Article Text

Download PDFPDF

Development and validation of a simple NAFLD clinical scoring system for identifying patients without advanced disease
  1. S A Harrison1,
  2. D Oliver2,
  3. H L Arnold1,
  4. S Gogia2,
  5. B A Neuschwander-Tetri2,3,4
  1. 1
    Division of Gastroenterology and Hepatology, Brooke Army Medical Center, San Antonio, USA
  2. 2
    Department of Internal Medicine, Saint Louis University, St Louis, USA
  3. 3
    Saint Louis University Liver Center, St Louis, USA
  4. 4
    Division of Gastroenterology and Hepatology, Saint Louis University, St Louis, USA
  1. Dr S A Harrison, 3841 Roger Brooke Drive, Fort Sam Houston, TX 78234, USA; Stephen.Harrison{at}


Background: Clinical predictors of advanced non-alcoholic liver disease (NAFLD) are needed to guide diagnostic evaluation and treatment.

Methods: To better understand the demographics of NAFLD and risk factors for advanced disease, this study analysed 827 patients with NAFLD at two geographically separate tertiary medical centres.

Results: The cohort was 51% female and had a median body mass index (BMI) of 33 kg/m2; 3% had a normal BMI. Common co-morbidities included hypertension (60%) and diabetes (35%); insulin resistance was present in 91% and advanced fibrosis in 24% of patients. When comparing patients with no fibrosis or mild fibrosis to those with advanced fibrosis, BMI ⩾28 kg/m2, age >50 years, and aspartate transaminase/alanine aminotransferase (AST/ALT) ratio ⩾0.8, a quantitative assessment check index (QUICKI) score <0.294 (equivalent to homeostatasis model assessment (HOMA) >6.2) and the presence of diabetes mellitus (DM) were individually associated by univariate analysis with odds ratios (ORs) of ⩾2.4 for advanced fibrosis. Based on the results of forced entry logistic regression analysis, three variables were combined in a weighted sum (BMI ⩾28 = 1 point, AAR of ⩾0.8 = 2 points, DM = 1 point) to form an easily calculated composite score for predicting advanced fibrosis called the BARD score. A score of 2–4 was associated with an OR for advanced fibrosis of 17 (confidence interval 9.2 to 31.9) and a negative predictive value of 96%.

Conclusions: Insulin resistance and its co-morbidities are often present in patients with NAFLD. An easily calculated score based on readily available clinical data can reliably exclude the presence of advanced fibrosis in these patients, particularly among non-diabetics.

View Full Text

Statistics from

Non-alcoholic fatty liver disease (NAFLD) is a well-recognised liver disease that is associated with underlying insulin resistance and has been linked with the growing epidemic of obesity and the metabolic syndrome. Evidence suggests that up to 30% of the United States population has NAFLD,1 while the prevalence of non-alcoholic steatohepatitis (NASH), the more progressive form of NAFLD, ranges from approximately 3%2 3 in the general population to 25–70% among morbidly obese patients.49

Data concerning the clinical characteristics of NAFLD patients have been obtained mainly from several relatively small, single-centre studies with varying results.1014 While most patients appear to be middle aged and obese at the time of diagnosis, studies differ as to a male or female predominance and the prevalence of hypertension and diabetes. A recent large, multi-centre study of 733 predominantly Caucasian patients with NAFLD from the USA, UK, Australia and Italy showed a mean age of 48 years, with a male prevalence of 53% and a mean homeostasis model assessment (HOMA) score of 5.53.15 Demographic data differentiating fatty liver patients alone from NASH patients remain poorly characterised.

Natural history studies suggest that measurable fibrosis progression occurs in 32–37% of patients over 3–6 years1618 and up to 12% of patients will progress to cirrhosis over 8–10 years.19 Identifying patients with more advanced fibrosis is important so that potential therapeutic interventions may be selectively targeted. Clinical scoring systems for identifying patients with NASH with more advanced fibrosis have been proposed,5 13 15 20 but not validated in subsequent study populations. Additionally, these scoring systems can be cumbersome and difficult to apply in everyday clinical practice.

The aim of this study was to define the clinical, laboratory and histopathological characteristics of a large group of patients with biopsy-proven NAFLD from two separate tertiary care medical centres and develop a simple, easily applied, clinical scoring system for identifying patients with advanced fibrosis with subsequent validation of this scoring system.


A retrospective analysis of patient characteristics, clinical data and histopathological information was performed on patients diagnosed with NAFLD from two separate tertiary care hospitals: Brooke Army Medical Center in San Antonio, Texas, and Saint Louis University, St. Louis, Missouri. The information was collected on all patients with available data who were determined to have biopsy evidence of NAFLD between May 2001 and September 2005. Patients were excluded from the analysis if they had a history of alcohol ingestion of >40 g per day for men or >20 g per day for women (an amount that is within the range generally accepted in differentiating non-alcoholic from alcoholic related liver disease), or evidence of viral hepatitis, autoimmune hepatitis, cholestatic liver diseases such as primary biliary cirrhosis and primary sclerosing cholangitis, drug-induced liver disease, haemochromatosis, Wilson’s disease or α1-antitrypsin deficiency. Liver biopsies were reviewed at each institution during the course of clinical care by the site pathologist. Both site pathologists followed established histopathological criteria for the diagnosis of NASH as previously published.21 A more recently described NAFLD scoring system was developed to facilitate analysis of treatment trial biopsies22 but there is no difference between this system and the older system used here in scoring advanced fibrosis. Biopsy slides were not re-reviewed for this study. A total of 426 patients were included in the analysis from Brooke Army Medical Center and 401 patients were included from Saint Louis University. Patient characteristics and clinical laboratory values were assessed in the entire cohort and then compared between patients with simple fatty liver and NASH and subsequently NASH patients with no fibrosis, stage 1–2 fibrosis, and stage 3–4 fibrosis were compared. Patients were identified as having diabetes if they received insulin or oral hypoglycaemic medications or carried a diagnosis of diabetes mellitus in their medical record. HbA1c levels were recorded. Hyperlipidaemia was defined as a total cholesterol or triglyceride level of >200 mg/dl or a history of taking antihyperlipidaemic medication. Hypertension was defined as a blood pressure of ⩾130/85 in the medical record or a history of taking antihypertensive medication. Body mass index (BMI) was calculated using the formula: weight (in kilograms)/height (in metres2). A BMI of ⩾30 was considered obese. Insulin resistance was measured by the quantitative assessment check index (QUICKI): 1/[log (fasting plasma insulin)+log (fasting plasma glucose)]. Scores of <0.35 (equivalent to a HOMA >1.8) were considered consistent with insulin resistance.

Statistical analysis

Clinical and laboratory characteristics of subjects diagnosed with NAFLD were described using frequencies (per cents) for categorical level parameters and measures of central tendency (mean and median) and variability (range and standard deviation) for continuous level parameters. Statistically significant differences in clinical and laboratory risk factors between the stages of fibrosis were determined using the χ2 test and analysis of variance (ANOVA). The Kruskal–Wallis test, an alternative to ANOVA, was used when assumptions for parametric testing were not met.

Risk assessment of various clinical parameters for advanced fibrosis was determined by univariate analysis. Receiver operator characteristic (ROC) areas under the curve (AUROC) values were generated for these variables as well. Clinical parameters found to be strongly associated with an increased risk of advanced fibrosis (OR >2.4) by univariate analysis were entered into a stepwise logistic regression model for the purpose of establishing and evaluating a composite index. A cross-validation analysis was used to evaluate and compare the number of misclassifications found using subsets of randomly selected cases from the total sample used for defining a clinical scoring system. This was performed in order to estimate the reliability of the scoring system for other datasets. Cases were randomly assigned to one of two subsets. Four cross-validation procedures were conducted. Two of the cross-validation procedures were performed by randomly assigning patients into one of two subsets (the total sample was split by randomly assigning 66% to one subset and 33% to another). Two additional cross-validation procedures were conducted by randomly assigning approximately 50% to each subset. The results of one of the cross validation procedures are detailed in this paper. Statistical significance was defined as alpha <0.05 (two-tailed). All statistical analyses were performed using SPSS version 13.0.


The median age of NAFLD patients at the time of biopsy diagnosis was 49 years (range, 17–95), with a slight majority being female (table 1). Median body mass index (BMI) was 33 kg/m2 (21–60). Only 3% of patients had a normal BMI of <25 whereas 23% of patients were overweight (BMI 25–29) and 74% were obese with a BMI of ⩾30. Extreme obesity has been defined as a BMI of ⩾40 and this characterised 16% of the cohort. The most frequently encountered clinical co-morbidity was hypertension, identified in 60% of patients, followed by diabetes with a prevalence of 35%. Insulin resistance was evaluated in a subset (n = 313) and identified in 91% of patients. What defines a normal serum alanine aminotransferase (ALT) level is debated and Prati et al23 suggest that a “healthy” upper limit for women is 19 U/l and for men is 30 U/l. Indeed, only 4.3% of women and 3.9% of men in this cohort with NAFLD on a liver biopsy had ALT levels within these gender-specific healthy ranges. If the commonly used upper limit of 40 U/l is used, then 27% (n = 109) of women and 11% (n = 41) of men would have had normal ALT levels. The median serum ALT level was 69 U/l (9–769), and serum aspartate aminotransferase (AST) was 48 U/l (11–445). Charts of the five patients with ALT and AST values >400 U/l were scrutinised for other causes of liver disease and none was identified. Median fasting glucose and triglyceride levels were 104 mg/dl (67–339) and 190 mg/dl (25–2184), respectively.

Table 1 Demographic and laboratory characteristics of patients with non-alcoholic fatty liver disease

Because distinguishing NASH from simple steatosis currently requires a liver biopsy, the clinical and laboratory features of patients with fatty liver alone on liver biopsy were compared to patients with histopathological evidence of NASH (table 2). Patients with NASH were significantly older (50 vs 47 years), heavier (BMI 34 vs 32 kg/m2), predominantly female (54 vs 39%), hypertensive (65 vs 54%), had diabetes (40 vs 15%), were insulin resistant (QUICKI 0.30 vs 0.32), and had a higher ALT (83 vs 68 U/l) and AST (62 vs 42 U/l).

Table 2 Demographic and laboratory characteristics of patients with fatty liver alone versus non-alcoholic steatohepatitis (NASH)

An essential distinction in evaluating patients with fatty liver disease is the identification of those with advanced fibrosis because these are the patients at greatest risk for developing complications of cirrhosis. A total of 669 of the 827 patients had biopsy proven NASH. Of these, 71 patients (12%) had no fibrosis (stage 0), 353 (58%) stage 1–2 fibrosis, and 182 (30%) stage 3–4 fibrosis (table 3). Comparison of the clinical and laboratory data between fatty liver alone and the different fibrosis stages of NASH demonstrated that age was significantly higher in those with stage 3–4 fibrosis (p<0.001). Women were over-represented in the group with advanced fibrosis, accounting for only 36% of those with NASH but no fibrosis and 63% of those with stage 3–4 fibrosis (p<0.001).

Table 3 Comparison of clinical data between fatty liver alone and different stages of non-alcoholic steatohepatitis (NASH) fibrosis

Although serum ALT is often used by clinicians as an indicator of more advanced or severe liver disease, the mean ALT level was 80 U/l in those with no fibrosis but was actually lower at 67 U/l in those with advanced fibrosis (p<0.001). The mean serum AST was 47 U/l in those with no fibrosis but 70 U/l in those with stage 3–4 fibrosis (p<0.001). Correspondingly, the AST/ALT ratio was elevated at 1.20 in those with stage 3–4 disease compared to 0.60 in those with no fibrosis (p<0.001). The prevalence of diabetes was increased in patients with advanced fibrosis. Whereas 23% of patients with no fibrosis had diabetes, 60% of patients with advanced fibrosis had this diagnosis (p<0.001). In parallel with this observation, HbA1c values were significantly higher in those with fibrosis (p<0.001). Interestingly, no significant differences were noted between fatty liver alone and stage 0 NASH for all potential predictors evaluated except for hypertension and QUICKI score. The prevalence of hypertension was higher among patients with stage 0 NASH when compared to those with fatty liver disease alone and the QUICKI score was lower among stage 0 NASH patients indicating a higher prevalence of insulin resistance in patients with non-fibrotic NASH compared to fatty liver disease alone.

Table 4 provides risk assessments for the major clinical parameters identified as potential risk factors for stage 3–4 fibrosis compared to fatty liver alone combined with NASH stage 0–2. Age ⩾50 years increased the risk of severe fibrosis by 2.7 whereas female gender increased the risk only 1.9-fold. Increased BMI was also a major risk factor for advanced fibrosis. Subjects who had a BMI ⩾28 kg/m2 had 2.4 times higher risk of severe fibrosis compared to those with a lower BMI. Patients with AST/ALT ratios ⩾0.8 had a 9.3 times higher risk of having severe fibrosis compared to those with lower ratios. An AST/ALT ratio of ⩾1.0 also conferred a significant risk (OR 7.75, CI 5.1 to 11.8, p<0.001) as it did in a smaller series,13 but the ratio of 0.8 was chosen as a cut-off for multivariate analysis because of its greater odds ratio by univariate analysis and by ROC analysis. The presence of diabetes also significantly increased the risk of having advanced fibrosis, much more so than the risk associated with hypertension. Haemoglobin A1c values ⩾6.0% were associated with an odds ratio of 1.8 for advanced fibrosis indicating that hyperglycemia per se was not a major risk for advanced fibrosis in this series.

Table 4 Risk assessments of clinical parameters for advanced fibrosis

To further increase the clinical utility of readily available parameters in predicting NASH fibrosis, the variables predicting at least a 2.4 times higher risk of fibrosis among NASH patients were combined to form a composite index. Three variables, BMI ⩾28, AST/ALT Ratio ⩾0.8, and Diabetes, were combined to form the BARD score and were obtained using forced-entry logistic regression analysis. The score was weighted based on the β coefficients obtained from logistic regression analysis (table 5). BMI was given 1 point, AST/ALT ratio 2 points, and the presence of diabetes 1 point. A score ranging from 0 to 4, defined by the presence of these clinical and laboratory parameters, was developed. The highest score would be in a subject with a BMI of ⩾28 kg/m2, AST/ALT ratio ⩾0.8, and with a history of diabetes mellitus. The score was then dichotomised into two categories: patients with a score of 0–1 and patients with a score of 2–4. A forward stepwise logistic regression analysis was used to predict the risk of advanced fibrosis (stage 3–4). The AUC ROC was found to be 0.81 (fig 1) with an OR of 17 (CI 9.2 to 31.9) for detecting stage 3–4 fibrosis using the BARD score. The positive predictive value (PPV) and negative predictive value (NPV) were 43% and 96%, respectively.

Figure 1 Fatty liver disease alone plus non-alcoholic steatohepatitis (NASH) with fibrosis stages 0–2 vs NASH with fibrosis stages 3–4. AUC, area under the curve.
Table 5 Forced-entry logistic regression (multivariate) of the aspartate aminotransferase/alanine aminotransferase (AST/ALT) ratio, a diagnosis of diabetes and body mass index (BMI), the three variables left standing in a backward stepwise logistic regression model as predictors of advanced fibrosis

The composite score for sensitivity and specificity (AUC), as determined by ROC analysis, for each of the validation subsets (example shown in table 1) for prediction of advanced fibrosis was compared to the AUC for the total sample (AUC = 0.81). The AUC for all of the subsets ranged from 0.81 to 0.83 suggesting a stable model for prediction of advanced fibrosis using the BARD score. The PPV and NPV for validation subset 1 was 42% and 97%, respectively. The PPV and NPV for validation subset 2 was 44% and 95%, respectively.

Subsequently, the BARD score was re-evaluated utilising a cohort of 160 patients randomly split approximately evenly between patients with isolated fatty liver and patients with NASH. The AUC was found to be 0.78 with a PPV and NPV of 27% and 97%, respectively.

Clinical and laboratory data were also assessed between NAFLD patients who had diabetes (n = 259) and those who did not (n = 486) by univariate analysis. Age ⩾50 years increased the risk of advanced fibrosis by 2.8 (CI 1.6 to 4.9) among patients with diabetes, compared with 1.8 (CI 1.2 to 2.9) for those without diabetes. A BMI ⩾28 kg/m2 had a similarly higher risk for advanced fibrosis in patients with diabetes (OR: 2.7, CI 0.9 to 8.6) compared with those without diabetes (OR: 1.8, CI 0.7 to 4.4). The most surprising finding was the OR of an AST/ALT ratio ⩾0.80 in patients without diabetes, which was found to be 15.6 (CI 8.4 to 28.9), compared with 4.3 (CI 2.5 to 7.4) for patients with diabetes. When the BARD score was applied, the ROC was 0.80 for patients without diabetes and 0.53 for patients with diabetes, likely as a result of the increased OR for AST/ALT among patients without the disease.

The BARD score was also compared to the recently published NAFLD fibrosis score in a group of 92 patients. The AUC for the NAFLD fibrosis score was found to be 0.74, with a PPV and NPV of 26% and 91%. This suggests that the easily calculated BARD score is at least equivalent to the more complex NAFLD fibrosis score in excluding patients with advanced fibrosis.


Non-alcoholic fatty liver disease is reaching near pandemic proportions in parallel with the well-documented increase in the prevalence of obesity and its other complications. While the prevalence of NASH, the more aggressive form of NAFLD, is about 15% of all NAFLD, the enormous number of people that this represents suggests that NASH is now the most common form of chronic liver disease. Because of the magnitude of this health problem and the uncertainty over appropriate screening, evaluation and treatment, it is essential that clinical data accurately describing patients with histopathological evidence of NAFLD be collected in large cohorts of patients. Furthermore, development of simple to use clinical algorithms that readily identify patients at risk for advanced fibrosis are needed.

This study represents the largest demographic and histopathological assessment of adult patients with biopsy-proven NAFLD described to date. The median age at time of diagnosis is in line with previous studies1115 and the number of women and men with NAFLD was equal. Although NASH was initially described as a disease predominantly affecting women,24 some small series challenged this idea12 and it was not confirmed in later studies.14 25 In retrospect, even very early studies of liver histopathology suggested that there would be no gender difference.26 In contrast, women may be at increased risk for having more advanced forms of NAFLD and this study found that women were over-represented in the cohort with NASH and even more so in those with NASH and advanced fibrosis.

In the current study, most patients (74%) were obese, but 3% had normal weight, confirming previous data showing that a fraction of patients with NASH are neither overweight nor obese.12 15 Insulin resistance is a common complication of obesity and is usually present in patients with NAFLD. A subset of patients was assessed for the presence of insulin resistance using the product of the fasting insulin times the fasting glucose. Expressed either as the HOMA score or the preferable inverse log-transformed QUICKI score,27 this product is a reasonable estimation of insulin resistance27 28 and has been validated in patients with NAFLD.29 Like most measures representing a spectrum from normal to morbid, specific values for the HOMA and QUICKI that distinguish healthy states from pathological insulin resistance are debated. One study suggested that HOMA values over 1.57 or equivalent QUICKI values less than 0.357 indicate pathological insulin resistance.28 Using a more stringent QUICKI of 0.35 as a cut-off suggested that 9% of patients in this cohort did not have insulin resistance as determined by a one-time measurement. Although the current paradigm is that most patients have NASH as a manifestation of insulin resistance as is found here, some disorders such as hypobetalipoproteinaemia are recognised causes of NASH30 31 but may not be associated with insulin resistance.32 Alternatively, insulin resistance may have been present at the time of the liver biopsy but not at a later time when the blood was sampled for measurement of glucose and insulin levels.

The prevalence of hypertension in this cohort of patients (60%) was higher than previous studies,10 12 15 but the prevalence of diabetes mellitus is in line with others.1015 Perhaps the most intriguing finding among the clinical data is the relatively normal median HbA1c (5.9%). This suggests that while 35% of patients are diabetic, the vast majority have “prediabetes”, in the form of hyperinsulinaemia, rather than hyperglycaemia.

Predictors of advanced fibrosis among NASH patients that have been identified in other studies vary but include age,4 7 13 20 33 obesity,7 13 20 AST/ALT ratio >1,13 the presence of significant insulin resistance,5 34 35 diabetes mellitus,7 13 14 33 hypertension,5 33 hypertriglyceridaemia,20 and an ALT >40 U/l in one study5 and >80 U/l in another.20 Our study reinforces the significance of older age, elevated BMI, elevated AST/ALT ratio, and the presence of diabetes and hypertension in distinguishing patients who have mild disease from those with more advanced stages of fibrosis. In addition, female gender was found to be an independent predictor of advanced fibrosis, in agreement with Chitturi et al.14 However, the odds ratio associated with female gender was not as high for advanced disease as other clinical variables and thus was not included in the scoring system.

The degree of insulin resistance, as indicated by lower QUICKI scores, was associated with the degree of fibrosis (table 3). This is an area of uncertainty because some studies have identified such a correlation5 34 35 while others have not.36 Experimentally, improving insulin sensitivity with thiazolidinediones decreases hepatic stellate cell activation37 and using these drugs as antifibrotics is currently being subjected to clinical investigation.

Combining a BMI ⩾28 kg/m2, an AST/ALT ratio ⩾0.8 and diabetes into the BARD score allows for the reliable identification of patients without significant fibrosis. The high negative predictive value (96%) indicates that patients not achieving a positive score are at low risk for having advanced fibrosis. Additionally, patients with a positive BARD score (⩾2 points) are 17 times more likely to have advanced fibrosis than those without a positive score. Interestingly, the BARD score seems to be more accurate at predicting advanced fibrosis in non-diabetics compared with diabetic patients. Because the BARD score gives more weight to the AST/ALT ratio, an easily remembered simple application of the score is that patients with NAFLD who either have an elevated AST/ALT ratio (>0.8), or are both obese and have diabetes, are at sufficient risk for having advanced fibrosis to warrant further evaluation. When applied to patients suspected of having NAFLD or found to have NAFLD by imaging but with a negative BARD score, further invasive evaluation with a liver biopsy may not be needed as these patients are at low risk for having advanced fibrosis, despite having clinical markers such as ALT elevations and insulin resistance. The ability to non-invasively distinguish those patients likely to have advanced disease from those less likely will enable health care providers to target those patients most in need of a liver biopsy to verify clinical suspicions and initiate appropriate therapeutic treatment strategies.

Using clinical scoring systems such as the BARD score or serum markers for liver fibrosis or a combination of clinical scores and serum markers will allow the use of percutaneous liver biopsies to become more focused, thus addressing the quandary of whom to biopsy when NASH is suspected. Care providers must keep in mind that the liver biopsy still has an important role in excluding unsuspected causes of liver disease, even when NAFLD is considered on clinical grounds. Other studies have shown that up to a third of patients suspected of having NAFLD had another cause of liver enzyme elevations identified by liver biopsy.3843 Thus if other diseases such as autoimmune hepatitis, Wilson’s disease in younger patients or haemochromatosis in older patients have not been excluded with confidence, then the liver biopsy continues to have an important role in the evaluation of these patients.

Limitations of this retrospective cross-sectional study are recognised. All study patients did not have complete laboratory data to include in the analysis and a single laboratory was not used to evaluate all patients. Consequently, not all patients have fasting insulin or glucose values, for example. Referral bias is likely present given the nature of the two academic medical centres and proclivity to attract more advanced NAFLD patients. Due to the retrospective nature of the study, as well as the large number of patient charts reviewed, the pathology slides were not re-reviewed for concordance, potentially hindering the reproducibility of the histopathological interpretation.

In summary, the present study provides a detailed descriptive clinical and histopathological analysis of a large cohort of biopsy proven NAFLD patients from two distinct regions of the United States. These data support previous smaller studies demonstrating that older age, elevated BMI, rising AST/ALT ratios, diabetes and female gender are all associated with more advanced forms of NASH and should prompt the clinician to consider a more aggressive approach to diagnosis and treatment in patients noted to have these clinical characteristics. A simple scoring system created and subsequently validated within this cohort of patients, which relies on only three of these clinical features, may assist clinicians in targeting those patients in need of liver biopsies to confirm the diagnosis and guide management, particularly among patients without diabetes.


The authors appreciate the constructive comments of Drs Elizabeth M Brunt (Washington University, St Louis) and Bruce R Bacon (St Louis University, St Louis) in designing this study and analysing the data.


View Abstract


  • See Commentary, p 1441

  • Competing interests: None.

  • Ethics approval: This study was approved by both the Brooke Army Medical Center and Saint Louis University Institutional Review Boards in June 2005 and December 2002, respectively.

  • The opinion or assertions contained herein are the private views of the authors and are not to be construed as official or reflecting the view of the US Department of the Army or the US Department of Defense.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles