Development, external validation, and comparative assessment of a new diagnostic score for hepatic steatosis

We used data from population-based studies to determine the accuracy of the Fatty Liver Index (FLI) and the Hepatic Steatosis Index (HSI) in determining individual risk of hepatic steatosis. We also developed a new risk scoring system and validated all three indices using external data.

We used data from the Study of Health in Pomerania (SHIP; n=4,222), conducted in North-eastern Germany, to validate the existing scoring systems and to develop our own index. Data from the South German Echinococcus Multilocularis and Internal Diseases in Leutkirch (EMIL) study (n=2,177) were used as an external validation data set. Diagnostic performance was evaluated in terms of discrimination (area under the receiver operating characteristic curve (AUC)) and calibration plots. We applied boosting for generalized linear models to select relevant diagnostic separators.

The FLI accurately discriminated patients with fatty liver disease from those without (AUC=0.817) but had poor calibration, in that predicted risks differed considerably from observed risks, based on SHIP data. The FLI performed well in discrimination and calibration in the analysis of EMIL data (AUC=0.890). The HSI performed worse than the FLI in analysis of both data sets (SHIP: AUC=0.782 and EMIL: AUC=0.841), showing an extremely skewed calibration. Our newly developed risk score had a good performance in the development data set (SHIP: AUC=0.860) and also good discrimination ability in the validation data (EMIL: AUC=0.876), but it had low calibration based on the validation data set.

We compared the ability of the FLI, HSI, and our own scoring system to determine the risk of hepatic steatosis using two population-based data sets (one for the development of our own system and one for validation). In the development and independent replication data set, all three indices discriminated well between patients with and without hepatic steatosis, but the predicted risks did not match well with the observed risks, when applied to external data. Scoring systems for fatty liver disease could depend on methodological standardization of ultrasound diagnosis and laboratory measurements.