Potential conflict of interest: Nothing to report.
Physical activity (PA) is commonly recommended for nonalchoholic fatty liver disease (NAFLD) patients. However, there is limited evidence on the independent role of PA in NAFLD. The aim of this study was to examine the association between PA and NAFLD. We conducted a cross-sectional study of a subsample (n = 375) of the Israeli National Health and Nutrition Survey. Exclusion criteria were any known etiology for liver disease. Participants underwent an abdominal ultrasound examination; biochemical tests, including leptin, adiponectin, and resistin; and the noninvasive biomarker SteatoTest and anthropometric evaluations. A semiquantitative food frequency questionnaire and a detailed PA questionnaire were administered. Three hundred forty-nine patients (52.7% men, 30.9% primary NAFLD) were included. The NAFLD group engaged in less aerobic, resistance, or other kinds of PA (P ≤ 0.03). The SteatoTest was significantly lower among subjects engaging in any PA or resistance PA at least once a week (P ≤ 0.01). PA at least once a week in all categories was associated with a reduced risk for abdominal obesity. Adjusting for sex, engaging in any kind of sports (odds ratio [OR] 0.66, 95% confidence interval [CI] 0.44-0.96 per 1 standard deviation increment in PA score) and resistance exercise (OR 0.61, 95% CI 0.38-0.85) were inversely associated with NAFLD. These associations remained unchanged after adjusting for homeostasis model assessment, most nutritional factors, adiponectin, and resistin. Only the association with resistance PA remained significant with further adjustment for body mass index (OR 0.61, 95% CI 0.44-0.85). Adding leptin or waist circumference to the model eliminated the statistical significance. Conclusion: Habitual leisure-time PA, especially anaerobic, may play a protective role in NAFLD. This association appears to be mediated by a reduced rate of abdominal obesity. (HEPATOLOGY 2008;48:1791-1798.)
Nonalcoholic fatty liver disease (NAFLD) has been recognized as a very prevalent disease that can progress to cirrhosis1 and represents a major health burden. However, a widely accepted treatment has yet to be found.
Insulin resistance, obesity and especially abdominal obesity represent the most important risk factors for the development of NAFLD.2–4 Because lifestyle modification, including weight reduction and physical activity (PA), has been shown to reduce many of the risk factors for NAFLD,5 it has become the primary treatment modality.6 While the benefits of gradual weight reduction in NAFLD are well established,7, 8 the role of PA as a potential treatment for NAFLD has been tested in only a few studies with a small number of patients, usually in combination with diet and weight reduction.9, 10 Thus, the specific effects of PA per se warrant further study. The results of one study demonstrated a significant reduction in aminotransferase levels in patients with NASH who adhered to an aerobic exercise program, regardless of weight loss.11 In a recent cross-sectional study in male patients, cardio-respiratory fitness was found to have an inverse association with prevalence of NAFLD, irrespective of body mass index (BMI).12 In another cross-sectional study, an inverse correlation was found between habitual physical activity and intrahepatic fat content. This association did not change after adjusting for accepted risk factors, but abdominal obesity was not measured in that study.13
To date, epidemiological evidence for an independent association between physical activity and NAFLD is limited. Furthermore, no study has adjusted for nutritional intake, which may be an important confounder. The present study is a large-scale population-based study designed to explore the role of PA in NAFLD and to identify the type of PA that should be recommended.
ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; CI, confidence interval; GGT, gamma-glutamyltransferase; HOMA, homeostasis model assessment; NAFLD, nonalchoholic fatty liver disease; OR, odds ratio; PA, physical activity; SD, standard deviation.
Subjects and Methods
The study population consisted of 799 participants (aged 24-70 years), a subgroup from the First Israeli National Health and Nutrition Survey (the MABAT Survey).14 As described elsewhere,4, 15, 16 375 subjects were recruited into the study, after which individuals with any of the following were excluded from it: alcohol consumption ≥30 g/day in men or 20 g/day in women (measured by a detailed questionnaire), presence of HBsAg or anti-HCV antibodies, fatty liver suspected to be secondary to hepatotoxic drugs, inflammatory bowel disease, prior surgery that could cause fatty liver, or celiac disease.
All patients underwent measurements of weight, height, and waist and hip circumference following a uniform protocol as described4, 15, 16 and BMI was calculated (20-25 kg/m2).
Each participant underwent biochemical testing following a 12-hour fast for liver enzymes, serum lipid profile, and fasting serum glucose and serum insulin levels. The homeostasis model assessment (HOMA) was calculated as fasting serum insulin (μU/mL) × fasting plasma glucose (mmol/L) / 22.5. Frozen serum samples from all participants were stored at −80°C until analyses were conducted for leptin, adiponectin, and resistin (using commercially available human adipocytokines ELISA kits [Linco Research, St. Charles, MO]) and for the noninvasive biomarker SteatoTest for the quantitative assessment of liver steatosis (BioPredictive, Paris, France).17
Fatty liver was diagnosed via abdominal ultrasound using standardized criteria.18 Ultrasound was performed in all subjects with the same equipment and by the same operator as described.4, 15, 16
A face-to-face interview was conducted by the same trained interviewer. The questionnaire included demographic data, health status, current alcohol intake, and PA. A second questionnaire was a detailed semiquantitative food frequency questionnaire adapted to the Israeli population and composed of 111 food items with specified serving sizes as described.15 Both questionnaires were developed by the Food and Nutrition Administration, Ministry of Health.
PA Questionnaire: Structure and Analysis.
The detailed PA questionnaire assessed leisure physical activity over the past year. It included the type of PA (running, cycling, swimming, aerobic exercise in a fitness room, dancing, instrumental and noninstrumental resistance training, and light exercise such as yoga), frequency and duration, and length of time that the subject engaged in PA. A score was determined that took into account the length of time that the subject engaged in PA. It was calculated as a multiple of minutes per week for each of four time periods as follows: up to 6 months (minutes/week × 1); 7 months to 1 year (× 2); 1-5 years (× 3); and more than 5 years (× 4). This score, as a type of weighted average, lent greater importance to physical activities that were performed for a longer length of time. According to the scientific statement from the American Heart Association, due to major differences in physiological responses during aerobic exercise compared with resistance-anaerobic exercise, these two general types of activities need to be dealt with separately when developing exercise recommendations,19 so we also performed separate analyses for aerobic PA and the resistance-anaerobic PA.
Statistical analyses were performed using SPSS version 13 software (SPSS Inc., Chicago, IL). Continuous variables are presented as the mean ± standard deviation (SD). To test differences in continuous variables between the two groups, the independent samples t test or the Mann-Whitney test were performed. Associations between nominal variables were performed using the Pearson chi-square test. A multivariate logistic regression analysis was performed to test the adjusted association between PA and NAFLD or between PA and abdominal obesity (waist circumference >88 cm in women and >102 cm in men). P < 0.05 was considered statistically significant for all analyses.
Characteristics of the Study Population and Comparison Between Subjects Diagnosed as NAFLD versus Normal Liver (Table 1)
Table 1. Characteristics of the Study Population and a Comparison Between Patients with NAFLD and Individuals with Normal Livers (Mean ± SD, Unless Otherwise Stated)
Parameter (Normal Range)
Total Population (n = 349)
NAFLD (n = 108)
Normal Liver (n = 241)
Mean age (years)
50.7 ± 10.4
51.4 ± 9.8
50.4 ± 10.7
Male sex (%)
Mean BMI (kg/m2)
27.2 ± 4.5
30.2 ± 4.6
25.8 ± 3.8
0.32 ± 0.2
0.47 ± 0.2
0.25 ± 0.2
Any type of PA (%)
Aerobic PA (%)
Resistance PA (%)
Both types of PA (%)
Three hundred forty-nine volunteers who met the inclusion and exclusion criteria were included in the analysis. The prevalence of primary NAFLD, diagnosed via ultrasound, was 30.9% (n = 108) (95% confidence interval [CI] 26%-36%). Detailed information on the study population has been described elsewhere.4, 15, 16
Three hundred and forty serum samples were available for the SteatoTest. The adipocytokines tests were completed in 343 serum samples for leptin, and 337 for adiponectin and resistin.
The percentage of subjects engaging in any kind of PA was 37.0% (31.6 % men and 42.6% women). Aerobic PA was practiced by 28% (26.7% men and 29% women) and resistance PA by 20% (11.8% men and 29% women).
The NAFLD group had a significantly higher percentage of men and a higher BMI (P < 0.001 for both). There was no significant difference in age between the groups.
The NAFLD group had significantly fewer subjects who engaged in resistance PA (13% versus 23%, P = 0.03). In the entire study population, 10.9% (n = 38) engaged in both aerobic and resistance PA. Of these, only 4.6% were in the NAFLD group, compared with 13.7% in the normal liver group (P = 0.04). The results would have been similar if a SteatoTest cutoff of 0.3 (above S0) had been used, rather than ultrasound, to classify by NAFLD or normal. The rate of performance of physical activity in all categories was lower in subjects with SteatoTest above 0.3, reaching statistical significance for resistance PA (P = 0.001) and for any type of PA category (P = 0.01), but still not for aerobic PA (P = 0.14).
The NAFLD group, compared with subjects with normal livers, engaged in less than half the amount of any kind of PA, a little more than half the amount of aerobic PA and only one-third the amount of resistance PA (Fig. 1). Similar results were observed when the score was compared between groups (P ≤ 0.03 for all).
Comparison Between Subjects Who Engaged in PA Less than Once a Week versus at Least Once a Week (Table 2)
Table 2. Comparison Between Subjects Who Engaged in PA Less than Once a Week (<1) Compared with at Least Once a Week (≥1)
Any Type of PA
<1 (n = 229)
≥1 (n = 120)
<1 (n = 262)
≥1 (n = 87)
<1 (n = 280)
≥1 (n = 69)
Values are expressed as the mean ± SD unless otherwise stated.
The percentage of men engaging in any kind of PA or resistance PA weekly was lower than women. No age differences were observed. In all PA categories, the prevalence of NAFLD tended to be lower in subjects who engaged in PA weekly, but the difference reached statistical significance only in the resistance category (18.8% versus 33.9%, P = 0.02).
BMI was significantly lower among subjects engaging in any type or resistance PA weekly. The rate of abdominal obesity was almost half in subjects engaging in any category of PA at least once a week. After adjusting for age, sex, and BMI, the high frequency of PA in all categories was still associated with a reduced risk for abdominal obesity (Fig. 2).
Serum triglycerides were lower among subjects engaging in any type or resistance PA at least once a week. Fasting serum insulin and glucose or HOMA did not differ between the groups (P ≥ 0.07, data not shown).
No significant difference was observed in serum levels of alanine aminotransferase (ALT) and aspartate aminotransferase (AST) (data not shown). However, gamma-glutamyltransferase (GGT) levels were lower among subjects with a higher frequency of PA in all categories (14.6 ± 12.2 versus 16.6 ± 11.8, P = 0.02 for all PA; 14.0 ± 9.0 versus16.6 ± 12.8, P = 0.04 for aerobic PA; 14.4 ± 14.4 versus 16.3 ± 11.3, P = 0.01 for resistance PA).
The Association Between PA and Steatosis by SteatoTest.
A significant negative correlation was found between the SteatoTest and the score of any type of PA (r = −0.12, P = 0.03) and with the score of resistance PA (r = −0.12, P = 0.03), but not with score of aerobic PA (P = 0.12). The SteatoTest was significantly lower only among subjects engaging in any type of PA or resistance PA at least once a week (Table 2).
The Association Between PA and Adipocytokines.
Because there were significant sex differences in leptin and adiponectin levels (P < 0.001), the associations were categorized by sex as depicted in Table 2. Serum resistin levels were significantly lower among subjects who engaged in resistance PA at least once a week, only when men and women were analyzed together (24.9 ± 16.4 versus 34.7 ± 32.0, P = 0.04).
A multivariate analysis was performed to test for an association between PA and adipocytokines levels above the fourth quartile, after adjusting for sex and age. The performance of resistance PA at least once a week was associated with higher levels of serum adiponectin (odds ratio [OR] 1.84, 95% CI 1.004-3.36). Any kind of PA (OR 0.52, 95% CI 0.28-0.94), aerobic PA (OR 0.46, 95% CI 0.23-0.92), and resistance PA (OR 0.44, 95% CI 0.22-0.89) at least once a week, were associated with lower levels of serum leptin.
Multivariate Analysis of the Association Between PA Measurements and NAFLD (Table 3)
Table 3. Multivariate Analysis of the Independent Association Between NAFLD and Physical Activity Score
Parameters Adjusted for in the Model
Model 1: Sex
Model 2: Sex and BMI
Model 3: Sex and HOMA
Model 4: Sex and Leptin
Model 5: Sex and Adiponectin
Model 6: Sex and resistin
Values are expressed as OR (95% CI) unless otherwise stated.
Score of any type of PA (413.2 [1 SD])
Score of aerobic PA (321.6 [1 SD])
Score of resistance PA (162.5 [1 SD])
Adjusting for sex, engaging in any type of PA, and resistance PA were inversely associated with NAFLD. The association with resistance exercise remained significant with further adjustment for BMI, but the association with any type of PA became borderline significant.
To further identify possible biological mechanisms, the association between PA and the prevalence of NAFLD was examined after adjusting for sex plus potential mediators. Adjusting for adiponectin, resistin, and HOMA/fasting serum insulin did not significantly modify the association between PA and prevalence of NAFLD when each was put into the regression model individually. However, adding leptin or waist circumference to the model eliminated the statistical significance of the PA–NAFLD relationship in all categories (P ≥ 0.07 for all).
The Association Between PA and NAFLD After Adjusting for Nutrition
We previously reported15 an association between intake of soft drinks and all types of meat and increased risk for NAFLD. Indeed, subjects who engaged in PA less than once a week had a higher consumption of soft drinks and meat compared with at least once a week (Table 2). Therefore, these factors were included in the multivariate analysis as potential confounders (along with sex). Adjusting for consumption of soft drinks or meat did not affect the association between the score of resistance PA and NAFLD (OR=0.61 per 1 SD, [0.44-0.85, P=0.04] for both). The association with the score of any type of PA was not affected by adjusting for meat consumption (OR=0.66 per 1 SD, [0.44-0.96, P=0.04]), but became borderline significant after adjusting for soft drink consumption (P=0.055). The association with the score of aerobic PA remained non-significant. No significant association was found between the frequency of PA in all categories and dietary intake of calories, carbohydrates, total fat or protein (data not shown). However, subjects who engaged in any kind of PA or aerobic PA less than once a week had a higher consumption of saturated fat and a lower consumption of monounsaturated fat as a percentage of total fat. Subjects who engaged in resistance PA less than once a week had a higher consumption of polyunsaturated fat and a lower consumption of monounsaturated fat as a percentage of total fat (Table 2). Adding the percent of saturated/monounsaturated/polyunsaturated fat intake into the multivariate analysis did not affect the association between any kind or resistance PA with NAFLD (P ≤ 0.03 for all, data not shown).
Comparison Between NAFLD Patients Who Engaged in PA versus NAFLD Patients Who Did Not (Table 4)
Table 4. Comparison Between Patients with NAFLD Who Engaged in Any Type of PA with Those Who Did Not Perform
PA (n = 67)
Sedentary (n = 41)
Values are expressed as the mean ± SD unless otherwise stated.
Mean age (years)
52.4 ± 10.1
49.7 ± 9.1
Male sex (%)
Abdominal obesity by waist circumference (%)
30.0 ± 4.4
30.7 ± 4.8
ALT (U/L) (5-39)
26.2 ± 14.3
28.3 ± 12.0
AST (U/L) (5-40)
24.4 ± 6.9
23.6 ± 5.9
GGT (U/L) (6-28)
16.7 ± 7.3
23.3 ± 16.9
Glucose (mg/dL) (70-110)
98.8 ± 21.9
98.4 ± 27.0
Insulin (μU/mL) (5-25)
25.7 ± 9.5
32.0 ± 16.2
Triglycerides (mg/dL) (50-175)
143.9 ± 65.2
163.4 ± 67.5
Among the NAFLD patients, those who engaged in PA (of any kind and amount) had lower fasting serum insulin levels and a lower rate of abdominal obesity even though they had a similar BMI. Moreover, PA remained associated with a lower rate of abdominal obesity irrespective of BMI, sex, and age (OR 0.25, 95% CI 0.07-0.93). Serum GGT levels were significantly lower among NAFLD patients engaging in PA, but no difference was observed in ALT and AST levels.
PA has long been considered a cornerstone of a healthy lifestyle. Although its protective role in cardiovascular and metabolic diseases is well established,19 its place and importance in NAFLD still requires scientific support and clarification. This is the first epidemiological study testing the association between NAFLD and lifestyle from the perspective of both nutritional intake and habitual PA.
In the present study, NAFLD patients were shown to engage in less physical activity. Regular weekly PA was associated with a lower prevalence of NAFLD that reached statistical significance only in the resistance PA category. This difference was accompanied with lower BMI, abdominal obesity, and serum triglycerides. GGT was also lower among subjects who engaged in PA at a higher frequency in all categories. GGT was previously demonstrated to be independently associated with PA20 in a large sample of British women. Furthermore, the present study demonstrated that a higher physical activity score is associated with a lower rate of NAFLD, implying that this relationship goes beyond the effect of potential mediators such as adiponectin, resistin, insulin resistance, and nutritional factors.
Subjects engaging in regular weekly PA tended to have a healthier diet in terms of fat composition and a lower soft drinks intake. Nutrition in general, fat composition,21, 22 and carbohydrates15, 23, 24 in particular, have been shown before to be associated with NAFLD. Thus, when evaluating the role of PA in NAFLD in observational studies, it is imperative to adjust for nutritional habits. In the present study, the association remained unchanged after adjusting for most nutritional factors including soft drinks and meat consumption that were shown to be risk factors for NAFLD in the same population. Only the consumption of soft drinks weakened the association between any kind of PA and NAFLD, rendering it borderline significant, but the association with resistance PA was not affected. The strongest mediators in the association between PA and NAFLD appeared to be body fat measures, waist circumference, and serum leptin, which is strongly linked to fat mass in humans.25
Any type of PA was associated with a reduced risk for NAFLD adjusting for sex, but the association reached only borderline significance after adjusting for BMI. Resistance exercise was associated with a reduced risk for NAFLD after adjusting for sex and with further adjustment for BMI. Surprisingly, aerobic PA in itself had the weakest association with NAFLD. This may stem from the use of self-reported PA that could lead to misclassification bias. Because participants were informed of their ultrasound results only after they completed their questionnaires, there is no reason to believe that NAFLD was the source of a differential recall bias, so a true association might only be underestimated. Furthermore, occupational activity can have a significant effect on the amount of daily activity. In this study, occupational activity was evaluated only in general terms by the question: “Regarding your level of physical activity at work, what main sort of activity is involved (sitting, standing, walking, light physical activity, heavy physical activity?)” and no differences were found among them. Therefore, this study did not attempt to cast doubts on the known health benefits of aerobic PA.26 However, the results do emphasize the importance of resistance training. Adding leptin or abdominal obesity to the multivariate model eliminated the statistical significance of the PA–NAFLD relationship in all PA categories.
The association between PA and NAFLD was also shown to be somewhat quantitative, because a significant negative correlation was found between the SteatoTest, a biochemical surrogate for liver steatosis,17 and the score of any type of PA and with the score of resistance PA. Steatotest was also significantly lower among subjects engaging in any type of PA or resistance PA at least once a week. In support of this finding, a recent study reported an inverse correlation between the habitual physical activity index and the intrahepatic fat content.13
The association of PA with adipocytokines may explain, at least in part, the PA–NAFLD association. The adipocytokines, leptin, adiponectin, and resistin have all been implicated in the development of hepatic steatosis, inflammation, and fibrosis in patients with NAFLD. Serum leptin levels27 and resistin28, 29 positively correlate with hepatic steatosis, whereas plasma adiponectin levels are inversely associated with steatosis.30
In the present study, resistance PA was significantly associated with higher levels of adiponectin and lower levels of leptin. Any kind of PA or aerobic PA were significantly associated with lower levels of leptin.
Adjusting for adiponectin and resistin did not affect the PA–NAFLD association, but adjusting for leptin eliminated the association. It is not surprising that leptin is a mediator in the PA–NAFLD association, because leptin serum concentrations have been shown to be reduced by long-term aerobic exercise and resistance exercise.31–34 Thus, exercise may improve leptin resistance,31, 32 which has been shown to be associated with the pathogenesis of hepatic steatosis.35
All categories of PA were associated with a lower rate of abdominal obesity irrespective of BMI, sex, and age. Similarly, NAFLD patients that engaged in physical activity had lower rates of abdominal obesity compared with sedentary patients, although no difference was seen in BMI. When abdominal obesity was added to the multivariate analysis, the PA–NAFLD association was eliminated, suggesting that central adiposity is one of the major mediators of the potential benefit associated with higher levels of PA. This observation is supported by other studies in which abdominal obesity was found to be a mediator of the PA–NAFLD association. In a random sample of British women, ALT levels decreased linearly with increasing duration of moderate or vigorous physical activity. However, the association of physical activity with ALT was weakened after adjusting for measures of adiposity, either BMI or waist.20 In a recent study, an inverse association was found between cardio-respiratory fitness categories and the prevalence of NAFLD. Whereas fitness and BMI were independent of each other in their associations with the prevalence of NAFLD, the addition of waist circumference to the regression model attenuated the association.12 This is in line with the fact that abdominal obesity has been shown to be a major risk factor for NAFLD, of greater importance than BMI,4, 36 and is consistent with previous studies demonstrating that exercise-induced weight loss is associated with a preferential reduction in abdominal fat37, 38 and that, at any given weight, individuals who exercise more have less visceral fat than those who are sedentary.39
The suggested effect of PA on NAFLD may stem from other mechanisms as well. Exercise alone, in the absence of any change in body weight or composition, may enhance insulin sensitivity and glucose homeostasis.40 PA appears to result in insulin-receptor up-regulation in muscle tissue and hence increased delivery of glucose and insulin to the muscles.41 Exercise also has a beneficial effect on free fatty acid metabolism by enhancing whole-body lipid oxidation.42 Hepatic triglyceride accumulation was shown to decrease with exercise intervention43 and hepatic free fatty acid uptake was lower in trained compared with untrained subjects.44
An interesting finding in the present study is the somewhat stronger association of resistance PA with NAFLD. In recent years, increasing attention has been paid to resistance training as a useful adjunctive tool of exercise.19, 45 A recent study showed that resistance training, without a concomitant weight loss diet, significantly improved insulin sensitivity and fasting glycemia and decreased abdominal fat.46 Tsuzuku et al.47 demonstrated that noninstrumental resistance training, using body weight as a load, appears to be effective in decreasing visceral fat and improving metabolic profiles, without weight loss. The results of a randomized trial comparing the effect of aerobic versus resistance training on coronary risk factors demonstrated that only the resistance training group showed a reduction in total body fat with an associated increase in lean body mass.48 A meta-analysis comparing aerobic training with weight training concluded that weight training resulted in greater increases in fat-free mass.49 An increase in muscle mass may improve insulin sensitivity by increasing the available glucose storage area, thereby reducing the amount of insulin required to maintain a normal glucose tolerance.50
The 2007 update of the American Heart Association dealing with resistance exercise concludes that resistance training should be viewed as a complement to aerobic exercise.19 This recommendation should probably be considered for NAFLD patients as well.
The major limitation of this study is its cross-sectional design, which does not allow causal inference. It is possible that NAFLD patients tend to be less physically active because of their liver disease or other obesity-associated disorders. Prospective studies are necessary to test the temporal sequence of these associations and whether they can be considered as causal.
In conclusion, higher leisure time PA is inversely associated with NAFLD. This association is independent of HOMA, nutritional intake, adiponectin and resistin and for resistance PA also of BMI. Therefore, this study supports the notion that lifestyle changes in NAFLD should include an overall increase in PA. More studies are necessary to determine whether resistance PA is better than aerobic training for NAFLD.