Validity of fatty liver disease indices in the presence of alcohol consumption

Abstract Background & aims Non-alcoholic fatty liver disease (NAFLD) and alcohol-related liver disease frequently coexist. While several blood-based indices exist for the detection of NAFLD, few studies have examined how alcohol use possibly impacts their diagnostic performance. We analysed the effects of alcohol use on the performance of indices for detecting fatty liver disease (FLD). Methods We included participants from the Cardiovascular Risk in Young Finns Study (Finnish sample) and KORA study (German sample) who underwent abdominal ultrasound or magnetic resonance imaging, respectively, for detection of FLD and had serum analyses available for calculation of Fatty Liver Index (FLI), Hepatic Steatosis Index (HSI), Lipid Accumulation Product (LAP), and Dallas Steatosis Index (DSI). Alcohol use was estimated by questionnaires as mean daily consumption and binge drinking (Finnish sample only). Predictive performance for FLD was assessed according to alcohol consumption. Results The study included 1426 (Finnish sample) and 385 (German sample) individuals, of which 234 (16%) and 168 (44%) had FLD by imaging. When alcohol consumption was <50 g/day, all indices discriminated FLD with area under the receiver operating characteristics (AUROC) of 0.82–0.88. AUROCs were 0.61–0.66 among heavy drinkers (>50 g/day). AUROCs decreased to 0.74–0.80 in the highest binge-drinking category (>2 times/week). Alcohol use correlated with FLI and LAP (r-range 0.09–0.16, p-range <.001–.02) in both samples and with DSI (r = 0.13, p < .001) in the Finnish sample. Conclusions Indices perform well and comparably for detection of FLD with alcohol consumption <50 g/day and with different binge-drinking behaviour.


Introduction
Fatty liver disease (FLD) is an increasing healthcare concern and is linked to increased mortality, severe comorbidity, decreased quality of life, and a significant economic burden on society [1,2]. Although some of the pathogenesis and risk factors are well known, the increasing prevalence of FLD suggests a need for improved prevention and interventions [3]. Alcohol, obesity, and type 2 diabetes (T2DM) are known significant etiological factors for FLD. Traditionally, FLD has been dichotomized as either non-alcoholic fatty liver disease (NAFLD) or alcohol-related liver disease (ArLD) [4,5]. However, the two conditions often coexist [6,7], and the European Association for the Study of the Liver (EASL) states a need for studies to reveal the impact of alcohol use on diagnostic tests for FLD [8].
The diagnosis of FLD can be made by imaging studies. Ultrasound (US) is the recommended first-line tool in clinical practice and magnetic resonance proton density fat fraction (MRI-PDFF) is the most accurate non-invasive method [9]. However, population screening using imaging methods is time-and resource consuming. Hence, several different index scores based on routine blood tests and biometric characteristics have been developed to detect NAFLD [4,10]. Fatty Liver Index (FLI), based on waist circumference (WC), body mass index (BMI), and serum levels of gamma-glutamyl transferase (c-GT) and triglycerides (TG), is one of the most validated indices and exhibits an area under the receiver operating characteristic (AUROC) from 0.72-0.97 for detecting NAFLD [11][12][13][14][15][16][17][18][19][20][21]. The Hepatic Steatosis Index (HSI; based on sex, BMI, alanine transaminase (ALT) and aspartate transaminase (AST), and T2DM), Lipid Accumulation Product (LAP; based on WC, TG, and sex) and Dallas Steatosis Index (DSI; based on age, sex, diabetes status, hypertension, fasting glucose, ALT, BMI, and ethnicity) are other indices that also have been used in this context [13,16,17,[20][21][22][23][24]. To date, no study has assessed the performance of FLD indices for detection of steatosis in ArLD [9]. Accordingly, we sought to investigate whether various levels of alcohol use impact the diagnostic performance of several fatty liver indices.
In the present cross-sectional study, we aimed to assess the impact of alcohol consumption on the diagnostic performance of FLI, HSI, LAP, and DSI in the detection of FLD in two population-based cohorts from Finland and Germany.

Study samples
The present study is a cross-sectional study based on two samples from The Cardiovascular Risk in Young Finns Study [25] (Finnish sample) and The Cooperative Health Research in the Region Augsburg study (KORA) [26,27] (German sample). In both samples, study design and procedures were approved by local ethics committees or institutional review boards. All participants gave written informed consent.

The cardiovascular risk in Young Finns Study
The Cardiovascular Risk in Young Finns Study is a follow-up study that started in 1980 [28]. Children and adolescents aged 3-18 years from five different centres in Finland were enrolled by random selection from the population register. The original sample included 4320 participants. Participants were followed up every 3-9 years with surveys, clinical examination, imaging studies, blood samples, or combinations thereof. In 2010-2012, abdominal US was performed on participants [29]. We included participants with available data on US, mean alcohol consumption, and c-GT in our study, which consisted of 1426 participants. US was used to define FLD, as either mild or definitive, using a validated protocol based on liver-to-kidney contrast, parenchymal brightness, deep beam attenuation and bright vessel walls, with the operator blinded for participants' characteristics [25,29,30]. Clinical data and blood samples collected at the same time were used for calculation of FLD indices.
Cooperative health research in the region Augsburg study (KORA) The KORA-MRI study is a sub-sample of the second follow up (KORA FF4 study, 2013-2014, N ¼ 2279) of the populationbased KORA S4 baseline survey, sampled in the region of Augsburg, Germany (KORA S4 study, 1999-2001, N ¼ 4261) [27]. The KORA-MRI study consisted of 400 participants who underwent whole-body MRI at 3 Tesla. Exclusion criteria were age >72 years, pregnancy/breastfeeding, history of stroke/myocardial infarction, or contraindication for gadolinium-enhanced MRI (e.g., implanted device, renal failure, allergy, claustrophobia) [26]. Hepatic fat content was measured as proton density fat fraction (MRI-PDFF) by a multi-echo Dixon technique on a volumetric interpolated breath-hold examination (VIBE) sequence. A cut-off point of mean 5.56% fat content at the level of the portal vein was used to define the presence of FLD [31]. Details of other covariates are described in Supplementary methods. Of the 400 participants, 14 were excluded due to missing MRI-PDFF-data and 1 was excluded because of revocation of participation. The final study sample consisted of 385 participants.

Fatty liver disease indices and alcohol consumption
We calculated FLI, HSI, LAP, and DSI for each participant based on variables obtained at the time of the imaging study according to the equations previously described (Supplementary Table 1) [11,22,24,32]. Based on established cut-offs, FLI was sub-grouped as <30, 30-60, and >60, HSI was sub-grouped as <30, 30-36, and >36, LAP was subgrouped as <23 or >23 for females and <30.5 or >30.5 for males, and DSI was sub-grouped as < À1.4, À1.4-0, and >0. Ethnicity was not documented, but we can safely assume that most participants were white in both Finnish and German study groups.
Binge drinking was recorded in the Finnish sample as the frequency of drinking at least six standard alcohol units per occasion and was divided into the following six sub-groups according to this frequency: Group 1, >2 times/week; Group 2, 1 time/week; Group 3, 2-3 times/month; Group 4, 1 time/ month; Group 5, 2-6 times/year; Group 6, More seldom. Binge drinking data were available for 1385 participants.

Statistical analyses
Groups were compared using the t-test or Mann-Whitney Utest for continuous variables (for normally and non-normally distributed variables, respectively), or Chi-Squared or Fisher's exact for categorical variables (Fisher's exact test was used when samples in at least one of the subgroups 5). Correlations were calculated using Spearman's correlation. Performance of indices was analysed by area under the receiver operating characteristic (AUROC) and sensitivities, specificities, positive predictive values (PPV) and negative predictive values (NPV). Binary logistic regression analysis was performed to calculate the modifying effect of alcohol use on the association between FLD indices and imagingbased FLD by including both the FLD index, alcohol use, and their interaction term in the regression model, with imagingbased FLD as the dependent variable. Statistical analyses were performed with SPSS for Windows 27.0 (SPSS, Chicago, IL). A p-value of <.05 was considered statistically significant.

Results
Of the 1426 participants (men 42.6%) in the Finnish sample, FLD was found in 234 (16%) individuals (male predominance, 67.1%); FLD was either mild (n ¼ 186) or definitive (n ¼ 48). Amongst 385 participants in the German sample, FLD was found in 168 (44%) participants. Characteristics for the study groups and by FLD status are shown in Table 1 reported binge drinking more frequently than once monthly. In this binge drinking group, prevalence of FLD was 27.5% (n ¼ 85), while prevalence of FLD was 12.7% (n ¼ 137) in those with less frequent binge drinking (p < .001).
All index scores were higher in the presence of FLD (Table 1). This same finding was consistent in all the alcohol subgroups ( Figure 2). The distributions of participants with low, intermediate, or high index score were also consistent between alcohol groups when alcohol consumption was <50 g/day (Figure 2).
Binge drinking modified the performance of indices in detection of FLD. AUROCs decreased with increasing binge drinking frequency. AUROCs ranged between 0.87-0.93 among non-binge drinkers, while AUROCs ranged between 0.70-0.80 among weekly binge drinkers (Supplementary Table 2). For comparison, AUROCs for WC to detect FLD ranged between 0.80-0.87 when alcohol use was <50 g/day; AUROCs ranged from 0.64-0.71 with alcohol use >50 g/day (Supplementary Table 3).
Performance measures for the various indices using established cut-offs are shown in Table 3. For FLI, HSI and DSI, which are divided into three-step scales, values are given for lower and higher cut-off points. When observing the low-risk groups in the Finnish sample, NPV was 97.5% (FLI), 97.6% (HSI), 96.1% (LAP), and 96.6% (DSI). Corresponding NPV results in the German sample were 94.6% (FLI), 96.0% (HSI), 88.7% (LAP), and 92.2% (DSI). In high-risk groups, PPVs in the Finnish sample were 45.6% (FLI), 41.5% (HSI), 26.6% (LAP),  In separate logistic regression analyses with FLD as the dependent variable, and the indices, alcohol use, and their interaction term as independent variables, we observed a significant interaction effect between alcohol use and FLI (p < .001), LAP (p ¼ .03), and DSI (p < .001) in the Finnish sample. In the German sample, a significant interaction effect was found for FLI (p ¼ .005), LAP (p < .001), and DSI (p < .001). For HSI, p-values were .39 and .32 (Finnish and German samples, respectively).

Discussion
FLD is highly prevalent worldwide and has serious life-threatening consequences, which emphasize the importance of increasing global awareness of the disease and in developing valid tools for clinical practitioners in early detection of persons at risk [33]. Early identification of FLD is essential, as steatosis can progress to NASH and fibrosis [34,35]. Alcohol use and binge drinking are associated with hepatic steatosis and disease progression, suggesting that significant overlap exists between NAFLD and ArLD [36,37]. The findings of the present cross-sectional study support the use of FLI, HSI, LAP, and DSI for detection of FLD when alcohol use is limited to <50 g/day. AUROC values ranged 0.82-0.90 for all indices when alcohol use was within these limits (i.e. excellent discrimination for FLD). Regarding sensitivity, specificity, and NPV, a dip could be seen among heavy drinkers; in particular, the NPV of a low index score weakened in the >50 g/ day alcohol group. Only 2.6% of the Finnish sample and up to 9.4% of the German sample reported such alcohol use.
Despite its limitations, US is recommended as a first-line tool for diagnosis of FLD in clinical practice [9]. MRI-PDFF is considered the most accurate non-invasive method in this context, but its higher price and poorer availability limit its use [9]. As a new finding, the results of the present study showed good performance of the four indices to detect both US-or MRI-PDFF-diagnosed FLD among non-drinkers and up to those who consume alcohol moderately. This is consistent with earlier findings of external validation for FLI and LAP using magnetic spectroscopy for detection of liver fat [13]. Previous studies suggest that the parameters included in the FLD-index algorithms may be influenced by alcohol consumption and binge drinking [38]. From this perspective, the present study provides novel and crucial data concerning the reliability of these indices in the context of alcohol consumption.
Even low alcohol intake in FLD is associated with increased risks for advanced liver disease [39,40], and binge drinking is related to an increased risk of liver disease independent of average alcohol intake and confounders [41,42]. In this regard, the present study provides novel evidence that binge-drinking behaviour did not impair the performance of the four FLD indices, thus increasing their validity in this context. Despite a well-established association between alcohol and liver disease, the results of the present study suggest that the correlation between alcohol and liver fat is smaller than the respective correlation of metabolic factors, such as WC and BMI. This further supports the usage of FLD indices regardless of alcohol consumption and the abandoning of strict alcohol criteria in the nomenclature and diagnosis of FLD. In addition, ArLD and FLD might have supra-additive harmful effects on the liver [43,44].
The prevalence of FLD amongst study participants varied significantly (16% and 44%, p < .001) between the two study groups. The differences between study groups may in part be explained by known etiological factors for FLD being    more prevalent in the German sample (e.g., degree of obesity, age, alcohol use, and diabetes). Furthermore, the modality used for FLD detection differed between the study groups (US vs. MRI-PDFF). The inclusion criteria in study enrolment also differed; although both original study samples included participants from the general population, further enrolment into imaging studies differed [26]. Both observed prevalence values matched previously reported values in western countries [4,45]. Sensitivity and specificity values of the four indices were generally similar between the Finnish and the German samples except for HSI, which had low specificity at the low-risk cut-off in the German sample. Among heavy drinkers, the Finnish sample had almost consistently a more profound reduction in sensitivity, but higher specificity compared to the respective values in the German sample. Along with the higher overall prevalence of FLD in the German sample, the PPVs of the four indices were also correspondingly higher in each group of alcohol consumption in the German sample. Our findings suggest that the various FLD indices perform well and are suitable for screening FLD in both high-and low-prevalence populations and with alcohol use up to 50 g/day. Comparison of correlation for commonly used obesity markers and degree of steatosis revealed strongest correlations for WC and WHR. It is noteworthy that these correlations and corresponding AUROC values were of comparable magnitude as the correlations between the indices and degree of steatosis, suggesting a possibility of using these simple body composition measures in screening for liver disease in low-resource settings. This is in accordance with a previous study that showed that WC and HC both show independent association with liver disease [46].
Interestingly, FLI has recently been suggested to mirror some of the risk of cardiovascular diseases [47,48]. This suggested association between FLD and cardiovascular disease further emphasizes the need for screening the population with FLD indices; the performances of these indices are only minimally impacted by moderate alcohol consumption.
The strengths of this study were two geographically differing cohorts with prevalence of FLD matching previously reported data, well-documented parameters, and objective measurements. Validation of the indices was successfully achieved in two different cohorts using both US and MRI-PDFF as criteria for FLD.
One of the limitations of this study was alcohol consumption data that relied on questionnaires. A more objective way to assess this parameter in future studies could be through blood tests for metabolites of alcohol use (e.g., phospatidylethanol [49]). Although the total number of participants was high compared to previous studies, both study groups were analysed individually, and sub-grouping according to alcohol use inevitably led to small counts in some of the sub-groups.
In conclusion, the four indices were valid in detecting FLD as measured by US and MRI-PDFF, suggesting that these indices may have a clinically relevant role in the primary diagnosis of this emerging health problem. Alcohol consumption up to 50 g/day and binge drinking had minimal impact on diagnostic performance of these indices. FLD indices offer an inexpensive and convenient tool to identify patients at risk when compared with imaging modalities such as US or MRI.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
The Young Finns Study has been financially supported by the following grants from the Academy