Introduction Liver biopsy is an accurate method for determining stage and grade of non-alcoholic fatty liver disease (NAFLD). Given risks and limitations of biopsy, non-invasive liver tests (NILT) are used to risk-stratify patients. Guidelines recommend tests based on routinely collected data such as NAFLD fibrosis score (NFS), APRI, Fib-4, AST/ALT ratio and BARD. Prevalence and severity of NAFLD and metabolic syndrome vary by ethnicity, yet NILTs have been developed in largely White populations. We hypothesised that scores that include metabolic syndrome parameters (NAFLD fibrosis score and BARD) are less accurate in South Asian patients, while tests relying on biochemical data alone (Fib-4, APRI and AST/ALT ratio) or liver stiffness by transient elastography are equally accurate in different ethnicities.

Method In a retrospective case review of patients with histologically-confirmed NAFLD, NILTs were calculated using clinical data within 1 week of the biopsy and compared to Kleiner fibrosis stage. For each NILT, we calculated markers of accuracy including sensitivity, specificity and area under receiver operator curve (AUROC) for predicting advanced (F3-4 vs F0-2) or mild (F0-1 vs F2-4) fibrosis.

Results 204 patients with biopsy-proven NAFLD and complete clinical datasets were identified. The largest ethnic groups were South Asian (n=93) and White (n=84). Although there were no significant differences in NASH stage or grade, or diabetes prevalence between the two groups, South Asians were younger (44 vs 52 years, p<0.0001), and fewer were obese by ethnicity-adjusted BMI ranges (60% vs 73%, p=0.01).

Area under the receiver operator curves were highest for NFS and Fib-4 in both main ethnic groups, but were consistently lower for all NILTs in South Asian than in White patients. NILTs were less sensitive at detecting advanced fibrosis in South Asians. Relative risk of correct diagnosis in White patients compared to South Asians was 1.86 (95%CI 1.4–2.6).

Fibroscan data were available for 48 patients, the accuracy of which was high in both White and South Asian patients. We derived and separately validated specific cut-off values for South Asian patients that increased the accuracy of NFS and Fib-4 to levels comparable to those seen in White patients.

Conclusion Blood-test based NILTs are less accurate in South Asian patients, irrespective of whether the test includes metabolic parameters. This is not the case with transient elastography. Ethnicity should be considered when devising risk-stratification algorithms for NAFLD.

Disclosure of Interest None Declared