Blood Magnesium Level and Risk of Hepatocellular Carcinoma in a Prospective Liver Cirrhosis Cohort

Hepatocellular carcinoma (HCC) is a common lethal malignancy, and its incidence is increasing at the global level, especially in high-income counties (1). In the United States, liver cancer incidence, primarily HCC, has tripled since the early 1980s (2). Liver cirrhosis is a well-known risk factor for HCC, with an annual incidence of 2% to 4% (3, 4). Current surveillance tools, e.g., ultrasound and alpha-fetoprotein (AFP), have limited sensitivity and specificity for early HCC detection (5). Primary prevention and early detection to decrease HCC-related mortality require new biomarkers to inform risk stratification, screening, and treatments (3).

Magnesium is an essential mineral for ion transportation, DNA replication and repair, cell proliferation, and signaling (6, 7). The liver is a major organ for magnesium metabolism. Magnesium deficiency induces inflammatory responses (8), genomic instability, and changes in gene expression and cell programming that subsequently may contribute to liver malfunction and carcinogenesis (9). Previous studies have shown that higher magnesium intake was associated with a lower risk of metabolic dysfunction–associated steatotic liver disease (MASLD; ref. 10), mortality due to liver diseases (11), and the risk of HCC incidence and mortality (12, 13). Patients with liver cirrhosis usually have low magnesium levels compared with healthy controls (14). Limited evidence suggests that serum magnesium at diagnosis of HCC is lower than before diagnosis (15). However, this previous study is cross-sectional in nature (15). Few studies have yet examined whether low blood magnesium is longitudinally associated with the risk of HCC (16), especially not in those with liver cirrhosis, and no study has examined the associations with changes in liver biomarkers.

Therefore, we developed a new prospective cohort of patients with liver cirrhosis with available baseline serum magnesium level using a real-world electronic health data repository and investigated the role of serum magnesium in predicting risk of HCC. We hypothesized that a lower magnesium level would be associated with a higher risk of HCC and faster decrease of liver function in the patients with cirrhosis.

We designed a new prospective cohort study based on the existing infrastructure of the Mass General Brigham (MGB) Biobank, a regularly updated data repository of >140,000 consented patients within the MGB healthcare system (including Brigham and Women’s Hospital, Massachusetts General Hospital, and other affiliated hospitals; ref. 17) Linked longitudinal electronic medical record (EMR) data and self‐reported survey data were made available in a research data repository. Based on the International Classification of Diseases (ICD) ninth and tenth versions, we identified 2,930 patients with at least one diagnosis of liver cirrhosis (ICD-9: 571.2, 571.5, and 571.6; ICD-10: K70.3, K74.3-6; ref. 18) by January 28, 2022 and used their first diagnosis date as the baseline of this Harvard MGB Biobank Liver Cirrhosis Cohort. ICD codes have been validated in capturing liver disease encounters in large heath systems and administrative databases and are consistent with the Expert Panel Consensus Statement on EMR-based liver disease research (18, 19). We excluded patients with liver cancer history (N = 475) or without blood magnesium assessment (N = 1,025) at the time of cirrhosis diagnosis, leaving 1,430 patients with liver cirrhosis for analysis. Baseline characteristics between participants with and without blood magnesium record were compared. All patients enrolled in the MGB Biobank provided informed consent. This study was conducted in accordance with the Declaration of Helsinki. The Human Research Committee of MGB approved the Biobank research protocol (2009P002312 and 2015P000983).

The MGB Biobank Portal provides visibility into EMR data including laboratory test results. We accessed all available test results of magnesium in serum or plasma under the standardized Logical Observation Identifiers Names and Codes system. We converted all result units into mg/dL. Baseline magnesium levels of each patient were calculated as the mean of all test results within ±180 days of the first cirrhosis diagnosis. Magnesium deficiency was defined as blood magnesium <1.70 mg/dL (20).

We identified incident HCC based on at least one record of ICD-9 codes 155.0 and 155.9 or ICD-10 codes C22.0 and C22.9. We identified incident liver transplant based on at least one record of ICD-9 code V42.7 and ICD-10 code Z94.4. We accessed available EMR liver blood test results, including alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), direct bilirubin, total bilirubin, creatinine, international normalized ratio (INR), albumin, and gamma-glutamyl transferase (GGT; ref. 21) Because gamma-glutamyl transferase was only tested in <50% patients possibly as follow-up to other liver biomarker tests, it was not included in the final analysis. Model for end-stage liver disease (MELD) score was calculated as 3.8 × loge [total bilirubin (mg/dL)] + 11.2 × loge (INR) + 9.6 × loge [creatinine (mg/dL)] + 6.4 × (etiology: 0 alcoholic, 1 otherwise; ref. 22).

Demographic and clinical data included age, sex, race, ethnicity, smoking status, alcohol intake, physical activity, body mass index (BMI), type 2 diabetes, viral hepatitis status, estimated glomerular filtration rate (eGFR), and magnesium prescriptions, accessed from the MGB Biobank EMR. Questions on smoking status, alcohol intake, and physical activity were included in the Health Maintenance and Health Information Survey. Weight and height were measured at the encounter and calculated into BMI as weight (kg)/height² (m²). Type 2 diabetes was identified by ICD-10 code E11. 9. Hepatitis B virus (HBV) positive status was defined using the laboratory testing results of the HBV surface antigen. Hepatitis C virus (HCV) positive status was defined using the laboratory testing results of the HCV antibody. Results of the tests (positive/negative) were recorded into the EMR. Magnesium prescription included magnesium, magnesium carbonate, magnesium chloride, magnesium gluconate, magnesium oxide, and magnesium sulfate under the standardized RxNorm system.

Statistical analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC). P values < 0.05 were regarded as significant for two-sided tests. For the descriptive analyses, continuous variables were compared with means (SD), and categorical variables were compared with numbers (percentages).

Person-time was calculated for each patient from the date of cirrhosis diagnosis (baseline) until the first diagnosis of HCC, death, or end of follow-up (January 28, 2022), whichever occurred first, in the standard Cox proportional hazards models. In the competing risk model with liver transplant as the alternative outcome, participants were censored at the first record of liver transplant, death, or end of follow-up, and person-time was calculated accordingly. Upon checking the outcome distribution and proportional hazards assumption by the interaction term of magnesium and the time of follow-up, we found that the proportional hazard assumption was violated (P = 0.002; Supplementary Fig. S1) and thus stratified the models by time.

Blood magnesium was categorized into inversed quartile categories based on its distribution in the analytic cohort, dichotomized into deficiency (<1.70 mg/dL) and sufficiency (≥1.70 mg/dL) groups, and examined as continuous variables for the tests of linear trend (P-trend). Cox regressions were used to determine the HR and 95% confidence intervals (CI) of developing HCC, adjusting for age (year) and sex (male and female), and other a priori selected potential confounders, including race (White, Black, and other), smoking status (never, past, and current), alcohol intake (never, past, and current), BMI (<25, 25 to <30, and ≥30 kg/m²), physical activity (<75, 75 to <150, and ≥150 MET-minutes/week), type 2 diabetes (yes and no), HBV test positive (yes and no), HCV test positive (yes and no), MELD score (<10 and ≥10), and ALT (U/L; in quartiles). In the 1-year lagged analysis, we excluded HCC cases within 1 year following cirrhosis diagnosis to address potentially reverse causation.

As a secondary exposure, we examined the association between magnesium prescription and HCC risk. In sensitivity analyses, we excluded: (i) patients with HBV/HCV infection, (ii) patients with alcoholic liver disease (ALD; ICD-9: 571.0; ICD-10: K70), (iii) patients without MASLD, and (iv) patients who received liver transplant. We tested the interaction between blood magnesium and potential modifiers.

Based on the repeated laboratory assessments at each visit, after the cirrhosis diagnosis, generalized estimating equation models were used to estimate the association between baseline magnesium level and 12-year longitudinal changes in liver biomarkers, including ALT, AST, ALP, direct bilirubin, total bilirubin, and albumin. The same sets of covariates were included in the models.

This study was conducted in accordance with the Declaration of Helsinki. The Human Research Committee of MGB approved the Biobank research protocol (2009P002312, 2015P000983). Patient consent All patients enrolled in the Mass General Brigham Biobank provided informed consent.

Data underlying this article cannot be shared publicly due to the privacy of individuals that participated in the study. De-identified data and analytic methods may be shared in collaboration with a MGB investigator. Please contact the corresponding author for access to de-identified data upon reasonable request and approval.

Baseline characteristics of the main Harvard MGB Biobank Liver Cirrhosis Cohort are presented in Supplementary Table S1. Compared with those without blood magnesium record (N = 1,025), participants who were included in the study analysis (N = 1,430) were more likely to be male, receive magnesium prescriptions, and have higher MELD scores. No significant difference was observed for the record of liver transplant after liver cirrhosis (Supplementary Table S1). Baseline characteristics of the study population by quartiles of blood magnesium are presented in Table 1. Of the 1,430 patients with liver cirrhosis, 37% were female, and 87% were White. The median age at diagnosis of liver cirrhosis was 58.2 years. The highest quartile (Q4) had blood magnesium ≥2.10 mg/dL, compared with the lowest quartile (Q1) of <1.80 mg/dL. No significant trend of differences was observed for age, sex, race, or type 2 diabetes. Patients with lower blood magnesium levels were more likely to have lower education, drink alcohol, but less likely to smoke, have obesity (BMI ≥ 30 kg/m²), or be physically active. In addition, patients with lower blood magnesium levels have lower percentage of HBV positive but higher HCV positive tests. Interestingly, patients with lower blood magnesium levels seem to have lower ALT, AST, ALP, albumin, INR, creatinine, and MELD scores, although they are more likely to receive liver transplant (Table 1).

During a median follow-up period of 4.26 years, 109 patients developed HCC (Supplementary Fig. S1). In the multivariable adjusted model, patients with cirrhosis in the lowest quartile (median = 1.70 mg/dL) were found to have a 97% higher risk of HCC than those in the highest quartile of blood magnesium (median = 2.21 mg/dL; HR_Q1vs.Q4 = 1.97; 95% CI, 1.12–3.45; P-trend = 0.02; Table 2). When using blood magnesium clinical cut-off, comparing deficiency (magnesium < 1.70 mg/dL; N = 158) versus sufficiency (≥1.70 mg/dL; N = 1282), the fully adjusted HR was 1.93 (95% CI, 1.12–3.30). Results remained essentially the same after introducing competing risk with liver transplant. For example, the HR comparing magnesium deficiency versus sufficiency was 2.15 (95% CI, 1.22–3.77; Table 2). This association remained robust in the 1-year lag analysis (e.g., deficiency vs. sufficiency HR in the normal Cox model = 2.18; 95% CI, 1.11–4.28).

After excluding 237 patients with HBV/HCV infection, the pattern toward an association between lower blood magnesium and higher HCC risk remained in the same direction without reaching statically significance (HR_Q1vs.Q4 = 1.84; 95% CI, 0.88–3.87; P-trend = 0.11; Table 3). After excluding 293 patients with at least one diagnosis of alcoholic liver disease, the association remained significant (HR_Q1vs.Q4 = 2.14; 95% CI, 1.09–4.20; P-trend = 0.05; Table 3). After restricting to 682 patients who were diagnosed with MASLD, the trend toward a positive association between lower blood magnesium and higher HCC risk remained without reaching statically significance, especially in the competing risk model with liver transplant (HR_Q1vs.Q4 = 1.27; 95% CI, 0.48–3.36; Table 3). After excluding 133 patients who received liver transplants during follow-up, the association remained significant (HR_Q1vs.Q4 = 2.61; 95% CI, 1.26–5.43; P-trend = 0.009; Table 3). No interaction was found between magnesium and age, sex, race, smoking status, BMI, type 2 diabetes, or MELD score (all P-interaction >0.1; Supplementary Table S2). In the sensitivity analyses, no significant difference in risk was found comparing patients with liver cirrhosis who received magnesium prescription versus not (multivariable adjusted HR = 1.62; 95% CI, 0.79–3.28; Supplementary Table S3).

Lower blood magnesium was associated with a faster increase in ALT, direct bilirubin, and total bilirubin, marginally associated with AST, but not associated with ALP or albumin (Table 4). In the multivariable adjusted generalized estimating equation model, comparing the lowest (Q1) versus the highest (Q4) magnesium quartiles, the mean difference in the increase rate (coefficient of the magnesium × time interaction term) was 4.35 (95% CI, 1.06–7.63; P-trend = 0.02) U/L per year for ALT, 6.46 (95% CI, 0.28–12.6; P-trend = 0.04) U/L per year for AST, 0.18 (95% CI, 0.01–0.35; P-trend = 0.02) mg/dL per year for direct bilirubin, and 0.21 (95% CI, 0.03–0.39; P-trend = 0.02) mg/dL per year for total bilirubin.

We identified blood magnesium level as an independent predictor for HCC among patients with cirrhosis in this EMR-based prospective cohort study. A lower blood magnesium level was associated with a higher risk of developing incident HCC. If confirmed, our findings may potentially enable better identification of high-risk patients for HCC among patients with cirrhosis and inform better management strategies for liver cirrhosis.

A hospital-based case–control study of 130 patients with liver cirrhosis with HCC and 161 patients with cirrhosis without HCC in Italy found that in pre-HCC serum samples, per 1 mg/dL higher concentration of magnesium was associated with lower odds of HCC (OR = 0.05; 95% CI, 0.02–0.16), and that magnesium deficiency (<1.70 mg/dL) was associated with higher odds of HCC (OR = 3.44; 95% CI, 1.75–7.00; ref. 15). Our results, based on a prospective cohort design, are comparable with this previous study. We included a larger at-risk cirrhosis population and took into consideration more lifestyle covariates in our models, including smoking status, alcohol intake, and physical activity, as well as clinical factors including MELD score and liver transplant record. A recent study in patients with MASLD found that high serum magnesium is associated with lower risk of HCC (HR per SD higher = 0.88; 95% CI, 0.80–0.97) using a similar EMR-based cohort approach (16). Our results are also in line with prospective cohort studies in which a higher magnesium intake was associated with a lower risk of HCC incidence among the general population (12, 13). Collectively, blood magnesium may be an important marker for liver tumorigenesis. Keeping sufficient magnesium intake and circulating magnesium concentration might be clinically important in the primary prevention of HCC.

Moreover, our study for the first time reported that lower magnesium was associated with a faster increase in the ALT, AST, and bilirubin levels in patients with cirrhosis following diagnosis. The longitudinal data included rich EMR laboratory test results that allowed us to examine the change in liver biomarkers in relation to baseline magnesium levels and provide reference for cirrhosis monitoring and management. To date, few data existed on the association between circulating magnesium and liver biomarkers. A randomized phase II trial in patients with drug‐induced liver injury found that magnesium isoglycyrrhizinate could normalize ALT (23). Magnesium deficiency might induce inflammatory response, activation of leukocytes, increased intracellular calcium, and upregulation of NF-κB (24), which may cause liver injury. The liver synthesizes albumin, which is an important transporter of magnesium in circulation (25). When liver function decreases, albumin generation from liver reduces which may, in turn, lead to lower magnesium combined with albumin in blood (26). Although we did not find a significant association between magnesium and the longitudinal change in albumin, future studies are warranted to confirm our findings.

In addition to the potential interplay between magnesium homeostasis and liver function, several other mechanisms might explain the higher HCC risk in magnesium deficiency. First, magnesium is an essential enzyme cofactor involved in DNA repair, cell-cycle progression, cell proliferation, differentiation, and apoptosis (9). For HCC, importantly, magnesium might alter cell proliferation by blocking the MAPK signaling pathway, which is frequently activated in HCC (27, 28). Second, the gut microbiome and bile acid secretion, strongly implicated in liver carcinogenesis (29, 30), might be sensitive to magnesium intake and potentially the circulating magnesium. For example, studies have shown that magnesium treatment could alter gut microbial metabolites (e.g., imidazole propionate) and median chain fatty acids (31, 32). In addition to the metabolic-attributed risk, underlying causes such as alcoholic and viral cirrhosis might cause magnesium deficiency and confer a greater risk of HCC development (25). However, after excluding patients with cirrhosis with HBV/HCV infection and ALD in our study, the lower risk of HCC in higher magnesium levels persisted.

HCC risk varies widely in patients with liver cirrhosis from <0.2% to >5% per year (33). Current guidelines recommend a screening strategy including abdominal ultrasonography and blood AFP test in patients with cirrhosis every 6 months for HCC risk (34). Unfortunately, analysis of magnesium and AFP was limited by the small number of patients with AFP test results (<20%) provided in the MGB Biobank EMR. This small number might suggest difficulties in compliance with screening tests using AFP (35, 36). These data highlight the need for optimizing HCC surveillance use. If confirmed by future studies, our finding provides evidence for adding magnesium level to the HCC risk prediction models and clinical screening.

Our study has several limitations. First, only patients with measured magnesium levels were included, which might have introduced selection bias due to various reasons of ordering magnesium test at the clinic. Situations that might require magnesium assessment, e.g., using diuretic or heavy alcohol intake, are clinician dependent, without guideline or consistent practice. We compared the characteristics to those without magnesium tests and found that our study population was more likely to be male, receive magnesium prescriptions, and have higher MELD scores. Although these differences may limit the generalizability of the results to all patients with liver cirrhosis, the strong association between magnesium and HCC could inform etiologic research and lay ground for clinical cohort and trials of magnesium in which selection bias could be minimized. Second, the severity of cirrhosis impacts the magnesium metabolism and HCC risk (37), and we cannot rule out the potential influence of the severity of cirrhosis on our main findings. However, further adjustment for ALT (an indicator of liver function) and MELD (a measure of liver disease severity) did not materially change our findings. The competing risk model with liver transplant as the alternative event also did not materially alter the association between magnesium and HCC. Third, although we adjusted for potential confounders, residual confounding from other lifestyle factors, e.g., dietary intake, might still exist by the constraints of the MGB Biobank survey question availability. Fourth, the EMRs of magnesium prescription include any type of any amount at any time, and based on this definition, >80% patients had at least one magnesium prescription. Because of the limited number of patients without any prescription, there is not enough power to stratify the association between magnesium and HCC. Future studies with detailed information on patients’ actual intake and adherence to the supplement are warranted to examine the effect of magnesium supplementation. Last, this study was designed in a liver cirrhosis cohort that was highly enriched for White patients (87%). Future studies should investigate whether this finding could be generalized to the general population without cirrhosis and diverse ethnic populations.

Lower blood magnesium levels are associated with higher HCC risk and unfavorable liver biomarker changes, e.g., an increase in ALT, AST, and bilirubin, in patients with liver cirrhosis. Future studies should evaluate whether interventions to increase magnesium levels can alter HCC risk.

No disclosures were reported.

X. Zhang: Conceptualization, formal analysis, writing–original draft. L. Zhao: validation, writing–review and editing. Q. Dai: Conceptualization, writing–review and editing. T. Hou: Data curation, writing–review and editing. C.J. Danford: Writing–review and editing. M. Lai: Writing–review and editing. X. Zhang: Conceptualization, resources, data curation, supervision, methodology, writing–review and editing.

We thank the Mass General Brigham Biobank for providing health information data. X. Zhang is supported by NIH/NCI R21 CA238651, R21 CA252962, R37 CA262299, U01 CA259208, U01 CA272452 and American Cancer Society Research Scholar Grant (RSG NEC-130476) and American Cancer Society Interdisciplinary Team Award (PASD-22-1003396-01). The content is solely the responsibility of the authors and does not necessarily represent the official views of the funding agencies.