Hereditary hemochromatosis is transmitted as an

Transcription

1 STEATOHEPATITIS/METABOLIC LIVER DISEASE Hemochromatosis Genotypes and Risk of 31 Disease Endpoints: Meta-Analyses Including 66,000 Cases and 226,000 Controls Christina Ellervik, 1 Henrik Birgens, 2 Anne Tybjærg-Hansen, 3 and Børge G. Nordestgaard 4 Hemochromatosis genotypes have been associated with liver disease, diabetes mellitus, heart disease, arthritis, porphyria cutanea tarda, stroke, neurodegenerative disorders, cancer, and venous disease. We performed meta-analyses including 202 studies with 66,263 cases and 226,515 controls to examine associations between hemochromatosis genotypes C282Y/C282Y, C282Y/H63D, C282Y/wild type, H63D/H63D, and H63D/wild type versus wild type/wild type and 9 overall endpoints and 22 endpoint subgroups. We also explored potential sources of heterogeneity. For liver disease, the odds ratio for C282Y/C282Y versus wild type/wild type was 3.9 (99% confidence interval: ) overall, 11 (3.7-34) for hepatocellular carcinoma, 4.1 (1.2-14) for hepatitis C, and 10 (2.1-53) for nonalcoholic steatohepatitis. For porphyria cutanea tarda, the odds ratios were 48 (24-95) for C282Y/C282Y, 8.1 (3.9-17) for C282Y/H63D, 3.6 ( ) for C282Y/wild type, 3.0 ( ) for H63D/H63D, and 1.7 ( ) for H63D/wild type versus wild type/wild type. Finally, for amyotrophic lateral sclerosis, the odds ratio was 3.9 (1.2-13) for H63D/H63D versus wild type/wild type. These findings were consistent across individual studies. The hemochromatosis genotypes were not associated with risk for diabetes mellitus, heart disease, arthritis, stroke, cancer, or venous disease in the overall analyses; however, the odds ratio for C282Y/C282Y versus wild type/wild type was 3.4 (1.1-11) for diabetes mellitus among North Europeans. Conclusion: In aggregate, clinically ascertained cases who are homozygous for the C282Y mutation are associated with a 4-11 fold risk of liver disease, whereas all 5 hemochromatosis genotypes are associated with a 2-48 fold risk of porphyria cutanea tarda, and H63D/H63D is associated with a 4-fold risk of amyotrophic lateral sclerosis. These results, mainly from case-control studies, cannot necessarily be extrapolated to the general population. (HEPATOLOGY 2007;46: ) See Editorial on Page 960 Hereditary hemochromatosis is transmitted as an autosomal recessive trait characterized by iron accumulation in various organs, such as the liver, pancreas, heart, joints, and skin. 1,2 In 1996, the Abbreviations: CI, confidence interval; NIDDM, non insulin-dependent diabetes mellitus. From the 1 Department of Clinical Biochemistry, Herlev University Hospital, 2 Department of Hematology, Herlev University Hospital, 3 Department of Clinical Biochemistry, Rigshospitalet, Copenhagen University Hospital, and 4 Department of Clinical Biochemistry, Herlev University Hospital, University of Copenhagen, Copenhagen, Denmark. Received May 29, 2007; accepted June 19, Supported by the Danish Medical Research Council (Copenhagen County), Herlev University Hospital, the Research Fund at Rigshospitalet (Copenhagen University Hospital), a grant from the Ingeborg and Leo Dannins Foundation, and a grant from the foundation of Henry Hansen and his wife. Address reprint requests to: Børge G. Nordestgaard, Professor, Chief Physician, Department of Clinical Biochemistry, Herlev University Hospital, Herlev Ringvej 75, DK-2730 Herlev, Denmark. brno@heh.regionh.dk; fax: /4489. Copyright 2007 by the American Association for the Study of Liver Diseases. Published online in Wiley InterScience ( DOI /hep Potential conflict of interest: Nothing to report. Supplementary material for this article can be found on the HEPATOLOGY Web site ( HFE gene was identified, with homozygosity for C282Y being responsible for 83% of hereditary hemochromatosis in Europeans and with compound heterozygosity for C282Y and H63D explaining an additional 4%. 3 The genotypes C282Y/C282Y, C282Y/H63D, C282Y/wild type and, in some studies, H63D/H63D, but not H63D/wild type, are associated with evidence of iron overload. 4 Since 1996, the C282Y and H63D polymorphisms have been associated with various clinical endpoints, including well-known iron overload related diseases (liver disease, diabetes mellitus, heart disease, arthritis, hypogonadism, and porphyria cutanea tarda) and some hypothesized related diseases (stroke, neurodegenerative disorders, cancer, and venous disease; Supplementary Appendices 1-10). Most studies have involved only a few hundred cases, which are too few to assess reliably any genetic effects on the risk of disease. We conducted meta-analyses of the associations between hemochromatosis genotypes C282Y/C282Y, C282Y/H63D, C282Y/wild type, H63D/H63D, and H63D/wild type versus wild type/wild type and 31 different disease endpoints (9 overall endpoints and 22 endpoint subgroups); the analyses comprised a total of

2 1072 ELLERVIK ET AL. HEPATOLOGY, October 2007 studies including 66,263 cases and 226,515 controls (Supplementary Appendices 1-10). Materials and Methods Relevant Genotypes. The hemochromatosis genotypes C282Y/C282Y, C282Y/H63D, C282Y/wild type, H63D/H63D, and H63D/wild type versus wild type/ wild type were assessed. As many of the studies included in the present meta-analyses reported on only 1 of the 2 polymorphisms, we assessed C282Y/wild type and H63D/wild type in 2 ways: (1) only studies presenting all genotypes were involved (simple heterozygotes) and (2) all studies were included (mixed heterozygotes). In mixed heterozygotes, the compound heterozygous genotype C282Y/H63D was included with both C282Y/wild type and H63D/wild type. However, as the results mainly were similar for simple and mixed heterozygotes, we show only the results for the simple heterozygotes, but in the Results section, differences in the findings are mentioned. Search Strategy and Selection Criteria. Epidemiological genetic association studies published from January 1996 to December 31, 2006 and our own data on stroke 5 were sought by searches limited to humans on PubMed, EMBASE, and the Genetic Association Database ( geneticassociationdb.nih.gov) through scanning of the reference lists of articles identified for relevant studies, through hand searching of relevant journals, and through correspondence with authors (myocardial infarction and stroke if the tabular data were insufficient). The keywords used were HFE OR Cys282Tyr OR C282Y OR His63Asp OR H63D OR hemochromatosis in combination with words related to the associated diseases (Supplementary Table 1) in either an English abstract or an article written in English. Finally, all hits on HFE OR Cys282Tyr OR C282Y OR His63Asp OR H63D OR hemochromatosis were checked by hand. Studies were included if they provided tabular data on C282Y and/or H63D genotypes (Supplementary Appendices 1-9 and Supplementary Tables 2-10). Studies were excluded (Supplementary Appendices 1-10) if tabular data on C282Y and/or H63D genotypes were not provided, if controls had disease in the same organ as that examined in the meta-analysis (for example, cirrhosis controls for hepatocellular carcinoma), if cases were defined by their iron overload status and not by disease status related to an organ, if participants were duplicated or overlapped with reports already included in the meta-analysis, if there were no controls, or if data on historical controls were not provided by at least one of the authors of the study. Data Abstraction. The following information was abstracted from each study according to a fixed protocol: tabular data on genotypes, disease endpoints, mean ages of cases and controls, sex, geographical location, ethnic group of the participants (if the ethnicity was not stated explicitly, it was assumed to be the ethnic majority of that country), and study design and size (Supplementary Tables 2-10). We use study throughout this article as the number of samples, as many articles have more than one sample. The allele frequencies and Hardy-Weinberg equilibrium were calculated. Relevant clinical endpoints included liver disease (hepatocellular carcinoma, fibrosis, cirrhosis, hepatitis B, hepatitis C, viral hepatitis, alcoholic liver disease, nonalcoholic steatohepatitis, and other liver diseases), diabetes mellitus [type 1 (insulin-dependent diabetes mellitus), type 2 (non insulin-dependent diabetes mellitus), and gestational diabetes], heart disease (ischemic heart disease, myocardial infarction, cardiomyopathy, and other heart diseases), arthritis (osteoarthritis), porphyria cutanea tarda (sporadic and familial), stroke (ischemic stroke), neurodegenerative disorders (Alzheimer s disease, mild cognitive impairment, Parkinson s disease, amyotrophic lateral sclerosis, and multiple sclerosis), cancers other than hepatocellular carcinoma (breast cancer, colorectal cancer, hematological cancer, and other cancers), venous disease, and hypogonadism (Supplementary Appendices 1-10 and Supplementary Tables 2-10). For hypogonadism, there were fewer than 3 eligible studies (Supplementary Appendix 10). For 31 of these endpoints, at least 3 independent studies were available for a meta-analysis (Supplementary Tables 2-9). Statistical Analysis. Statistical analyses were performed with Stata (version 9.0) statistical software with the metan command, which automatically corrects for the zero cell problem by adding 0.5 to each cell. The odds ratios for genotypes C282Y/C282Y, C282Y/H63D, C282Y/wild type (simple and mixed heterozygotes), H63D/H63D, and H63D/wild type (simple and mixed heterozygotes) versus wild type/wild type were compared between cases and controls. Random effects summary measures were calculated with the method of DerSimonian and Laird. 6 Heterogeneity was assessed with the I 2 statistic, which describes the percentage of variation in the log odds ratios that might be attributable to genuine differences across studies rather than random error, 7 and a statistical test. 8 We used 99% confidence intervals in order to compensate for multiple comparisons for 5 genotypes. The publication bias was assessed with 2 statistical tests (Begg s and Egger s) 8 for meta-analyses comprising more than 10 studies; in the case of disparate results, a visual inspection of funnel plots overruled statistical tests. In meta-analyses with fewer than 10 studies, only a visual inspection of funnel plots was performed.

3 HEPATOLOGY, Vol. 46, No. 4, 2007 ELLERVIK ET AL As the frequencies of C282Y and H63D depend on the geographical location and ethnicity and as hemochromatosis disease penetrance is more pronounced in men than in women, 1 we focused the subanalyses on these stratifications in assessing clinical heterogeneity. The ethnic groups were defined as (1) people of European continental ancestry (more than 95% Europeans for mixed populations), including Russia, the United States, South Africa, and Australia, and (2) other ethnic groups (Asian, African, or other). Turks, Arabs, and Hispanics were not included in the European category. The methodological heterogeneity was assessed by stratification of the study design (prospective, case-control, and cross-sectional), study size (small studies had fewer than 200 cases, and large studies had 200 or more cases; for ischemic heart disease and myocardial infarction, small studies had fewer than 500 cases, and large studies had 500 cases or more), Hardy-Weinberg equilibrium/disequilibrium among controls, and source of controls (diseased versus healthy). If heterogeneity still persisted after these stratifications, we post hoc inspected the studies included in the analyses by hand. In the figures, only aggregate data are shown because of the large number of comparisons made. For individual studies, see Supplementary Figures 1-9 (the area of the squares for individual studies is inversely proportional to the variances of the log odds ratio estimates, the horizontal lines indicate confidence intervals, and the diamonds denote pooled estimates, the width of which represents the confidence intervals). Results A total of 202 studies (in 163 articles) including 66,263 cases and 226,515 controls covering 31 different clinical endpoints (9 overall endpoints and 22 endpoint subgroups) were included in the meta-analyses (Table 1 and Supplementary Tables 1-9). There were too few eligible studies for hypogonadism (Supplementary Appendix 10). One hundred five potential studies were excluded on the basis of the exclusion criteria previously described (Supplementary Appendices 1-10). Two hundred two studies (100%) provided data on C282Y genotypes, of which 97% had controls in Hardy-Weinberg equilibrium; 150 studies (74%) provided data on H63D genotypes, of which 94% had controls in Hardy-Weinberg equilibrium; and 120 studies (59%) provided data on the 2 polymorphisms combined. Thirty-one studies (15%) provided data for men separately, and 37 studies (18%) provided data for women separately. The studies varied in their geographical setting, with 85% being of European origin, 47% being of Northern European origin, and 19% being of Southern European origin. Thirty-two studies (16%) were prospective, 160 studies (79%) were casecontrol studies, and 10 studies (5%) were cross-sectional. Forty-eight studies (24%) had a study size of 200 cases or more (500 cases or more for ischemic heart disease and myocardial infarction). All but 4 studies had nondiseased controls (Supplementary Tables 4, 7, and 8). C282Y/C282Y. For liver disease, the odds ratio for genotype C282Y/C282Y versus wild type/wild type was 3.9 overall (99% confidence interval , I 2 39%, 95% confidence interval 0%-63%; Fig. 1). This was mainly due to an odds ratio for hepatocellular carcinoma of 11 (3.7-34, I 2 21%, 0%-55%), for hepatitis C of 4.1 (1.2-14, I 2 8%, 0%-73%), and for nonalcoholic steatohepatitis of 10 (2.1-53, I 2 16%, 0%-87%; Fig. 2). For porphyria cutanea tarda, the overall odds ratio was 48 (24-95, I 2 3%, 0%-61%). The stratified odds ratio for sporadic porphyria cutanea tarda was 61 (28-130, I 2 0%, 0%-55%), and for familial porphyria cutanea tarda, it was 47 (14-160, I 2 0%, 0%-68%). For these endpoints, there was high consistency of effects across studies (Supplementary Figs. 1 and 6). For other overall endpoints (diabetes mellitus, heart disease, arthritis, stroke, neurodegenerative disease, other cancer, and venous disease; Fig. 1) or endpoint subgroups (Fig. 2), the odds ratios did not differ significantly from 1.0. Stratifying diabetes mellitus in Northern Europeans only resulted in an odds ratio of 3.4 (1.1-11, I 2 49%, 0%-74%). C282Y/H63D. For porphyria cutanea tarda, the overall odds ratio was 8.1 (3.9-17, I 2 0%, 0%-47%; Fig. 1). The stratified odds ratio for sporadic porphyria cutanea tarda was 9.1 (3.5-24, I 2 0%, 0%-53%), and for familial porphyria cutanea tarda, it was 15 (4.6-51, I 2 0%, 0%-0%; Fig. 2). For other overall endpoints (liver disease, diabetes mellitus, heart disease, arthritis, stroke, neurodegenerative disease, and other cancer; Fig. 1) or endpoint subgroups (Fig. 2), the odds ratios did not differ significantly from 1.0. C282Y/Wild Type: Simple Heterozygotes. For porphyria cutanea tarda, the overall odds ratio was 3.6 ( , I 2 47%, 0%-73%), and for sporadic porphyria cutanea tarda, it was 4.7 (1.7-13, I 2 53%, 0%-79%; Figs. 1 and 2). For other overall endpoints (liver disease, diabetes mellitus, heart disease, arthritis, stroke, neurodegenerative disease, and other cancer; Fig. 1) or endpoint subgroups (Fig. 2), the odds ratios did not differ significantly from 1.0. C282Y/Wild Type: Mixed Heterozygotes. The data for mixed heterozygotes (not shown) did not differ substantially from those for simple heterozygotes. For mixed heterozygotes, the odds ratios for hepatocellular carcinoma was

4 1074 ELLERVIK ET AL. HEPATOLOGY, October 2007 Table 1. Characteristics of the Meta-Analyses Endpoint Studies (n)* Cases Controls C282Y C282Y, HWE H63D H63D, HWE Polymorphisms Combined Endpoint Subgroups (n) Male/Female Stratification Europeans Northern Europeans Southern Europeans Other Ethnicities 1/2/3 Design Study Size Above 200 Cases Liver disease 50 6,969 41, / /49/1 8 Diabetes mellitus 24 7,662 37, / /2/2 12 Heart disease 31 13,113 48, / /20/3 7 Arthritis 12 15,371 27, / /11/1 5 Porphyria cutanea tarda 23 1,146 14, / /23/0 0 Stroke 7 1,304 13, / /2/2 2 Neurodegenerative disease 24 3,826 5, / /24/0 6 Cancer 26 15,593 36, / /24/1 11 Venous disease 5 1,279 2, / /5/0 4 Total , , (97%) 150 (74%) 141 (94%) 120 (59%) 22 31/ (85%) 94 (47%) 39 (19%) 30 (15%) 32/160/10 48 (24%) The endpoint subgroups include liver disease (hepatocellular carcinoma, cirrhosis, hepatitis B, hepatitis C, alcoholic liver disease, and nonalcoholic steatohepatitis), diabetes mellitus [type 2 diabetes (non insulin-dependent diabetes mellitus)], heart disease (ischemic heart disease, myocardial infarction, and cardiomyopathy), arthritis (osteoarthritis), porphyria cutanea tarda (sporadic and familial), stroke (ischemic stroke), neurodegenerative diseases (Alzheimer s disease, mild cognitive impairment, Parkinson s disease, amyotrophic lateral sclerosis, and multiple sclerosis), cancer (breast cancer, colorectal cancer, and hematological cancer), and venous disease. The Europeans had continental ancestry [that is, Europe including Russia and populations of European origin [United States (if not mixed racial groups), South Africa, and Australia]. The Northern Europeans were from Finland, Norway, Sweden, Denmark, Germany, the Netherlands, Belgium, the United Kingdom, France, Austria, and Switzerland. The Southern Europeans were from Portugal, Spain, Italy, and Greece. The other ethnicities included mixed racial groups, blacks, Hispanics, Arabs, Turks, Asians, and Jews. HWE indicates Hardy-Weinberg equilibrium among controls. *n is a sample number and is equal to or more than the number of articles included as many articles have more than one sample. The design was as follows: (1) cohort, (2) case-control, and (3) cross-sectional. The stratification limit is 500 cases for heart disease.

5 HEPATOLOGY, Vol. 46, No. 4, 2007 ELLERVIK ET AL Fig. 1. Meta-analyses of studies of the association between the hemochromatosis genotypes and overall endpoints. Because of graphical restrictions, some graphical CIs may not be symmetrical; however, the listed odds ratios are. The power indicates the odds ratio that can be detected at 90% power at 2-sided P I square represents the study heterogeneity. CI indicates confidence interval. 1.6 ( , I 2 4%, 0%-66%), and for familial porphyria cutanea tarda, it was 3.3 ( , I 2 0%, 0%-70%). For nonalcoholic steatohepatitis with 7 studies, we observed an odds ratio of 3.3 ( , I 2 58%, 0%-85%). H63D/H63D. For porphyria cutanea tarda, the overall odds ratio was 3.0 ( , I 2 0%, 0%- 30%), and for sporadic porphyria cutanea tarda, the odds ratio was 3.1 ( , I 2 0%, 0%-55%; Figs. 1

6 1076 ELLERVIK ET AL. HEPATOLOGY, October 2007 Fig. 2. Meta-analyses of studies of the association between the hemochromatosis genotypes and endpoint subgroups. Because of graphical restrictions, some graphical CIs may not be symmetrical; however, the listed odds ratios are. The power indicates the odds ratio that can be detected at 90% power at 2-sided P I square represents the study heterogeneity. CI indicates confidence interval; and NIDDM, non insulin-dependent diabetes mellitus. and 2). For amyotrophic lateral sclerosis, the odds ratio was 3.9 (1.2-13, I 2 0%, 0%-0%); when we viewed the 4 individual studies participating (Supplementary Fig. 8), all individual odds ratios were above 1.0. For other overall endpoints (liver disease, diabetes mellitus, heart disease, arthritis, stroke, neurodegenerative disease, and other cancer; Fig. 1) or endpoint subgroups (Fig. 2), the odds ratios did not differ significantly from 1.0. H63D/Wild Type: Simple Heterozygotes. The odds ratios did not differ significantly from 1.0 for the overall

7 HEPATOLOGY, Vol. 46, No. 4, 2007 ELLERVIK ET AL endpoints (Fig. 1) or endpoint subgroups (Fig. 2), except for porphyria cutanea tarda in Europeans with an odds ratio of 1.9 ( , I 2 70%, 45%-84%) and for sporadic porphyria cutanea tarda in Europeans with an odds ratio of 2.2 ( , I 2 75%, 49%-87%). H63D/Wild Type: Mixed Heterozygotes. The data for mixed heterozygotes (not shown) did not differ substantially from those for simple heterozygotes. For mixed heterozygotes for amyotrophic lateral sclerosis with 4 studies, we observed an odds ratio of 1.8 ( , I 2 0%, 0%-79%). Power. We had 90% statistical power or more at 2-sided P 0.01 to detect the odds ratio differences from 1.0 as described previously (Figs. 1 and 2). For endpoints not differing from 1.0, we had 90% statistical power to detect odds ratios from ; this depended on the genotype and endpoint studied. Study Heterogeneity. In general, meta-analyses with odds ratios differing from 1.0 had low heterogeneity (I 2 ; Figs. 1 and 2). Heterogeneity among studies was most prevalent among heterozygous genotypes C282Y/wild type and H63D/wild type. When the analyses shown in Figs. 1 and 2 were stratified for Hardy-Weinberg equilibrium/disequilibrium among controls, sex (for meta-analyses on ischemic heart disease, myocardial infarction, and diabetes mellitus), geographical origin, study design, study size ( 200 cases versus 200 cases; for ischemic heart disease and myocardial infarction, 500 cases versus 500 cases), and healthy controls (Table 1), the results were mostly similar to those shown in Figs. 1 and 2 (data not shown); exceptions to this rule are described earlier in this article. These stratifications usually also reduced heterogeneity further (data not shown), as did stratification of the endpoint subgroups. Evidence for publication bias was seen only for C282Y/H63D on the risk of ischemic heart disease (P 0.03) and for H63D/wild type on the risk of hepatitis C (P 0.02). It was not possible to make all stratifications for all endpoints because of a lack of studies. Personal communication with authors to obtain tabular data did not change results for myocardial infarction or stroke but just provided more power to the analyses and the possibility for sex stratification. Discussion The present meta-analyses of 202 genetic association studies, involving a total of 66,263 cases and 226,515 controls (with every study s cases and controls counted only once), provides the most comprehensive assessment so far of the association between hemochromatosis genotypes and 31 disease endpoints. It provides evidence of associations between C282Y/C282Y and liver disease, C282Y/C282Y and diabetes mellitus in Northern Europeans, all 5 hemochromatosis genotypes and porphyria cutanea tarda, and H63D/H63D and amyotrophic lateral sclerosis. For most endpoints with aggregated odds ratios above 1.0, there was high consistency of the odds ratios above 1.0 across studies. Because almost 80% of the included studies were case-control studies, the results of these meta-analyses present effects for people already diagnosed with the disease studied, but that does not necessarily imply that the average person in the population at large with these genotypes will have similar effects. The associations of C282Y/C282Y with hepatocellular carcinoma and diabetes mellitus in Northern Europeans are consistent with the traditional description of clinically identified hereditary hemochromatosis, with the C282Y/ C282Y genotype resulting in iron accumulation in the liver and pancreas. 1,2 For the association of the C282Y/ C282Y genotype with hepatitis C, such persons could either be more susceptible to contracting an acute hepatitis C infection or be more susceptible to developing a chronic hepatitis C infection after an episode of acute infection. 9 A likely explanation for the association of C282Y/C282Y with nonalcoholic steatohepatitis is that steatosis has been shown to be a significant factor (along with alcohol) in the overt expression of hemochromatosis. 10 Another explanation for the association of C282Y/ C282Y with nonalcoholic steatohepatitis is insulin resistance 11 ; however, the pathogenesis of this disease is still not fully understood. The association of all hemochromatosis genotypes with porphyria cutanea tarda shows a clear dose-response relationship, with C282Y/C282Y conferring the highest risk and H63D/wild type conferring the lowest risk (Fig. 3). This observation is in accordance with porphyria cutanea tarda being associated with hepatic iron overload and the disease being responsive to iron-reduction therapy. 12 The close association between hepatitis C and porphyria cutanea tarda 13 and the fact that hepatocellular carcinoma is occasionally seen in patients with porphyria cutanea tarda 14 further support the idea that a common pathogenic factor could be iron overload mediated by increased intestinal absorption due to mutations in the HFE gene. Amyotrophic lateral sclerosis was associated with H63D/H63D and also with mixed H63D/wild type, but we did not find associations with other neurodegenerative diseases for any of the hemochromatosis genotypes. The molecular connection between iron dysregulation and neurodegeneration is still controversial, but recently identified anomalies in proteins linked with iron metabolism signify a critical role for this connection. 15 Also, our own recent large prospective study found a 2-3 fold risk of

8 1078 ELLERVIK ET AL. HEPATOLOGY, October 2007 Fig. 3. Meta-analysis of studies of the association between the hemochromatosis genotypes and porphyria cutanea tarda. The power indicates the odds ratio that can be detected at 90% power at 2-sided P I square represents the study heterogeneity. CI indicates confidence interval; and NIDDM, non insulin-dependent diabetes mellitus. ischemic stroke associated with H63D/H63D but not with any other hemochromatosis genotypes. 5 Because it is well documented that H63D and C282Y influence iron metabolism via different molecular mechanisms, 16 it is biologically plausible that H63D/H63D could be associated with diseases of the nervous system, including amyotrophic lateral sclerosis and stroke, whereas other hemochromatosis genotypes are not. However, the finding for amyotrophic lateral sclerosis could also be due to publication bias toward positive results, as we identified only 4 studies, all with an odds ratio nominally above 1.0. Some important iron overload relevant endpoint subgroups (such as hepatocellular carcinoma, familial porphyria cutanea tarda, and nonalcoholic steatohepatitis) are associated with the mixed heterozygous genotype C282Y/wild type but not with the simple heterozygous genotype. This may be due to more statistical power in the analyses of mixed heterozygotes versus simple heterozygotes, as more studies could be included in the mixed group; however, the results could also be artifacts from the influence of C282Y/H63D in the mixed group. This indicates the importance of having data on both C282Y and H63D because the compound heterozygotes introduce bias into studies of the mixed heterozygous associations as compound heterozygosity is more strongly related to hemochromatosis than single heterozygosity for C282Y. The study lacks sufficient power to draw conclusions with respect to some diseases, especially cancers. Furthermore, for arthritis, only osteoarthritis is considered in subgroup analyses, and the conclusions do not necessarily apply to other forms of arthritis, particularly chondrocalcinosis, which is reportedly associated with hemochromatosis. 17,18 The odds ratios that are shown were calculated with the random effects method, which incorporates heterogeneity that cannot be explained. This method provides estimates of average effects and larger confidence intervals than the fixed effects method, which provides the best effect estimates and narrow confidence intervals. We also chose the metan command in Stata statistical software instead of the meta command in order to include studies with a zero cell in a pair. These decisions resulted in conservative estimates. Almost 80% of the included studies were case-control studies, and this introduced possible ascertainment bias into the results. Thus, the results of these meta-analyses present effects for people already diagnosed with the disease studied, but that does not necessarily imply that the average person in the population at large with these genotypes will have similar effects. 19,20 The effects of genotypes on intermediate parameters and the risk of disease are often less in the general population than in case-control studies For most endpoints (except ischemic heart disease and myocardial infarction), we could limit analyses only to case-control studies and could not look at the effects in the general population separately. However, large population-based screening studies indicate that the disease risks are actually very low for these hemochromatosis genotypes in the population at large. 19,20,23 Two very large population-based cross-sectional studies of the penetrance of C282Y concluded that only C282Y homozygotes were more likely to report a history of liver disease than wild type/wild type. 19,23 Diabetes mellitus, heart disease, arthritis, and impotence were also assessed, but individuals with the various genotypes did not report a history of any of these diseases more than individuals with wild type/wild type. With this in mind, the genotypes are only one risk factor, and probably a weak one at that, for most of the disease endpoints studied. It was not possible to make all stratifications for all endpoints because of a lack of studies; for example, the influence of sex was studied for only diabetes mellitus, ischemic heart disease, and myocardial infarction. Other ethnic groups could not be reliably substratified into relevant groups in any meta-analysis, and stratification for Southern Europeans was mainly possible for the H63D polymorphism. In meta-analyses based on tabular data, it is not possible to control for confounding factors. However, because of Mendelian randomization, confounding by conventional risk factors is rarely a problem in studies of genetic factors. 24 The residual heterogeneity in the present metaanalyses after stratification may be explained by the many

9 HEPATOLOGY, Vol. 46, No. 4, 2007 ELLERVIK ET AL Table 2. Previous Meta-Analyses of the Association of C282Y and H63D Polymorphisms and Specific Endpoints C282Y (Mix of Homozygotes and Heterozygotes) H63D (Mix of Homozygotes and Heterozygotes) Author Endpoint Odds Ratio (95% CI), P Studies Cases/controls Odds Ratio (95% CI), P Studies Cases/controls Bonkovsky et al. 9 Hepatitis C NA, * 4 481/2759 NA 4 481/2759 Njajou et al. 29 Type 2 diabetes 1.0 ( ), NA / ( ), NA /2524 Halsall et al. 25 Type 2 diabetes 1.09 ( ), NA 9 NA Qi et al. 26 Type 2 diabetes 1.03 ( ), NA 17 NA 1.11 ( ), NA 13 NA van der A 27 Coronary heart 1.02 ( ), NA /NA 1.03 ( ), NA /NA disease Guerreiro et al. 28 Alzheimer s NA, /714 NA, /713 disease Guerreiro et al. 28 Parkinson s disease NA, /3465 NA, /2973 The data are abstracted as they are written in the articles. All the meta-analyses have appeared as part of an article with original results. CI indicates confidence interval; and NA, not available (the analysis was not performed). *Frequency of C282Y/C282Y among patients versus controls. For H63D/H63D, the odds ratio was 1.2 ( ). clinical differences in patient selection (differences in disease severity in case-control studies), the different diagnoses of endpoints (some World Health Organization based and other self-reported), and the limited number of studies available. Previous meta-analyses (Table 2) on type 2 diabetes, 25,26 coronary heart disease, 27 Alzheimer s disease, 28 and Parkinson s disease 28 have mainly been conducted for combined genotypes (that is, C282Y homozygosity and heterozygosity combined and H63D homozygosity and heterozygosity combined), except for 2 meta-analyses on hepatitis C 9 and type 2 diabetes. 29 The previous metaanalyses showed only an association for C282Y/C282Y with hepatitis C, 9 in accordance with the present metaanalyses, and an association for H63D/H63D with type 2 diabetes, 29 which we could not confirm. Previous metaanalyses did not document any of the other findings in the present comprehensive meta-analysis. In general, in all previous meta-analyses, there was a lack of reporting of many standardized data for a meta-analysis 30 (Table 2), and this makes interpretations and critical comparisons with the present report difficult. The primary purpose of this review is to present combined information. The strength of the evidence for the associations observed for the C282Y/C282Y genotype may be seen in the light of high odds ratios but with large confidence intervals, relatively low heterogeneity, consistency of results across study populations (Europeans, Northern Europeans, and Southern Europeans), consistency of odds ratios above 1.0 across studies, and consistency of results with the expected findings for that genotype. This review also underscores the need for much larger, coherent, and prospective studies to assess the associations of various diseases with hemochromatosis mutations in greater detail. Thus, the present meta-analyses do not necessarily represent a final or complete compilation of disorders that have been or may in the future be found to have an association with common hemochromatosis genotypes other than wild type/wild type. In addition, it is important to emphasize that disordered iron absorption or metabolism is not necessarily involved in the pathophysiology of some of the conditions studied; common hemochromatosis mutations may be surrogate markers for traits on 6p haplotypes that are otherwise unrelated to iron metabolism and the HFE gene. Finally, in future genetic association studies, reviewers should ensure that tabular data are reported (at least in the supplementary tables) in order to make future meta-analyses possible. In conclusion, clinically ascertained cases who are homozygous for the C282Y mutation are associated with a 4-11 fold risk of liver disease, whereas all 5 hemochromatosis genotypes are associated with a 2-48 fold risk of porphyria cutanea tarda, and H63D/H63D is associated with a 4-fold risk of amyotrophic lateral sclerosis. These results, mainly from case-control studies, cannot necessarily be extrapolated to the general population. References 1. Pietrangelo A. Hereditary hemochromatosis a new look at an old disease. N Engl J Med 2004;350: Beutler E, Bothwell TH, Charlton RW, Motulsky AG. Hereditary hemochromatosis. In: Scriver CR, Beaudet AL, Sly WS, Valle D, eds. The Metabolic and Molecular Bases of Inherited Disease. New York, NY: McGraw-Hill; 2001: Feder JN, Gnirke A, Thomas W, Tsuchihashi Z, Ruddy DA, Basava A, et al. A novel MHC class I-like gene is mutated in patients with hereditary haemochromatosis. Nat Genet 1996;13: Beutler E. Hemochromatosis: genetics and pathophysiology. Annu Rev Med 2006;57: Ellervik C, Tybjaerg-Hansen A, Appleyard M, Sillesen H, Boysen G, Nordestgaard BG. Hereditary hemochromatosis genotypes and risk of ischemic stroke. Neurology 2007;68:

10 1080 ELLERVIK ET AL. HEPATOLOGY, October DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986;7: Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ 2003;327: Egger M, Smith GD, Altman DG. Systematic Reviews in Health Care. London, England: BMJ Books; Bonkovsky HL, Troy N, McNeal K, Banner BF, Sharma A, Obando J, et al. Iron and HFE or TfR1 mutations as comorbid factors for development and progression of chronic hepatitis C. J Hepatol 2002;37: Powell EE, Ali A, Clouston AD, Dixon JL, Lincoln DJ, Purdie DM, et al. Steatosis is a cofactor in liver injury in hemochromatosis. Gastroenterology 2005;129: Chitturi S, George J. Interaction of iron, insulin resistance, and nonalcoholic steatohepatitis. Curr Gastroenterol Rep 2003;5: Alla V, Bonkovsky HL. Iron in nonhemochromatotic liver disorders. Semin Liver Dis 2005;25: Fargion S, Piperno A, Cappellini MD, Sampietro M, Fracanzani AL, Romano R, et al. Hepatitis C virus and porphyria cutanea tarda: evidence of a strong association. HEPATOLOGY 1992;16: Kordac V. Frequency of occurrence of hepatocellular carcinoma in patients with porphyria cutanea tarda in long-term follow-up. Neoplasma 1972; 19: Lee DW, Andersen JK, Kaur D. Iron dysregulation and neurodegeneration: the molecular connection. Mol Interv 2006;6: Feder JN, Penny DM, Irrinki A, Lee VK, Lebron JA, Watson N, et al. The hemochromatosis gene product complexes with the transferrin receptor and lowers its affinity for ligand binding. Proc Natl Acad Sci U S A1998; 95: Jones AC, Chuck AJ, Arie EA, Green DJ, Doherty M. Diseases associated with calcium pyrophosphate deposition disease. Semin Arthritis Rheum 1992;22: Jager HJ, Mehring U, Gotz GF, Neise M, Erlemann R, Kapp HJ, et al. Radiological features of the visceral and skeletal involvement of hemochromatosis. Eur Radiol 1997;7: Beutler E, Felitti VJ, Koziol JA, Ho NJ, Gelbart T. Penetrance of 845G a (C282Y) HFE hereditary haemochromatosis mutation in the USA. Lancet 2002;359: Andersen RV, Tybjaerg-Hansen A, Appleyard M, Birgens H, Nordestgaard BG. Hemochromatosis mutations in the general population: iron overload progression rate. Blood 2004;103: Juul K, Tybjaerg-Hansen A, Schnohr P, Nordestgaard BG. Factor V Leiden and the risk for venous thromboembolism in the adult Danish population. Ann Intern Med 2004;140: Tybjaerg-Hansen A, Jensen HK, Benn M, Steffensen R, Jensen G, Nordestgaard BG. Phenotype of heterozygotes for low-density lipoprotein receptor mutations identified in different background populations. Arterioscler Thromb Vasc Biol 2005;25: Adams PC, Reboussin DM, Barton JC, McLaren CE, Eckfeldt JH, McLaren GD, et al. Hemochromatosis and iron-overload screening in a racially diverse population. N Engl J Med 2005;352: Smith GD, Ebrahim S. Mendelian randomization: prospects, potentials, and limitations. Int J Epidemiol 2004;33: Halsall DJ, McFarlane I, Luan J, Cox TM, Wareham NJ. Typical type 2 diabetes mellitus and HFE gene mutations: a population-based case-control study. Hum Mol Genet 2003;12: Qi L, Meigs J, Manson JE, Ma J, Hunter D, Rifai N, et al. HFE genetic variability, body iron stores, and the risk of type 2 diabetes in U.S. women. Diabetes 2005;54: van der A DL, Peeters PH, Grobbee DE, Roest M, Marx JJ, Voorbij HM, et al. HFE mutations and risk of coronary heart disease in middle-aged women. Eur J Clin Invest 2006;36: Guerreiro RJ, Bras JM, Santana I, Januario C, Santiago B, Morgadinho AS, et al. Association of HFE common mutations with Parkinson s disease, Alzheimer s disease and mild cognitive impairment in a Portuguese cohort. BMC Neurol 2006;6: Njajou OT, Alizadeh BZ, Vaessen N, Vergeer J, Houwing-Duistermaat J, Hofman A, et al. The role of hemochromatosis C282Y and H63D gene mutations in type 2 diabetes: findings from the Rotterdam study and meta-analysis. Diabetes Care 2002;25: Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, Rennie D, et al., for the Meta-Analysis of Observational Studies in Epidemiology Group. Meta-analysis of observational studies in epidemiology: a proposal for reporting. JAMA 2000;283: