Phenotype, Body Composition, and Prediction Equations (Indian Fatty Liver Index) for Non-Alcoholic Fatty Liver Disease in Non-Diabetic Asian Indians: A Case-Control Study

Objective In this study, we have attempted comparison of detailed body composition phenotype of Asian Indians with non-alcoholic fatty liver disease (NAFLD) vs. those without, in a case controlled manner. We also aim to analyse prediction equations for NAFLD for non-diabetic Asian Indians, and compare performance of these with published prediction equations researched from other populations. Methods In this case-control study, 162 cases and 173 age-and sex-matched controls were recruited. Clinical, anthropometric, metabolic, and body composition profiles, and liver ultrasound were done. Fasting insulin levels, value of homeostasis model assessment of insulin resistance (HOMA-IR), and serum high sensitive C-reactive protein (hs-CRP) levels were evaluated. Multivariate logistic and linear regression analyses were used to arrive at prediction equations for fatty liver [Indian fatty liver index (IFLI)]. Results As compared to those without fatty liver, those with fatty liver exhibited the following; Excess dorsocervical fat (‘Buffalo hump’), skin tags, xanthelasma, ‘double chin’, arcus; excess total, abdominal and subcutaneous adiposity, and high blood pressure, blood glucose, measures of insulin resistance (fasting insulin and HOMA-IR values), lipids and hs-CRP levels. Two prediction equations were developed; Clinical [Indian Fatty Liver Index-Clinical; IFLI-C]: 1(double chin) +15.5 (systolic blood pressure) +13.8 (buffalo hump); and IFLI-Clinical and Biochemical (CB): serum triglycerides+12 (insulin)+1(systolic blood pressure) +18 (buffalo hump). On ROC Curve analysis, IFLI performed better than all published prediction equations, except one. Conclusion Non-diabetic Asian Indians with NAFLD researched by us were overweight/obese, had excess abdominal and subcutaneous fat, multiple other phenotypic markers, had higher insulin resistance, glycemia, dyslipidemia and subclinical inflammation than those without. Prediction score developed by us for NAFLD; IFLI-C and IFLI-CB, should be useful for clinicians and researchers.


Introduction
Non-alcoholic fatty liver disease (NAFLD) includes a spectrum of liver disorders characterized by accumulation of hepatic fat in absence of significant alcohol consumption (<20 gm/day) and other causes of liver diseases. It is most common cause of asymptomatic elevation of liver enzymes worldwide [1]. If fat accumulation continues, inflammation ensues in liver parenchyma, termed as non-alcoholic steatohepatitis (NASH), which has potential of progressing to cirrhosis.
Estimates based on imaging and autopsy studies suggest that about 20% to 30% of adults in the United States and other Western countries have NAFLD [2][3][4]. According to available sparse data in India, the prevalence of NAFLD ranges from 15-32% [5,6]; variations in estimates are due to rural/urban habitat, socio-economic stratum, and varying dietary habits in different regions of country. Some data indicate that hepatic triglyceride content (estimated by proton magnetic resonance spectroscopy) of Asian Indians living in USA is more when compared to white Caucasians [7].
Several studies have suggested the associations of NAFLD with obesity, abdominal obesity, dysglycemia and other components of the metabolic syndrome [8]. Excess of body fat, mainly abdominal fat [9], is related to NAFLD [10]. Lin et al [11] reported that the waist circumference (WC) is better than body mass index (BMI) for predicting liver steatosis in Taiwanese subjects. Fallo et al [12] studied 86 hypertensive obese adults (48 NAFLD and 38 controls) and showed WC as predictor for NAFLD. Damaso et al. [10] reported that the group of adolescents with NAFLD had significantly higher values of BMI, visceral and subcutaneous fat, insulin, and homoeostasis modal assessment for insulin resistance (HOMA-IR) in both genders, compared with control subjects living in Brazil.
It is now well established that Asian Indians have higher levels of body fat, more abdominal adipose tissue, less lean body mass (LBM) and higher magnitude of insulin resistance than American and European subjects [13,14]. Some data suggest that fat deposition at ectopic places (nape of the neck, excess dorsocervical fat; 'buffalo hump'; excess fat below chin; 'double chin', liver) correlates with the metabolic syndrome in Asian Indians [15]. Overall a phenotype of excess body fat, low muscle mass, and fat deposition, and ectopic fat deposition is characteristic for Asian Indians [16].
It is important to identify those at risk for development of NAFLD, so that appropriate investigations could be applied for evaluation and diagnosis and for management. An optimal prediction formula for NAFLD should include simple clinical parameters which could be used by clinicians. In this context, several prediction equations have been researched in other populations; North American [17], Finnish [18] and Italian [19] population.
While others and our group have investigated phenotype and body composition of diabetic and non-diabetic Asian Indians [20], detailed studies have not been carried out in persons with NAFLD. In this study, we have attempted comparison of phenotype of those with fatty liver with those without, in a case controlled manner. Further, we present prediction equations for Asian Indians, and compare performance of these prediction equations with published prediction equations from other populations.

Subjects
In this case control study, we recruited a 335 overweight/obese (238 males and 97 females) subjects (BMI !23kg/m 2 ), 162 (129 males and 33 females) with NAFLD (cases) and 173 (109 males and 64 females) without NAFLD (controls) at two clinical sites [Fortis Hospital and All India Institute of Medical Sciences, New Delhi, India] between May 2009 and October, 2014. The study was approved by the Institutional ethics committee at Fortis Hospital and All India Institute of Medical Sciences, New Delhi, India and written informed consent was obtained. NAFLD was defined by liver ultrasonography in those with alcohol intake of less than 20 gram/day [14]. Subjects with known type 2 diabetes mellitus (T2DM), cardiovascular disease (CVD), presence of other liver diseases (alcoholic liver disease, hepatitis virus infection, autoimmune hepatitis, primary biliary cirrhosis obstruction, drug-induced liver damage etc), severe organ damage in other organs, human immunodeficiency virus infection, pregnancy, lactation, or any pro-inflammatory state and patients on statins or fenofibrate were excluded from the study.

Ultrasound imaging
Presence of fat in liver was assessed with ultrasound using 3.5 MHz curvilinear probe (Siemens-G 60 S 2004, Germany). The definition of fatty liver was based on a comparative assessment of image brightness relative to the kidneys, in line with previously reported diagnostic criteria [24]. Severity of fatty liver was classified according to the brightness compared to kidneys, blurring of gall bladder wall, of hepatic veins and of portal vein. The radiologists performing the ultrasound were unaware of the clinical and laboratory results.

Body composition
Percentage body fat (%BF), LBM and bone mineral density (BMD) were estimated by using whole body dual-energy X-ray absorptiometry (DEXA) scan (Lunar Prodigy Advanced Whole Body DEXA system, GE Medical Systems) as previously [5].

Statistical Analysis
Data were entered in an Excel spreadsheet (Microsoft Corp, Washington, USA). The distribution of clinical, biochemical, anthropometry and body composition parameters was confirmed for approximate normality. We used mean and standard deviation to summarize the variables. The differences in biochemical anthropometry and body composition parameters in cases and controls were compared using the Student's t-test. Difference between proportions was tested using Chi-square test. Bivariate logistic regression was performed to identify significant predictors of NAFLD. After adjusting for age, sex, TG, TC, HDL-C, LDL-C, WC, FBG, ALT, AST, ALK, fasting insulin, HOMA and % BF, multivariable logistic regression was carried out to identify the independent risk factor(s) and to estimate odds ratio (OR) and 95% confidence interval. For all above, a p value of <0.05 was considered as statistically significant.
The levels of fasting blood glucose, serum TG, TC, LDL-C, VLDL, ALT, GGT, fasting insulin, HOMA-HR and hs-CRP were significantly higher in cases as compared to controls ( Table 2). The intra and inter assay percentage coefficient variables were 2.30% and 1.92% for insulin and 2.03% and 1.56% for hs-CRP, respectively. Body composition by Dual Energy X-ray Absorptiometry profiles were presented in Fig 2 and S1 Table. The values of right leg lean mass, right leg total mass, trunk fat percentage, trunk fat, total trunk mass, %BF and BF were significantly higher in cases as compared to controls, whereas % LBM and LBM were higher in controls.

Univariate and step-wise logistic regression model
Using Univariate logistic regression model (Table 3), significant risk factors associated with the development of NAFLD were; double chin, buffalo hump, skin tag, xanthelasma, systolic blood pressure, WC, BMI, %BF, TG, fasting insulin, HOMA-IR and Hs-CRP. Using a step-wise logistic regression model as shown in Table 4

Using Clinical variables (IFLI-C)
Three clinical variables found to be significant in the logistic model were used to calculate the prediction score for NAFLD. The simplest equation for estimation of clinical parameters is: 1(double chin) +15.5(systolic blood pressure) +13.8(buffalo hump) (maximum score = 28.7, minimum score = 0). Systolic blood pressure: 1: (>120/80 mmHg); 0: otherwise. Using a ROC analysis (Fig 3) a score cut-off of !1.0 ensured the best balance between sensitivity and specificity. The sensitivity, specificity and ROC area under the curve (95% CI) were 64.81%, 61.85%, and 65.0 (59.0-70.67), respectively. The probability of subject having NAFLD was more if score was !1 with positive likelihood ratio = 1.67 and negative likelihood ratio = 0.58.

Discussion
This detailed phenotype analysis of patients with NAFLD shows adverse body composition features; high body fat, truncal fat, truncal subcutaneous fat, low lean mass, and ectopic fat deposition; excess fat deposition below chin ('double chin') and over the nape of the neck ('buffalo hump'). Moreover, we present two prediction equations for NAFLD, one based on clinical parameters alone (IFLI-C), and other, a combination of clinical and biochemical parameters (IFLI-CB). Overall, previous studies have shown that Asian Indians have more fat in various abdominal fat depots. In a comparative study of Asian Indians vs. white Caucasians in USA; for similar value of BMI, migrant Asian Indians had significantly greater total abdominal fat and intraabdominal adipose tissue (IAAT) [28]. Some investigators have reported that truncal subcutaneous adipose tissue (SCAT; measured by subscapular and supra-iliac skinfolds and by magnetic resonance imaging) is thicker in South Asians than in White Caucasians [29,30]. Recently, we have also shown that pancreatic volume (surrogate marker of pancreatic fat) and liver span (surrogate marker of liver fat) correlate strongly to diabetes in non-obese individuals [31]. Thicker truncal subcutaneous fat, as shown by skinfolds, and more truncal fat, signifies overall increased truncal and abdominal adiposity in persons with NAFLD in the current study. These increase fat depots, consisting mostly of metabolically dysfunctional adipocytes [32], generate increased amount of non-esterified fatty acids contributing significantly to fatty liver. It is important to note low lean mass in those with NAFLD, which may also of significance to glucose metabolism, and has been shown to have genetic basis in Asian Indians [33].
'Buffalo hump' and 'double chin', both signifying excess fat deposition at unusual sites, have been described by us as phenotypic markers closely correlating with metabolic syndrome in Asian Indians [15]. Incidentally, 'buffalo hump' is also present in HIV-associated lipodystrophy after prolonged use of protease inhibitors [34], and is associated with overall state of insulin resistance. Further, 'double chin' is commonly associated with diabetes in patients with partial lipodystrophies [34]. In the present study, both phenotypic markers were associated with NAFLD. Of significant note, both these signs could easily be detected on simple visual examination.
Interestingly, there are differences in the diagnostic prediction of different non-invasive scores, which may be accounted by sample of different populations studies and use of variegated phenotypic and biochemical measurements. Overall, assessment of phenotype done by us has been extensive, and includes conventional and novel signs (acanthosis nigricans, 'buffalo hump' 'double chin') based on previous studies done by others and by us. Many of these signs/markers have not been taken in account by other investigators while researching prediction equations. Further, some prediction equations include subjects with T2DM, which we have strictly excluded them, because many other variables in patients with T2DM may confound the prediction. These include changes the weight following diet and exercise; inclusion of drugs such as metformin and thiazolidinediones and vitamin E, all of which may have effect on liver fat. It has been argued that other methods; magnetic resonance spectroscopy and liver biopsy are better tools for defining NAFLD, and could be considered as "gold standard". Conversely, ultrasonography is by far the most common method of diagnosing NAFLD in clinical practice and has a fair sensitivity (87%) and specificity (94%) in detecting hepatic steatosis [35]. In a recent meta-analysis, forty-nine (4720 participants) studies regarding ultrasonography for diagnosis of fatty liver were included. Interestingly, the overall sensitivity, specificity, positive likelihood ratio, and negative likelihood ratio of ultrasound for the detection of moderatesevere fatty liver, compared to histology (gold standard), were 84.8% (95% confidence interval: 79.5-88.9), 93.6% (87.2-97.0), 13.3 (6.4-27.6), and 0.16 (0.12-0.22), respectively. Further, the area under the summary receiving operating characteristics curve was 0.93 (0.91-0.95). Overall, sensitivity and specificity of ultrasound was similar to that of other imaging techniques (i.e., computed tomography or magnetic resonance imaging) [36]. In summary, ultrasonography for liver fat is simple to perform, non-invasive, cost-effective and does not entail any radiation hazard, and could also be used in the epidemiological studies. Hence, although not "gold standard", this method of investigation provides reasonable alternative to more expensive and difficult-to-perform diagnostic methods of NAFLD.
Two investigators have previously formulated prediction equations based on liver ultrasound estimated liver fat [17,19]. In first study (prediction equation, FLI), 216 subjects with and 280 without NAFLD were studied [17], while in second study (prediction equation, LAP) 588 Italian adults were studied [18]. The performances of both the prediction equations on ROC curve analysis were inferior than IFLI-C and IFLI-CB presented by us. NAFLD Liver Fat score was developed using most robust estimation of liver fat using magnetic resonance spectroscopy in 470 Finnish subjects (non-diabetic and patients with T2DM) [19]. This score, however, performed better than IFLI on ROC curve analysis.

Conclusion
Non-diabetic Asian Indians with NAFLD sampled by us, as compared to those without NAFLD, had adipose and insulin resistant phenotype. We also present prediction score for NAFLD; IFLI-C and IFLI-CB, which should be useful for clinicians and researchers. n, 588; definition of fatty liver was based on liver ultrasonography (18). 3 n, 359 non-diabetic, 111 type 2 diabetes; liver fat content was measured using proton magnetic resonance spectroscopy (19). doi:10.1371/journal.pone.0142260.t005 Supporting Information S1 Table. Body composition by Dual Energy X-ray Absorptiometry. (DOC)