Validation of the Dallas Steatosis Index to Predict Nonalcoholic Fatty Liver Disease in the UK Biobank Population

Introduction:

Nonalcoholic fatty liver disease (NAFLD) is a global public health crisis that affects one-quarter of the world population¹. Preventing either cardiometabolic or liver-related complications by achieving weight loss and resolving hepatic steatosis would be the central goal of a NAFLD screening program in the primary care setting. Despite the overwhelming prevalence and the multimodal impact on health posed by NAFLD, specialty society guidelines do not recommend screening for NAFLD in the general population², partly due to the as-yet unproven cost benefit.

One proposal to improve the cost-effectiveness is to screen only the highest prevalence populations such as patients with diabetes mellitus³. Unfortunately, this strategy would be of limited societal cost-benefit since it is not expected to actually identify NAFLD early enough to prevent the development of diabetes, which represents the majority of healthcare costs associated with NAFLD⁴.

To address this limitation, and using only clinical predictors routinely available in the primary care setting, we previously derived the Dallas Steatosis Index (DSI) as a tool to predict NAFLD as inferred using MR spectroscopy in the Dallas Heart Study (DHS)⁵. Here we undertook external validation of the DSI in the United Kingdom Biobank (UKB).

Methods:

This study conformed to the 1975 Declaration of Helsinki. De-identified UKB data was analyzed with institutional review board (Washington University in St. Louis #201905162) and UKB (Access Application #50689) approval. UKB participants underwent MR liver fat quantification by proton density fat fraction (PDFF) from 2014–2015 as part of the imaging enhancement protocol⁶.

We included patients with a successful PDFF measurement and excluded those with at-risk alcohol intake or a known alternative liver disease (>21 standard drinks per week for men, >14 standard drinks per week for women, patient report of infectious or noninfectious hepatitis). The primary outcome was hepatic steatosis (PDFF ≥5.5%).

The native DSI was calculated as previously published⁵. Missing laboratory data was supplemented by multiple imputations (x5). Decision curve analysis was performed by graphing the Net Benefit to screen using a risk model’s predicted risk compared to the decision to screen all patients⁷.

Results:

The analytic cohort consisted of 4,146 participants (53% female, 97% white, median BMI 26 kg/m2, median age 63 years). The prevalence of NAFLD (PDFF ≥5.5%) was 19% (n = 783). The DSI exhibits good discrimination (C statistic = 0.826; 95% CI 0.811–0.841).

The diagnostic operating characteristics of the DSI-risk categories are similar to those in the DHS derivation. The low-risk category applied to 46% of the UKB and had 91% sensitivity to exclude NAFLD with a 96% NPV. The high-risk category applied to 20% of the UKB and had 87% specificity to diagnosis NAFLD with a 50% PPV.

Decision curve analysis shows that there is separation of the net benefit curves using the DSI compared to screening all patients. Therefore, we propose a screening program (see Figure 1) that would use the predicted DSI risk to stratify patients for subsequent NAFLD screening.

Figure 1: A proposed NAFLD screening program using the Dallas Steatosis Index in Primary Care Clinics.

NAFLD, nonalcoholic fatty liver disease. The proposed screening program would start by using routinely available clinical predictors to calculate the pre-test probability and therefore assign a pre-test NAFLD risk category (low <20% predicted, intermediate 20–49%, ≥high 50% predicted). When the risk of NAFLD is low, then there is no screening offered. At intermediate or high risk of NAFLD, formal steatosis and fibrosis assessment is offered (e.g. transient elastography with controlled attenuation parameter or Fibrosis 4 score with routine liver ultrasound). All patients at more than minimal risk for clinically relevant hepatic fibrosis would then be triaged for hepatology consultation regardless of the steatosis result. But in the presence of low-risk of hepatic fibrosis and high risk of NAFLD by the DSI, aggressive cardiometabolic risk factor modification is still appropriate due to the suboptimal sensitivity for routine liver imaging to detect lower grades of steatosis. In the presence of low-risk of hepatic fibrosis and only an intermediate risk of NAFLD, the absence of steatosis has enough negative predictive value to justify a standard of care cardiometabolic risk factor modification.

Discussion:

The Dallas Steatosis Index (DSI) uses predictors readily available in primary care to predict NAFLD⁵, and a risk calculator is publically available (dsi.wustl.edu). Here we show that the DSI is a valid tool to predict NAFLD in participants of the UKB. We also propose a DSI-based NAFLD screening program.

Compared to screening programs aimed only at those with diabetes, a DSI-based screening program would be expected to provide an avenue for aggressive weight loss intervention, resolution of steatosis and therefore normalization of the incident diabetes risk before that cost burdensome complication becomes entrenched. Furthermore, compared to programs that would recommend screening all patients for NAFLD, a DSI-based screening program would avoid the costs of screening those at the lowest risk for NAFLD and would not provide inappropriately reassuring normal results if steatosis was not suggested by routine, low-sensitivity imaging studies.

We recognize the DSI does not discriminate among NAFLD patients from those with nonalcoholic steatohepatitis (NASH). However since the primary predictor for outcomes in NAFLD is hepatic fibrosis⁸, we propose the sequential application of the DSI and then non-invasive steatosis/fibrosis assessment to those at-risk. We acknowledge that our proposed screening program will need to be explicitly studied for cost-effectiveness and patient/provider acceptance.

In summary, the DSI remains a valid tool to predict NAFLD in the United Kingdom and the United States. Using the DSI to triage primary care patients for NAFLD screening should be considered once cost-effectiveness has been rigorously demonstrated. Other future direction include using the DSI as a diagnostic surrogate for NAFLD in epidemiologic studies and additional external validation in other racial/ethnic populations.