticated algorithms like Random Forest, they developed models that enhance the accuracy of MASLD and hepatic fibrosis diagnosis. The inclusion of demographic, metabolic, lipid and biochemical variables in these models highlights the complexity of MASLD and its association with metabolic syndrome.
This research holds potential for improving MASLD diagnoses by providing noninvasive and cost-effective alternatives to liver biopsy. However, further validation with liver biopsy data and integration into clinical practice are necessary. This approach could lead to earlier interventions, potentially reducing the progression of MASLD to more severe conditions like metabolic dysfunction-associated steatohepatitis (MASH) and cirrhosis. This study underscores the need for effective nondiagnostic tools to accurately identify which patients with MASLD have significant liver fibrosis. As the prevalence of MASLD continues to rise, the development of better diagnostic tools becomes increasingly important. The focus on liver fibrosis, rather than just liver fat, is important for early detection and management of these conditions.
Collapse