The incidence and risk factors of nonalcoholic fatty liver disease (NAFLD) have never been prospectively determined. To determine the frequency and risk factors of NAFLD and chronological ordering between NAFLD, weight gain, and features of insulin resistance, a historical cohort study was conducted in a Japanese workplace. A cohort free of previous liver injury, alcohol consumption of more than 140 g/wk, and hepatitis B or C infection (529 of 1,537 subjects), and a subcohort of 287 subjects free of insulin resistance–related features were identified. Elevated aminotransferases in nonalcoholics were used as a surrogate for NAFLD. High aminotransferases together with weight gain of more than 2 kg and insulin resistance–related features in the subcohort were sought for up to 5 years. The incidence of high aminotransferases was 31 per 1,000 person-years (71 events). A significant interaction occurred between age and sex in the development of high aminotransferases. In subjects younger than age 40 years, male sex (hazard ratio [HR]: 4.6), elevated body mass index (HR: 2.1), hypertension (HR: 2.6), and low high-density lipoprotein cholesterol (HR: 2.8) increased the risk of high aminotransferases, whereas age (HR: 0.6 for each 5 years) decreased the risk. In subjects older than age 40 years, glucose intolerance (HR: 5.3) was the only significant risk factor. In the subcohort, weight gain preceded high aminotransferases and other insulin resistance–related features, which appeared sequentially in order of low high-density lipoprotein cholesterol, hypertriglyceridemia/hypertransaminasemia/hypertension, and glucose intolerance. In conclusion, this cohort study clearly showed chronological ordering and an association between development of elevated aminotransferases and risk factors of NAFLD. (HEPATOLOGY 2005;41:64–71.)
Nonalcoholic fatty liver disease (NAFLD) is recognized as one of the most common causes of chronic liver diseases in industrialized countries. NAFLD is histologically characterized by excess accumulation of lipids in hepatocytes, either with or without inflammatory cell infiltration, necrosis, and fibrosis. NAFLD complicated by necroinflammation with or without fibrosis is termed nonalcoholic steatohepatitis,1 which is one of the major causes of cryptogenic cirrhosis.2, 3
The true frequency of NAFLD in the general population remains unknown. The prevalence of histologically confirmed fatty liver is only available from autopsy studies, and the estimates range from 13% to 24%.4–6 However, detailed information about alcohol consumption was not available in these reports, and the definition of fatty liver varied among studies. The prevalence of NAFLD determined by ultrasonography or laboratory data reported in the general population ranges from 2.8% to 13%.7–9 However, no prospective studies have been conducted in healthy populations to address the incidence of NAFLD.
Many cross-sectional or case series studies show that several conditions are strongly associated with NAFLD, including obesity,8, 10, 11 diabetes mellitus,10, 12 hyperlipidemia,8, 13 hypertension,14, 15 and insulin resistance.13, 15, 16 However, none of these relationships are based on prospective studies; therefore, the chronological ordering among them remains unclear.
The diagnosis of NAFLD is based on histological evidence of steatosis or fatty infiltration proven by imaging tests. However, approximately 80% to 90% of hypertransaminasemia may be explained by NAFLD once other causes such as chronic viral hepatitis and alcohol-induced liver disease are excluded.12, 17 Therefore, nonalcoholic hypertransaminasemia, in which viral or other causes of liver disease are also excluded, has been used as a noninvasive surrogate marker for NAFLD.7, 9 Thus, we used data on aminotransferases to characterize the prevalence, incidence, and risk factors for NAFLD in a cohort of healthy employees in a Japanese government office. The aims of this study were (1) to estimate the prevalence and incidence of nonalcoholic hypertransaminasemia, (2) to prospectively investigate the risk factors for development of nonalcoholic hypertransaminasemia, and (3) to assess the chronological ordering between nonalcoholic hypertransaminasemia, weight gain, and insulin resistance–related clinical features in a healthy population with a robust health management system.
NAFLD, nonalcoholic fatty liver disease; HDL, high-density lipoprotein; BMI, body mass index; ATP, Adult Treatment Panel.
Patients and Methods
Study Design and Data Source.
Data used in this study were obtained between 1997 and 2002 as part of routine health care for employees in a Japanese government office. Using these data, we conducted a historical cohort study to address our aims. Most of the employees work in sedentary positions or with only mild physical tasks related to government administration. The employees receive comprehensive yearly health examinations, including anthropometric measurements such as height, body weight, and body fat composition, blood pressure, laboratory tests including most routine blood work, a complete medical history and physical examination, and hepatitis B antigen and hepatitis C antibody levels every 5 years. The information is collected in a database established in 1997. The database includes (1) yearly health checkup data with the results of all subsequent examinations; (2) health-related information including past, present, and family medical history; and (3) lifestyle-related information such as alcohol consumption, smoking status, and exercise habits. The information in the database is updated at each annual health examination by questionnaire and physician interview. The total number of employees who had annual examinations from 1997 to 2001 was 1,546. The study population included those whose age was between 20 and 60 years at entry (n = 1,537). Hypertransaminasemia was defined as serum aspartate aminotransferase or alanine aminotransferase above the upper limit of normal; that is, >40 IU/L for aspartate aminotransferase and >35 IU/L for alanine aminotransferase. The term hypertransaminasemia is used in this article to refer to nonalcoholic hypertransaminasemia likely attributable to NAFLD.
The prevalence of hypertransaminasemia was defined as the number of persons who presented with hypertransaminasemia not associated with alcohol consumption (>140 g/wk) or other common liver diseases (including positive hepatitis B surface antigen or hepatitis C virus antibody) at entry divided by the total study population (1,537 subjects). The 95% confidence interval was calculated by Wilson's score method.
To determine the incidence of hypertransaminasemia between 1997 and 2002, we defined the cohort free of liver injury or disease by excluding those who had positive hepatitis B surface antigen or hepatitis C antibody (n = 45), those whose weekly alcohol consumption was greater than 140 g (n = 393), and those who had other common liver diseases (n = 7). We also excluded those who were evaluated only once without subsequent follow-up (n = 76) and those with previous diagnoses of fatty liver disease by imaging (n = 257) or history of hypertransaminasemia (n = 230) (Fig. 1). Therefore, the employees who had a diagnosis of fatty liver with any kind of imaging tests or history of hypertransaminasemia before entry, regardless of the time of the diagnosis (recent or long time previously), were excluded from this study. We then followed all employees for the development of hypertransaminasemia by using yearly health checkup data. The follow-up time in months was calculated by subtracting the date of yearly examination at the entry year from the date of yearly examination at which hypertransaminasemia was first detected. The employees who moved from the office for any reasons during the study period were censored at the last yearly examination that they took in the office. The cases that were diagnosed with any other specific liver diseases by subsequent follow-up were excluded and censored at the last yearly examination at which the liver enzymes were still normal.
To calculate cumulative incidence of hypertransaminasemia, we performed Kaplan-Meier analysis and expressed the data as proportions with 95% confidence intervals. The incidence of hypertransaminasemia was expressed per 1,000 person-years as well. The 95% confidence interval for the incidence (per 1,000 person-years) was from a formula using the assumption that the number of events follows a Poisson distribution.18
The risk factors for incident nonalcoholic hypertransaminasemia were prospectively assessed. We collected the following information about the study subjects at entry by using the database: history of hyperlipidemia, hypertension, low high-density lipoprotein (HDL) cholesterol, and glucose intolerance (or diabetes mellitus); anthropometric measurements (body mass index [BMI] and body fat composition); data on serum triglycerides, HDL cholesterol, total cholesterol, fasting blood sugar, and blood pressure; weekly alcohol consumption, smoking habits, regular exercise activity, and family history of diabetes mellitus. BMI was classified into two groups: less than 25 kg/m2 or 25 kg/m2 or more. A cutoff value of 25 kg/m2 is used to define obesity in the Japanese population.19 Body fat composition was measured by impedance methods after 8 hours of fasting and classified into two groups by using the upper limit of normal; that is, 20% for men and 25% for women. Alcohol consumption was classified into two groups: never or less than 70 g/wk, and between 70 g/wk and less than 140 g/wk. Smoking habits were classified into two groups: current nonsmoker (includes previous smokers), or current smoker. Exercise was classified into two groups: no exercise or less than once per week, and regular exercise equal to or more than once per week. History of metabolic disease and corresponding abnormal laboratory findings at entry (by using the criteria of the Adult Treatment Panel III [ATP III]20) were combined and analyzed as a single variable. The variables of hyperlipidemia, low HDL cholesterol, glucose intolerance/diabetes mellitus, hypertension, and family history of diabetes mellitus were analyzed as binary variables.
To analyze risk factors for development of hypertransaminasemia, we used Cox proportional hazards regression for each individual risk factor and developed models with multiple variables. First, the risk factors were individually analyzed. The model was then selected through backward elimination. In developing the model, the significance level of .2 was applied for removal of the variables. The magnitudes of the risks are expressed as hazard ratios with 95% confidence intervals. To assess interaction between age group and the other variables, we performed Cox proportional hazard regression in each age group (ages 20-39, ages 40-59). The interaction between age groups and each of the other variables were then assessed in Cox proportional hazards regression models by using interaction term in the model.
To determine the chronological ordering between hypertransaminasemia, body weight gain, and insulin resistance–related clinical features, we also defined a subgroup of people (287 of 529) by excluding 242 individuals who had any history of either hyperlipidemia (n = 179), low HDL cholesterol (n = 55), hypertension (n = 62), or glucose intolerance/diabetes mellitus (n = 74) alone or in combination at entry (Fig. 1). We followed them until their health checkups in 2002 for hypertransaminasemia as defined previously, body weight gain equal to or more than 2 kg from the baseline, elevation of serum triglycerides (>150 mg/dL), low HDL cholesterol (<40 in men and <50 in women), high blood pressure (>130/85 mm Hg), or elevation of fasting blood sugar (>110 mg/dL). The 2-kg cutoff value was chosen for weight gain because it is indicative of true weight gain as opposed to day-to-day fluctuation and yet is small enough to detect the onset of weight gain. Furthermore, 2 kg was the median value of body weight gain in the total study population. The definitions of hypertriglyceridemia, low HDL cholesterol, hypertension, and glucose intolerance were based on the criteria of the ATP III.20 The follow-up time to each event was calculated as noted previously.
To assess the chronological ordering between any pair of the outcomes, 2 × 2 tables of each pair of outcomes were constructed by comparison of the time to event (follow-up time). If the difference in follow-up time was 0 or could not be determined because of censoring of the earlier outcome, the subject was omitted. We compared the proportion of one outcome preceding the other with 0.5 by using a binomial test to determine whether the precedence occurred more than would be expected by chance. We also used a Cox proportional hazards model with multiple endpoints per subject and robust variance estimation to account for the multiple events.21 Chronological ordering of hypertransaminasemia versus the other endpoints was expressed as hazard ratios of each endpoint against hypertransaminasemia.
For all analyses in this study, we used JMP statistical software version 4.0.4 (SAS Institute Inc., Cary, NC) and S-PLUS 6.0.3 Release 2 for Windows (Insightful Corp., Seattle, WA) and considered differences statistically significant when the P values were less than .05.
Between 1997 and 2001, 529 individuals of the study cohort were identified in the population of 1,537 subjects (women, n = 185; men, n = 1,352) as shown in Fig. 1: 480 in 1997, 26 in 1998, 8 in 1999, 7 in 2000, and 8 in 2001. Using the definition of the study cohort, 1,008 (66%) were excluded. The characteristics of the study population are summarized in Table 1. Follow-up time ranged from 10 to 65 months (median, 60).
Table 1. Characteristics of the Study Population
Total n = 529
20s–30s n = 377
40s–50s n = 152
NOTE. Shown as means ± standard deviations or proportions in the columns. Data were missing in body fat composition (n = 61), total cholesterol (n = 158), triglycerides (n = 158), HDL cholesterol (n = 160), blood glucose (n = 159), alcohol consumption (n = 76), cigarette (n = 36), and exercise (n = 44), which are not counted in each calculation.
35 ± 8
30 ± 4
46 ± 4
Sex (female), %
Body mass index (kg/m2)
22 ± 3
22 ± 3
22 ± 2
Body fat composition (%)
22 ± 5
22 ± 5
21 ± 5
Blood pressure (mm Hg)
112 ± 14
111 ± 13
114 ± 16
71 ± 10
70 ± 10
74 ± 12
Total cholesterol (mg/dL)
189 ± 34
183 ± 31
199 ± 35
100 ± 53
95 ± 54
108 ± 50
HDL cholesterol (mg/dL)
56 ± 15
55 ± 15
56 ± 14
Blood glucose (g/dL)
88 ± 8
87 ± 7
90 ± 10
Alcohol consumption (70–140 g/wk), %
Cigarette (current smoker), %
Exercise (regular exercise), %
Family history (history +), %
The prevalence of hypertransaminasemia at entry was 9.3% (143 of 1,537; 95% CI: 7.8%, 11.0%) in the population: 10.5% (95% CI: 8.8%, 12.4%) in men and 0.5% (95% CI: 0.0%, 3.0%) in women. The prevalence in the age groups was 9.3% (95% CI: 7.2%, 11.8%) in the 20 to 39 age group and 9.3% (95% CI: 7.3%, 11.7%) in the 40 to 59 age group (Fig. 2A).
The incidence of hypertransaminasemia was 31 per 1,000 person-years (71 events; 95% CI: 21, 44). In comparison between different age groups, the cumulative incidence at 60 months was 14.7% (95% CI: 11.0%, 18.8%) in the 20 to 39 age group and 8.1% (95% CI: 4.6%, 14.1%) in the 40 to 59 age group, as shown in Fig. 2B.
By univariate analysis, the incidence of hypertransaminasemia was significantly associated with age, male sex, high BMI, high body fat, low HDL-cholesterol, and cigarette smoking (Cox proportional hazards regression, P < .05) (Table 2).
Table 2. Association Between Risk Factors and Incident Hypertransaminasemia in the Total Cohort and the Age Groups
NOTE. Age is a continuous variable and the others are binomial variables.
No. events, number of events; no. at risk, number of subjects at risk at entry; IR, incidence rate (×102, during total study period); HR, hazard ratio; 95% CI, 95% confidence interval.
The diagnosis before or at entry.
P value of interaction term in the model composed of age-group, each of listed variables, and interaction term (age group* the variable).
There was no event in subjects with glucose intolerance/diabetes mellitus, and the Cox proportional hazard model did not converge. There were missing data in body fat composition (n = 61), alcohol consumption (n = 76), cigarette (n = 36), and exercise (n = 44).
There was significant interaction between age groups and sex, hypertension, glucose intolerance/diabetes mellitus, and cigarette smoking, as shown in Table 2 (P < .05). Therefore, separate multivariable Cox proportional hazard regression models were created for the 20 to 39 age group and the 40 to 59 age group. The model chosen in the 20 to 39 age group is shown in Table 3. Male sex (HR: 4.7; 95% CI: 1.4, 15.3), elevated body mass index (HR: 2.2; 95% CI: 1.2, 4.0), hypertension (HR: 2.5; 95% CI: 1.1, 5.5), low HDL cholesterol (HR: 3.4; 95% CI: 1.8, 6.7), and regular exercise (HR: 1.9; 95% CI: 1.0, 3.4) were significantly associated with higher risk of hypertransaminasemia, whereas older age (every 5 years, HR: 0.6; 95% CI: 0.4, 0.9) was associated with lower risk for hypertransaminasemia, after adjusting for the other variables. In contrast, in the 40 to 59 age group, only glucose intolerance/diabetes mellitus was significantly associated with higher risk of hypertransaminasemia (HR: 5.3; 95% CI: 1.3, 21.2, after adjustment for male sex) (Table 3).
Table 3. Adjusted Hazard Ratio of the Risk Factors in the Different Age Groups
HR (95% CI)
HR (95% CI)
NOTE. Age is a continuous variable, and the others are binary variables.
The cumulative incidence of hypertransaminasemia, body weight gain, and insulin resistance–related clinical features is shown in Fig. 3. Binomial tests for each pair of events showed that weight gain preceded any other outcome (P < .0001 for each test) (Table 4). Low HDL-cholesterolemia preceded hypertransaminasemia (P < .05). Hypertransaminasemia preceded hyperglycemia (P < .0001). In multivariate Cox model analysis, the hazard ratios of the outcomes against hypertransaminasemia were 5.9 (P < .0001) for body weight gain, 1.5 (P = .053) for low HDL cholesterolemia, and 1.4 (P = .087) for hypertriglyceridemia, all of which are more likely (or tend) to occur than hypertransaminasemia for a given individual at any given time, and 0.22 (P < .0001) for glucose intolerance, indicating that it is less likely to occur than hypertransaminasemia (Fig. 3).
Table 4. Chronological Ordering Among Hypertransaminasemia, Body Weight Gain, and Insulin Resistance–Related Clinical Features
Body Weight Gain
Low HDL Cholesterol
NOTE. This table shows the results of binomial tests based on 2 × 2 tables for each pair of the outcomes, which were constructed by the comparison of the time to event (follow-up time in months) in each individual. In each pair of comparison, the variable in the first column indicates an earlier event, whereas the variable in the first row indicates a subsequent event. The numbers in each cell indicates the number of subjects preceding divided by the total number (percentage of preceding cases) after omitting the cases in which the follow-up time of both events was same or in which an earlier event was censored. The percentages of preceding cases in parentheses were then compared with .5, which P values are shown by symbols as follows:
Our results indicate that the prevalence of non-alcoholic hypertransaminasemia in a healthy Japanese population was 9.3% and the overall incidence of hypertransaminasemia was 31 cases per 1,000 person-years. In people 20 to 39 years of age, male sex, elevated body mass index, hypertension, low HDL cholesterol, and regular exercise increased the risk of developing hypertransaminasemia, whereas older age decreased the risk. In contrast, in people older than age 40 years, glucose intolerance/diabetes mellitus was the only significant factor associated with the incidence of hypertransaminasemia. Finally, we found that weight gain and low HDL cholesterolemia precede hypertransaminasemia and that hypertriglyceridemia, hypertransaminasemia, and hypertension appeared nearly simultaneously and preceded hyperglycemia. This cohort study prospectively addressed the frequency and risk factors of nonalcoholic hypertransaminasemia. This study also assessed the chronological ordering between body weight gain, hypertransaminasemia, and insulin resistance–related clinical features in a healthy population.
The prevalence of nonalcoholic hypertransaminasemia (as a surrogate of NAFLD) recently reported in the US population was 2.8% to 5.4%,7, 9 which is much lower than the frequency in our study. Our study was conducted in the population of 20- to 59-year-olds with a higher proportion of men, in whom NAFLD is more prevalent,7, 8 and the subjects in our cohort worked in sedentary occupations. Thus, these may partially explain our higher prevalence. A recent population-based study showed that the prevalence of nonalcoholic hypertransaminasemia was different among different ethnic groups even after adjusting for other confounders.9 Thus, difference in ethnicity also could be an explanation. Nevertheless, the point prevalence defined by a single blood test may underestimate the prevalence of NAFLD because the level of serum transaminases could fluctuate and fall within normal ranges at given times, and NAFLD could exist without the elevation of aminotransferases.22–24 Thus, the true prevalence and incidence of NAFLD could be even higher than the estimates determined by hypertransaminasemia.
Our results showed men to be more likely to develop nonalcoholic hypertransaminasemia in the younger (20-39 years of age) study participants. This finding is contrary to the finding based on several referral practice-based studies,25, 26 in which women dominated among patients with nonalcoholic steatohepatitis, which is a more advanced stage of NAFLD. In a recent population-based study, however, the prevalence of hypertransaminasemia was slightly higher in men (5.7% vs. 4.6%).7 The relative risk of men to women in developing hypertransaminasemia was 3.7 (95% CI: 1.6-8.4) in our study, which is higher than the calculated odds ratio of 1.3 based on the previous US population-based study.7 Because our cohort contains more subjects in the 20- to 39-year age group than in the 40- to 59-year age group, and working women may have different characteristics than homemakers (healthier), our estimate may be artificially higher than the true relative risk in the general population. Also, our laboratory did not apply sex-specific cutoffs of aminotransferases, and this may to some extent explain our higher estimate. In the other recent population-based study, the sex difference was eliminated by adjusting for waist–hip ratio.9 Thus, other confounders unmeasured in our study may explain our higher estimate. The lack of sex effect (rather opposite effect) in the group older than age 40 years may be explained by the loss of a protective effect of female hormones.
In this study, we found an age difference in the development of hypertransaminasemia; younger age was associated with a greater rate of developing hypertransamiansemia. The prevalence of hypertransaminasemia was similar between age groups. Because we selected our study cohort to be free of any preexisting liver injury, the older population in our study cohort was a more “selected” population compared with the younger population, and may be much healthier than the average population in the same age group.
A significant association existed between regular exercise and increased risk of hypertransaminasemia. One possible explanation for this association would be the fact that a substantial number of subjects who developed hypertransaminasemia in the office had already received health consultation (recommendation of diet and regular exercise) as a standard practice when they gained significant weight, which preceded hypertransaminasemia and features of metabolic syndrome.
In this study, we also showed the chronological ordering between hypertransaminasemia, weight gain, and insulin resistance–related features. In most cases, weight gain preceded hypertransaminasemia and all insulin resistance–related features. The development of hypertransaminasemia followed low HDL cholesterolemia, appeared approximately simultaneously with hypertriglyceridemia and hypertension, and preceded glucose intolerance. The presence of low HDL cholesterol together with hypertriglyceridemia has been shown to relate to insulin resistance and is used as the conventional marker for the dyslipidemia of obesity and the metabolic syndrome.27, 28 Therefore, we interpret our findings to indicate that hypertransaminasemia likely due to NAFLD may appear during development of insulin resistance. In this study, body weight gain clearly preceded the development of hypertransaminasemia and components of the insulin resistance syndrome. Although our study could not address whether weight gain causes NAFLD or metabolic syndrome, previous literature has provided strong evidence favoring causal relationship between obesity and these conditions. Because body weight gain occurred more in the younger generation (data not shown), body weight control in the younger generation may be more efficient in prevention of NAFLD, and probably cryptogenic cirrhosis as well as the metabolic syndrome.
Our study has limitations related to the study population and methodology. Our study population is homogenous in terms of their ethnicity, ages, sex (male-dominant population), and possibly socioeconomic status compared with the general population. Because the prevalence of NAFLD may differ among ethnic groups, cautious inference may be required about our frequency estimates. Another limitation is that we used elevation of transaminases as a surrogate without a proof of fatty liver; considering that NAFLD may exist without elevation of liver enzymes, our frequency estimates could have been underestimated. Also, we did not apply sex-specific cutoffs of aminotransferases. This could be a potential drawback and may cause underestimation of frequency of hypertransaminasemia in women and artificially raise the HR of male sex in development of hypertransaminasemia.
In summary, we prospectively determined the frequency and the risk factors of nonalcoholic, nonviral hypertransaminasemia, most certainly due to NAFLD, and the chronological ordering between weight gain, hypertransaminasemia, and insulin resistance–related clinical features in a healthy ethnically homogenous population. Our data showed that weight gain clearly precedes the development of hypertransaminasemia and features of metabolic syndrome (insulin resistance) syndrome. This study may have important implications in understanding the pathophysiology of NAFLD and designing preventive strategies as well as future studies of this condition.