Validation of the Fatty Liver Index for identifying non‐alcoholic fatty liver disease in a Kenyan population

Fatty Liver Index (FLI) is a simple clinical scoring system estimating non‐alcoholic fatty liver disease (NAFLD). It is validated in European‐descent and Asian populations, but not in sub‐Saharan Africans. The aim of this study is to evaluate the validity of the FLI for predicting NAFLD in a population from Kenya.


INTRODUCTION
Non-alcoholic fatty liver disease (NAFLD) is emerging as one of the most prevalent causes of liver diseases worldwide [1,2].Paralleling the obesity epidemic, NAFLD is expected to become the leading cause of end-stage liver disease in adults and children in the coming decades [3].The global prevalence of NAFLD is currently estimated to be 24% [2], but scant data available from sub-Saharan African countries suggest that NAFLD prevalence might be lower in sub-Saharan Africa compared to the Americas, Europe or the Asia and Pacific regions [1].In Nigeria, NAFLD prevalence is estimated to be 1.2%-4.5% in individuals without diabetes mellitus (DM) and 9.5% to 16.7% in individuals with DM [4,5].Using ultrasonography scanning, we have recently shown that in a Kenyan population of adult men and women (743 individuals, 59.1% women, mean age 38.0 years, and 3.2% with DM), NAFLD prevalence was 15.9%, with the majority of participants having mild hepatic steatosis (94.1%) [6].
NAFLD has a large spectrum of histological manifestations, which include different levels of steatosis (5% and more) [7], hepatic inflammation (i.e., non-alcoholic steatohepatitis), fibrosis, cirrhosis and sometime associated hepatocellular carcinoma, all being in the absence of an excessive alcohol intake (<20 g and 30 g a day for women and men, respectively) [7].Recent evidence suggests that NAFLD would play a major role in the onset and progression of cardiovascular risk factors and diseases [8,9].Scientific and clinical attention is now directed towards an early detection of NAFLD, in order to prevent and manage cardiometabolic diseases.
Liver biopsy is recognised as the gold standard to diagnose and characterise NAFLD, which cannot be routinely implemented due to well-known limitations including cost, invasiveness and risk of complications [10,11].In high income settings, non-invasive imagery techniques such as ultrasound scanning, ultrasound elastography, computed tomography and magnetic resonance elastography and spectroscopy are increasingly used as alternative screening tools for NAFLD [11][12][13][14].Magnetic resonance imaging (MRI) and proton magnetic resonance spectroscopy (MRS) are increasingly accepted as a non-invasive alternative to liver biopsy, but are extremely time consuming and costly [15,16], whereas computer tomography exposes participants to radiation which limits its repetitive use to monitor the disease [16].Ultrasonography has been previously suggested as a safe, accessible and inexpensive method to detect hepatic steatosis and have been validated in various populations of different ages and body compositions [13,[17][18][19].Comparing ultrasound to MRS in 50 adults of various ethnic origins aged between 21 and 69 years old (14% women, 18% of them being overweight and 54% being obese), Mehta et al. have described ultrasonography as a useful and valid screening tool for NAFLD, showing an overall sensitivity of 85.4% and a specificity of 100% [17].However, the reported accuracy of ultrasonography to detect NAFLD carries important inter-observer variability [17], and has not always been consistent in the literature [20,21], nor tested specifically for SSA populations.
Over the past decades, clinical indices (including Fatty Liver Index [FLI], NAFLD Fibrosis score or FIB-4) have been proposed as alternative tools to detect NAFLD [12], but so far, they have only been validated in a few specific populations [12,22].In many sub-Saharan African countries with limited clinical settings, the external validity of clinical indices to detect NAFLD remains uncertain, as they have never been tested in sub-Saharan African populations.
FLI is a non-invasive scoring system based on anthropometric measurements (body mass index [BMI, kg/m 2 ] and waist circumference [WC]), triglycerides (TG) and gammaglutamyl-transferase (GGT) [23].Developed in 2006 by Bedogi et al. based on data from the Dionysos Nutrition & Liver Study (496 participants with and without suspected liver diseases), it is now recognised as an easy-to-use, economical and practicable clinical tool to diagnose NAFLD.FLI has been validated in larger epidemiological studies, with imaging studies and liver biopsy [12,22,24], but its validity has never been tested in a sub-Saharan African population.
The aim of the present study was to validate the accuracy of FLI and to establish its optimal cut-off points in a Kenyan population of men and women, by using liver ultrasound scanning as a diagnosis test [13] for NAFLD.

METHODOLOGY
A community-based cross-sectional study was conducted from August 2005 to January 2006 in rural and urban communities from Kenya.A total of 1473 individuals were invited to participate.All interested individuals were then registered for participation and gave their informed consent prior to their inclusion, as approved by the National Ethical Review Committee in Kenya and the National Committee on Biomedical Research Ethics in Denmark.Inclusion criteria were: ≥18 years of age and self-reporting as or culturally related to either Luo, Kamba or Maasai ethnic groups.Exclusion criteria were: pregnancy (n = 9), serious illnesses such as malaria, inability to walk unassisted and severe mental disease (n = 5), for a total inclusion of 1459 participants.Following an overnight fast (≥8 h), anthropometric measurements (weight, height, WC and hip circumference) were carried out.BMI and waist-to-hip ratio (WHR) were calculated, and blood pressure in the sitting position was measured using the average of two measurements.Fasting blood samples were collected to measure glucose, insulin, lipid profile (TG, low-density-lipoprotein cholesterol [LDL-C], very low-density lipoprotein cholesterol [VLDL-C], total cholesterol [TC] and high-density-lipoprotein cholesterol [HDL-C]), and liver profile (albumin, aspartate transaminase [AST], alanine transaminase [ALT], GGT and total bilirubin).Participants were then asked to perform a 2-h standard 75-g oral glucose tolerance test (OGTT), except for those previously diagnosed with DM who had a fasting venous blood glucose level ≥10.0 mmol/L [25].Homeostasis model assessment of insulin resistance (HOMA-IR) was calculated as described by Matthews et al. [26].Hypertension, dyslipidemia, DM, impaired glucose tolerance (IGT) or impaired fasting glucose (IFG) were either self-reported by participants or diagnosed during the assessment, using the most recent clinical guideline at the time [25,27,28].
Ultrasound scanning was used for the assessment of abdominal adipose tissue distribution (visceral and subcutaneous layers), and hepatic tissue with regards to steatosis, using a portable scanner (Aquila Basic Unit, Pie Medical Equipment; Esaote, Maastricht, the Netherlands) with a 3C.RS 3.5/5.0MHz curved transducer (Probe Article no.410638 Curved Array HiD probe R40; Pie Medical Equipment, Maastricht, the Netherlands).To assess hepatic steatosis, two trained operators (DLC and Andreas W. Hansen [mentioned in the Acknowledgement section]) obtained ultrasonography images on both sides of the main liver lobes.All liver sweeps were made during a deep inspiration, the first sweep being done from lateral to medial, second sweep from cranial to caudal and third sweep being a left transverse scan.Using a standardised approach, a semiquantitative grading system was used to characterise hepatic steatosis as absent, mild moderate or severe [13].The scoring system to detect hepatic steatosis was based on three criteria: increased echo reflectivity of the liver parenchyma compared to the right kidney cortex, decreased visualisation of the intra-hepatic vasculature and attenuation of ultrasound beam.The observers scored on a 4-point scale each criterion and summed, resulting in a liver fat score which ranged between 3 and 12.A score <5 was considered a normal liver, a score between 5 and 7 a liver with mild steatosis, a score between 8 and 10 as moderate and a score ≥11 was classified as severe hepatic steatosis.The liver scoring criteria used in this study has been described in more details elsewhere [13], and has been shown to be valid for the detection of liver steatosis in adults [13,29,30].In the context of this study, any steatosis (mild to severe), thus a score ≥5, was considered positive for ultrasound diagnosed NAFLD.
Abdominal visceral adipose tissue (VAT) was estimated by measuring its thickness using ultrasound scanning, from the spine to the linea alba, whereas abdominal subcutaneous adipose tissue (SAT) was measured from the abdominal muscles to the skin, with minimal pressure on the skin.Use of ultrasound scanning for VAT and SAT estimation has been shown to be a reliable method when validated against the gold standard computed tomography [31][32][33].
Participants were interviewed for demographic data, alcohol and tobacco consumption.Excessive alcohol consumption threshold was 20 g/day and 30 g/day for women and men, respectively [7].Dietary patterns were assessed using one interactive 24 h recalls, and cardiorespiratory fitness (mLO 2 /min/kg) was estimated using individual heart rate (HR) response to a submaximal step test.Habitual physical activity energy expenditure was measured with combined accelerometry and HR monitoring, with individual calibration of HR using information from the step test [34,35].
After converting TG units from mmol/l to mg/dL, FLI was calculated using Bedogi et al. formula [23]: All participants gave written or thumbprint informed consent.Ethical permission was obtained from the National Ethical Review Committee in Kenya (SSC Protocol No. 936), and consultative approval was given by the Danish National Committee on Biomedical Research Ethics in Denmark.More details concerning the methodology have been published elsewhere [36].

Statistical analyses
All statistical analyses were performed using Stata/MP 15.0 (Stata Corp, College Station, USA).Continuous variables were tested for normality using histogram and normal quantile plot, and they are presented as means ± standard deviation (SD) if normally distributed and as median ± quartile 1 (Q1) and quartile 3 (Q3) in the case of non-normally distributed continuous variable.Proportions (n, %) were used for categorical variables.Differences in participants with NAFLD or without NAFLD were determined by using Student's t-test, while the Chi-Squared test was used to test for differences in proportions.In case of non-normally distributed data, differences between groups were tested using the Kruskall-Wallis test.
The predictive value of FLI, each individual component (BMI, WC, TG and GGT) of the index, other anthropometric measurements (WHR, VAT and SAT) and HOMA-IR (validated proxy of insulin resistance [26]) to diagnose NAFLD was estimated using area under the receiver operator characteristic curve (AUROC).An AUROC score of 0.500 denoted no discriminatory power, 0.501-0.699denoted poor discriminatory ability, 0.700-0.799was considered acceptable discriminatory ability, 0.800-0.899denoted excellent discriminatory ability, and 0.900-1 was considered outstanding discriminatory ability [37].Comparisons between AUROC of FLI, its components and anthropometric measurements were conducted using the De Long method [38].
Sensitivity, specificity, positive likelihood ratio (LR+), negative likelihood ration (LRÀ) of 10-value intervals of the  significantly older and a higher proportion of participants were from urban areas.Individuals with NAFLD presented with higher BMI, WC, WHR and abdominal VAT and SAT levels (all p < 0.05).A total of 65 NAFLD participants (69.2%) were diagnosed with obesity (≥30 kg/m 2 ), compared to 20 (3.7%) non-NAFLD participants ( p < 0.05).Despite similar liver enzyme levels (AST, ALT, total bilirubin and GGT), NAFLD participants had higher fasting insulin ( p < 0.05 for both).They also had more insulin resistance (HOMA-IR of 1.9 compared to 1.0 in the non-NAFLD group), and more dysglycaemia (10.6% with IGT in the NAFLD group vs. 1.6% in the non-NAFLD, p < 0.05).A total of 7.2% of NAFLD participants were diagnosed with DM vs. 1.4% in the non-NAFLD group ( p < 0.05).Cardiorespiratory fitness was lower in the NAFLD group ( p < 0.05).
The AUROC of the FLI for predicting NAFLD (identified by liver ultrasound) was 0.80 (95% CI 0.74-0.85),which was significantly higher than individual parameters GGT or TG (p < 0.05), but not compared to anthropometric parameters BMI (AUROC of 0.83, 95% CI 0.79-0.88)and WC (AUROC of 0.81, 95% CI 0.76-0.87)(Figure 1).When compared to other anthropometric measurements and HOMA-IR, FLI predictive capacity was higher than WHR, VAT and HOMA-IR but not higher than ultrasoundmeasured SAT (AUROC of 0.82, 95% CI 0.76-0.87p < 0.05) (Table 2).The estimated Kappa value for FLI was 0.45, which signifies a moderate agreement between FLI and liver ultrasound to predict NAFLD.The optimal cut-off established by Youden index analysis was FLI ≥ 6.2.This was corroborated by examining the diagnostic accuracy of the 10-value intervals of the FLI (Table 3), and showing that the optimal cut-off was estimated to be FLI ≥ 10, with a sensitivity of 54.3%, specificity of 91.8 and a Youden Index of 0.45.

DISCUSSION
This study demonstrates that FLI is a good predictor of ultrasound-diagnosed NAFLD, and that the optimal cut-off point in the Kenyan population is ≥6.1, which is considerably lower compared to what is currently used in Europeandescent and Asian populations (FLI ≥ 60) [23,24,41].
FLI was developed by Bedogni et al. in 2006 [23] and since then, its predictive accuracy for NAFLD has been tested in several other studies [24,41,42].Using liver ultrasound in a population of 2652 participants from the Netherlands (mean age of 76.3 ± 6.0 years, 60.1% females, and mean BMI of 27.2 ± 4.2 kg/m 2 ), Koehler et al. have reported that FLI could accurately identify NAFLD (AUROC of 0.81, 95% CI of 0.79-0.83).However, recommended cut-offs of FLI ≥ 60 [23]  In the present study, when compared to individual anthropometric parameters BMI and WC, FLI did not present a better predictive capacity to detect NAFLD.This is consistent with a previous study published by Kim et al., which showed that BMI and WC were also good predictors of NAFLD (AUROC of 0.81, 95% CI 0.76-0.87 and AUROC of 0.79, 95% CI 0.73-0.85,respectively), compared to FLI (AUROC of 0.79, 95% CI 0.73-0.84) in an adult Korean population [41].However, Huang et al. showed that FLI predictive capacity (AUROC 0.83, 95% CI 0.83-0.84)was significantly greater than individual components of the scoring system FLI (BMI, WC, GGT and TG) [24].
FLI was developed using data from a European (Italian) population, from which NAFLD-associated specific components were identified [23].Adiposity and associated cardiometabolic diseases including NAFLD tend to present differently according to ethnic background.For example, risk of mortality, DM and cardiovascular diseases develop with lower BMI and WC in Asian populations, which justifies the use of ethnic-specific cut-offs in these populations [43].Growing evidence suggests that risk factors for NAFLD, its progression and severity are also greatly influenced by ethnicity, which involves genetic and epigenetic background, lifestyle components such as diet, physical activity and cardiorespiratory fitness, socio-economic variables and other environmental factors [6,[44][45][46].Only a few studies have examined the ethnic differences in the underlying mechanisms involved in the association between adiposity, insulin resistance, NAFLD and the risk of cardiovascular diseases [46][47][48].Despite having an increased risk of T2DM and cardiovascular diseases, African-descent populations have consistently been shown to have lower levels of TG, compared to their European descent counterparts [49,50].African-descent populations are also known to present a dissociative relationship between TG and insulin resistance, a phenomenon commonly referred to as the Triglyceride Paradox [49,50].Several studies have also demonstrated that for a given level of BMI or WC, African populations tend to have lower visceral adiposity and hepatic steatosis, but a higher level of insulin resistance [36,[50][51][52].
In the present study, we showed that Kenyan individuals diagnosed with NAFLD were generally overweight (BMI between 25.0 and 29.9 kg/m 2 ), as we found mean BMI values of 27.6 ± 5.5 and 27.3 ± 6.4 kg/m 2 for Kenyan male and female, respectively (p = 0.8).They also presented with higher WHR values (0.85 ± 0.06 for men and 0.95 ± 0.06 for women), higher WC in women (97.5 ± 16.2 cm), but a WC within normal in men (88.7 ± 12.7 cm) [43].Interestingly, they also presented with a lower level of TG, when compared to other populations diagnosed with NAFLD, where FLI has previously been tested for [23,24,41,42].The optimal cut-off given by Youden Index analysis for FLI was 6.1, which is noticeably lower than what has previously been reported in validation studies carried out in Europeandescent and Asian populations [23,24,41,42].This could be due to the FLI algorithm, based on metabolic parameters in which the association with NAFLD might vary according to ethnicity.
The complex and multidirectional interactions involved in the pathogenesis of obesity-associated cardio-metabolic diseases including NAFLD are not yet understood, particularly in understudied ethnic groups such as the Kenyan population.Differences in cardiorespiratory fitness and diet, among other factors, could also partly explain some of the ethnic differences noted.In this study, despite a similar level of physical activity energy expenditure, participants diagnosed with NAFLD presented with a lower cardiorespiratory fitness level (37.1 ± 5.7 vs. 42.2 ± 7.6 mLO 2 /min/kg, p < 0.05).Cardiorespiratory fitness has previously been shown to have a protective effect against NAFLD [53], thus a greater fitness level in some of our participants might have acted as a protective effect, even beyond anthropometric measures of adiposity or FLI parameters.
To our knowledge, this is the first study to validate the diagnostic accuracy of the FLI scoring system for NAFLD in a sub-Saharan African population.However, some limitations should be acknowledged.The present study used liver ultrasound scanning as the baseline diagnostic test for NAFLD, which is not recognised as a gold-standard diagnosis technique, and could thus lead to misdiagnosis of NAFLD.Albeit having demonstrated good discriminative capacities to distinguish between mild, moderate and severe hepatic steatosis, previous studies have reported that liver inflammation, that is, fibrosis might affect ultrasound images and lead to misclassification [16].In this study, only a small percentage (7.4%) of the participants presented with moderate steatosis and no participant were diagnosed with severe ultrasound features of steatosis, which suggest a relatively low risk of developing high level of hepatic inflammation and fibrosis.On the other hand, the predictive capacity of ultrasonography to detect lower grades of liver steatosis is known to be inferior compared to higher levels (moderate to severe) [21,54].Reliability of ultrasound scanning for the detection of NAFLD is modest.A previous meta-analysis including 49 studies and 4720 participants showed a kappa index ranging from 0.54 to 0.92 for intrarater reliability and from 0.44 to 1.00 for inter-rater reliability [21].In this study, we only used two experienced blinded observers to detect NAFLD, but misdiagnosis related to the low reliability of ultrasound scanning cannot be excluded.Nevertheless, several studies have shown the validity of ultrasound scanning to assess hepatic steatosis and have recommended this method, especially in limited settings such as urban and rural Kenya [13,30].Still, the relatively low number of participants diagnosed with NAFLD needs to be acknowledged.Finally, medical conditions affecting the liver such as human immunodeficiency virus or viral hepatitis were not screened during this study, and might have affected parameters used for the FLI such as GGT, adipose tissue distribution or liver ultrasound results of included participants.
In conclusion, despite suggested differences in the mechanism associating FLI parameters (TG, BMI and WC-proxies of adiposity) and NAFLD in sub-Saharan African populations, FLI is a valid scoring system to use in adult Kenyans.However, this index is not superior to anthropometric parameters such as BMI or WC to predict NAFLD.Considering the relative complexity of assessing FLI in many sub-Saharan African countries, BMI and WC may be more useful in limited settings.Still, these conclusions need to be interpreted with caution since NAFLD was diagnosed using liver ultrasonography and participants presented mostly with mild level of liver steatosis.

F I G U R E 1
Predictive capacities to detect non-alcoholic fatty liver disease of FLI and associated components.BMI, body mass index; FLI, Fatty Liver Index; GGT, gamma-glutamyl transferase; TRIG, triglycerides; WC, waist circumference.
T A B L E 1 Baseline characteristics of 640 adult Kenyans presented as n (%), mean (SD) or medium (IQR).
RESULTSFrom the 1459 individuals included in the community-based study, liver ultrasound data were obtained and analysed in 866 participants.Two participants were excluded from assessment of hepatic steatosis as kidney's ultrasound images could not be obtained for comparison, 24 participants were excluded due to excessive alcohol consumption, and 200 participants were excluded due to missing values for of NAFLD detected by ultrasound (score ≥ 11).Baseline characteristics for individuals with and without ultrasounddiagnosed NAFLD are presented in Table 1.The mean age of participants was 37.4 ± 0.4 years, and 58.7% were female.Compared to the non-NAFLD group, the NAFLD group (71.2% female with a mean age of 39.4 ± 1.4 years old) wasNote: All data are presented as mean (SD) or proportion (%).Abbreviations: ALT, alanine aminotransferase; AST, aspartate aminotransferase; CRF, cardio-respiratory fitness; GGT, gamma-glutamyl transferase; HDL-C, high-density lipoprotein; HOMA-IR, homeostasis model assessment was calculated as (fasting plasma insulin Â fasting plasma glucose)/22.5;LDL-C, low-density lipoprotein; PAEE, physical activity energy expenditure; TG, triglycerides; VLDL-C, very low-density lipoprotein.a Participants with an alcohol-consumption that does not exceed the criteria for NAFLD.b Presented as median (IQR 25%-75%).
[24]insufficient sensitivity in this elderly European-descent population.Similarly, Huang et al. have demonstrated FLI capacity to predict NAFLD in a middle-age and elderly population of Chinese men and women (mean age 58.6 ± 9.7 years, 69.0% females, and mean BMI 25.1 ± 3.3 kg/m 2 )[24].Predictive capacities to detect NAFLD of body fat distribution parameters and insulin resistance index HOMA-IR.Diagnostic accuracy of Fatty Liver Index by 10-value intervals.