We sought to develop a clinically useful index comprising standard and physiologically relevant variables to predict the probability of significant hepatic fibrosis in subjects with chronic hepatitis C virus (HCV) infection. Fibrosis was graded as mild (stages F0 or F1) or significant (stages F2–F4). Thirty-five clinical and laboratory parameters were analyzed initially in 176 patients with detectable HCV RNA to derive a fibrosis probability index (FPI) to predict significant fibrosis. This index then was validated in a second group of 126 subjects. Among 18 variables associated with severe fibrosis on univariate analysis, multiple logistic regression analysis identified age, aspartate aminotransferase (AST), total cholesterol level, insulin resistance (by homeostasis model), and past alcohol intake as independent predictors of significant fibrosis. The area under the receiver operating characteristic (ROC) curves was 0.84 for the initial cohort and 0.77 for the validation cohort. In the initial cohort, the sensitivity of the FPI based on these five predictors was 96%, and the negative predictive value was 93% at a score of ≥0.2. At scores ≥0.8, the FPI was 94% specific and had a positive predictive value of 87%. In conclusion, an FPI using routinely assessed markers and incorporating a measure of insulin resistance can reliably predict the probability of significant hepatic fibrosis in most patients with chronic HCV infection. Such an index should prove useful to guide decision making regarding the need for liver biopsy, and potentially for avoiding or deferring biopsy in a large proportion of patients with mild liver disease. (HEPATOLOGY 2004;39:1239–1247.)
Chronic infection with the hepatitis C virus (HCV) runs a protracted course, and symptoms that impair quality of life are not related to the severity of the hepatic fibrosis. Although the outcome of chronic hepatitis C infection is variable, between 4% and 22% of patients progress to cirrhosis over 20 years.1 The extent of hepatic fibrosis is the major determinant of adverse clinical outcomes in chronic HCV infection; liver failure, hepatocellular carcinoma, and portal hypertension are almost exclusively confined to those with stage 3 or stage 4 fibrosis.2–4 Conversely, although impressive advances have been achieved in the efficacy of antiviral therapy, this is based on interferon-based therapies that have a high rate of unpleasant or dangerous adverse effects and involve considerable cost and inconvenience. Hence, the decision to treat or not to treat an individual patient requires a careful consideration of the risk of cirrhosis versus the likelihood of response.
The National Institutes of Health, French and Asia-Pacific consensus guidelines currently recommend antiviral therapy for those with moderate to severe stages of fibrosis,5–7 whereas those with mild (stage 1) or no fibrosis reasonably can remain untreated. Liver biopsy remains the gold standard for the assessment of hepatic fibrosis, but is a procedure associated with discomfort and rare but serious complications. None of the currently available noninvasive markers for hepatic fibrosis are sufficiently accurate to replace liver biopsy, and some involve specialized biochemical tests.8–10 A recent study identified readily assessable variables to predict the absence of severe fibrosis.11 However, those with alcohol intake >30 g/day were excluded, thereby limiting the usefulness of the index in clinical practice. In another recent study, Wai et al.12 suggested that the aspartate aminotransferase (AST)/platelet ratio index (APRI) may be useful in the prediction of hepatic fibrosis stage. However, previous studies did not consider the influence of certain pertinent clinical, metabolic, and viral factors germane to fibrosis progression in the liver. In particular, host factors including the body mass index, waist/hip ratio, and insulin resistance, which may influence hepatic fibrogenesis, were not considered.13–17 These factors affect hepatic steatosis, which has been well documented as a correlate of fibrosis progression in chronic HCV infection.13–15 Likewise, interaction between long-term alcohol ingestion, hepatic steatosis, and fibrosis progression have been documented,18 but these were not considered in detail by earlier studies.9, 11, 12, 19
We undertook this study to assess the usefulness of a large number of routinely available clinical and laboratory indices to predict the severity of hepatic fibrosis in a clinical practice setting. The purpose was to derive a useable index that could predict the presence or absence of significant fibrosis and to demonstrate how such an index could be applied according to different clinical settings, such as the liver clinic setting with a high prevalence of significant fibrosis, and the community where the prevalence may be lower. A key feature of this study was to include metabolic (anthropometric indices, insulin resistance) and toxic (alcohol) factors that may accelerate fibrosis and other factors (antioxidant level, use of complementary and alternative medicine) that may reduce fibrosis progression.20–24
HCV, hepatitis C virus; FPI, fibrosis probability index; AST, aspartate aminotransferase; ROC, receiver operating characteristic; APRI, AST/platelet ratio index; HOMA-IR, insulin resistance by the Homeostasis model assessment; PPV, positive predictive value; NPV, negative predictive value.
All consenting patients infected with HCV who underwent a liver biopsy at two teaching hospitals of the University of Sydney (Westmead Hospital, May 1999 through August 2002, and Royal Prince Alfred Hospital, March 2001 through July 2002) were enrolled prospectively. The protocol was approved by the Human Ethics Committees of the Western Sydney Area Health Service, the Central Sydney Area Health Service, and the University of Sydney. Written informed consent was obtained. All patients were positive for HCV antibody by third-generation enzyme-linked immunosorbent assay and had detectable serum HCV RNA. Other causes of chronic liver disease were excluded, including concurrent hepatitis B virus infection (i.e., positive for hepatitis B surface antigen), drug-induced liver disease, autoimmune hepatitis, primary biliary cirrhosis, primary sclerosing cholangitis, hemochromatosis, α-1 antitrypsin deficiency, and Wilson's disease, as described elsewhere.25 Patients with human immunodeficiency virus coinfection and diabetes mellitus receiving insulin therapy were excluded. No patient had clinical evidence indicative of portal hypertension (splenomegaly or varices) or hepatic decompensation (jaundice, ascites, encephalopathy, or variceal hemorrhage) at the time of biopsy.
Sociodemographic data and clinical and historical details were collected, including country of birth, age at infection and age of biopsy, estimated duration of infection, risk factor(s) for acquiring HCV infection, average alcohol intake (g/day) in the 6 months preceding liver biopsy, and alcohol intake in the distant past (more than 6 months before liver biopsy). The major risk factors for HCV infection assessed were intravenous drug use, blood transfusion, tattoos, and body piercing. Cases with no obvious risk factor were termed sporadic. The duration of infection was estimated from the transfusion date or the date of first reported parenteral exposure to that of liver biopsy; it could not be calculated in patients with sporadic infection. Estimates of alcohol intake were made at patient interviews by two experienced physicians. For the purpose of analysis, patients were graded for alcohol intake as follows: grade 0 (<10 g alcohol/day), 1 (10–40 g/day), or 2 (>40 g/day).
The use of complementary and alternative medicines was assessed by direct interview, including separate consideration of antioxidant preparations, such as vitamin preparations that included vitamin E, fish oil, or Silymarin. A person was deemed a regular user of antioxidants if they used any of these compounds at least three times weekly.
Data on waist/hip ratio (waist circumference at umbilicus/hip circumference at the maximal circumference over the buttocks), body mass index, peripheral stigmata of chronic liver disease, and presence of hepatomegaly were recorded. Body mass index was calculated as weight in kilograms/(height in meters)2.
Laboratory Tests of Liver Disease and Virological Markers.
Serum alanine aminotransferase, AST, γ-glutamyl transpeptidase (GGT), albumin, globulin (total protein minus albumin level), bilirubin, platelet count, international normalized ratio, plasma glucose, and total cholesterol were determined on the day of liver biopsy by automated procedures in the clinical pathology laboratory. Blood was collected after a 12-hour overnight fast. APRI was calculated as 100 × AST/platelet count (109/L; AST was expressed as ratios of upper limit of normal as described by Wai et al12). Serological analysis for hepatitis B surface antigen and core antibody was performed by enzyme-linked immunosorbent assay (Sanofi Diagnostic Pasteur, Marne-la-coquette, France). Hepatitis C virus genotyping was performed with a second-generation reverse hybridization line probe assay (Inno-LiPA HCV II; Innogenics, Zwijndrecht, Belgium), and HCV RNA quantitation was carried out by Roche Amplicor HCV monitor version 2.0 (Roche Diagnostics, Branchburg, NJ). We categorized viremia into two groups (low vs. high viral load), with high viral load being >850,000 IU/mL. Serum insulin was determined by radioimmunoassay (Phadaseph insulin RIA; Pharmacia and Upjohn Diagnostics AB, Uppsala, Sweden). Serum c-peptide was estimated by a competitive immunoassay (IMMULITE; Diagnostic Products, Los Angeles, CA). Insulin resistance was calculated from fasting serum insulin and plasma glucose determinations using the homeostasis model assessment (HOMA-IR) method: fasting insulin (mU/mL) × plasma glucose (mmol/L) / 22.5.26 Insulin resistance calculated by this method correlates closely with the gold standard of hyperinsulinemic euglycemic clamp method.27 Red cell glutathione was measured by reverse-phase high-performance liquid chromatography,28 whereas high-performance liquid chromatography with ultraviolet detection was used for vitamin E determination.29
Hepatic histological results were scored semiquantitatively as described by Scheuer.30 Fibrosis was scored as follows: F0, no fibrosis; F1, portal fibrosis; F2, periportal or portal septa but intact architecture; F3, architectural distortion but no obvious cirrhosis; and F4, definite cirrhosis. Fibrosis was classified as negligible or minimal (F0 or F1) or significant (F2–F4), because F2 is generally chosen as the threshold for antiviral therapy.7
The two patient cohorts studied were the initial cohort used to develop the predictive index for significant fibrosis and a second (validation) cohort in whom the derived index was assessed prospectively. Logarithmic transformation was applied to input variables that exhibited significant departure from normality. In the initial cohort, univariate logistic regression was used to quantify the association between each of the 35 clinical and laboratory variables (Table 1) and the presence of significant hepatic fibrosis (F2–F4). Using the univariate predictors as input variables, multiple logistic regressions with forward stepwise selection of variables identified the independent predictors of significant fibrosis in the best fitting model. A fibrosis probability index (FPI) to calculate the predicted probability of significant fibrosis was constructed using this model. Receiver operating characteristic (ROC) curves were constructed for this index to illustrate its diagnostic ability in identifying those with significant fibrosis in both the initial and the validation cohorts. The areas under the ROC curves were computed. The sensitivity, specificity, positive predictive value (PPV), and negative predictive values (NPV) were computed for a variety of score cutoff points of the FPI. All analyses were carried out using the statistical software package SPSS for Windows, version 10 (SPSS Inc., Chicago, IL). A significance level of 5% was used throughout.
Table 1. Clinical and Laboratory Variables Assessed in Relation to the Stage of Hepatic Fibrosis in Univariate Models
Insulin resistance by the homeostasis model assessment.26
Age at biopsy, gender, country of birth, ethnicity, age at infection, duration of infection, risk factor for acquiring HCV infection, use of antioxidants, current and past average alcohol intake, waist/hip ratio, body mass index, peripheral stigmata of chronic liver disease, and presence of hepatomegaly
Liver function tests (serum ALT, AST, AST/ALT ratio, γ-glutamyl transpeptidase, albumin, globulin, bilirubin, and international normalized ratio), platelets, APRI,* glucose, insulin, c-peptide, HOMA-IR,† total cholesterol, vitamin E, glutathione, ferritin, and anti-hepatitis B core antibody
HCV genotype and HCV RNA titer
Characteristics of Study and Validation Cohorts.
The study involved 302 consenting subjects, including 176 in the initial cohort and 126 in the validation cohort to formulate the FPI. Baseline characteristics of the initial cohort are listed in Tables 2 and 3. Significant fibrosis was evident in 84 patients (48%) in the initial cohort, and 10 patients (6%) had cirrhosis. In the validation cohort, significant fibrosis was present in 74 patients (59%), and 17 patients (13%) had cirrhosis.
Table 2. Clinical Characteristics of the Initial Cohort of HCV-Infected Patients According to Fibrosis Stage* (n = 176)
Mild Fibrosis (F0 or F1, n = 92)
Significant Fibrosis (F2–F4, n = 84)
Values are expressed as frequency or mean ± SD.
38 ± 10
44 ± 8
21 ± 7
22 ± 7
Duration of infection
17 ± 10
21 ± 8
Country of birth
Present alcohol intake
Past alcohol intake
Intravenous drug use
Tattoos, body piercing
Use of antioxidants
Body mass index
25.9 ± 4.5
26.9 ± 5.1
0.9 ± 0.1
0.9 ± 0.1
Peripheral stigamata of chronic liver disease
Table 3. Laboratory Characteristics of the Initial Cohort of HCV-Infected Persons According to Fibrosis Stage* (n = 176)
Mild Fibrosis (F0 or F1, n = 92)
Significant Fibrosis (F2–F4, n = 84)
Abbreviation: ALT, alanine aminotransferase.
Values are expressed as median (interquartile range) or frequency.
Univariate analysis revealed that age at biopsy, duration of infection, past alcohol intake (>6 months preceding), hepatomegaly, alanine aminotransferase, AST, γ-glutamyl transpeptidase, globulin level, bilirubin, platelet count, APRI, ferritin, plasma glucose, serum insulin level, HOMA-IR, total cholesterol, viral load, and genotype were all associated with the stage of hepatic fibrosis (Tables 2 and 3). Using data from 170 subjects (the remaining six had missing values), multiple logistic regression analysis identified age at biopsy, AST, HOMA-IR, total cholesterol level, and past alcohol intake as the independent predictors for significant fibrosis (F2–F4; Table 4). An FPI was constructed using these independent predictors to determine the probability (0.0–1.0) of a patient having significant fibrosis.
Table 4. Best-Fitting Multiple Logistic Regression Model for the Prediction of Significant Hepatic Fibrosis (Stages F2–F4) in the Initial Cohort* (n = 170†)
95% CI for Odds Ratio
Abbreviations: LnAST, logc AST.
Fibrosis Probability Index = e*/1 + e*, where * = −10.929 + (1.827 × LnAST) + (0.081 × Age) + (0.768 × past alcohol use graded as 0–2) + (0.385 × HOMA-IR) − (0.447 × Cholesterol).
The remaining six subjects had missing variables.
LnAST (per U/L)
Age (per year)
HOMA-IR (per unit)
Past alcohol intake
Cholesterol (per 1 mmol/L)
Diagnostic Accuracy of the FPI.
Based on the 170 subjects in the initial cohort, the area under the ROC curve for the FPI was 0.84 (Fig. 1A), which represents the overall diagnostic accuracy of the model in predicting significant fibrosis. Thus, on random selection of an individual with significant fibrosis versus mild fibrosis, the score of the former will be greater than that of the latter individual 84% of the time.
Table 5 illustrates the sensitivity, specificity, PPV, and NPV of the FPI for the initial cohort at different cutoff levels. At scores of 0.2 and higher, the index had a high sensitivity (96%) and powerful NPV (93%). Thus, 80 of the 83 patients with significant fibrosis in the initial cohort were correctly identified using 0.2 as the cutoff value; there were only three false negatives (none of these patients had cirrhosis). Although the specificity was low, this still would have translated to avoiding liver biopsies in 38 of the 87 persons (44%) with mild fibrosis (22% of the entire cohort). However, if a higher predictive score were to be used as the cutoff for biopsy, the specificity of the index improves markedly. Thus, at a score of ≥0.8, the specificity (94%) and positive predictive value (87%) of the index was high. Using this value, liver biopsy would have been avoided in 82 of the 87 patients with mild fibrosis. In the 42 patients (25% of total) with this score, there were only five false positives.
Table 5. Diagnostic Accuracy of the FPI in Predicting Significant Fibrosis (Stage 2–4) for the Initial Cohort at Various Cutoff Points*
Cutoff Point of the FPI
Prevalence of significant fibrosis, 48%.
Testing the Validity of the FPI.
The FPI was evaluated prospectively in a separate cohort of 126 persons (59% had significant fibrosis). The ROC curve for the index in the validation cohort had an area under the curve of 0.77 (Fig. 1B). Table 6 illustrates the sensitivity, specificity, PPV, and NPV of the index in the validation cohort. As with the initial cohort, the specificity and PPV of scores ≥0.8 was high (98% and 97%, respectively). Using the lower cutoff value of 0.2, the index had a sensitivity of 85% and an NPV of 69%. Liver biopsy would be avoided in 25 of the 52 patients (48%) with mild fibrosis (20% of the entire validation cohort). At this score, 11 of 73 patients with significant fibrosis also would be predicted to have mild disease; all had stage 2 fibrosis. The specificity of the index in the validation cohort improved to 98% at the cutoff score of ≥0.8. At this value, liver biopsy could have been avoided in 51 of the 52 patients with mild disease.
Table 6. Diagnostic Accuracy of the FPI in Predicting Significant Fibrosis (Stages F2–F4) for the Validation Cohort at Various Cutoff Points*
Cutoff Point of the FPI
Prevalence of significant fibrosis, 59%.
In this study, a fibrosis probability index was constructed using five routinely assessed historical and laboratory parameters in patients with chronic hepatitis C being treated at a hospital-based liver clinic. The markers selected by multivariate regression were age at biopsy, previous alcohol intake, serum AST, total cholesterol, and HOMA-IR. Univariate analysis had identified 18 factors associated with significant hepatic fibrosis, but several are interrelated, such as serum AST, ALT and GGT, blood glucose, serum insulin, c-peptide, and HOMA-IR. Although viral load and genotype were associated with advanced fibrosis, they were not independent predictors in the multivariate model, findings that are consistent with numerous published reports.3, 31–35 Although the relationship of age, alcohol intake, and total cholesterol level to fibrosis stage have been noted previously,3, 11, 32–37 inclusion of insulin resistance is a highly novel and potentially important finding. The FPI derived from these five markers was shown to have sufficiently high sensitivity and specificity to aid clinical decision making regarding the need for and timing of liver biopsy.
Given the protracted and uncertain natural history of chronic HCV infection and the incomplete efficacy (∼50%) of current antiviral viral therapy, together with its adverse effects and costs, current treatment recommendations for HCV infection target those at increased risk for adverse clinical outcomes. In turn, this is related to the stage of hepatic fibrosis, which is determined principally by the duration of infection, host factors, and exposure to exogenous compounds that could influence hepatic injury and fibrogenesis. Among host factors, the metabolic milieu, in particular obesity, insulin resistance, and hepatic steatosis, can serve as cofactors in accelerating liver injury.13–17 Of the environmental factors, alcohol consumption adversely influences disease outcome,3 whereas anecdotal evidence suggests that complementary and alternative medicines, particularly antioxidant intake, may be hepatoprotective, especially in forms of liver injury that are mediated by oxidative stress (e.g., alcoholic liver disease, nonalcoholic fatty liver disease, and HCV).21–24, 38 Further, in several community and patient populations, 30% to 40% of patients ingest some form of complementary and alternative medicines21; this was confirmed in the present study. The present prospective study was undertaken in an attempt to derive practical clinical guidelines for liver biopsy in liver clinic populations that take into account all known variables that could influence disease severity.
Insulin resistance and past alcohol consumption both were common in the HCV-infected population, and both were shown to be important in the prediction of fibrosis stage. Recent studies have demonstrated that insulin resistance is associated with more rapid fibrosis progression in chronic HCV infection.16, 17 Hyperinsulinemia, the hallmark of insulin resistance, directly can stimulate hepatic stellate cells mitogenesis and collagen synthesis39 and causes the upregulation of connective growth factor.40 In contrast, intake of complementary and alternative medicines was not shown to have an independent impact on hepatic fibrosis. These data strongly suggest that any physiologically relevant fibrosis predictive model applicable to clinic patients with chronic HCV infection must consider insulin resistance as well as alcohol intake. The FPI constructed here has the advantage of not requiring additional less common laboratory assessments and appears physiologically relevant. Other studies have included simple liver function tests and more sophisticated markers of matrix turnover to predict fibrosis stage. However, the ability of these markers (AST/alanine aminotransferase ratio, serum hyaluronic acid, matrix metalloprotease-2, tissue inhibitors of metalloprotease-1 and -2) to predict milder stages of hepatic fibrosis is poor.41–43 In a recent large study, the area under the curve of the ROC using more sophisticated biochemical markers (including α2-macroglobulin, apolipoprotein A1, and haptoglobin) was 0.733 to 0.76619; in this respect, the ROC of the FPI for both our initial and validation cohorts compared favorably (area under the curve, 0.84 and 0.77, respectively).
In clinical practice, the present index can be used to aid decision making on the timing of and need for liver biopsy. In turn, the application and the best-suited cutoff of the predicted probability score will depend on the prevalence of mild versus significant fibrosis in the cohort of interest and the clinical question. This study involved subjects from a liver clinic setting with a high prevalence of significant fibrosis (approximately 50%), similar to the prevalence noted in other reports from liver clinic.3, 31, 37 In this setting, most patients routinely would be recommended to undergo a liver biopsy. Hence, the role of the FPI in this setting is to identify those with a low risk of significant fibrosis to avoid unnecessary biopsies. A low cutoff value (high sensitivity and NPV) should be chosen, and subjects with an FPI of less than 0.2 would not be recommended to undergo a liver biopsy. When this is applied to our two cohorts, we would have been able to avoid unnecessary biopsies in 44% to 48% of the subjects with mild fibrosis (20%–22% of the entire cohort). Moreover, none of the subjects with stage 3 or 4 fibrosis in the validation cohort would have been misclassified. As reported earlier, in only 2% of patients with stage 2 fibrosis did liver-related complications develop over 5 years, and none of these patients required transplantation or died of liver-related causes.4 This therefore may be regarded as a less critical error than failing to identify patients with F3 or F4 fibrosis (cirrhosis).
In a recent study by Wai et al.,12 the application of a novel index, the APRI, in the liver clinic setting (47% with significant fibrosis) resulted in a similar percentage of unnecessary biopsies being avoided in their cohort. We included APRI in our analysis and found that it was not an independent predictor of significant fibrosis in the best-fitting multivariate model. The correlation coefficient of APRI with the stage of fibrosis was poor (Spearman's r = 0.4; P < .001) and the area under the curve for the ROC of APRI in predicting significant fibrosis was 0.76 in our initial cohort, compared with an area under the curve of 0.84 using the FPI. The AST level reflects both the activity and stage of HCV infection,44 and there is a large overlap of platelet count values between those with mild and significant fibrosis.12 Hence, a scoring system that relies only on these two variables, without considering other important clinical data such as age, was shown to have limitations in the present patient cohort.
In a community setting, the prevalence of significant fibrosis is low and the rate of fibrosis progression seems to be slower.1, 33 Many such patients would not be referred for a liver biopsy or already harbor a disinclination for liver biopsy. In this setting, the role of the FPI would be to select out the minority of individuals who actually do have significant fibrosis. Hence, a high score (high specificity and positive predictive value) should be chosen; at scores of ≥0.8, the FPI has a specificity of 94% to 98 %. Forns et al.11 recently constructed another predictive model using readily available laboratory markers, including GGT, cholesterol, and platelet count, to predict the absence of significant fibrosis in such a population. In their validation group (the prevalence of significant fibrosis was 26%), using the higher cutoff point, the positive predictive value was 66%. In comparison, although the present FPI was developed in a setting with a high prevalence of significant fibrosis, the positive predictive value (using a high cutoff point of ≥0.8; Table 5) would be better at 73% if it is applied to such a cohort (with prevalence of significant fibrosis at 26%). We have also applied Forns et al.'s predictive formula to our initial cohort, and the area under the curve of the ROC was 0.76 compared with 0.84 using our index. Unlike the FPI, Forns et al. did not include measures of important profibrotic determinants, such as body mass index or insulin resistance.13–17 Further, patients with a reported intake of alcohol >30 g/day were excluded, limiting the usefulness of the index in most Western populations of HCV-infected patients. By contrast, 38% of subjects in the initial cohort of our study had alcohol intake >40 g/day, similar to some other published cohorts of HCV subjects.33, 45, 46 It is likely that the documented differences in the rate of fibrosis progression in the community and liver clinic cohorts partly may relate to the prevalence of subjects with heavy alcohol intake in these cohorts. Hence, the inclusion of measures of alcohol intake confers greater accuracy to this index and enables it to have a much broader application.
It should be noted that, as illustrated by the differences observed in the area under the ROC curve values in our initial and validation cohorts, FPI indices depend integrally on patient numbers, the mix of fibrosis stages, and possibly other unknown etiopathogenic factors (genetic and environmental) that may be different in geographically distinct HCV populations. Thus, large centers seeking to derive similar noninvasive indices of hepatic fibrosis ideally need to derive their own formula and continually refine the model.
In conclusion, we describe a predictive model for assessing the probability of significant hepatic fibrosis in a hospital liver clinic population that was derived from a prospective analysis of all known factors that could influence disease progression in chronic HCV infection, including the newly recognized metabolic determinants. The fibrosis probability index demonstrated good sensitivity and specificity and is derived from standard, readily assessable laboratory and clinically relevant parameters. The usefulness of the index in the longitudinal monitoring of patients with mild disease and its ability to assist in the timing of biopsy needs to be examined prospectively.
The authors thank Prof. Daya Naidoo, Dr. Ora Lux, and Mr. Chris Salonikas (Department of Clinical Chemistry, Prince of Wales Hospital, Sydney) for performing the glutathione analyses; Dr. Karen Byth for advice in statistical analysis; and Ms. Keshni Lata Sharma, Jasmin Canete, and Seng Kee Teo for their assistance in data collection.