Quantifying risk modifiers of hereditary hemochromatosis using genomic and electronic health record data

Abstract

Background & Aims

Hereditary hemochromatosis is an autosomal recessive disorder of excessive iron accumulation. Early diagnosis enables treatment before organ damage. The C282Y variant in the HFE gene is the main cause, but its penetrance of only 20% limits its utility for population-wide screening. We aimed to identify and quantify novel genetic and non-genetic modifiers of C282Y-related disease from electronic healthcare records, and thereby partly explain its incomplete penetrance.

Methods

We carried out a cohort study on data from 420,543 individuals in the FinnGen project, for whom genotype information and healthcare records were available. We performed both standard and interaction genome-wide association study analyses for hemochromatosis and fitted statistical models including age, sex, 21 million genetic variants, preceding diagnoses, and blood donation history as predictors. Results were validated using data from the UK Biobank.

Results

We identified three novel fine-mapped variants within 4 Mb of the HFE gene. Of these, variant rs181949568 in the CASC15 gene remained significant in the multivariable model (odds ratio 7.25, 95% CI 3.63–28.87, p = 1.96 × 10^-8). We found that donating blood at least twice a year is likely sufficient to reduce the risk of C282Y homozygotes (male risk 0.16, 80% CI 0.13–0.19) to that of C282Y–H63D compound heterozygotes (male risk 0.018, 80% CI 0.015–0.023). Additionally, the S65C variant protects against severe disease (incidence ratio 0.328, 95% CI 0.192–0.562).

Conclusions

We demonstrated that use of large-scale electronic health record data allows for precise quantification of individual-level risk, which we present as risk tables to support clinical practice. Furthermore, our findings suggest that hemochromatosis may be under-recognized in Finland.

Impact and implications

Because the factors influencing the penetrance of the C282Y variant in hemochromatosis remain incompletely understood, a study leveraging newly available large-scale healthcare and genetic data is warranted. We present the findings of our study as an individual-level risk table designed for practicing clinicians, summarizing the combined effects of key variables most frequently observed in the dataset. Our results suggest that asymptomatic individuals who are homozygous for C282Y could significantly reduce their risk of developing hemochromatosis by donating blood just twice a year.

Graphical abstract

Highlights

• •We identified a novel Finnish-specific risk variant for hemochromatosis in the CASC15 gene.
• •We quantified the large protective effect of blood donation on hemochromatosis risk.
• •We confirmed the effects of HFE C282Y and H63D on the risk of hemochromatosis.
• •We identified that that HFE S65C variant protects from severe disease.
• •Our findings suggest underdiagnosis of hemochromatosis in Finland.

Introduction

Hereditary hemochromatosis is an autosomal recessive disorder in which excess iron accumulates in organs such as the liver, heart and joints. As iron in certain forms can induce oxidative stress, this buildup may lead to organ damage. Common consequences of hemochromatosis include liver disease, cirrhosis, congestive heart failure, diabetes, and arthritis.¹ The prevalence of hemochromatosis is highest in populations of Nordic or Celtic origin, with approximately one in every 220 to 250 individuals affected.²

The C282Y variant of the HFE gene is thought to downregulate the activity of hepcidin, whose absence in turn allows for uninhibited release of iron from enterocytes and macrophages into the circulation.³ Even though the homozygous C282Y variant has a prevalence of about 5 in 1,000 individuals of Northern European descent,⁴ only a small fraction of these individuals develop the disease. For instance, in the UK Biobank (UKBB), the penetrance in C282Y homozygotes was estimated to be 25.3% for males and 12.5% for females.⁵ By contrast, cumulative incidences up to age 80 of 56.4% and 40.5% for males and females, respectively, have been reported.⁶

Early diagnosis is important, so that treatment can be started before significant organ damage has occurred. The most common and effective treatment is removal of excess iron through venesections. Blood donation can also be used as treatment in the maintenance phase in case no organ damage has occurred. Moreover, blood donation may prevent organ damage among C282Y homozygotes, as in a recent study no indication of organ damage was detected in frequent blood donors homozygous for C282Y (Finnish Red Cross Blood Service, unpublished data).

Previously, Gallego et al. compared the penetrance of C282Y homozygotes and C282Y–H63D compound heterozygotes in the American eMERGE cohort.⁷ Pilling et al. investigated the effect of HFE variants on the baseline prevalence and incidence of hemochromatosis-related diseases and found significant differences in the incidence of associated conditions during a 7-year follow-up, as well as in the prevalence of associated conditions among males, when comparing C282Y homozygotes with C282Y wild-type individuals.⁸ They also studied the effect of genetic scores of four iron markers, cirrhosis, osteoarthritis and type 2 diabetes on the penetrance of HFE variants on hemochromatosis-related diseases and found significant associations between the scores and related diseases among C282Y homozygotes.⁵

In this paper, we use genetic and healthcare registry data to model the incidence and severity of hereditary hemochromatosis in the FinnGen cohort of 500,000 Finnish individuals (release 12). Our aim was to identify novel genetic variants and their interactions that may explain the low penetrance of HFE mutations. This information may help determine whether genetic screening for single nucleotide polymorphisms (SNPs) associated with hemochromatosis risk is warranted in Finland. We identified new SNPs and interactions associated with the diagnosis of hemochromatosis and present our findings as tables characterizing individual-level hemochromatosis risk.

Materials and methods

Materials

The FinnGen project⁹ recruited circa 520,000 individuals, aged ≥18 years, within the period of 2017–2023 through various Finnish biobanks, which cover roughly 10% of the Finnish population. For each participant, 21,331,644 imputed genetic variants, various laboratory measurements and other health registry data were available.

The UKBB¹⁰ recruited circa 500,000 individuals, aged 40–69 years, between 2006–2010. For each participant, genomic data, consisting of 96 million variants, together with extensive questionnaire, omics, and healthcare registry data, were available.

In FinnGen, the hemochromatosis diagnosis (endpoint E4_IRON_MET) is based on the ICD-10 code E83.1 and the ICD-9 code 2750. Individuals with metabolic disorders (ICD-10 codes E70–E90) were excluded from controls. The diagnoses came from the death register (follow-up from 1969 to 2021) and hospital discharge register (follow-up for inpatient and outpatient registers are 1969–28 April 2023 and 1998–28 April 2023, respectively). Although there is no official nationwide recommendation in Finland, a diagnosis of primary HFE hemochromatosis is typically made by a physician in patients with transferrin saturation >45%, ferritin above the laboratory- and sex-specific reference limit (often 200 μg/L for females and 300 μg/L for males), and an HFE genotype of either C282Y homozygosity or C282Y–H63D compound heterozygosity.¹¹ In the UKBB data we defined hemochromatosis similarly based on the ICD-10 codes.

Since menstruation reduces the risk of hemochromatosis through blood loss, it is important to include menopausal status in hemochromatosis models. We defined females aged ≤50 years as premenopausal.

We used the number of venesections performed as the measure of severity of hemochromatosis. The Nomesco code TPH00 and the SPAT code SPAT1074 (Finnish primary care outpatient procedures) capture venesection procedures in FinnGen, resulting in 9,421 venesections from 1,177 patients between 1998 and 2023. However, we excluded individuals with either polycythemia vera (1,236 individuals) or porphyria cutanea tarda (35 individuals) from the disease severity analyses as venesection is also used to treat these conditions.

Blood donation histories were available for 48,243 individuals between 2000 and 2021. We used the number of donations in the previous 5 years as a predictor of hemochromatosis incidence, survival and disease severity.

Ethical approval for this study (Ethical Committee statement number for the FinnGen study is Nr HUS/990/2017) was provided by the Coordinating Ethics Committee of the Hospital District of Helsinki and Uusimaa (HUS), and informed consent was obtained from all study participants (details are provided in the supplementary methods).

Statistical analyses

The additive hemochromatosis genome-wide association study (GWAS) was performed with Regenie, using age, sex, genotyping batch and the first 10 principal components as covariates on FinnGen data. Fine-mapping of causal variants was done using SuSiE. The extended MHC region (GRCh38, chromosome 6, 25–34 Mb) was excluded from fine-mapping due to its complex linkage disequilibrium structure. The association analyses between hemochromatosis and alleles of 15 HLA genes, as well as the genome-wide search of putative variants that additively interact with C282Y were performed with Regenie.

We modeled the individual-level hemochromatosis risk as the probability of hemochromatosis diagnosis at the end of follow-up given by a binary response logistic regression or an XGBoost classification model. The latter has the advantage of automatically selecting important variables and their interactions, as well as the functional form of the response. The survival from hemochromatosis was modeled using a Cox proportional hazards model. We used the number of venesections performed as a proxy for disease severity, modeled using either Poisson or negative binomial models – or their zero-inflated versions – since most individuals had no venesections, and even among hemochromatosis cases, fewer than half had received venesection.

We specified three versions of the incidence and survival models, differing by the set of predictors used:

• •Baseline model with only C282Y and H63D as genetic variables;
• •Comparison model that consisted of all relevant genetic and non-genetic prediction variables in common with FinnGen and UKBB;
• •Best FinnGen model that contained all relevant genetic and non-genetic FinnGen predictors.

We included interactions of C282Y with all other predictors that did not induce collinearity among the predictors (variance inflation factor <5). The SNPs were modeled additively, except for C282Y and H63D, which were modeled as categorical variables as they exhibited non-additive effects in previous literature. This allowed all nine combinations of these two variants. In the Best FinnGen model we also included correlated previous diagnoses of other diseases as predictors, see the supplement on the selection of diseases. For the disease severity model, we used the same predictors as in the Best FinnGen model except the preceding diagnoses were excluded to avoid fitting problems since venesection was quite rare.

Model selection was based on performance on held-out data, see supplementary methods. A fixed-effect meta-analysis was performed between effect sizes of comparison models of FinnGen and UKBB.

As a sensitivity analysis, we refitted the Comparison model and the Best FinnGen model using data restricted to cases with at least one venesection procedure in addition to a register-based hemochromatosis diagnosis. We also fitted an alternative version of the disease severity model, in which the response included only venesections from the first 2 years.

Further details, including the used R packages and their versions, can be found in the supplementary methods.

Results

The FinnGen cohort contained 520,210 individuals, and after the exclusions described in Fig. 1, 420,543 individuals remained. The UKBB included 502,547 individuals, and after exclusions, 318,778 individuals remained. Individuals with missing information were filtered out. Demographic information for the FinnGen and UK Biobank cohorts after exclusions is shown in Table 1. The prevalence of hemochromatosis in the UKBB was approximately 2.4 times that in FinnGen. The mean age of participants at baseline was also higher in the UKBB, reflecting the minimum recruitment age of 40. The male–female ratio was more balanced in the UKBB, while a clear majority of the participants were female in both cohorts.

Fig. 1.

The exclusions performed.

Exclusions in (A) the FinnGen cohort and in (B) the UKBB cohort. UKBB, UK Biobank.

Table 1.

Demographic statistics (after exclusions have been performed) in FinnGen and UKBB cohorts.

Variable	FinnGen			UKBB
	Overall	Male	Female	Overall	Male	Female
n	420,543 (100)	177,752 (42.3)	242,791 (57.7)	318,778 (100)	138,606 (43.5)	180,172 (56.5)
Hemochromatosis = case (%)	477 (0.1)	258 (0.1)	219 (0.1)	1,272 (0.4)	772 (0.6)	500 (0.3)
Blood donor (%)	53,499 (12.7)	21,003 (11.8)	32,496 (13.4)
Baseline age (median [IQR])	53.50 [37.08, 65.11]	57.02 [41.62, 66.85]	50.56 [34.76, 63.47]	57.00 [49.00, 62.00]	57.00 [49.00, 63.00]	57.00 [49.00, 62.00]
Age at diagnosis (median [IQR])	56.15 [46.86, 65.83]	55.84 [46.34, 66.51]	56.47 [47.29, 65.14]	64.03 [56.81, 70.69]	62.98 [56.00, 69.70]	65.45 [58.99, 72.10]
Donations in 5 previous years (median [IQR])	5 [3, 8]	6 [3, 10]	4 [2, 7]
Lifetime venesections (median [IQR])	3 [1, 9]	3 [1, 10]	2 [1, 7]

UKBB, UK Biobank. In FinnGen the distribution of age at diagnosis is shown for 477 individuals, the number of donations for 44,259 and the number of lifetime venesections for 452. In the UKBB, the age at diagnosis is available for 1,272 individuals.

GWAS analyses

A Manhattan plot of the hemochromatosis GWAS on FinnGen data (477 cases and 420,066 controls) is shown in Fig. 2, with the QQ-plot in Fig. S1 and lead variants in Table S1. The Manhattan plot revealed only a single peak on chromosome 6 at the HFE gene, ranging from approximately 21.9 Mb to 34.1 Mb. Fine-mapping of the GWAS hits produced three high-quality credible sets, whose representative SNPs (Causal1–Causal3) are listed in Table S2 and were chosen for further statistical analyses.

Fig. 2.

Manhattan plot of the hemochromatosis GWAS in FinnGen with 477 cases and 420,066 controls.

(A) Full Manhattan plot showing that the only peak resides near the HLA region in chromosome 6. (B) Manhattan plot zoomed in to the peak region. The HLA area and the extended HLA area are shown as gray boxes. The genome-wide significance threshold of 5 × 10^-8 is shown as a gray horizontal line. The annotation in (A) shows the gene at the peak, and annotation in (B) shows the SNPs used as covariates in the statistical models. GWAS, genome-wide association study; SNPs, single-nucleotide polymorphisms.

Because the extended HLA region was excluded from fine-mapping, we separately selected all independent significant lead SNPs from the HLA region. This resulted in SNPs HLA1 and HLA2 being included for further analysis (see supplemental results and Fig. S2). In addition, we selected SNPs C282Y, H63D,¹² S65C,¹³ H1-2, and HLA3¹⁴ based on prior literature.

As the HLA genes were not associated with hemochromatosis when adjusting for C282Y, we did not include any HLA alleles in the statistical models (see supplementary results, Fig. S3 and Table S3 for details).

We also used Regenie to find putative interactions with C282Y that could be associated with hemochromatosis. This GWAS was restricted to chromosome 6, as all significant associations in the basic GWAS were observed there, see Fig. S4A. All the significant interactions with C282Y were found in region 25,897,390–26,135,271 Mb, including the known interactions C282Y–H63D and C282Y–H1-2. Since we wanted to find new interactions, we performed the GWAS again, this time adjusted for the SNPs H63D, S65C, H1-2 and HLA3, and restricted to range chr6:24.8–29.7 Mb (see Fig. S4B and Table S4), resulting in 67 significant SNPs. We tested the utility of these SNPs in a later statistical model (the “All interactions” model) of hemochromatosis incidence.

SNPs

The selected SNPs, excluding the 67 C282Y-interaction SNPs, are described in more detail in Table S5 and Fig. 3, stratified by hemochromatosis status.

Fig. 3.

Information on the genetic variants putatively related to a registry diagnosis of hemochromatosis.

(A) Hemochromatosis prevalence in FinnGen and the UKBB stratified by sex. (B) MAFs stratified by the hemochromatosis status. (C) Observed minor allele dosages as bars, dosages expected based on MAF in red dots and the HWE test result among cases. The p values of SNPs significantly deviated from equilibrium are indicated in bold and italic. (D) Dosage combinations of C282Y and H63D occurring in FinnGen. (E) Dosage combinations in the UKBB. HWE, Hardy-Weinberg equilibrium; MAF, minor allele frequency; SNP, single nucleotide polymorphism; UKBB, UK Biobank.

Fig. 3A shows that the HC prevalence was more than twice as high in the UKBB as in FinnGen. In both cohorts, hemochromatosis was more common among males. Minor allele frequencies (MAFs) differed substantially between cases and controls, especially for C282Y and H1-2, except for H63D in the UKBB and HLA3 in FinnGen. The variants C282Y, H1-2, and H63D are not in Hardy-Weinberg equilibrium among cases, suggesting their importance in predicting future hemochromatosis diagnoses. Fig. 3D,E demonstrate that individuals who are homozygous for C282Y do not carry H63D mutations, and vice versa. Thus, the C282Y and H63D alleles are unlikely to occur on the same chromosome.

In FinnGen, 75% of hemochromatosis cases carried at least one mutation in C282Y or H63D, while in the UKBB the fraction was 88.6%. The penetrance of the homozygote C282Y and the compound heterozygote C282Y–H63D in FinnGen was 20.2% and 2%, respectively, while in the UKBB the corresponding values were 24.8% and 1.6%. Fig. 4 shows survival curves for hemochromatosis in FinnGen, stratified by sex and C282Y–H63D genotype.

Fig. 4.

Kaplan–Meier curves of time to hemochromatosis diagnosis stratified by sex and HFE genotype.

Model performance and selection

As the more complex models of hemochromatosis incidence did not significantly improve performance (see supplementary results, Fig. S5 and Table S6), we concentrated on the logistic regression models for further analysis. For disease severity, we tested Poisson and negative binomial models, as well as their zero-inflated versions. Based on Akaike information criterion values and the instability of the alternative models on the training data, the zero-inflated Poisson model was selected.

Inference

The odds ratios of the meta-analyzed logistic Comparison model and the Best FinnGen model; the hazards ratios of the meta-analyzed survival Comparison model and the Best FinnGen model; and the incidence ratios of the venesection model are shown in Fig. 5 and Table S7, with 95% CIs. Variable importance measures are shown in Fig. S7 and 8 and in Table S8 and 9.

Fig. 5.

Effect sizes of the hemochromatosis incidence (odds ratio from logistic regression), survival (hazard ratio from Cox model) and severity models (incidence ratio).

Full data was used to fit the models. For disease severity the number of venesections were modeled using a zero-inflated Poisson model with the same predictors for both the count and zero components. The 95% CIs are shown as horizontal lines. The minimum gap in years to previous disease diagnoses is indicated in parentheses. Interaction between two variables is denoted by the letter x. Note that, for example, the effect of being H63D–C282Y compound heterozygous is the sum of the effects of variables H63D heterozygote, C282Y heterozygote and their interaction. The effect alleles for the novel variants are given after the RSIDs. The age at the end of follow-up was standardized with units given in standard deviations.

Both logistic and Cox versions of the Best FinnGen model have similar effect sizes, except for the preceding diagnoses, the number of blood donations in the last 5 years and the premenopausal female status. Also, the meta-analyses of the logistic Comparison models and Cox Comparison models yielded similar effect sizes. The venesection model gave notably different effect size estimates compared to the rest of the models for predictors age, C282Y heterozygosity and homozygosity, C282Y–H63D compound heterozygosity, H63D homozygosity and S65C. Notably, S65C has the second largest protective effect from severe disease, after premenopausal status (incidence ratio 0.328, 95% CI 0.192–0.562).

Although the effect size of a single blood donation over 5 years is small, note that the unit corresponds to one donation. Thus, if a male donates the maximum of 30 times in 5 years, the odds ratio is raised to the power of 30. For example, with an odds ratio of 0.9 for a single donation, the odds ratio for 30 donations would be 0.9³⁰ ≈ 0.04, representing the strongest protective effect among all predictors.

Prior diagnoses of malignant neoplasm, hypothyroidism, hypertension, cholelithiasis, and arthrosis (see Table S10 for the required minimum time gap between the disease and hemochromatosis events) were associated with increased risk. Note that the venesection model includes fewer predictors than the Best FinnGen model, as the small number of venesections in the data permits fitting only models with a limited number of predictors.

We confirmed the large effect sizes of the C282Y and H63D SNPs, as well as their interaction. Similarly, we observed a significant risk-increasing effect of age (except in the Best FinnGen logistic model), and found that females were less likely to develop hemochromatosis. However, we also identified a risk-increasing interaction between postmenopausal status in females and C282Y homozygosity. In addition, the SNP Causal1 showed a clear risk effect in all FinnGen models, independent of C282Y, as demonstrated in the logistic regression for hemochromatosis among non-C282Y carriers (Fig. S9, Table S11). However, we were unable to replicate this finding in the UKBB due to the rarity of this SNP.

Sensitivity analyses, where the hemochromatosis cases were additionally required to have at least one venesection procedure (Fig. S10 and 11 and Table S12 and 13), showed good agreement with the respective original models. The results of the venesection model, where the venesections were from the first two years only, are shown in Fig. S14 and Table S14. Although the model contained noise, the main results of the original model still held.

Individual-level risk estimates

We modeled the individual-level hemochromatosis risk as the probability of hemochromatosis diagnosis during follow-up given by a logistic regression. To evaluate the utility of the prediction models, we computed the hemochromatosis risks for selected predictor combinations for the Comparison and Best FinnGen logistic models.

Fig. 6 and Table S15 highlight the large risk increasing effect of C282Y in the Comparison model, along with the large effect of sex.

Fig. 6.

Risk table of the FinnGen comparison incidence model.

The hemochromatosis risk (x-axis) with both 80% and 95% CIs for different predictor combinations. Predictor variables not shown were held fixed at mode or mean. The predictor combinations were ordered by the maximum risk over sexes. In parentheses the counts of predictor combinations in full data are shown (predictors not shown were ignored when counting).

In the risk table of the Best FinnGen model in Fig. 7 and in Table S16 the large protective effect of blood donation and the risk effect of C282Y and Causal1 SNPs were seen. Also, premenopausal females had much lower hemochromatosis risk than postmenopausal females.

Fig. 7.

Risk table of the Best FinnGen incidence model.

The hemochromatosis risk (x axis) with both 80% and 95% CIs for different predictor combinations. Predictor variables not shown were held fixed at mode or mean. The predictor combinations are ordered by the maximum risk over sexes. In parentheses the counts of genetic predictor combinations in full data are shown (predictors not shown were ignored when counting). The donations column gives the blood donations in the last 5 years.

For C282Y homozygotes without the Causal1 variant the hemochromatosis risk decreased significantly. For non-donors the risk estimates were 0.161 (80% CI 0.133–0.195), 0.086 (0.058–0.126), and 0.151 (0.123–0.184), for males, premenopausal, and postmenopausal females, respectively. For individuals who donated 10 times in 5 years, the respective estimates were 0.027 (80% CI 0.009–0.074), 0.013 (0.004–0.039), and 0.025 (0.009–0.068), for males, premenopausal, and postmenopausal females, respectively.

Moreover, for C282Y homozygotes who donated the maximum number of donations in 5 years (20 for males and 15 for females), hemochromatosis risks were 0.004 (80% CI 0.000–0.090), 0.005 (0.000–0.055), and 0.009 (0.001–0.098), for males, premenopausal and postmenopausal females, respectively. Although the risks were decreased, the confidence intervals widened.

Discussion

In this large study, we identified a novel hemochromatosis risk variant (rs181949568) in the CASC15 gene, demonstrated a protective role of the S65C variant against severe disease, and quantified the effect of blood donation on hemochromatosis risk. However, we did not identify any variants that strongly interact with C282Y.

The subpopulations of blood donors and non-blood donors in FinnGen appear to differ substantially. For example, the hemochromatosis prevalence was 0.12% in non-donors and 0.07% in donors (two-sample test for equality of proportions with continuity correction, p = 0.00359, 95% CI for the difference [-0.07% to -0.02%]). There is also a marked difference in C282Y prevalence in hemochromatosis cases with MAF being 39.9% in non-donors and 64.1% in donors (p = 4.024 × 10^-5, 95% CI for the difference [-36.4% to -12.8%]). This may reflect individuals diagnosed with hemochromatosis informing their relatives of the risk, prompting relatives to start donating blood as a preventive measure. This hypothesis is consistent with the observed enrichment of first-degree relatives among blood donors.¹⁵

Even though, for example, Pilling et al. have fitted separate models for males and females, in our data, the onset and prevalence of hemochromatosis did not differ substantially between sexes (Welch two-sample t-test for mean diagnosis age: males 56 years, females 55.2 years, p = 0.514). Therefore, we fitted a single model with sex as a predictor, which maximized statistical power given the relatively small number of hemochromatosis cases. Furthermore, because the minimum recruitment age in the UKBB was 40 years, it was not feasible to stratify females into pre- and postmenopausal groups in the UKBB data or in the Comparison models.

The 67 SNPs putatively interacting with C282Y discovered by Regenie did not significantly improve the prediction performance, see Fig. S6. Hence, we did not further investigate their individual contribution or mechanism.

Large differences in MAFs between cases and controls were observed in both C282Y and H1-2 in both FinnGen and the UKBB, suggesting their utility in predicting the hemochromatosis diagnosis. However, including both SNPs in the logistic models caused problems with collinearity, with the linkage disequilibrium between H1-2 and H63D being R² = 0.23, see Fig. S2. Hence, we excluded H1-2 from the logistic models, since it complicated the interpretation of the effect sizes. However, the SNP H1-2 was retained in the “All interactions” model, where it did not show a large effect, see Fig. S8 and Table S9 of variable importances, and in the non-C282Y-carrier model, where its effect was insignificant, see Fig. S9.

The MAF was greater in the UKBB for both C282Y (FinnGen 3.7%, UKBB 7.4%) and H63D (FinnGen 11%, UKBB 15.1%). However, in hemochromatosis cases, the MAF of H63D was higher in FinnGen. This is probably a consequence of the fact that the frequency of C282Y homozygotes in cases is much higher in the UKBB, Fig. 3D,E. As C282Y and H63D variants cannot exist in the same chromosome, this causes lower MAF of H63D in the UKBB in cases.

The relatively low proportion of HFE mutation carriers among hemochromatosis cases (75%) may reflect secondary hemochromatosis due to other underlying conditions or non-HFE hemochromatosis caused by variants in the HJV, HAMP, TFR2, or SLC40A1 genes.

The Causal1 SNP is located in the CASC15 (cancer susceptibility 15) gene, which has been demonstrated to be associated with hepatocellular carcinoma.¹⁶ In analyses restricted to non-C282Y carriers, this variant showed an effect on hemochromatosis independent of C282Y (Fig. S9) and was not in linkage disequilibrium with C282Y (Fig. S2). However, the variant is rare: all three identified causal SNPs had carrier frequencies of 0.3% in FinnGen and were even less frequent in the UKBB, leading to their exclusion from the Comparison models.

The SNP HLA3, located in the HLA region within the HLA-DQA1 gene, has been reported by Clancy et al. to be strongly associated with blood donor status.¹⁴ Neither the variant itself nor any allele of HLA-DQA1 (Fig. S3) was associated with hemochromatosis. However, a weak protective interaction between HLA3 and C282Y was observed in both FinnGen and the UKBB, although it did not reach the significance threshold.

Fig. 7 suggests that at least 10 blood donations per 5 years is enough to reduce the hemochromatosis risk in C282Y homozygotes to the level observed in compound heterozygotes. This is feasible because, in Finland, men may donate blood up to four times per year and women up to three times per year. An exception is women aged 18–25, who are advised to donate only once per year; however, they are not at risk of hemochromatosis.

Given that the penetrance of the C282Y variant is approximately 20% in FinnGen, additional factors must influence disease expression. Previous studies have proposed genetic modifiers of C282Y (e.g. Pilling et al.⁵), but we did not observe large interaction effects. Environmental factors are also relevant: alcohol intake and red meat consumption have been associated with hemochromatosis,³ but these data were unavailable in FinnGen. Future studies should include alcohol consumption, as it may partly explain the incomplete penetrance.

The prevalence of hemochromatosis in our data was 0.1% in FinnGen and 0.4% in the UKBB, while the MAFs of C282Y were 3.7% in FinnGen and 7.4% in the UKBB. The disparity between the ratio of prevalences and the ratio of MAFs suggests that hemochromatosis is underdiagnosed in Finland. This is further supported by the predictive value of the preceding diagnoses in our models.

Pilling et al. proposed that large proportions of C282Y homozygotes eventually develop hemochromatosis.⁵ This claim was based on survival curves from Gallego et al., which suggested a penetrance by age 90 of nearly 50% in males and 25% in females. However, that analysis included only 98 homozygotes, and the sex-stratified survival curves lacked confidence bands and numbers at risk, limiting the reliability of the estimates. In a comparable analysis in FinnGen, stratified by sex and C282Y dosage, we did not observe a significant difference in hemochromatosis-free survival between male and female homozygotes (Fig. 4). The hemochromatosis-free survival at age 90 was 66% for males (5 at risk; 95% CI 58%–75%) and 66% for females (7 at risk; 95% CI 59%–74%).

A weakness of our study is the reliance on registry data. Although the registry data in FinnGen is comprehensively available, with some registers dating back to the 1960s, the data may still contain missing or incorrectly entered values. As the ICD-10 code E83.1 is not fully specific due to the lack of official nationwide diagnostic criteria, some variation in diagnosis was observed. However, detailed sensitivity analysis and validation showed the robustness of our results.

Due to the Causal1 variant being rare in the UKBB and the lack of blood donation information, these findings could not be replicated. But as the effects of other variants were remarkably similar in FinnGen and the UKBB, which represent genetically quite distinct populations, and as the sensitivity analyses supported our key findings, the results should be broadly applicable. Given that the biology related to the HFE variants should operate similarly across populations, the protective effect of blood donation should also be generalizable.

Our findings underscore the importance of blood donation in reducing hemochromatosis risk. Therefore, special attention should be given to blood donors diagnosed with hemochromatosis or at risk of developing the disease. Given that genetic data is becoming increasingly available, our results highlight the need to make C282Y homozygotes aware of their risk and the possibility of maintaining their health and helping others through blood donation. At least eight blood services have allowed shorter donation intervals for HFE variant carriers or hemochromatosis patients.¹⁷ Encouraging therapeutic donors to see themselves as donors rather than patients may benefit both the blood service and the donors, fostering commitment and positive advocacy.¹⁸ In addition, iron supplementation, routinely offered to frequent donors by the blood service in Finland, should not be given to C282Y homozygotes.

In conclusion, although we did not identify strong effects among the newly discovered modifiers of the C282Y variant, we found a novel variant, Causal1, with a risk effect independent of C282Y. We also demonstrate that large-scale electronic health record data enable precise quantification of individual-level risk, which is particularly useful because such results can be applied to clinical practice more directly than population-level effects. While our findings are already relevant for clinical care, incorporating data on alcohol consumption and iron intake could further improve risk precision. Additionally, our results suggest that hemochromatosis is under-recognized in Finland. We also recommend that blood services be aware of donors carrying hemochromatosis risk variants. Recent evidence indicates that blood donors are highly willing to receive health-related information;¹⁹ conversely, carriers of risk variants should be informed about the potential benefits of blood donation.

Abbreviations

GWAS, genome-wide association study; MAF, minor allele frequency; SNP, single nucleotide polymorphism; UKBB, UK Biobank.

Authors’ contributions

JT, JC, JR and MA designed the study; FinnGen performed the hemochromatosis GWAS and finemapping while JT performed all other analyses; FÅ provided medical expertise; JT wrote the initial version of the manuscript. All authors read and critically commented the manuscript.

Data availability

The FinnGen individual-level data is not publicly available. But researchers can apply for the health register data from the Finnish Data Authority Findata (https://findata.fi/en/permits/) and for individual-level genotype data from Finnish biobanks via the Fingenious portal (https://site.fingenious.fi/en/) hosted by the Finnish Biobank Cooperative FINBB (https://finbb.fi/en/). Information on how to download FinnGen summary statistics can be found from https://finngen.gitbook.io/documentation/data-download.

Code availability

All analysis R code will be made available at GitHub: https://github.com/FRCBS/hemochromatosis_modeling_article.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process

During the preparation of this work the authors used Microsoft Copilot to improve language and readability. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.

Financial support

Jarkko Toivonen was funded by the State Research Funding.

Conflict of interest

The authors declare no conflicts of interest pertaining to this manuscript.

Please refer to the accompanying ICMJE disclosure forms for further details.

Appendix A Supplementary data

The following are the Supplementary data to this article: