Circulating Ferritin in Patients with Nonalcoholic Fatty Liver Disease: A Systematic Review and Meta-analysis

Abstract

Objectives

To synthesize data on circulating ferritin between patients with histologically confirmed nonalcoholic fatty liver disease (NAFLD) and non-NAFLD controls.

Methods

A systematic literature search was conducted in PubMed, Scopus, and the Cochrane Library. Thirty-one studies comprising data on 5631 individuals (2929 biopsy-proven NAFLD patients and 2702 controls) were included in the meta-analysis.

Results

Higher circulating ferritin levels were observed in NAFLD patients than in controls [standardized mean difference (SMD) 1.14; 95% confidence interval (95% CI) 0.73–1.55], in patients with simple nonalcoholic fatty liver (NAFL) than in controls (SMD 0.57; 95% CI 0.34–0.80), in patients with nonalcoholic steatohepatitis (NASH) than in controls (SMD 0.95; 95% CI 0.69–1.22), and in NASH than in NAFL patients (SMD 0.62; 95% CI 0.25–0.99). There was moderate-to-high heterogeneity among studies in the above pairs of comparisons (I² = 68–97%); no risk of publication bias was observed by Egger's test (P = 0.81, P = 0.72, P = 0.59, P = 0.42, respectively). The heterogeneity was reduced in the subgroup of biopsy-proven controls in all pairs of comparisons (I² = 0–65%). The heterogeneity was also reduced after excluding studies with the Newcastle–Ottawa Scale (NOS) score <7 (n = 10) for the comparison of NAFLD patients vs. controls (I² = 54%, P = 0.02). The meta-regression analysis revealed that the male ratio was positively associated with ferritin SMD in the comparison between NAFLD patients and controls and accounted for 32.7% (P = 0.002) of the heterogeneity in this pair of comparison.

Conclusions

Circulating ferritin was higher in NAFLD (or NAFL or NASH) patients compared with controls. Higher levels of circulating ferritin were also associated with the severity of the disease, which, however, should be cautiously interpreted.

PROSPERO registration ID: CRD42022354025.

Graphical abstract

Nonalcoholic fatty liver disease (NAFLD) is considered a leading cause of chronic liver disease worldwide, with a prevalence of 30% in the general population.¹ This clinicopathological entity is defined as hepatic steatosis diagnosed by histology or imaging, while excluding secondary causes of liver steatosis (e.g. viral, autoimmune, iatrogenic).^2,3 The NAFLD spectrum includes nonalcoholic fatty liver (NAFL) and nonalcoholic steatohepatitis (NASH), with varying degrees of hepatocyte inflammation and necrosis, which may evolve into fibrosis, cirrhosis, and/or hepatocellular carcinoma in a few patients.⁴ The pathogenesis of NAFLD is multifactorial, with insulin resistance (IR) playing a central role, interacting with other contributing factors such as adipokines, oxidative stress, the intestinal microbiome and even iron overload.⁵ NAFLD leads to hepatic and extra-hepatic morbidity, with cardiovascular diseases and malignancies being the two primary causes of mortality.⁶ Although there is to date no approved medication for NAFLD, the multifactorial nature of the disease may imply the need of combination therapy in order to target multiple contributing factors.⁷

Concerning diagnosis, histological confirmation following a liver biopsy is the gold standard.² However, liver biopsy is an invasive method, so there is risk of complications and even a low but substantial risk of death. The risk of systematic bias, due to the evaluation of a small part of a large organ, and inter-observer variability also exist.⁸ Bearing in mind the high prevalence of the disease, performing a liver biopsy on all NAFLD patients and repeating the procedure during the follow-up is neither safe nor cost-effective.⁹ In this regard, the research is shifted towards noninvasive diagnostic methods, either imaging techniques or noninvasive biomarkers.^8,10

Ferritin is an acute-phase protein that increases in inflammatory states and is a biomarker of iron storage.¹¹ Considering inflammation as a substantial driver of NAFLD and also due to the association of ferritin with IR, ferritin has been studied in NAFLD; however, high circulating ferritin levels have been reported in NAFLD by some,^12,13 but not all authors.^{14, 15, 16, 17, 18}

Thus, the primary aim of this systematic review and meta-analysis was to quantitatively synthesize existing evidence concerning the circulating levels of ferritin in NAFLD patients in comparison to non-NAFLD controls. Secondary aims were the comparison of circulating ferritin levels between NAFL or NASH patients and controls, as well as between NAFL and NASH patients.

Methods

Literature Search

This systematic review and meta-analysis was conducted according to a protocol pre-registered in the PROSPERO registry (CRD42022354025). The reporting guidelines of the Meta-analysis of Observational Studies in Epidemiology (MOOSE) were followed to conduct the systematic review and to write this article.¹⁹

A systematic literature search in the electronic databases PubMed, Scopus, and Cochrane Library (starting September 18, 2022) was performed by two independent reviewers (E.M. and O.M.) using the following search string: ((“non-alcoholic fatty liver disease” [MeSH Terms]) OR (non-alcoholic fatty liver disease) OR (nonalcoholic fatty liver disease) OR (non alcoholic fatty liver disease) OR (“fatty liver” [MeSH Terms]) OR (fatty liver) OR (fatty liver disease) OR (nonalcoholic fatty liver) OR (non-alcoholic fatty liver) OR (non alcoholic fatty liver) OR NAFLD OR FLD OR steatohepatitis OR (non-alcoholic steatohepatitis) OR (non alcoholic steatohepatitis) OR (nonalcoholic steatohepatitis) OR NASH OR steatosis OR (liver steatosis) OR (hepatic steatosis) OR (metabolic associated fatty liver disease) OR (metabolic dysfunction associated fatty liver disease) OR (metabolic dysfunction-associated fatty liver disease)) AND ((“ferritins” [MeSH Terms]) OR (“iron” [MeSH Terms]) OR (ferritin) OR (iron)) for PubMed, and similar queries for Scopus and Cochrane Library, based on the syntactic requirements of each database. There were no limitations on the publication date or language. Search was extended up to July 20, 2023, through activating alerts in PubMed (“My NCBI”), Scopus (“Alerts”), and Cochrane Library (“Saved Search Alert”), which were based on the same search string.

Hand (manual) searching was also performed, including the reference lists of the eligible articles and the abstract books of five major Hepatology and Hematology Congresses during the last 5 years (2018–2022): the European Association for the Study of the Liver (EASL), the American Association for the Study of Liver Diseases (AASLD), the Asian-Pacific Association for the Study of the Liver (APASL), the European Hematology Association (EHA), and the American Society of Hematology (ASH) congresses.

Inclusion and Exclusion Criteria

Observational studies (cross-sectional, case-control, and cohort studies) providing comparative data on circulating (serum or plasma) ferritin levels for patients with NAFLD (NAFL and/or NASH) and non-NAFLD controls were considered eligible for this systematic review and meta-analysis. Inclusion criteria were: 1) histological confirmation of NAFLD patients; and 2) comparative data on circulating ferritin levels for NAFLD (NAFL and/or NASH) patients vs. non-NAFLD controls.

Exclusion criteria were: 1) patients with other liver disease (e.g. viral hepatitis, alcoholic fatty liver disease, autoimmune hepatitis, drug-induced liver injury, Wilson's disease, alpha-1 antitrypsin deficiency) or coexistence of NAFLD with other liver diseases; 2) patients' overlap in two or more studies; 3) patients with hemochromatosis; 4) patients with any acute disease; 5) studies for which additional data were required, but the corresponding authors did not respond or were unwilling to provide them; 6) other types of articles, including animal or cell line studies, reviews, editorials, guidelines, opinions, commentaries, hypotheses, book chapters, or case reports; letters to the editor were considered only if original data were reported in them.

Data Extraction

The retrieved articles were imported into a reference manager software (EndNote, Clarivate Analytics, Philadelphia, PA, USA). Proceeding to the stage of screening, after manual deduplication, the titles and abstracts were assessed by two independent reviewers (E.M. and O.M.) for inclusion in the eligibility stage or exclusion based on the aforementioned criteria. Next, the potentially eligible studies were read on a full-text level. Disagreements between the reviewers were referred for discussion with the participation of the supervisor (S.A.P.) until a consensus was reached. The supervisor (S.A.P.), qualified in systematic review and meta-analysis, guided the reviewers throughout the procedure and adjudicated on any disagreement between them.

Subsequently, for each eligible study, the following parameters were extracted: 1) characteristics of the publication (first author's surname; country of origin; year of publication; design of the study); 2) characteristics of the study population (e.g. pediatric/adolescent participants; morbidly obese participants; the performance of liver biopsy in the controls; inclusion of patients with NASH-related cirrhosis); 3) patient and control characteristics (numbers per group, sex, age, body mass index [BMI], number of patients with type 2 diabetes [T2DM]); 4) method of ferritin measurement; 5) histological system used for the grading and staging of NAFLD; 6) laboratory measurements (ferritin, aspartate aminotransferase [AST], alanine aminotransferase [ALT] and gamma-glutamyl-transferase [GGT] levels in mean ± standard deviation [SD]) and 7) homeostasis model assessment - IR (HOMA-IR) in mean ± SD.

When essential additional data were required (e.g. ferritin levels) and the attempt to communicate with the corresponding author(s) failed or the corresponding author(s) were not willing to provide them, the study was excluded. In cases of possible overlap between the populations of two or more studies, only data from the larger study were extracted. When necessary, statistical transformations (e.g. standard error of the mean to SD) were made by standard formulas or calculated via the online tool StatsToDo (https://www.statstodo.com/index.php), and data from figures were calculated via the web-based tool WebPlotDigitizer (https://automeris.io/WebPlotDigitizer/). In the case of articles written in languages other than English, Greek, French, and German (some or all authors are familiar with these languages), the retrieved articles were first translated with both Google Translate (https://translate.google.gr) and the online translator DeepL (https://www.deepl.com/translator); however, confirmation of the consistency of the translated data was asked by the corresponding authors of all the respective articles.

Quality Assessment

The quality of the included studies was assessed by two independent reviewers (E.M. and O.M.) applying the Newcastle–Ottawa Scale (NOS) (Ottawa Hospital Research Institute, Ottawa, ON, Canada). Disagreements between the reviewers were referred for discussion with the participation of the supervisor (S.A.P.) until a consensus was reached. According to NOS, the quality assessment was rated on a scale from 0 (the lowest score; poor quality) to 9 (the highest score; high quality).

Outcomes

The main outcome of this systematic review and meta-analysis was the standardized mean difference (SMD) of circulating (serum or plasma) ferritin levels between NAFLD patients and controls. In the case of available data for classification of NAFLD patients into NAFL and NASH groups, further comparisons were performed, leading to secondary outcomes, i.e. SMD between: 1) NAFL patients and controls; 2) NASH patients and controls; and 3) NASH and NAFL patients.

Statistical Analysis

The statistical analysis was conducted with the R software (R for Windows, version 4.4.1; the R Foundation for Statistical Computing, Vienna, Austria) and Review Manager (Revman, version 5.4.1; Cochrane Collaboration, London, UK). A random-effects inverse-variance model was applied for the calculation of the SMD of ferritin levels, with 95% confidence intervals (95% CI), due to the expected high heterogeneity among studies. Statistical tests were two-sided, and the level of significance was set at a P-value <0.05. The evaluation of heterogeneity was performed with the I² test. Publication bias was evaluated with the Egger's test and the visual inspection of the funnel plots. We also performed subgroup, sensitivity, and meta-regression analyses. Subgroup analysis was performed for the comparisons between: 1) studies with histological confirmation of NAFLD in controls vs. studies without histological confirmation in controls; 2) studies evaluating liver histology by NAFLD activity score (NAS)²⁰ vs. studies evaluating liver histology by other histological scoring systems (e.g. Brunt,²¹ etc.); 3) studies containing vs. not containing patients with NASH-related cirrhosis. Sensitivity analysis for primary and secondary outcomes was also conducted after the exclusion of studies: 1) of lower quality, as assessed by NOS (NOS <7); 2) with morbidly obese populations; 3) with metabolic dysfunction-associated fatty liver disease (MAFLD) populations; 4) with extreme ferritin levels; 5) with possible data inconsistency. Meta-regression was also performed in order to regress male percentage, T2DM percentage, age, BMI, or HOMA-IR on ferritin SMD based on aggregate-level data.

Results

Literature Search

We initially retrieved 2946 articles from Scopus, 2088 from PubMed, and 43 from the Cochrane Library, as well as 282 additional articles added through hand searching in: 1) key international conferences (n = 17); 2) automatic alerts set in PubMed, Scopus, and the Cochrane Library (n = 257); 3) references to the included articles (n = 8). After the exclusion of duplicates (n = 1549), 3084 articles were excluded at the stage of screening, and 695 articles were excluded at the stage of eligibility. The attempt to request additional data from the corresponding authors led to the exclusion of 31 articles since the data were considered crucial (e.g. ferritin levels were neither reported nor feasible to be calculated). On the contrary, the authors of 16 articles provided the requested data; their contribution is acknowledged in the acknowledgment section. At the end, 31 studies were included in this systematic review and meta-analysis.^{22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52} It is highlighted that none of the initially retrieved congress abstracts was finally included in the systematic review and meta-analysis since none met the prespecified inclusion and exclusion criteria and/or the corresponding author did not provide us with the requested essential data. A flowchart summarizing the stages of identification, screening, eligibility, and final selection of studies, based on the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines,⁵³ is presented in Figure 1.

Figure 1.

A flowchart presenting the literature search process, according to the PRISMA statement. PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

Characteristics of the Included Studies

The 31 included studies were published between 2003 and 2022 and reported data from 5631 individuals (2929 NAFLD patients and 2702 controls).^{22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52} The main characteristics of these studies are presented in Table 1. Fourteen studies were carried out in Europe, eleven in Asia, four in Africa, and two in America (1 in Northern and 1 in Southern America). Sixteen studies were case-control, thirteen were cross-sectional and two were cohort (Table 1). Five studies were performed on morbidly obese populations subjected to bariatric surgery.^{25,33,40,50,51} Liver biopsies on controls were performed in nine studies.^{25,28,31,38,40,41,47,50,51} Unexpectedly, the method of measurement of circulating ferritin levels was not reported in most studies. They were reportedly measured by enzyme-linked immunosorbent assay (ELISA) in two studies, electrochemiluminescence immunoassay (ECLIA) in one, chemiluminescence immunoassay (CLIA) in three, immunoassay method (not specified) in one, and nephelometric method in one (Table 1). Only one article written in a language other than English (French) was finally included in this meta-analysis (Table 1).²³

Table 1.

Main Characteristics of Studies Included in the Systematic Review and Meta-analysis.

First author, year, origin [Reference]a	Study design	Method of ferritin measurement	Histological definition	Biopsy in controls	Cirrhosis (n)	NOS score	Additional information
Boga, 2015, Turkey22	Case-control	na	NAS	No	na	7
Chellali, 2019, Algeria23	Case-control	na	NAS	No	na	5	In French language
Colak, 2012, Turkey24	Case-control	na	NAS	No	na	7
Dali-Youcef, 2019, France25	Case-control	CLIA	NAS	Yes	na	6	Morbidly obese population
Di Rosa, 2013, Italy26	Case-control	na	NAS	No	0	6
Domenici, 2013, Brazil27	Case-control	na	Brunt	No	na	5
El Nakeeb, 2017, Egypt28	Case-control	ELISA	Brunt	Yes	0	5
Erhardt, 2011, Germany and Italy29	Case-control	Na	Brunt	No	na	7
Feldman, 2021, Austria30	Cohort	Na	NAS	No	11	6
Garcia-Monzon, 2015, Spain31	Cross-sectional	Na	Brunt	Yes	0	6
Hanafy, 2019, Egypt32	Case-control	ECLIA	NAS	No	0	6
Hernández-Aguilera, 2021, Spain33	Cohort	Immunoassay	NAS	No	2	5	Morbidly obese population
Hoki, 2015, Japan34	Case-control	Na	NAS	No	na	6
Kaya, 2013, Turkey35	Case-control	Na	Brunt	No	0	8
Koruk, 2003, Turkey36	Case-control	Nephelometric method	Brunt	No	0	6
Lang, 2019, Germany37	Cross-sectional	na	NAS	No	7	5
Malaguarnera, 2006, Italy38	Cross-sectional	na	Brunt	Yes	na	5
Marmur, 2018, Sweden39	Cross-sectional	na	NAS	No	na	6
Marques, 2021, Portugal and Spain40	Case-control	na	NAS	Yes	18b	7	Morbidly obese population
Nagaya, 2010, Japan41	Cross-sectional	na	NAS	Yes	20b	5
Parikh, 2015, India42	Case-control	na	Brunt	No	0	8
Polyzos,2012, Greece43	cross-sectional	CLIA	NAS	No	0	7
Priya, 2010, Libya44	Cross-sectional	na	na	No	na	7
Sazci, 2008, Turkey45	Cross-sectional	na	Brunt	No	na	7
Tarantino, 2011, Italy46	Cross-sectional	na	na	No	na	6
Tsuchiya, 2010, Japan47	Cross-sectional	na	na	Yes	na	5
Valenti, 2012, Italy48	Case-control	na	NAS	No	38c	6
Vecchiet, 2005, Italy49	Cross-sectional	CLIA	Brunt	No	na	6
Vuppalanchi, 2014, USA50	Cross-sectional	ELISA	NAS	Yes	1	6	Morbidly obese population
Wu, 2022, China51	Cross-sectional	Na	NAS	Yes	na	7	Morbidly obese population
Yoneda, 2009, Japan52	Case-control	Na	Brunt	No	2	5

Abbreviations: CLIA, chemiluminescence immunoassay; ECLIA, electrochemiluminescence; ELISA, enzyme-linked immunosorbent assay; F, fibrosis stage; na, not available; NAS, NAFLD activity score; NOS, Newcastle–Ottawa Scale.

^aReferences are sorted according to the first author's surname.

^bSum of Fibrosis stages 3 and 4, the latter being cirrhosis.

^cSum of Fibrosis stages 2, 3 and 4, the last being cirrhosis.

The main demographic and laboratory characteristics per group of each study included in the systematic review and meta-analysis are presented in Supplementary Table 1, including circulating ferritin levels, sex, age, BMI, waist circumference, number of patients with T2DM, liver function tests (AST, ALT, and GGT levels), and HOMA-IR. Seventeen of the 31 studies included all groups of interest (controls, NAFL, and NASH patients); data were not provided separately for NAFL and NASH in nine studies, and five studies consisted only of NASH patients. In summary, comparative data on ferritin levels between NAFLD patients (n = 2929) and controls (n = 2702) were retrieved from 31 studies, between NAFL patients (n = 629) and controls (n = 690) from 17 studies, between NASH patients (n = 1087) and controls (n = 1105) from 22 studies, and between NAFL patients (n = 906) and NASH patients (n = 629) from 17 studies (Supplementary Table 1).

Quality of Included Studies

The NOS scores of the 31 included studies are presented in Table 1 and in detail in Supplementary Table 2. Specifically, the NOS score was 5 in 9 studies, 6 in 12 studies, 7 in 8 studies, and 8 in 2 studies. The mean (±SD) NOS score was 6.1 ± 0.91.

Outcomes

Higher circulating ferritin (Table 2; Figure 2a–d) were observed in: 1) NAFLD patients than controls (SMD 1.14; 95% CI 0.73–1.55); 2) NAFL patients than controls (SMD 0.57; 95% CI 0.34–0.80); 3) NASH patients than controls (SMD 0.95; 95% CI 0.69–1.22); and 4) NASH patients than NAFL patients (SMD 0.62; 95% CI 0.25–0.99). There was high heterogeneity among studies in all comparisons (I² was 68–97%; Table 2; Figure 2a–d). Egger's test showed no significant publication bias in all pairs of comparisons (Table 2). However, the funnel plots of the comparisons between NAFLD vs. controls, NAFL vs. controls, and NASH vs. NAFL patients showed a degree of visual asymmetry (Table 2; Supplementary Figures 1a1, 1b1, 1d1). In order to investigate the source of heterogeneity, subgroup, sensitivity, and meta-regression analyses were performed.

Table 2.

Ferritin SMD Between Groups in all Studies and After Sensitivity Analysis.

Comparison	All studies	Excluding studies with morbidly obese patients	Excluding studies with NOS <7	Excluding Nagaya et al. study41	Excluding Wu et al. study51	Excluding Hanafy et al.32 and Valenti et al. studies48
NAFLD vs. control	1.14 (0.73–1.55); P < 0.00001	1.26 (0.79, 1.72); P < 0.00001	0.77 (0.57, 0.98), P < 0.00001	1.17 (0.75, 1.59); P < 0.00001	1.16 (0.74, 1.58), P < 0.00001	0.84 (0.67, 1.00), P < 0.00001
	I2 = 97%; P < 0.00001	I2 = 97%; P < 0.00001	I2 = 54%; P = 0.02	I2 = 97%, P < 0.00001	I2 = 97%; P < 0.00001	I2 = 79%; P < 0.00001
	Egger's test P = 0.806	Egger's test P = 0.989	Egger's test P = 0.460	Egger's test P = 0.888	Egger's test P = 0.840	Egger's test P = 0.024
NAFL vs. control	0.57 (0.34–0.80); P < 0.00001	0.59 (0.38, 0.79); P < 0.00001	0.42 (0.01, 0.83); P = 0.05	0.59 (0.35, 0.82); P < 0.00001	0.58 (0.34, 0.82); P < 0.00001	not applicable
	I2 = 68%; P < 0.0001	I2 = 39%; P = 0.08	I2 = 62%; P = 0.03	I2 = 70%; P < 0.0001	I2 = 70%; P < 0.0001
	Egger's test P = 0.7196	Egger's test P = 0.2459	aEgger's test P = 0.6599	Egger's test P = 0.5709	Egger's test P = 0.7161
NASH vs. control	0.95 (0.69–1.22); P < 0.00001	1.06 (0.84, 1.27); P < 0.00001	0.88 (0.54, 1.22); P < 0.00001	0.99 (0.72, 1.25); P < 0.00001	0.97 (0.70, 1.25); P < 0.00001	not applicable
	I2 = 84%; P < 0.00001	I2 = 67%; P < 0.0001	I2 = 78%; P < 0.0001	I2 = 84%; P < 0.00001	I2 = 84%; P < 0.00001
	Egger's test P = 0.589	Egger's test P = 0.854	aEgger's test P = 0.595	Egger's test P = 0.769	Egger's test P = 0.573
NASH vs. NAFL	0.62 (0.25–0.99); P = 0.001	0.87 (0.33, 1.41); P = 0.001	1.37 (0.23, 2.50); P = 0.02	0.66 (0.28, 1.05); P = 0.0007	0.66 (0.26, 1.06); P = 0.001	not applicable
	I2 = 89%; P < 0.00001	I2 = 91%; P < 0.00001	I2 = 96%; P < 0.00001	I2 = 90%; P < 0.00001	I2 = 90%; P < 0.00001
	Egger's test P = 0.425	Egger's test P = 0.513	aEgger's test P = 0.186	Egger's test P = 0.356	Egger's test P = 0.471

Sensitivity analyses results are presented in SMD (95% Confidence Interval); P-value, I² test; P-value, Egger's test P-value (if P > 0.05, no publication bias is presented).

Abbreviations: NAFLD, nonalcoholic fatty liver disease; NAFL, nonalcoholic fatty liver; NASH, nonalcoholic steatohepatitis, NOS, Newcastle–Ottawa Scale; SMD, standardized mean difference.

^a<10 studies.

Figure 2.

Forest plots presenting the quantitative synthesis of circulating ferritin SMD between: a) NAFLD patients and controls; b) NAFL patients and controls; c) NASH patients and controls; d) NASH and NAFL patients, in the sum of the included studies. SMD, standardized mean difference; NAFLD, nonalcoholic fatty liver disease; NAFL, nonalcoholic fatty liver; NASH, nonalcoholic steatohepatitis.

Concerning subgroup analysis, firstly, studies were classified into those with biopsy-proven controls and those without biopsy-proven controls (Table 3; Supplementary Figure 2a1, 2b1, 2c1) for the comparisons: 1) NAFLD patients vs. controls; 2) NAFL patients vs. controls; 3) NASH patients vs. controls, but not between NASH and NAFL, since the control group was not included in the last pair of comparison (all patients were biopsy-proven). Similar results were observed between studies with and without liver biopsy in the controls, with statistically significant results in all comparisons. However, the nonbiopsy-proven subgroup revealed higher SMD in all comparisons, and the difference was statistically significant between groups in all pairs of comparisons. The heterogeneity remained high in the subgroup without biopsy-proven controls (I² was 71–98%), but was lower in the subgroups with biopsy-proven controls (I² was 0–65%). Egger's test revealed no significant publication bias in both subgroups (Table 3, Supplementary Figure 1a2.1, 1a2.2, 1b2.1, 1b2.2, 1c2.1, 1c2.2).

Table 3.

Ferritin SMD in Subgroup Analysis Based on Biopsy in Controls, Histological Definition and Presence of NASH-cirrhosis.

Comparison	Biopsy in controls		Histological definition		NASH-cirrhosis
	Νo biopsy in controls	Βiopsy in controls	NAS	Other than NAS	No cirrhosis	Cirrhosis
NAFLD vs. control	1.46 (0.94, 1.98); P < 0.00001	0.38 (0.17, 0.59); P = 0.0004	1.35 (0.66, 2.05); P = 0.0001	0.85 (0.71, 1.00); P < 0.00001	0.82 (0.60, 1.03); P < 0.00001	1.71 (0.54, 2.89); P = 0.004
	I2 = 98%; P < 0.00001	I2 = 46%; P = 0.07	I2 = 98%; P < 0.00001	I2 = 23%; P = 0.23	I2 = 45%; P = 0.11	I2 = 99%; P < 0.00001
	Egger's test P = 0.744	Egger's test P = 0.380	Egger's test P = 0.964	Egger's test P = 0.165	Egger's test P = 0.662	Egger's test P = 0.754
	P for difference = 0.0002		P for difference = 0.17		P for difference = 0.14
NAFL vs. control	0.87 (0.54, 1.20); P < 0.00001	0.29 (0.11, 0.46); P = 0.001	0.58 (0.23, 0.93) P = 0.001	0.64 (0.27, 1.01); P = 0.0008	2.40 (0.87, 3.94); P = 0.002	0.40 (−0.02, 0.82); P = 0.06
	I2 = 71%; P < 0.00001	I2 = 0%; P = 0.47	I2 = 76%; P < 0.0001	I2 = 65%; P = 0.02	I2 = 96%; P < 0.00001	I2 = 77%; P = 0.002
	aEgger's test P = 0.228	aEgger's test P = 0.181	Egger's test P = 0.919	aEgger's test P = 0.198	aEgger's test P = 0.559	aEgger's test P = 0.284
	P for difference = 0.002		P for difference = 0.81		P for difference = 0.01
NASH vs. control	1.23 (0.95, 1.52); P < 0.00001	0.42 (0.07, 0.78); P = 0.02	0.77 (0.23, 1.31); P = 0.005	1.12 (0.85, 1.39); P < 0.00001	1.37 (0.87, 1.87); P < 0.00001	0.80 (−0.04, 1.65); P = 0.06
	I2 = 81%; P < 0.00001	I2 = 65%; P = 0.005	I2 = 91%; P < 0.00001	I2 = 67%; P = 0.002	I2 = 77%; P = 0.002	I2 = 94%; P < 0.00001
	Egger's test P = 0.617	aEgger's test P = 0.924	Egger's test P = 0.262	aEgger's test P = 0.117	aEgger's test P = 0.827	aEgger's test P = 0.542
	P for difference = 0.0005		P for difference = 0.25		P for difference = 0.26
NASH vs. NAFL	not applicable	not applicable	0.30 (−0.06, 0.67); P = 0.11	1.98 (0.73, 3.22); P = 0.002	2.40 (0.87, 3.94); P = 0.002	0.40 (−0.02, 0.82); P = 0.06
			I2 = 84%; P < 0.00001	I2 = 95%; P < 00001	I2 = 96%; P < 0.00001	I2 = 77%; P = 0.002
			Egger's test P = 0.182	aEgger's test P = 0.010	aEgger's test P = 0.411	aEgger's test P = 0.937
			P for difference = 0.01		P for difference = 0.01

Subgroup analyses results are presented in SMD (95% Confidence Interval); P-value, I² test; P-value, Egger's test P-value, P-value for difference between subgroups.

Abbreviations: NAFLD, nonalcoholic fatty liver disease; NAFL, nonalcoholic fatty liver; NASH, nonalcoholic steatohepatitis; NAS, NAFLD activity score; SMD, standardized mean difference.

^a<10 studies.

Secondly, studies were classified into those having implemented NAS²⁰ vs. other histological scoring systems for liver histology, e.g., the Brunt histological system.²¹ For the comparisons: 1) NAFLD patients vs. controls; 2) NAFL patients vs. controls; 3) NASH patients vs. controls, the results were statistically significant in all subgroups, the heterogeneity was high (I² was 65%–98%), with the exception of the subgroup of other histological scoring systems in the comparison of NAFLD vs. control, which was 23% (nonstatistically significant), and the SMD difference between groups was not statistically significant (Table 3, Supplementary Figure 2a2, 2b2, 2c2, 2d1). However, the NAS subgroup of the comparison between NASH vs. NAFL showed a nonsignificant SMD, whereas the difference was significant in the other than NAS histological system subgroup. No publication bias was shown in Egger's test, except for the other NAS histological system subgroup for the comparison NASH vs. NAFL; however, this pair of comparison consists of five studies (<10), so this result should be cautiously interpreted (Table 3; Supplementary Figure 1a3.1, 1a3.2, 1b3.1, 1b3.2, 1c3.1, 1c3.2, 1d2.1, 1d2.2).

The classification for the third subgroup analysis was based on NASH-related liver cirrhosis (NASH-cirrhosis); thus, studies were classified into those containing and those not containing patients with NASH-related liver cirrhosis for the comparisons: 1) NAFLD patients vs. controls; 2) NAFL patients vs. controls; 3) NASH patients vs. controls; 4) NASH vs. NAFL patients (Table 3; Supplementary Figures 2a3, 2b3, 2c3, 2d2). For all the comparisons, except from the NAFLD vs. controls (Table 3), the results were significant for the subgroup not containing patients with liver cirrhosis but marginally not significant for the subgroup containing patients with NASH-cirrhosis. The heterogeneity remained high, with the exception of the subgroup without patients with NASH-cirrhosis in the pair of NAFLD patients vs. controls (I² was 45%). Egger's test revealed no significant publication bias (Table 3; Supplementary Figures 1a4.1, 1a4.2, 1b4.1, 1b4.2, 1c4.1, 1c4.2, d3.1, 1d3.2).

The pre-specified sensitivity analyses, according to the protocol registered in the PROSPERO, consisted of the exclusion of studies with morbidly obese populations and exclusion of studies with NOS score <7. Sensitivity analysis after excluding studies with morbidly obese populations (n = 5) had minimal to null effect on SMD difference, on the heterogeneity, which remained high (apart from the pair of NAFL vs. control: I² = 39%, P = 0.08), and on Egger's test, which provided nonsignificant result in all pairs of comparisons (Table 2; Supplementary Figure 1a5,1b4,1c5,1d4; Supplementary Figure 3a2,3b2,3c2,3d2). Ten studies were evaluated with NOS score ≥7 for the comparison of NAFLD vs. controls, in which SMD remained higher in NAFLD than controls, but the heterogeneity dropped to 54% (Table 2, Supplementary Figure 3a1,3b1,3c1,3d1); Egger's test remained non-significant (Table 2, Supplementary Figure 1a6,1b5,1c6,1d5). There were only minimal changes in the rest of the comparisons, with the comparison of NAFL vs. controls after excluding studies with NOS <7 presenting a marginal significance in the SMD difference (P = 0.05). Again, most results of this sensitivity analysis should be carefully interpreted due to the small number of the included studies.

Some post-hoc (not prespecified) sensitivity analyses were deemed necessary following the evaluation of the results of this meta-analysis (Table 2, Supplementary Figure 3a3, 3a4, 3a5, 3b3, 3b4, 3c3, 3c4, 3d3, 3d4). After excluding the Nagaya et al. study,⁴¹ since the calculated SD values were extremely high, or the Wu et al. study,⁵¹ which was referred to MAFLD rather than NAFLD patients, a minimal to null effect on heterogeneity and publication bias was shown for all pairs of comparisons. Last but not least, after excluding the studies of Hanafy et al.³² and Valenti et al.⁴⁸ because of reporting extreme values of ferritin levels, as compared to other studies, SMD difference and heterogeneity were not affected, but publication bias (Egger's test P = 0.02) was revealed (Table 2, Supplementary Figure 1a7-9, 1b6-7, 1c7-8, 1d6-7). The last sensitivity analysis was tested only for the comparison of NAFLD patients vs. controls, since the two excluded studies did not report ferritin levels separately for NAFL and NASH patients.

Meta-regression analysis was also performed in our attempt to identify potential sources of heterogeneity. The preselected variables were the percentage of males, percentage of T2DM, age, BMI, and HOMA-IR. Male sex was positively associated with ferritin SMD when comparing NAFLD patients with controls, explaining 32.7% of the heterogeneity of this comparison. BMI was negatively correlated with ferritin SMD in the comparison of NASH vs. controls, explaining 63.3% of the heterogeneity of this comparison. Significant associations were not observed between ferritin SMD and other variables in any other pair of comparisons. The results of meta-regression analysis are summarized in Table 4, and bubble plots of significant results are presented in Supplementary Figure 4a–b.

Table 4.

Meta-regression Analysis Between Ferritin SMD and Variables of Interest in Different Groups of Comparisons.

Comparison	Sex (male%)	T2DM (%)	Age (years)	BMI (kg/m2)	HOMA-IR
NAFLD vs. control	4.685 (1.900, 7.474);	−1.161 (−5.411, 3.090);	−0.016 (−0.096, 0.064);	−0.059 (−0.146, 0.029);	−0.160 (−0.594, 0.273);
	0.002	0.557	0.688	0.178	0.440
	Adjusted R square: 32.7%	Adjusted R square: 0.00%	Adjusted R square: 0.00%	Adjusted R square: 3.62%	Adjusted R square: 0.00%
NAFL vs. control	0.4232 (−1.620, 2.466);	a–2.013 (–5.594, 1.568);	0.003 (−0.025, 0.031);	−0.016 (−0.043, 0.011);	a0.125 (–0.043, 0.293);
	0.651	0.208	0.806	0.210	0.121
	Adjusted R square: 0.00%	Adjusted R square: 34.76%	Adjusted R square: 0.00%	Adjusted R square: 0.00%	Adjusted R square: 99.98%
NASH vs. control	0.252 (−2.194, 2.699);	a–2.403 (−6.326, 1.520);	−0.002 (−0.044, 0.039);	−0.058 (−0.092, −0.024);	a–0.023 (−0.205, 0.160);
	0.828	0.191	0.919	0.003	0.777
	Adjusted R square: 0.00%	Adjusted R square: 16.88%	Adjusted R square: 0.00%	Adjusted R square: 63.32%	Adjusted R square: 0.00%
NASH vs. NAFL	1.427 (−13.013, 15.867);	a–0.991 (−5.021, 3.038);	−0.021 (−0.185, 0.142);	−0.060 (−0.200, 0.081);	−0.066 (−0.275, 0.144);
	0.832	0.555	0.786	0.380	0.499
	Adjusted R square: 0.00%	Adjusted R square: 0.00%	Adjusted R square: 0.00%	Adjusted R square: 0.00%	Adjusted R square: 0.00%

Meta-regression analyses results are presented in Beta (95% CI); P-value, adjusted R square (%).

Abbreviations: BMI, body mass index; CI, confidence interval; HOMA-IR, homeostasis model assessment; NAFLD, nonalcoholic fatty liver disease; NAFL, nonalcoholic fatty liver; SMD, standardized mean difference; T2DM, type 2 diabetes mellitus.

^a<10 studies.

Discussion

This meta-analysis demonstrated that higher circulating ferritin SMD was associated with the presence of NAFLD and the severity of the disease; ferritin SMD of NAFL and NASH patients were higher in comparison to the respective control groups and in NASH patients compared to NAFL patients (Table 2, Figure 2a–d).

However, heterogeneity was moderate to high in most pairs of comparisons; thus, the results of this meta-analysis should be cautiously interpreted. Our attempts to explain potential sources of heterogeneity provided some interesting results. Heterogeneity was lower (low to moderate) in the subgroup of studies with histological confirmation in the controls (Table 3, Supplementary Figure 2a1, 2b1, 2c1), possibly implying that heterogeneity may be partly attributed to the lack of biopsy in the controls. Indeed, in studies without histological confirmation in the controls, a type of classification bias may have been committed, i.e. some patients with NAFLD may have been classified as controls due to the low accuracy of imaging modalities (especially liver ultrasonography).⁵⁴ Of course, it is acknowledged that, due to ethical considerations, a liver biopsy could not be performed on all non-NAFLD controls for the sake of research. The results of other subgroup analyses should be cautiously interpreted due to the small number of studies in most pairs of comparisons. The sensitivity analysis after the exclusion of studies with morbidly obese patients had a null to minimal effect on the results. After excluding studies with presumably lower NOS score (<7), the heterogeneity was reduced for the pair of comparison NAFLD vs. controls, possibly implying that low-quality studies may increase the heterogeneity at least in this pair of comparison. Excluding studies with extreme ferritin values did not substantially affect the heterogeneity among the studies. The meta-regression analysis revealed that the male ratio was positively associated with ferritin SMD between NAFLD patients and controls and could explain about one-third of the heterogeneity observed in this pair of comparison.^55,56 BMI was negatively correlated with ferritin SMD in the comparison NASH vs. control, explaining about two-thirds of the heterogeneity of this pair of comparison^{57, 58, 59}; this may also imply that a part of the ferritin SMD difference observed in this pair of comparison may be attributed to the lower BMI of controls vs. NASH patients.

Publication bias was not a major concern in this meta-analysis. Egger's test revealing publication bias in the other than NAS histological system subgroup for the comparison of NASH vs. NAFL should be interpreted with caution since 5 studies were included and Egger's test performs better when ≥10 studies are included.⁶⁰ Funnel plots' asymmetry may also be partly explained by the high heterogeneity among studies.⁶¹

Increased ferritin has been proposed to affect NAFLD by various mechanisms. Higher ferritin may indicate higher iron stores; excess iron seems to adversely affect all major liver cells, i.e. hepatocytes, Kupffer cells, and hepatic stellate cells, thus promoting hepatic steatosis, inflammation, and fibrogenesis, as is elsewhere reviewed in detail.⁶² In this regard, hereditary hemochromatosis, characterized by iron and ferritin overload, typically manifests NAFLD.⁶³ Furthermore, high ferritin has been associated with higher IR, mitochondrial dysfunction, oxidative stress, and ferroptosis, a form of programmed cell death caused by iron-dependent lipid peroxidation^43,64,65; although all these mechanisms have been implicated in the pathogenesis of NAFLD,^66,67 more focused mechanistic studies are required to elucidate the specific association of ferritin as their inducer and/or mediator in the pathogenesis of NAFLD.

The results of this meta-analysis warrant further studies evaluating circulating ferritin as a noninvasive index of NAFLD, NAFL, and/or NASH. Since a single parameter may hardly provide accurate enough results towards the noninvasive diagnosis of NAFLD and/or its severity, circulating ferritin may be evaluated in algorithms in combination with other parameters (e.g. adiponectin, leptin, and/or TNF-α), which have provided favorable results in relevant meta-analyses.^{68, 69, 70} This is considered important in the quest for alternative means of diagnosis, staging, and risk stratification of NAFLD since the gold standard of diagnosis, liver biopsy, cannot be applied to all NAFLD patients due to the high prevalence of the disease.^{8,71, 72, 73} Indeed, circulating ferritin has already been incorporated into some relevant noninvasive algorithms for NAFLD. The FibroMeter NAFLD (Echosens Paris, France) is a noninvasive index of fibrosis in NAFLD patients, implementing ferritin along with age, body weight, AST, ALT, platelet count, and glucose.^74,75 In a systematic review and meta-analysis of diagnostic accuracy studies, the FibroMeter NAFLD showed sensitivity of 72% and specificity of 83% for the diagnosis of advanced fibrosis (F ≥ 3).⁷⁶ The NAFIC score is a noninvasive test introduced for the diagnosis of NASH (vs. NAFL), implementing ferritin, fasting insulin, and type IV collagen 7S levels⁷⁷; however, this test needs further external validation by independent groups.

This meta-analysis could not show that reducing ferritin might have therapeutic potential in NAFLD because it is a meta-analysis of observational studies. However, since iron depletion has been introduced as an appealing therapeutic target to limit the progression of NAFLD to its more advanced forms,⁶² this meta-analysis warrants the setting of clinical trials to investigate the therapeutic potential of iron-depleting medication in patients with NASH. In line, until we have robust relevant results, it may be prudent that patients with NAFLD avoid excess dietary iron, e.g. by reducing the red meat or processing meat in the diet.⁷⁸ In this regard, a meta-analysis of four interventional studies suggested that phlebotomy reduced ALT and HOMA-IR in patients with NAFLD⁷⁹; nonetheless, further studies, preferably with paired liver biopsies, are needed to show the efficacy and safety of phlebotomy in patients with NAFLD.

A meta-analysis on the same topic, consisting of 14 studies, was published in 2017.⁸⁰ However, the previous meta-analysis included a smaller number of studies (n = 14) than our meta-analysis; the diagnosis of NAFLD was based on both liver biopsy and liver ultrasonography, i.e. possibly committing an information bias; hand searching had not been performed; non-English literature had been excluded; and there was not extensive analysis to elucidate the sources of high heterogeneity among studies.⁸⁰ A more recent systematic review of the topic did not end with a meta-analysis, i.e. a quantitative synthesis of the included studies.⁸¹

This meta-analysis, however, has certain limitations: 1) the inclusion of observational studies cannot show causality between circulating ferritin levels and NAFLD or its severity; 2) despite extensive analysis to identify the sources of high heterogeneity among studies, our results could only partly explain the high heterogeneity; 3) the comparison of ferritin levels between patients with NAFLD and no or mild hepatic fibrosis (F0–F1) vs. patients with NAFLD and significant or advanced hepatic fibrosis/cirrhosis (F2–F4) was not feasible, due to the small number of studies with relevant data; 4) subgroup analysis based on sex was not feasible since most studies did not provide data separately for males and females; 5) subgroup analysis according to different methods of ferritin measurement was not performed due to the lack of relevant data; 6) the proposed update in the terminology from NAFLD to MAFLD⁸² or metabolic associated steatotic liver disease, was not integrated,⁸³ due to the lack of relevant data, since these terms may not be used interchangeably.⁸⁴ In this regard, the study of Wu et al.,⁵¹ providing data for MAFLD patients, was excluded in a sensitivity analysis without notable difference; 7) the comparisons of NASH vs. NAFL, NAFL vs. non-NAFLD controls and NASH vs. non-NAFLD were secondary aims of this meta-analysis and their quantitative synthesis was based on the studies derived from the predefined inclusion criteria, based on which studies without a non-NAFLD control group were not eligible for this meta-analysis; 8) this meta-analysis was not designed to investigate the association between high ferrum (HFE) gene mutations (referring to hemochromatosis) and NAFLD, since hemochromatosis was an exclusion criterion; however, there are studies supporting that HFE gene mutations are not associated with NAFLD or NASH⁸⁵ or that HFE gene mutations are observed in a small percentage of NAFLD patients, especially in specific populations (e.g. Indian).⁸⁶

In conclusion, higher circulating ferritin was shown in patients with NAFLD (or NAFL or NASH) compared with non-NAFLD controls, supporting an association of circulating ferritin with the disease. Furthermore, higher circulating ferritin was also shown in patients with NASH compared with patients with NAFL, implying an association of circulating ferritin with the severity of the disease, which should be cautiously interpreted since this was a secondary aim of this meta-analysis. This meta-analysis warrants further investigation of ferritin in the setting of a noninvasive diagnosis of the disease, and as a potential therapeutic target.

CRediT authorship contribution statement

Eleftheria Makri: Methodology; Investigation; Data curation; Validation; Formal analysis; Visualization; Writing-original draft; Writing-review & editing.

Myrsini Orfanidou: Investigation; Data curation; Validation; Formal analysis; Visualization; Writing-review & editing.

Evangelia S. Makri: Data curation; Validation; Writing-review & editing.

Antonis Goulas: Validation; Writing-review & editing.

Evangelos Terpos: Validation; Writing-review & editing.

Stergios A. Polyzos: Conceptualization, Methodology, Investigation; Data curation; Writing-original draft; Writing-review & editing; Supervision.

All authors approved the final version to be submitted and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Conflicts of interest

The authors declare that they have no competing financial interests or personal relationships that could have influenced this study.

Acknowledgments

We are sincerely grateful to the following authors for their willingness to provide us additional data that were necessary for this systematic review and meta-analysis: Dr. Emanuele Albano (University of Eastern Piedmont, Novara, Italy), Dr. Elmar Aigner (Paracelsus Medical University, Austria), Dr. Tomislav Bulum (Merkur University Hospital, Croatia), Dr. Shivakumar Chitturi (Canberra Hospital, Australia), Dr. Evangelos Chologitas (Medical School of National & Kapodistrian University of Athens, Greece), Dr. Teresa Cristina A. Ferrari (Universidade Federal de Minas Gerais, Brazil), Dr. Sonja Lang (University Hospital of Cologne Germany), Dr. Amedeo Lonardo (Azienda Ospedaliero-Universitaria, Ospedale Civile di Baggiovara, Modena, Italy), Dr. Charalampos Papadopoulos (Democritus University of Thrace, Greece), Dr. Pathik Parikh (Royal Free Hospital, United Kingdom), Dr. Michael Pavlides (University of Oxford, United Kingdom), Dr. Vibha Singhal (Massachusetts General Hospital and Harvard Medical School, USA), Dr. Dhastagir Sultan Sheriff (Al Arab Medical University, Libya), Dr. Per Stål (Karolinska Institutet, Sweden), Dr. Karjpong Techathuvanan (Vajira Hospital, Navamindradhiraj University, Thailand) and Dr. Pamela Valva (Instituto Multidisciplinario de Investigaciones en Patologías Pediátricas, Argentina).

Funding

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Supplementary Figure 1

Funnel plots in the sum of included studies for the comparisons between: a) NAFLD patients and controls; b) NAFL patients and controls; c) NASH patients and controls; d) NASH and NAFL patients.

Supplementary Figure 2

Forest plots presenting the quantitative synthesis of circulating ferritin standardized mean difference (SMD) between: a) NAFLD patients and controls; b) NAFL patients and controls; c) NASH patients and controls; d) NASH and NAFL patients, according to the subgroup analysis.

Supplementary Figure 3

Forest plots presenting the quantitative synthesis of circulating ferritin standardized mean difference (SMD) between: a) NAFLD patients and controls; b) NAFL patients and controls; c) NASH patients and controls; d) NASH and NAFL patients, according to sensitivity analysis.

Supplementary Figure 4

Bubble plots of the association between ferritin standardized mean difference (SMD) and: a) the percentage of males between NAFLD patients and controls, and b) BMI between NASH patients and controls, according to the meta-regression analysis.