Interaction of obesity, metabolic syndrome and Framingham risk on steatohepatitis among healthy Taiwanese: population-based nested case-control study

We collected subjects from January 2002 to December 2002 at one health examination center of National Taiwan University Hospital. We collected in total 5,797 adult subjects (41.4% women) from January 2001 to December 2001 from the health-screening program in a tertiary hospital in Taiwan. The protocol for this study was approved by the board of National Taiwan University Hospital, and informed consent was obtained. All participants enrolled voluntarily. This study was of a nested case control design. After excluding those with HBV and HCV viral infection, heavy alcohol drinkers, and those with a medication history of diabetes or hyperlipidemia, we defined the cases of steatohepatitis as follows: NASH (non-alcoholic steatohepatitis) was defined by ultrasound evidence of fatty liver, fatty infiltration, parenchymal liver disease, liver cirrhosis, liver function profile, and alanine tranferase (ALT) more than 1.5 fold (60 IU/mL). We defined two control groups, matched by the age and gender of the subjects. The first control group was defined by fatty liver or filtration, along with an ALT concentration less than 1.5 fold of normal range. The second control group was defined by ultrasound as having no evidence of either fatty liver or infiltration patterns. We eliminated recall bias, yielding an efficient approach with matching confounding factors such as age, sex and drinking status.

Details of medical history such as medication, hospitalization and smoking status were asked in structural questionnaires. Standardized procedures for physical examination, such as anthropometric measures and blood pressure, were performed. Blood pressure was measured by trained medical assistants, with subjects remaining in a resting position. Body mass index (BMI) was calculated as weight (in kilograms)/height (in meters) square.

The procedures for blood sampling and analytic methods have been mentioned in a previous paper [9]. In brief, blood samples were collected from each participant in a fasting status for least 12 hours. Serum total cholesterol levels were measured using the CHOD-PAP method (Boehringer Mannheim, Germany). HDL-C was measured following precipitation of apolipoprotein B-containing lipoproteins with phosphotungstic acid and magnesium ions (Boehringer Mannheim, Germany) [10].

Triglyceride concentrations were measured by the GPO-DAOS method (Wako Co., Japan). All of the lipids mentioned above were measured using an Hitachi 7450 automated analyzer (Hitachi, Japan). LDL-C concentrations were calculated using the Friedewald formula [11]. CRP was measured by automated nephelometric immunoassay using a Beckman Array instrument (Beckman Array 360 system, Canada). The peripheral blood cell counts were measured by a blood cell counter (Sysmex Cell Counter NE-8000, TOA Medical Electronics, Co., Ltd., Kobe, Japan). All the measures from both samples were carried out in a single hospital. The coefficient of variation was 5%.

Smoking status was stratified into never, abstinence and current categories. Metabolic syndrome status was defined by criteria as defined in the Third Adult Treatment Panel of the National Cholesterol Education Program [12], modified for use with BMI data and waist circumference cut points [13]. Therefore, three of following five criteria were grounds for definition: 1) blood pressure of at least 130/85 mmHg or treatment for hypertension; 2) serum triglyceride of at least 150 mg/dL; 3) HDL cholesterol of <40 mg/dL in men and <50 mg/dL in women; 4) fasting glucose of 110 mg/dL or more; and 5) waist circumference > = 90 cm in men, > = 80 in women or BMI of 27 kg/m2 or greater [14]. Framingham risk scores were calculated among the population, and the population was stratified by high-moderate risk (> = 10%) and low risk (<10%) of cardiovascular disease probability for the future 10 years. High CRP was classified as more than 0.30 mg/dL, which was the 75th percentile cut-off value among the study population. Also, high white blood cell counts and lymphocytes were specified by the top quartile in the study population.

We analyzed study subjects and two groups of controls from a cohort of the healthly check-up population. Stratified by these three groups, continuous variables were presented as the mean ± standard deviation, and categorical data were presented as contingency tables. We tested the differences by ANOVA and chi-square tests. Conditional logistic regression was applied to estimate the odds ratios and corresponding 95% confidence intervals [CI] for cases of fatty liver with abnormal liver function. Proportional odds estimation was applied if the assumption was not rejected.

The risk factors for fatty liver with abnormal liver function, after adjusting for age, gender and alcohol drinking status were determined. We estimated the prediction model ability of metabolic syndrome by the area under a receiver operating characteristic (ROC) curve [15] to separate those who were cases from those who were not, i.e. discrimination between case/control status. The area-under-curve statistics showed the probability that the model assigned a high risk to those who were cases than to those who were not. We also compared the areas under two receiver operating curves between genders graphically. Then, we calculated how closely predicted outcomes agreed with actual outcomes by calibration ability. Population attributable risks (PAR) were estimated by total population prevalence and respective odds ratios. PAR described the percentage of the risk reduction if the risk factor was eliminated from the population. This is a useful indicator to evaluate the impact on the overall occurrence of steatohepatitis status [16]. Conditional logistic regression analysis was used to estimate the odds ratio and 95% confidence interval [CI] of various atherosclerotic and inflammatory risk factors to predict steatohepatitis. Because the score test for the proportional odds assumption was not significant in ordinal categorical outcomes (chi-square = 0.003, df = 1, P = 0.96), we summarized the endpoints as binary steatohepatitis vs. non-steatohepatitis groups in high ordered interaction effects. We tested with the Hosmer-Lemeshow chi-square statistic [17], dividing the population risk for case status, plotting figures to compare the difference between predicted and observed prevalence. The small chi-square test statistics indicated good calibration, and values exceeding 20 indicated significant lack of calibration.

High-dimensional interactions of various atherosclerotic risk factors were performed by a MDR (multifactor dimensionality reduction) program [18, 19]. We included the following binary risk factors into the model: smoking, obesity, high blood pressure, high triglyceride, low HDL, high fasting glucose, metabolic syndrome, high CRP, high lymphocyte, high white cell count, and moderate-high Framingham risk. The above risk factors were highly correlated and interacted with each other. MDR methods can handle high ordered interaction effects. The model with the lowest prediction error and highest cross-validation consistency was selected for each number of factors considered. The reported cross-validation consistency was the number of cross-validation intervals of a particular risk factor combination as chosen by a MDR average across 10 runs. The average classification and prediction errors were the averages across all cross-validation intervals and all runs [19]. Statistical analyses were performed using SAS, version 9.1 (SAS Institute, Inc., Cary, NC) and STATA, version 9 (Stata Corp., College Station, Texas).

We collected a total of 124 subjects with steatohepatitis, matching by individual age, gender status, collecting 124 subjects with fatty liver only (without abnormal liver function) and 124 normal subjects. We found that there were significant differences among these three groups. Subjects with steatohepatitis had the highest body weight, body mass index, fasting glucose, blood pressure, triglyceride, cholesterol and the lowest HDL cholesterol concentrations. Also, steatohepatitis had the highest inflammation markers such as C-reactive protein, lymphocyte values, Framingham scores and predicted coronary risks. The continuous values decreased in the fatty liver group, and the lowest inflammation markers and Framingham risks were in the normal group (Table 1). The distribution of white blood cell counts and granulocytes was not statistically significant among the three groups.

The frequency distribution of various atherosclerotic risk factors and metabolic syndrome components are described in Table 2. The highest prevalence rates of smoking, hypertension, diabetes, obesity, high blood pressure, hypertriglyceridemia, hyperglycemia, low HDL and metabolic syndrome were found among the subjects with steatohepatitis. Also, the prevalence rates decreased significantly in steatohepatitis, fatty liver and normal groups. For the atherosclerotic and inflammatory risk profiles (Figure 1), we found that Framingham risk, CRP values and lymphocytes counts decreased with steatohepatitis, fatty liver only and the normal pattern.

To investigate the relative importance of risk factors on steatohepatitis status, we estimated two parameters. The OR of metabolic syndrome status was the highest (OR = 9.9), followed by high glucose status (OR = 4.5) and obesity (OR = 3.6), all statistically significant. Also, the ranking of the area under the ROC curve was similar as OR, the highest with metabolic syndrome (area = 0.80), followed by obesity (area = 0.748) and high glucose level (area = 0.726). Metabolic syndrome was the highest population attributable risk (0.59) among the atherosclerotic risk factors (Table 3).

High-ordered risk factor interaction models predicting steatohepatitis was estimated by an MDR program. For high-order risk factor interactions, MDR showed significant interaction using a three factors model, including obesity, metabolic syndrome and Framingham risk status (Table 4). Compared with other factor combinations, the three risk factor interaction model had the lesser average prediction error (22.6%), the higher average cross-validation consistency (6.3) and the lower average prediction error (24.3%). Also, we found in higher order interaction (> = 6 factors), metabolic syndrome components, instead of metabolic syndrome itself, played a significant role in predicting steatohepatitis (Table 4). We constructed the contingency tables under the final best-fitted three-risk factor model and simplified each risk factor combination into high and low risk groups (Table 5). Comparing subjects with no risk factors, OR increased as the number of risk factors increased from 1 to 3 (OR = 3.0 with one risk factor, 17.5 with two risk factors, 10.8 with three risk factors, respectively).