Fatty liver biomarkers and insulin resistance indices in the prediction of non‐alcoholic fatty liver disease in Ghanaian patients

Abstract Background Scant West African data on non‐alcoholic fatty liver disease (NAFLD) means there is little representation of this population in the modelling used to derive biomarkers and predictive indices for risk stratification of patients for the presence of hepatic steatosis. This study evaluates the performance of the fatty liver index (FLI), hepatic steatosis index (HSI) and triglyceride‐glucose (TyG) index and its derivatives in predicting ultrasound detected NAFLD in a locally resident population of Ghanaian participants. Methods and Findings A post hoc analysis of data from a cross sectional assessment of NAFLD and cardiovascular risk was performed. Data from 210 participants without significant alcohol intake, or secondary causes of fatty liver and not on steatogenic drugs was evaluated. A structured questionnaire had been used to collect demographic data, medical and drug history. Anthropometry, blood sampling for liver chemistry and fasting lipids were performed. Hepatic steatosis was detected by ultrasonography. A retrospective analysis involving multivariate binary logistic regression assessed FLI, HIS, TyG (and its derivatives) as predictors of NAFLD with p < .05 considered statistically significant. Sensitivity, specificity, predictive values, likelihood ratios were calculated and accuracy of the proxies evaluated from area under the receiver operating characteristics curve (AUROC). All the biomarkers and indices were significantly associated with NAFLD (p ≤ .001). All the lipid and fatty liver indices assessed performed acceptably as predictors of NAFLD. FLI (AUC = 0.8, 95% CI [0.74–0.87]), TyG‐WC (AUC = 0.81, 95% CI [0.75–0.88]) and TyG‐WHtR (AUC = 0.81, 95% CI [0.74–0.88]) performed best at predicting NAFLD. Whilst in all cases the markers had good specificity (>90%) they lacked sufficient sensitivity with FLI having the highest sensitivity of 36.7%. Their overall accuracy was greater than 70% in each case. Conclusion The overall accuracy of HSI, FLI, TyG index and its derivatives (TyG WHtR, TyG BMI, TyG WC) was acceptable for predicting NAFLD in this population. Given their performance in this study and in light of their low cost, accessibility, easy interpretation and non‐invasive nature; they are suitable tools for screening in the Ghanaian population.


| INTRODUC TI ON
With the growing recognition of non-alcoholic fatty liver disease (NAFLD) as a precursor for chronic hepatic diseases and as a significant risk factor for extra-hepatic, systemic disease outcomes, 1 there have been growing efforts to develop efficient, cost-effective, minimally invasive, accurate methods for screening patients for NAFLD.
While liver biopsy for histopathology is continually upheld as the gold standard, it is acknowledged that this is not without its shortcomings, including sampling errors, high cost, risk of complications and variability in pathological diagnosis. 2Ultrasonography, despite its shortcomings in detecting steatosis under 20%, remains relevant in the diagnosis of NAFLD. 3,4forts to identify useful risk stratifying proxies using blood biomarkers and anthropometric measurements have led to the development of numerous indices that are proposed to predict the presence or otherwise of hepatic steatosis.The scant West African data on NAFLD means there is little by way of representation of this population group in the modelling used in deriving these predictive indices.While the utility of these indices has been validated to various degrees in populations of European descent [5][6][7] and Asian populations, 8,9 to our knowledge, no study has specifically assessed their comparative performance in predicting ultrasound defined NAFLD in a wholly West African resident population.The fatty liver index (FLI) 10 and hepatic steatosis index 11 were developed as potential predictors of NAFLD and they have performed better than most other biomarkers in this respect. 12The triglyceride-glucose (TyG) index 13 is a lipid related index that was initially developed to aid in the assessment of insulin resistance.The role of insulin resistance in the pathogenesis of hepatic steatosis forms part of the rationale behind the increasing use of the terminology metabolic dysfunction-associated steatosis liver disease (MASLD) in patients with steatosis and features of cardiometabolic dysfunction. 14However, the TyG index (together with its derivatives) has joined the FLI and hepatic steatosis index to become some of the most frequent non-invasive tests or biomarkers utilised in evaluating the likelihood of fatty liver disease. 15veloped in the Italian Dionysos nutrition and liver study, 10 and validated against ultrasonography, the use of FLI as a simple yet accurate marker for NAFLD has been widely assessed and its use endorsed by professional bodies. 4Its validity is, however, still debated, and a recent large meta-analysis found that while it is effective in stratifying the risk of NAFLD, it failed to sufficiently diagnose or exclude NAFLD. 16Derived from the findings of a large cross-sectional study in South Korea, the HSI is easier to calculate and does not require the input of lipid related variables.As such, it potentially represents an economical alternative to FLI. 11 NAFLD is increasingly recognised as a growing threat in Sub-Saharan Africa 17,18 with the age-standardised prevalence of NAFLD ranging from 5.0%-7.5% to 10.1%-12.5%.Global Burden of Disease data estimated an annual percentage change (EAPC) increase in Western sub-Saharan Africa of 0.69 (95% CI [0.63-0.75]),with Ghana having an estimated annual percentage increase of 1.26-1.5. 19There are already some indications that there are differences in the utility of non-invasive tests (NITs) for NAFLD between patients of Ghanaian descent living in Europe and other ethnicities living in the same region. 20This study aimed at evaluating the performance of insulin resistance indices and steatosis biomarkers derived from various clinical, anthropometric and biochemical measures in predicting NAFLD in a population of Ghanaian participants resident in Ghana.

| Study Design, Site and Sampling
This work is a sub study of a cross sectional assessment of NAFLD and cardiovascular risk carried out over a 10-month period from April 2016 to February 2017.Utilising a consecutive sampling approach, 310 patients were consented for the study.Eighty two met one or more of the exclusion criteria.18 failed to complete all required aspects of the study resulting in a final group of 210 recruited patients from among adult patients (>18 years) of Ghanaian descent who were attending general medical outpatient clinics at the Cape Coast Teaching Hospital, a tertiary health facility in Cape Coast, Central Region, Ghana.

| Exclusion criteria
The 210 participants had been screened for significant alcohol intake using the Alcohol Use Disorders Identification Test 21 (AUDIT) and those with scores of ≥7 or self-reported alcohol consumption of >14 units per week for women and 21 units for men were excluded from the study.Patients diagnosed with hepatitis B, C or HIV Given their performance in this study and in light of their low cost, accessibility, easy interpretation and non-invasive nature; they are suitable tools for screening in the Ghanaian population.

K E Y W O R D S
fatty liver index, Ghana, hepatic steatosis index, non-alcoholic fatty liver disease, triglyceride glucose index infection or known liver disease were excluded.Patients suffering from thyroid disease, chronic hemolytic disorders, malignancy or on chemotherapy drugs, corticosteroids, valproate or antiretroviral drugs were also excluded.

| Ethical Considerations
Ethical approval for the recruitment of patients into the NAFLD and cardiovascular risk study was obtained from the Institutional Review Board of the University of Cape Coast (UCCIRB/CHAS/2015/19).
Permission was also obtained from the Cape Coast Teaching Hospital for the study (CCTH/G/002/93-15).Informed consent was obtained from all participants and documented.The recommendations guiding physicians in biomedical research involving human subjects issued by the World Medical Declaration of Helsinki (2013) 22 were applied during that study.This paper presents a poc hoc analysis of anonymised data generated by that study.

| Data collection
Demographic data, medical and drug histories had been obtained using a structured questionnaire.Significant use of alcohol was ruled out based on both self-reported usage as well as the administration of the AUDIT questionnaire. 21Height was measured to the nearest 0.5 cm using a stadiometer (Seca® mechanical column scale with stadiometer), and weight in light clothes was measured to the nearest 0.1 kg using an electronic scale (Omron® BHF-510).Body mass index (BMI) was computed as the weight in kilogrammes divided by the square of the height in metre (kg/m 2 ).
Waist circumference (WC) was measured in centimetres with an inelastic tape measure at the midpoint between the lower margin of the last rib and the top of the iliac crest.Waist-to-height ratio (WHtR) was computed by dividing the waist by the height.Hip circumference was measured around the widest portion of the buttocks.Waist-to-hip ratio (WtH) was computed by dividing WC by the hip circumference.

| Sample collection, analysis and imaging
All study subjects had under gone phlebotomy after an overnight fast (of at least 8 hours).Hepatitis B and C serological status was ascertained by commercially available Micropoint® Hepatitis B surface antigen Gold Rapid and Micropoint® hepatitis C Gold Rapid Test kits, respectively, for the qualitative detection of hepatitis B surface antigen and antibodies to HCV in the serum.Serum was used for the measurement of transaminases and lipids in a Chemistry auto-analyzer (ChemWell® from Awareness Technology) using Randox laboratories clinical chemistry kits.Total cholesterol analysis was carried out using an enzymatic colorimetric test, the CHOD-PAP method, high density lipoprotein cholesterol (HDL) by precipitation method and triglyceride (TG) estimation employed enzymatic hydrolysis of TG with lipases.Low density lipoprotein cholesterol (LDL) was calculated using Friedwald's formula: LDL = (Total Cholesterol − HDL)-TG/2.2as all the levels of TG were below 5.0 mmol/L.Abdominal ultrasound scans had been carried out by a radiologist with a Philips ClearVue 350® ultrasound machine to identify sonographic evidence of fatty liver, utilising the criteria set by Hamaguchi et al which were a brighter hepatic echo pattern in comparison to the kidneys, deep attenuation and vascular blurring; present either alone or in combination (p 2710). 23

| Biomarkers and Indices
The selected biomarkers and indices were calculated as follows.was considered statistically significant.The accuracy, sensitivity, specificity, positive and negative predictive values as well as positive and negative likelihood ratios and AUROC were calculated for each insulin resistance index and steatosis biomarker.AUROC of ≥0.7 to <0.8 was considered acceptable, ≥0.8 to <0.9 was considered good, and ≥0.9 was considered excellent.Post hoc analyses on the fitness of the logistic regression models were conducted using the Hosmer-Lemeshow goodness-of-fit test.

| The study population
The 210 participants were predominantly female, with a female-tomale ratio of 2. ) were not significant compared to those without (Table 3).6).With the exception of the TyG BMI, the p values for all the models using the Hosmer Lemeshow test were >.05 indicating a good fit for the models.

| Liver steatosis indices and insulin resistance related biomarkers as predictors of NAFLD
When the data were segregated by sex the overall trend among the predictors was maintained in females, who made up the majority of participants.FLI performed slightly better in females than it had in males.In the solely male group, TyG-BMI and FLI were the best predictors of NAFLD with an AUC = 0.79 (Figures 1 and 2).

| DISCUSS ION
This analysis assessed the performance of various biomarkers derived from clinical, anthropometric and lipid related measures as predictors of NAFLD among an undifferentiated pool of participants from medical outpatient clinics at a teaching hospital in Ghana.
Our results demonstrated that variables such as waist-to-hip ratio, waist-to-height ratio, diastolic and systolic blood pressures, BMI, fasting plasma glucose (FPG) and TG were significantly associated with NAFLD.Postulated to be driven by insulin resistance among other things, 24 the relationship between fatty liver and obesity has been continually reaffirmed. 25,26Despite the relationship between FPG and NAFLD, diabetes mellitus was not found to be a predictor of steatosis in our patient cohort, which is at variance with that of a Turkish study. 27This might have to do with the common use of medication such a metformin and pioglitazone among diabetics which are known to affect fatty liver.Our work found a significant association between NAFLD and the anthropometric measurements we assessed, which included BMI, WtH and WHtR.As a predictor of NAFLD, BMI remained significant even after adjustment and after controlling for smoking, triglycerides, high density lipoprotein and medication.We found no studies that had previously evaluated BMI as a predictor of NAFLD in West Africa.However, the association between BMI and NAFLD has been suggested by findings in other sub regional studies.In Nigeria 28,29 and in Sudan, 30 BMI, was found to be significantly higher among subjects with NAFLD than in those without NAFLD.Among diabetics 31 and participants with HIV 32 in Nigeria with NAFLD; however, this difference in BMI was not found to be significant probably because of the effect of their underlying diseases and the associated treatments in those sub populations.
One previous Nigerian study assessing the utility of novel indices like TyG index, TyG WC, TyG BMI and TyG WHtR evaluated their efficacy as predictors of metabolic syndrome (MetS) and it was found that all these indices significantly identified metabolic syndrome in the participants. 33Despite the lack of previous studies assessing the TA B L E 5 Bivariate and multivariate binary logistic regression of lipid related and fatty liver indices as predictors of NAFLD.Despite the observation of adequate specificities of 85% and 91.6% respectively, for the European study 5 and the current one, sensitivity in the European population was much higher than was found in our study (61% vs. 36.7%,respectively).While the low sensitivity of FLI for detecting NAFLD in our population (36.7%) might raise questions about its suitability as a screening tool for patients at risk of NAFLD in our local context, the proportion of overweight and obese participants should be taken into account.FLI has been found to be a poor predictor of steatosis in obese patients. 35Approximately one  19 No explanation was given for these differences, but it raised a question that this study attempts to answer relating to the performance of these non-invasive tests in the Ghanaian population.
Several works have reviewed the performance of predictive fatty liver biomarkers for the presence of steatosis.Despite the variations in performance of these biomarkers within different studies, there are widely noted trends.A French study found a relatively poor predictive utility of some biomarkers for NAFLD as diagnosed by ultrasonography, their AUCs were 0.56 (0.48-0.64) for FLI, 0.65 (0.58-0.73) for HSI, and 0.63 (0.54-0.71) for TyG. 7Although we cannot directly compare the performance of these variables between studies, we describe our finding of 0.8 (0.74-0.87) for FLI as good (AUC 0.8-0.9),and 0.74 (0.66-0.81) for HSI, 0.76 (0.68-0.82) for TyG index as acceptable (AUC >0.dependent studies. 36In that meta-analysis, TyG index performed better in females than in males, a finding similar to that of our study.In the second meta-analysis, which involved 121,975 participants from 17 studies, higher TyG index was found to be an independent predictor of NAFLD. 15No African data were included in either of the meta-analyses cited.Overall, in our study TyG-WC and TyG-WHtR performed best at predicting NAFLD, adding credence to the argument that TyG-related parameters (combining both TyG and anthropometric measures) could be a better predictor of hepatic steatosis compared them with TyG index on its own. 9,37In summary, our work has found HSI, FLI and TyG and related indices (TyG index, TyG WC, TyG WHtR satisfactorily predict NAFLD meeting the cut-off for acceptability (AUROC ≥0.7).
The key limitation of this study is the use of ultrasonography for detecting NAFLD.Although it is less sensitive in detecting steatosis less than 20%, it remains the most widely utilised clinical tool in routinely assessing liver structure 3 Data were entered in Microsoft excel® and exported to Stata®, version 15; StataCorp, software for analysis.Descriptive statistics was used to describe measures of central tendencies, frequencies and percentages.Pearson's χ 2 -test was used to find associations between categorical variables.Student t-test (independent twosample) was used to compare the means of scale variables with two levels of category.Cohen's d, a standardised effect size measure, was used to determine the magnitude of the difference between the means of the independent variables, with a Cohen's d ≥ 0.8 indicative of a large effect size.Binary logistic regression was used to determine the relationship between relevant clinical, anthropometric parameters, insulin resistance indices and steatosis biomarkers and the presence of Fatty liver index (FLI) = e (0.953×ln(triglycerides)+0.139×BMI+0.718×ln(GTP)+0.053×WC−15.745)∕ 1 + e (0.953×ln(triglycerides)+0.139×BMI+0.718×ln(GTP)+0.053×WC−15.745)× 100 Triglyceride − Glucose (TyG) index = Ln Fasting triglyceride (mg ∕ dL) × fasting glucose (mg ∕ dL) ∕ 2 TyG − BMI = TyG × BMI TyG − WC = TyG × WC TyG − WHtR = TyG × WHtR Triglyceride ∕ HDL ratio (TG − HDL) = TG (mmol ∕ l) ∕ HDL (mmol ∕ L) NAFLD.Those variables with a p ≤ .25 were considered for multivariate binary logistic regression analysis.Models were adjusted for relevant sociodemographic and other risk factor variables.A p < .05

7 )
in our study.In a separate study among Japanese subjects, the AUCs of all anthropometric and lipid-related indices were greater than 0.5, indicating that all have certain predictive values for NAFLD.8This finding mirrors our own, with TG HDL in females (AUC = 0.645) being the least predictive in our study.In that Japanese study, TyG index-related parameters had overall good predictive values [0.79-0.81]for TyG index, 0.84 [0.87-0.89]for TyG BMI, 0.88 [0.88-0.89]for TyG WC and 0.87 [0.87-0.88]for TyG WHtR) which is similar to our finding that TyG related indices performed well in our population.The value of TyG index as a novel biomarker is highlighted in the findings of two meta-analyses.Ling et al demonstrated a strong positive association and a significant dose-response relationship between TyG index and NAFLD in a meta-analytic study involving 105,365 participants from 12 in-

and anthropometric parameters as predictors of NAFLD
Characteristics of study participants.