Prevalence of Fatty Liver Disease and Hepatic Iron Overload in a Northeastern German Population by Using Quantitative MR Imaging

Hepatic steatosis had a prevalence of 42.2% (1082 of 2561), whereas liver iron overload was found in 17.4% (447 of 2561) of participants; although liver fat content is associated with changes connected with the metabolic syndrome, liver iron content is associated with mean serum corpuscular hemoglobin, male sex, and age.

Abstract

Purpose

To quantify liver fat and liver iron content by measurement of confounder-corrected proton density fat fraction (PDFF) and R2* and to identify clinical associations for fatty liver disease and liver iron overload and their prevalence in a large-scale population-based study.

Materials and Methods

From 2008 to 2013, 2561 white participants (1336 women; median age, 52 years; 25th and 75th quartiles, 42 and 62 years) were prospectively recruited to the Study of Health in Pomerania (SHIP). Complex chemical shift–encoded magnetic resonance (MR) examination of the liver was performed, from which PDFF and R2* were assessed. On the basis of previous histopathologic calibration, participants were stratified according to their liver fat and iron content as follows: none (PDFF, ≤5.1%; R2*, ≤41.0 sec⁻¹), mild (PDFF, >5.1%; R2*, >41 sec⁻¹), moderate (PDFF, >14.1%; R2*, >62.5 sec⁻¹), high (PDFF: >28.0%; R2*: >70.1 sec⁻¹). Prevalence of fatty liver diseases and iron overload was calculated (weighted by probability of participation). Clinical associations were identified by using boosting for generalized linear models.

Results

Median PDFF was 3.9% (range, 0.6%–41.5%). Prevalence of fatty liver diseases was 42.2% (1082 of 2561 participants); mild, 28.5% (730 participants); moderate, 12.0% (307 participants); high content, 1.8% (45 participants). Median R2* was 34.4 sec⁻¹ (range, 14.0–311.8 sec⁻¹). Iron overload was observed in 17.4% (447 of 2561 participants; mild, 14.7% [376 participants]; moderate, 0.8% [20 participants]; high content, 2.0% [50 participants]). Liver fat content correlated with waist-to-height ratio, alanine transaminase, uric acid, serum triglycerides, and blood pressure. Liver iron content correlated with mean serum corpuscular hemoglobin, male sex, and age.

Conclusion

In a white German population, the prevalence of fatty liver diseases and liver iron overload is 42.2% (1082 of 2561) and 17.4% (447 of 2561). Whereas liver fat is associated with predictors related to the metabolic syndrome, liver iron content is mainly associated with mean serum corpuscular hemoglobin.

^© RSNA, 2017

Online supplemental material is available for this article.

With the increasing prevalence of obesity, diabetes, and metabolic syndrome, nonalcoholic fatty liver disease and liver iron overload have become the major causes of chronic liver disease in the general population. Lipotoxicity because of fat overload–inferring steatohepatitis and oxidative processes that occur in the presence of hepatic iron overload may lead to hepatocyte injury (1). Average annual health care use and costs are 26% higher in individuals with fatty liver disease (2). Early assessment of liver fat and liver iron by using quantitative imaging techniques has great potential for early detection and early intervention. In this way, progression to liver injury, inflammation, fibrosis, and ultimately irreversible stages of cirrhosis with an increased risk of hepatocellular carcinoma, liver failure, and portal hypertension may be avoided (3), lowering health care costs and probably mortality.

Core biopsy with semiquantitative histologic grading is the standard of reference for determining hepatic fat and liver iron deposition. However, biopsy sampling variability is problematic and can lead to inaccurate results (4,5). In addition, biopsy is invasive and ethically inappropriate for screening or epidemiologic studies (6). Alternative radiologic imaging techniques, such as ultrasonography (US) or conventional computed tomography (CT), cannot reliably quantify hepatic fat, particularly when fat content is low, and both are inaccurate for estimation of iron content (7).

Noninvasive magnetic resonance (MR) imaging techniques such as emerging confounder-corrected chemical shift–encoded (CSE) MR imaging showed excellent promise for reliable quantification of liver fat and iron content (8–10). By accounting for all known signal confounders, CSE MR imaging methods can provide accurate and precise estimates of proton density fat fraction (PDFF). PDFF is a fundamental metric of tissue triglyceride concentration that is increasingly accepted as an imaging biomarker for quantifying liver fat content (11–13). Measured by using CSE MR imaging, PDFF is highly reproducible across MR systems (14,15) and across imaging parameters (16). Accurate estimation of PDFF requires consideration of confounders such as T1 bias, T2* bias, noise bias, eddy currents, and the multipeak spectral complexity of fat (13).

In addition, CSE MR imaging methods also demonstrated excellent promise for quantifying liver iron content through estimation of R2* (R2* = 1/T2*). R2* is well known to have a linear relationship to hepatic iron concentration (10,17). Confounder-corrected estimates of R2* that are inherently corrected for the presence of fat can be generated with simultaneous estimation of PDFF (18,19). Accurate estimation of R2*, like PDFF, requires spectral modeling of fat to avoid false-positive diagnosis of iron overload (19). Furthermore, the use of complex CSE MR imaging methods allows R2* fitting by using complex data, naturally avoiding bias from noise floor effects, which are problematic for R2* estimation methods that use magnitude-based fitting (19).

Data regarding the distribution of hepatic PDFF and R2* in the general population for the detection of hepatic steatosis and liver iron overload are sparse. Such data are clinically warranted to define risk cohorts for prevention and to understand distributions in different populations. In addition, it is important to understand the associations of hepatic steatosis and liver iron overload to predict steatohepatitis (alcoholic steatohepatitis and nonalcoholic steatohepatitis) and improve algorithms for diagnosis and possible intervention in patients who have the highest risk of developing cirrhosis.

Therefore, the purpose of this study was to quantify liver fat and liver iron content by measurement of confounder-corrected PDFF and R2* and to identify clinical associations for fatty liver disease and liver iron overload and their prevalence in a large-scale population-based study.

Materials and Methods

The Study of Health in Pomerania (SHIP) is part of the Community Medicine Research Net of the University of Greifswald, Germany, which is funded by the Federal Ministry of Education and Research (grant numbers 01ZZ9603, 01ZZ0103, 01ZZ0403, 01ZZ0701, 03ZIK012), the Ministry of Cultural Affairs, and the Social Ministry of the Federal State of Mecklenburg-Western Pomerania. Whole-body MR imaging was supported by a joint grant from Siemens Healthcare (Erlangen, Germany) and the Federal State of Mecklenburg-Western Pomerania. The University of Greifswald is a member of the Center of Knowledge Interchange program of Siemens AG.

The authors had complete control of the data and the information submitted for publication.

Study of Health in Pomerania

The current analysis was performed by using data from two independent cohorts of the population-based SHIP. The SHIP study population consisted of adult German residents of Western Pomerania in Northeast Germany. From the total population of Western Pomerania, which included 213 057 inhabitants from a 1996 census, we drew a two-stage stratified random cluster sample of adults aged 20–79 years (20,21).

The net sample consisted of 6265 eligible participants. Of these, 4308 people participated in the baseline SHIP study between 1997 and 2001 (response rate of 68.8%). A total of 2333 people who participated in the baseline SHIP study volunteered for the second follow-up examination (referred to as SHIP-2). SHIP-2 was conducted between 2008 and 2012 (response rate of 54.1%).

In addition, a separate stratified random sample of 8016 adults (20–79 years) was drawn for the SHIP-TREND cohort (20) between 2008 and 2012, and 4420 participants in this sample were willing to participate (response rate, 55.1%).

SHIP-2 and SHIP-TREND collected a wide range of data including simple medical examinations, US examinations, computer-assisted personal interviews, anthropometry, echocardiography, dental examinations, self-reported medications, and laboratory data (see Table E1 [online] for a complete list and description of variables used for the present analysis).

In addition, participants of SHIP-2 and SHIP-TREND had the opportunity to undergo a whole-body MR examination if they agreed and had no contraindications to MR imaging. Absolute exclusion criteria from MR imaging participation included cardiac pacemakers, contraindicated metallic implants, extensive tattoos, and known pregnancy. Relative contraindications included the following: claustrophobia; artificial heart valves; stents; prostheses; and bone plates, screws, or nails. In participants with relative contraindications to MR imaging, the principal investigator determined patient eligibility on an individual basis.

Study Population

The population-based study including whole-body MR imaging was approved by the local institutional review board of SHIP, and written informed consent was obtained from all volunteers before study inclusion.

The current analysis was conducted in all participants who completed the whole-body MR examination, which included a confounder-corrected CSE MR imaging of the liver and complete covariate predictor variable datasets. An overview of the included participants is given in Figure 1. Overall, we enrolled 2561 participants: 881 patients from SHIP-2 and 1680 patients from SHIP-TREND. The study population included 1224 men (424 SHIP-2 participants and 800 SHIP-TREND participants) and 1337 women (455 SHIP-2 participants and 882 SHIP-TREND participants) with a mean age of 52.1 years ± 13.8 (standard deviation; mean age of SHIP-2 participants, 55.2 years ± 12.7; mean age of SHIP-TREND participants, 50.5 years ± 14.0).

Figure 1:

Flowchart of analysis procedure and participants included in the study.

MR Imaging Technique and Image Analysis

We performed 1.5-T MR imaging (Magnetom Avanto; Siemens Healthcare AG) by using a 12-channel phased-array coil. The whole-body MR imaging protocol included a three-echo CSE MR imaging acquisition of the upper abdominal organs (22). Three echoes were generated in a single repetition time, and all data were acquired in a single breath hold. The following imaging parameters were used: repetition time (msec)/echo time 1 (msec)/echo time 2 (msec)/echo time 3 (msec), 11.0/2.4/4.8/9.6; flip angle, 10°; one signal average; bandwidth, 1065 Hz/pixel; 224 × 126 × 32 matrix; field of view, 410 × 308; section thickness, 6.0 mm (interpolated to 64 sections by using a zero-padding method); and monopolar readout. Parallel imaging (generalized autocalibrating partially parallel acquisition) with an effective acceleration factor of 1.5 was used for acceleration to ensure complete acquisition during a total breath-hold time of 19 seconds.

Following image acquisition, MR datasets were processed by using an offline reconstruction algorithm written in Matlab (Mathworks, Natick, Mass) to estimate PDFF and create R2* maps (19). PDFF maps were generated on a pixel-by-pixel basis with correction for known confounders that included T2* decay (23,24), T1 bias (25,26) with static T1 values from the literature (27), noise bias (25), eddy currents (25), and multipeak spectral complexity of fat (24). The use of complex-valued (ie, magnitude and phase) images enabled PDFF mapping over the entire 0%–100% PDFF range. In addition, R2* maps were generated simultaneously with PDFF. Because fat, water, and R2* are estimated simultaneously, the R2* estimate is corrected for the presence of fat, just as fat is corrected for R2* signal decay (19,28). Parametric maps of PDFF and R2* were reconstructed for each volunteer. Postprocessing took approximately 7 minutes for each patient when performed on a MacBook Pro Mid 2012 (2.6GHz Core i7; 16GB RAM, 1600 Mhz DDR3; Apple, Cupertino, Calif). The algorithm used for reconstruction is not vendor-specific and versions of this algorithm have been used on data from Philips Healthcare, Siemens Healthcare, and GE Healthcare (29). At this point, the exact code used for this work is not publicly available, although similar modeling is included in the International Society for Magnetic Resonance in Medicine fat-water toolbox (http://www.ismrm.org/workshops/FatWater12/data.htm).

For image analysis, a representative section through the center of the liver was chosen by using the first in-phase sequence for anatomic orientation. We avoided sections that were degraded by partial-volume effects or motion artifacts. Operator-defined selection of the entire liver was performed in one section by using the region-of-interest tool in Osirix (version 4.6; Pixmeo, Bernex, Switzerland). The central portal vein and inferior vena cava were excluded. The region of interest was assigned to the same section of the PDFF map and R2* map by using the copy-and-paste function for perfect colocalization.

The images from each patient were analyzed independently by two observers (J.P.K., with 11 years of experience in abdominal MR imaging, and C.H., a medical student without experience in MR imaging who reviewed the images following instruction) by using the same procedure. The purpose of double reading was to determine interreader variability of PDFF and R2* estimates from PDFF and R2* maps. The results from J.P.K. were defined as the standard of reference and used for correlations. The observers were blinded to each other’s results. The results of C.H. only served to evaluate the interobserver variability of the method.

Prevalence of Hepatic Steatosis and Iron Overload

After determination of PDFF (as percent) and R2* (as sec⁻¹), patients were classified by using defined cutoffs of liver fat and liver iron content. Cutoffs of PDFF and R2* were on the basis of histopathologic calibrations. These calibrations were defined in an external study described elsewhere (19). Hardware, image acquisition, and image reconstruction in the calibration study were the same as in the current study (19). The following threshold values for PDFF were used: PDFF of 5.1% or less, grade 0 (no fat content); PDFF greater than 5.1%, grade 1 (mild fat content); PDFF greater than 14.0%, grade 2 (moderate fat content); and PDFF greater than 28.0%, grade 3 (high fat content). The following thresholds were used for iron content: R2* of 41.0 sec⁻¹ or greater, grade 0 (no overload); R2* greater than 41.0 sec⁻¹, grade 1 (mild overload); R2* greater than 62.5 sec⁻¹, grade 2 (moderate overload); and R2* greater than 70.1 sec⁻¹, grade 3 (severe overload).

Calibration was exclusively based on histopathologic grading of liver fat and iron content and did not include biochemical findings such as measurement of triglycerides or iron content.

Correlations of Liver Fat and Liver Iron Content

Data mining was performed by using the 165 explanatory variables of the general SHIP project as mentioned above (Table E1 [online]). Associations between these variables and liver fat content or iron content were established by using the SHIP-TREND cohort and verified with data from SHIP-2.

Statistical Analysis

Interobserver variability for both PDFF and R2* was evaluated by using Bland-Altman analysis of agreement between observers. In addition, concordance-correlation coefficients were calculated.

Median PDFF and R2* and their ranges were calculated in the study population. To account for selective participation, we computed inverse probability weights. Weights were computed from two separate logistic regression models. First, we used a model that accounted for nonresponse to the baseline examination by using sociodemographic predictors (age, sex, and region where patients live). Second, weights were used to account for potentially selective MR imaging subresponse on the basis of a wide range of baseline measures (age, sex, educational level, marital status, quality of life, fatty liver on the basis of previous US, diabetes, and biomarkers associated with the metabolic syndrome). The two weights were multiplied with each other.

To identify associations of liver fat and liver iron content, we used the statistical tool of boosting to select the most predictive variables (30). Boosting is a machine-learning technique that combines many simple weak models to build a strong model that well predicts the outcome of interest. Weak learners are added to the model and are weighted so that the weak learner’s accuracy is improved. This results in so-called boosting of unexplained variance and underlies the excellent predictive performance of boosting algorithms (31). Generalized linear models were boosted with the mboost package (32) implemented in the software package R applying Poisson distribution (Appendix E1 [online]). Boosting of generalized linear models for SHIP-TREND data identified the variables that were most strongly associated with liver fat and liver iron content. These associations were validated with data from the SHIP-2 cohort. Final analysis of associations of liver fat and liver iron content with the selected variables was performed by using quasi-Poisson distribution, including an over-dispersion parameter to improve estimates of standard errors. We applied variable selection in the dataset from the baseline cohort of SHIP-TREND and validated the findings externally with data from the independent SHIP-2 cohort. For this we recalculated the associations found in the model-building data with the validation data and checked if results were similar. All results presented are validated findings that used all datasets with complete data. We chose a small shrinkage factor (also called learning rate, parameter ν in R package mboost [R Foundation for Statistical Computing, Vienna, Austria]) to obtain a small set of predictors (ν = 0.00004 for liver fat and ν = 0.000015 for liver iron). Boosting was performed without weighting for probability of participation in the MR imaging study.

For illustration of unadjusted relationships of sex-specific relationships between age and liver fat content, waist-to-height ratio and liver fat content, and age and waist-to-height ratio we applied restricted cubic splines implemented in the Stata rcspline command (Stata, College Station, Tex) with the default parameterization.

All analyses were performed with Stata (Stata Corporation) and R (R Foundation for Statistical Computing).

Results

MR imaging and reconstructions of PDFF and R2* were successfully performed (Fig 2) in 2812 participants. Concordance-correlation coefficients between the two observers were 0.995 for PDFF (n = 2800) and 0.986 for iron (n = 2803). There was excellent agreement between the two observers for region-of-interest–based determination of PDFF and R2* as demonstrated by Bland-Altman analysis (Fig 3a). Mean bias between observers was 0.07% points for PDFF (limits of agreement, ±1.32% points) and 0.17% points for R2* (limits of agreement, ±4.82% points) (Fig 3b). After exclusion of 251 participants because of incomplete predictor variables, a total of 2561 patients were included in the analysis. Participant demographics, MR imaging data, and relevant clinical findings are summarized in Table 1. Data in Table 1 are weighted by probability of participation in the MR imaging study.

Figure 2:

Examples of healthy participants without fatty liver disease and without liver iron overload (upper row), of a participant with high liver fat content (middle row) as shown in the PDFF map, and a participant with high liver iron content as indicated in the R2* map (bottom row). Confounder-corrected multiecho CSE MR imaging is a straightforward method to accurately estimate liver fat and iron content.

Figure 3:

Bland-Altman difference plots of agreement to determine interreader variability for, A, liver PDFF and, B, liver R2*. The dashed line indicates the 1.96-fold of the difference’s standard deviation.

Table 1.

Characteristics of the Study Participants after Exclusion of Participants because of Incomplete Predictor Variables

Note.—Data are medians and data in parentheses are quartiles unless otherwise noted. Values are weighted by probability of participation in MR imaging study.

* Data are numbers; data in parentheses are percentages.

^†Systolic blood pressure of 140 mm/Hg or greater, or diastolic blood pressure of 90 mmHg or greater, or administration of antihypertensive medication.

Median PDFF was 3.9% (range, 0.6%–41.5%). Distributions of PDFF, R2*, and their combination are presented in Figure 4. On the basis of the semiquantitative histopathologic classification of PDFF, hepatic steatosis was diagnosed in 39.5% of participants (1012 of 2561 participants). Steatosis was classified as mild in 27.2% (696 participants), moderate in 10.6% (272 participants), and severe in 1.7% (44 participants). Prevalence of hepatic steatosis weighted by probability of participation in the whole SHIP study was 42.2% (1082 participants; men, 50.9% [623 men]; women, 34.7% [464 women]). Median R2* was 34.4 sec⁻¹ (range, 14.0–311.8 sec⁻¹). Iron overload was identified in 17.5% of participants (449 of 2561) and classified as mild in 14.9% (381 participants), moderate in 0.7% (18 participants), and high in 2.0% (50 participants). Prevalence of liver iron overload weighted by probability of participation was 17.4% (447 participants; 333 male participants [27.2%]; 120 female participants [9.0%]). Men had 54.4% higher liver fat content and 20.9% higher liver iron content compared with women. Sex-related distribution of PDFF and R2* is depicted in Table 1. A combination of hepatic steatosis and liver iron overload was observed in 229 of 2561 participants, corresponding to 8.9% of the study population (prevalence of 9.3%).

Figure 4:

Frequency distribution of liver fat content (PDFF, top), liver iron content (R2*, middle), and their combination (PDFF vs R2*) in the setting of SHIP (bottom). The bottom shows a distribution of the semiquantitative histologic classification A–D, defined as: A, no; B, low; C, moderate; and, D, high tissue content.

We checked whether the associations we had found in the model-building cohort were also valid in the validation data. We found that all associations were also statistically significant in the validation data, and estimates were similar and lay within the 95% confidence interval of the estimates from the model-building data. Therefore, we used the combined dataset of both cohorts for the final estimates. The most consistent validated associations of liver fat and liver iron content are shown in Table 2. For men and women together, liver fat content was significantly associated with waist-to-height ratio (Fig 5), serum alanine transaminase, serum triglycerides, serum glucose, serum uric acid, and hypertension. For men, a similar picture emerged but without serum glucose and with diastolic blood pressure instead of general hypertension. In women, waist circumference was more strongly associated with liver fat than in men. No significant association with hypertension remained in the sex-specific analysis. Validated predictors of liver iron content are shown in Table 3. The analysis for all volunteers revealed a 14% lower liver R2* in women than in men. In men, only mean corpuscular hemoglobin was associated with liver iron in both cohorts. In women, age was also associated with liver iron.

Table 2.

Results of Poisson Regression for Liver Fat Content

Note.—Data in parentheses are 95% confidence interval. Factor change is the exponentiated coefficient of Poisson regression, that is, with one unit change of a variable, liver fat content changes by the specified factor, for hypertensives by 15%. Estimates are calculated with the variables that had been selected by boosting and using a standard multivariate Poisson model with quasi-Poisson distribution.

* Systolic blood pressure ≥ 140 mmHg or diastolic blood pressure ≥ 90 mmHg or antihypertensive medication.

Figure 5:

Sex-specific relationships between, A, age and liver fat content (PDFF), B, waist-to-height ratio and PDFF, and, C, age and waist-to-height ratio. Graphs show functions of restricted cubic splines.

Table 3.

Results of Poisson Regression for Liver Iron Content

Note.—Data in parentheses are 95% confidence interval. Factor change is the exponentiated coefficient of Poisson regression. Estimates are calculated with the variables that had been selected by boosting and using a standard multivariate Poisson model with quasi-Poisson distribution.

Discussion

We assessed PDFF and R2* to determine the prevalence of hepatic steatosis and liver iron overload in a large epidemiologic study. In addition, we investigated associations between liver PDFF or R2* and demographic, behavioral, and laboratory data.

The double reading of PDFF and R2* maps by both an experienced observer and an inexperienced observer revealed excellent agreement, confirming that interpretation of these quantitative maps is fairly observer-independent. For example, with data from the second reader for the calculation of the weighed prevalence values, we obtain similar results: 44.2% (1132 of 2561) instead of 42.2% (1082) for hepatic steatosis and 17.9% (458 of 2561) instead of 17.4% (447 participants) for iron overload. This result demonstrates the robustness of the method. In addition, we assessed PDFF and R2* in the context of a population-based study, which demonstrated the feasibility of MR imaging screening for hepatic steatosis and liver iron for both research and clinical applications.

By using CSE MR imaging and known PDFF biopsy calibrations, we identified hepatic steatosis in 42.2% of the participants, which is more than twice the estimated worldwide prevalence of 20% (33). In addition, this value is also higher than the 29.9% prevalence identified by US in the baseline SHIP study (34). However, our results are in line with the findings of the population-based Dallas Heart Study. In the Dallas Heart Study, hepatic fat content was assessed in 2287 participants by using MR spectroscopy (35). Szczepaniak et al (35) reported a median liver fat content of 4.7% (25% and 75% quantiles, 2.7% and 8.6%) and found hepatic steatosis (fat content, >5%) in 34.3% of participants. Median fat content in the Dallas Heart Study and SHIP is similar. In both SHIP and the Dallas Heart study the prevalence of fatty liver disease is high. However, the frequency of a fatty liver disease was approximately 8% higher in SHIP (42.2%) compared with the Dallas Heart study (34.3%). This small difference might be attributable to differences in ethnicity, age of participants (Dallas Heart study median, 44.9 years; SHIP median, 52 years), and geographic origin.

The prevalence of obesity in Germany is stable but high (36). Compared with other European countries, Germany is ranked in the middle third with an obesity rate of approximately 20% among individuals age 18 or older (37). A recent study that included 7116 volunteers compared the prevalence of obesity between the new and old federal states of Germany and found no significant differences (36). Therefore, we may assume that the rate of fatty liver diseases as an expression of obesity is likely similar throughout Germany. In the near future, we will be able to confirm this assumption by using data from the largest population-based study in Germany, the National Cohort, which will investigate the rate of fatty liver, among other pathologic conditions, in various German states (38).

Schwenzer et al (39) assessed T2* decay of liver, spleen, and pancreas in a cohort of 129 participants without any evidence of iron overload, hepatitis, or inflammatory or malignant disease. They reported a mean T2* value of liver tissue of 28.1 msec ± 7.1 (range, 13.6–45.9 msec), which corresponded to a mean R2* of 35.6 sec⁻¹, which is similar to the value of 34.4 sec⁻¹ we found in this population-based study. On the basis of serum ferritin levels, Schwenzer et al observed iron overload in approximately 7% of participants (39). The absolute frequency of hepatic iron overload should be interpreted carefully. A previous analysis of pathologic findings in 5224 patients who underwent liver transplantation identified iron overload in 8.4% of participants, which was classified as mild iron overload in 5.6% and excessive iron content in 2.8% (40). The prevalence of iron overload in our study was higher (17.4%) on the basis of the R2* threshold of 41 sec⁻¹. The threshold in our study is based on histopathologic classification (19). This discrepancy in the measured frequency of iron overload may be because of differences in the respective patient populations or technical differences and choice of thresholds in the assessment of iron overload.

In our analysis, 9.3% of participants (238 of 2561) were found to have both hepatic steatosis and liver iron overload. Recent studies indicated that liver iron may play a crucial role in the development of nonalcoholic steatohepatitis (41–45). Therefore, combined hepatic steatosis and liver iron overload in our population could be used to identify participants at risk for both steatohepatitis and liver fibrosis. This observation also provides further evidence that concomitant presence of increased liver fat and liver iron may increase the risk of developing nonalcoholic steatohepatitis.

In our study, data mining revealed associations of hepatic fat and iron content. Liver fat content was associated with variables related to the metabolic syndrome, namely obesity (here assessed quantitatively by waist-to-height ratio), hypertension, serum triglycerides, and serum glucose levels. Moreover, we found a strong association of hepatic steatosis with alanine-transaminase and serum uric acid levels, which was previously reported (46,47). These observations confirm the generally accepted hypothesis that hepatic steatosis is an important component of the metabolic syndrome.

Epidemiologic studies describe an age-related increase in liver fat content up to age 40–50 years, followed by a decline in the elderly population (48), which was confirmed by our results. In addition, we found a sex-specific difference in liver fat accumulation: whereas in men liver-fat content increased continuously between age 20 and 50 years, the period during which liver fat content increased in women began 20 years later, extending from 40 to 65 years. This can be attributed in part to the tight association between liver fat content and waist-to-height ratio. The data-mining approach did not identify age as strongly associated with liver fat content, which is attributable to the fact that liver fat content is more closely related to waist-to-height ratio than to age.

Mean corpuscular hemoglobin was identified to be the most predictive marker of liver iron content in both men and women. In women, liver iron content was also found to increase with age. Women have 10% lower iron content, which is likely because of menstruation. This probably also explains the increase in liver iron content in older women when menstrual periods stop after menopause.

Clinically interesting variables such as alcohol consumption, diabetes (on the basis of increased HbA1c levels), and cholesterol levels were not identified as predictors of hepatic steatosis and liver iron overload, as demonstrated by the finding that inclusion of these variables did not significantly improve the model fit.

Strengths of our study are the large sample size and the population-based approach. In addition, statistical analysis for evaluation of clinical associations of hepatic fat and iron content was performed by using the hypothesis-free approach of boosting with 165 variables, which selects associations by means of the most consistent correlations.

However, this study had limitations. Assessment of liver fat and liver iron was performed by using MR imaging instead of biopsy. However, ethical concerns preclude the use of liver biopsy in a population-based study. The results presented here are dependent on the definitions of fatty liver and liver iron overload; in other words, they are dependent on the accuracy of the thresholds used for PDFF and R2*. In addition, the thresholds used in our study were on the basis of histopathologic calibrations performed with the same conditions as the volunteer study (ie, same MR imager and pulse sequence protocol). Further, the cutoffs used for assessment of nonalcoholic fatty liver disease and iron overload reflect the amount of tissue content and did not present cutoffs that are necessarily clinically relevant. There is a technical limitation: the number of echoes used for accurate assessment of PDFF and R2* is a subject of some controversy. Recent studies recommend the use of six echoes or more to ensure accurate assessment of tissue fat and tissue iron (49). The use of six echoes is based on the rationale that longer echo trains provide better estimation of R2*, and therefore, more accurate correction for R2* when attempting to measure tissue fat and iron. We agree that use of some additional echoes might have improved the signal-to-noise ratio performance of PDFF and R2* estimation. However, accurate separation of water and fat signal, including T2* correction, only requires three echoes (23,24,50). For these reasons, we are confident that our data on both liver fat and liver iron are accurate. In addition, the data presented here were generated from a white northeastern German cohort and may not be generalizable to other populations. Finally, this is a cross-sectional evaluation with limited outcomes data.

To conclude, quantification of liver fat and iron overload by using quantitative CSE MR imaging is feasible for epidemiologic research. In the general white northeastern German population, hepatic steatosis had a prevalence of 42.2% (1082 of 2561), whereas liver iron overload was found in 17.4% (447 of 2561) of participants. Combined hepatic steatosis and liver iron overload was observed in 9.3% of participants. Whereas liver fat content is associated with changes connected with the metabolic syndrome, such as waist-to-height ratio, alanine transaminase, uric acid, serum triglycerides, and blood pressure, liver iron content is associated with mean serum corpuscular hemoglobin, male sex, and age.

Advances in Knowledge

• ■ In a white German population, hepatic steatosis and liver iron overload have a prevalence of 42.2% (1082 of 2561 participants) and 17.4% (447 of 2561 participants), respectively.

• ■ The combination of hepatic steatosis and liver iron overload was observed in 9.3% (238 of 2561) of participants.

• ■ Liver fat content is associated with waist-to-height ratio, alanine transaminase, uric acid, serum triglycerides, and blood pressure.

• ■ Liver iron content is correlated with mean corpuscular hemoglobin.

• ■ Quantitative chemical shift–encoded MR imaging allows simultaneous quantification of liver fat and iron overload in a single examination and can be used for epidemiologic research.

Implication for Patient Care

• ■ The prevalence of fatty liver disease and liver hemosiderosis may inform design of future studies.