Back to 2018 Program and Abstracts

---

COMPARISON OF PREDICTION MODELS OF ADVANCED FIBROSIS IN MORBIDLY AND NON-MORBIDLY OBESE PATIENTS WITH NONALCOHOLIC FATTY LIVER DISEASE
Tavankit Singh¹, Cathrine M. Frakes¹, Rocio Lopez¹, Arthur McCullough^1
1. Cleveland Clinic Foundation, Cleveland, OH

Introduction: NAFLD has a high prevalence in morbidly obese patients. The current methods to detect advanced fibrosis in this subgroup of patients have their own drawbacks with liver biopsy being invasive, fibroscan being difficult to interpret in obese patients and the prediction models developed using demographic information and laboratory values not been validated in morbidly obese patients. We designed this study to develop a model for prediction of advanced fibrosis exclusively in morbidly obese patients and compare it with other prediction models (APRI, BARD, FIB-4 and NAFLD fibrosis scores). The newly developed prediction model was then compared to existing prediction models in a cohort of patients who were not morbidly obese.
Methods: Patients with morbid obesity (BMI >40 kg/m² or >35 kg/m² with diabetes) and NAFLD proven by biopsy between 2006-2015 were included and split into groups for model building (training cohort) and model validation (validation cohort) in a 2:1 ratio. Multivariable logistic regression analysis was performed to build the prediction model. An automated stepwise variable selection method performed on 1000 bootstrap samples was used to choose the final model. Internal model validation was measured by area under receiver operating curve (AUROC). The new score was then compared to other scores in a cohort of patients with BMI <35. p value of <0.05 was considered statistically significant.
Results: 1969 morbidly obese patients were included in the initial analysis. Both the cohorts had similar clinical profiles, laboratory values and patient demographics and 7% patients had advanced fibrosis.

After analysis, the model to predict advanced fibrosis was defined as follows:
Z=0.1218+(1.5873 if diabetic)+1.1829 if hypertensive requiring medication) – (2.0675X ln(platelet count)+ 1.8330X ln(AST)
This value was then converted into a probability distribution with a value between 0 and 100 by the following formula: [e^z/ (1+e^z)]X 100
AUROC of the obese fibrosis score (OFS) was 0.855 and 0.787 in training and validation cohorts respectively. This model was superior to APRI, FIB-4, BARD, NAFLD fibrosis scores in both the cohorts. When tested in non-morbidly obese patients, OFS and FIB-4 scores were superior to APRI, BARD and NAFLD fibrosis scores in ability to predict advanced fibrosis but there was no statistically significant difference noted between OFS and FIB-4 score.
Conclusion: In the obese population, the obesity fibrosis score is superior to all the currently existing prediction models. In non-obese population, OFS is similar to FIB-4 score in predicting presence of advanced fibrosis thus making it a more reliable prediction model for use in NAFLD patients across all grades of obesity.

Scores for Prediction of Advanced Fibrosis in obese patients: Receiver Operating Characteristics Analysis

SCORE	Training Set AUC (95% CI)	Validation Set AUC (95% CI)
APRI	0.794 (0.752, 0.836)	0.748 (0.672, 0.823)
BARD	0.666 (0.621, 0.711)	0.685 (0.620, 0.751)
FIB-4	0.790 (0.750, 0.830)	0.723 (0.647, 0.798)
NFS	0.725 (0.678, 0.773)	0.635 (0.543, 0.726)
OFS	0.855 (0.823, 0.886)	0.787 (0.726, 0.848)