Study selection and study characteristics:
A total of 10472 articles were obtained by preliminary search. After screening the titles and abstracts, 546 articles were selected for full-text screening, of which 523 articles were excluded due to the reasons reported in the PRISMA chart (Figure 1). In the end, this study included 24 studies (3 prospective cohort studies, 16 retrospective cohort studies, and 5 case-control studies) of 23 articles. Perform NOS scores on these articles. The basic characteristics of included literature could be seen in Table 1. The research includes studies from Asia (Japan 13,17,18,19,20,23,24,26,28,29,32,33 and China 22, South Korea25), Europe (UK21, Italy30,31,36), and North America (United States10,27,34,35,37) with 1004284 NAFLD cases and 3610 HCC cases.
Table 1. Baseline characteristics of all the studies included in the meta-analysis.
Study
|
Country
|
Study design
|
Number of patients diagnosed with NAFLD
|
Number of HCC cases
|
Adjustments
|
Effect value
|
NOS
scores
|
Tokushige
2013[17]
|
Japan
|
case-control
|
574(histological)
|
41
|
age、gender 、fibrosis、other metabolic RFs
|
OR
|
7
|
Tobari
2019[18]
|
Japan
|
prospective cohort
|
857(Imaging or histological)
|
119
|
age、gender、alcohol、cirrhosis、other metabolic RFs
|
OR
|
9
|
Seko
2015[19]
|
Japan
|
retrospective cohort
|
312(histological)
|
6
|
age、gender、fibrosis、other metabolic RFs
|
HR
|
9
|
Seko
2017[20]
|
Japan
|
retrospective cohort
|
238(histological)
|
10
|
age、gender、PNPLA3、fibrosis、other metabolic RFs
|
HR
|
9
|
Liu
2014[21]
|
UK
|
case-control
|
375(histological or others)
|
100
|
age、gender、PNPLA3、fibrosis/ cirrhosis 、other metabolic RFs
|
OR
|
7
|
Lee
2017[22]
|
China
|
retrospective cohort
|
18080(coding)
|
494
|
age、gender、ALT、BP、TC、DM、statin、metformin、aspirin
|
HR
|
9
|
Kogiso
2020[23]
|
Japan
|
retrospective cohort
|
365(histological)
|
26
|
age、gender、fibrosis、other metabolic RFs
|
HR
|
8
|
Kimura
2018[24]
|
Japan
|
retrospective cohort
|
301(histological)
|
9
|
age、gender、fibrosis、other metabolic RFs
|
OR
|
8
|
Kim
2017[25]
|
Korea
|
retrospective cohort
|
8721(ultrasound)
|
23
|
age、gender、fibrosis、other metabolic RFs
|
HR
|
8
|
Kawamura
2012[26]
|
Japan
|
retrospective cohort
|
6508(ultrasound)
|
16
|
age、gender、other metabolic RFs
|
HR
|
8
|
Kanwal
2018[13]
|
US
|
retrospective cohort
|
296707(coding)
|
490
|
age、race、other metabolic RFs
|
HR
|
8
|
Kanwal
2020[27]
|
US
|
retrospective cohort
|
271906(coding)
|
253
|
age、gender、race、DM、BP、dyslipidemia、BMI
|
HR
|
9
|
Ito
2019[28]
|
Japan
|
retrospective cohort
|
246(ultrasound or clinical)
|
15
|
age、gender、other metabolic RFs
|
HR
|
6
|
Hashimoto
2009[29]
|
Japan
|
prospective cohort
|
382(histological
)
|
34
|
age、fibrosis、other metabolic RFs
|
OR
|
8
|
Grimaudo
2020[30]
|
Italy
|
prospective cohort
|
471(histological or clinical)
|
13
|
gender、age、PNPLA3、fibrosis、other metabolic RFs
|
HR
|
9
|
Donati
2017[31]
|
Italy
|
retrospective cohort
|
765
|
132
|
age、gender、BMI、DM、fibrosis、PNPLA3
|
OR
|
8
|
Akuta
2018[32]
|
Japan
|
retrospective cohort
|
402(histological)
|
9
|
age、gender、fibrosis、other metabolic RFs
|
HR
|
7
|
Azuma
2019[33]
|
Japan
|
case-control
|
182(histological or imaging)
|
22
|
age、gender、fibrosis、other metabolic RFs
|
OR
|
9
|
Ioannou
2019[34]
|
US
|
retrospective cohort
|
7068(coding)
|
1278
|
age、gender、DM、BMI、PLT、ALB 、AST/ALT
|
HR
|
9
|
Yang
2020[35]
|
US
|
retrospective cohort studies
|
354(coding)
/6630(coding)
|
30/291
|
age、gender、race、other metabolic RFs
|
HR
|
9
|
Sorrentino2009[36]
|
Italy
|
case-control
|
482(histological)
|
71
|
age、gender、other metabolic RFs
|
HR
|
7
|
Corey
2017[37]
|
US
|
case-control
|
244(histological or clinical)
|
94
|
age、gender、other metabolic RFs
|
OR
|
9
|
Ascha
2010[10]
|
US
|
retrospective cohort
|
195(histological or imaging)
|
25
|
age、gender、smoking、alcohol、BMI、DM
|
HR
|
9
|
Abbreviations: CI, confidence interval; HR, risk ratio; OR ratio; NAFLD, non-alcoholic fatty liver disease; RF, risk factors; DM, diabetes; AFP, metformin; ALT, glutamate transaminase; AST, aspartate Transaminase;PNPLA3: PNPLA3 genotypes metabolic
RFs include: BMI, DM, BP, lipids, blood tests (total bilirubin, ALB, AST, ALT, ALP, γ-GTP, PLT, clotting enzyme duration)
Risk factors and HCC in NAFLD
All pooled data were shown in Figure 2. By combining the HRs of univariate analysis, the statistically significant factors were: male, low platelet count, advanced liver fibrosis, diabetes, hypertension (see Figure 2a). There were 4 factors statistically significant by combining the ORs of multivariate analysis: male, older age, diabetes, advanced liver fibrosis (see Figure 2b). 4 factors were statistically significant by combining the HRs of multivariate analysis: male, low platelet count, diabetes, advanced liver fibrosis (see Figure 2c). The results of publish bias assessment were shown in Suppl. Table 3.
Male A total of 14 observational studies suggested that the risk of HCC in NAFLD was associated with gender. 6 of the studies conducted a univariate analysis to obtain HRs [pooled HR=1.63, 95%CI (1.33-2.01), P<0.00001] (Figure 2, Figure 3a), with 5 studies conducting multivariate analysis to obtain HRs [pooled HR=1.79, 95%CI (1.46-1.21), P<0.00001] (Figure 2, Figure 5a). 5 studies conducted multivariate analysis to obtain ORs [pooled OR=4.38, 95%CI (2.93-6.57), P<0.00001] (Figure 2, Figure 4a). According to Egger test results (Suppl. Table 3) and forest plots, the combined values had good homogeneity, and there was a certain publication bias in pooled multivariate analysis OR values but there was no obvious publication bias in both pooled univariate and multivariate analysis HRs. Thus, we concluded that the gender factor male was an important risk factor for the development of NAFLD to HCC.
Diabetes Diabetes was diagnosed by any of the following criteria: (i) classic symptoms of hyperglycemia and random plasma glucose≥200 mg/dl; (ii) fasting plasma glucose≥126 mg/dl; (iii) 2-h post-glucose (oral glucose tolerance test)≥200 mg/dl; (iv) HgbA1C≥6.5%38. 13 studies mentioned this factor. The type of diabetes included in 4 of the studies was type 2 diabetes30,31,32,34 and the others were unspecified diabetes. There were 6 studies conducting univariate analysis to obtain HR values [pooled HR=2.58, 95%CI (1.40-4.75), P<0.00001] (Figure 2, Figure 3b). 7 studies conducted multivariate analysis to obtain HRs [pooled HR=1.64, 95%CI (1.13-2.36), P=0.008] (Figure 2, Figure 5b) and 4 studies conducted multivariate analysis to obtain ORs [pooled OR=3.65, 95%CI (2.32-5.75), P<0.00001] (Figure 2, Figure 4b). According to Egger test results, there was no significant publication bias (Suppl. Table 3). The results suggested that NAFLD patients with diabetes had a higher risk to suffer from HCC.
Advanced liver fibrosis There were 10 studies that mentioned advanced liver fibrosis as a risk factor. When the included participants underwent liver biopsy, the NAFLD pathological stage was evaluated according to the classification of Brunt et al39. Fibrosis ≥F3 was defined as advanced fibrosis20,21,23,24,28,29,31,32. Otherwise, the severity of liver fibrosis was assessed by two noninvasive markers, NAFLD fibrosis score and fibrosis-4 score16,33. 3 studies conducted the univariate analysis to obtain HRs [pooled HR=21.32, 95%CI (8.74-52.02), P<0.00001] (Figure 2, Figure 3c). 4 studies conducted multivariate analysis to obtain HR values [pooled HR=11.98, 95%CI (4.93-29.12), P<0.00001] (Figure 2, Figure 5c). And 6 studies conducted multivariate analysis to obtain OR values [pooled OR=5.15, 95%CI (2.66-9.95), P<0.00001] (Figure 2, Figure 4c). There was no obvious publication bias in both pooled multivariate analysis HRs and ORs (Suppl. Table 3). Thus, we concluded that advanced fibrosis was a significant predictor of HCC in NAFLD, which could substantially increase the risk of HCC.
Older age A total of 11 observational studies involved in the factor, older age. The results of the pooled univariate analysis HRs were not statistically significant (Figure 2). Pooled multivariate analysis of HRs and ORs results showed that factor older age could increase the risk of HCC in NAFLD by 1.16-fold and 3.62-fold, respectively (Figure 2, Figure 4d, Figure 5d). Results were publication-biased and heterogeneous. The high heterogeneity might be due to different research "abnormal" cut-off points. According to a large retrospective study by Lee et al, which included 18080 NAFLD patients, the 10-year cumulative incidences of HCC were shown to be increasing along with age levels: 18-45 years (0.19%; 95% CI, 0-0.57), 46-55 years (1.31%; 95%, CI, 0-2.86), 56-65 years (3.80%; 95% CI, 1.02-6.59), and >65 years (6.20%; 95% CI, 3.20-9.20) 22. Despite the large heterogeneity between studies, we still considered older age to be an important predictor of HCC.
Low platelet count There were 5 studies included in total. 3 of the studies set the threshold for platelet count at 150×109/L 27,29,37, 1 at 200×109/L33 and 1 at 190×109/L 31. The results of pooled univariate analysis of HRs showed that low platelet count increased HCC risk 13.53-fold [HR=13.53, 95%CI (6.35-28.84), P<0.00001] (Figure 2, Figure 3d). The results of pooled multivariate analysis of HRs showed that low platelet count increased HCC risk 7.39-fold [HR=7.39, 95%CI (3.47-15.74), P<0.00001] (Figure 2, Figure 5e). The results showed good homogeneity and no publication bias (Figure 5e, Suppl. Table 3).
Hypertension 9 studies did research in the factor hypertension. The results of pooled multivariate analysis of HRs were not statistically significant (Figure 2). Pooled univariate analysis of HRs results showed that factor hypertension could increases the risk of HCC in NAFLD by 3.14-fold [HR=3.14, 95%CI (1.32-7.50), P=0.01] with no obvious publication bias (Figure 3e, Suppl. Table 3).
Sensitivity analysis
We limited the analysis to studies judged to be at low risk of bias and conducted sensitivity analysis by comparing the difference between the point estimate and the interval estimate of the combined effect size, when different effect models were compared. The results showed that the combined effect size conclusions of all factors had no significant change, indicating that the meta-analysis results of various indicators in this study were stable. The high levels of heterogeneity between studies for some factors, as indicated by the high I2 values, were explored. These were felt to be due to the variation in study design, particularly around the range of populations and outcomes studied, leading to clinical heterogeneity. Our research pooled univariate analysis HRs, multivariate analysis ORs and HRs. There was a consistent direction of effect. Based on the objective of the review, pooling using meta-analysis was still felt to be appropriate.
Other risk factors
Other factors, such as low albumin, patatin-like phospholipase domain containing 3 (PNPLA3) genotype, abnormal liver enzymes, overweight, and dyslipidemia had also been reported as possible risk factors, but there were too few reports to be pooled or the pooled results not statistically significant.We conducted a systematic review of the risk factors.
low albumin A total of 5 studies mentioned low albumin levels as a risk factor 26,28,29,30,35. The studies by Kawamura et al26 and Grimaudo et al30 conducted univariate analysis, and the results showed that low albumin levels can increase HCC risk by 2.18-fold and 1.26-fold respectively, but both were not statistically significant. The research of Yang et al35 included 2 cohort studies. One included 354 patients with NASH cirrhosis over 10 years of follow-up at the Mayo Clinic, and the other included 6630 patients who enrolled on the liver transplantation waiting list due to NAFLD. The results of the 2 cohort studies both showed that low albumin levels increased the risk for NAFLD developing HCC [HR=0.48, 95% CI (0.36-0.68), P<0.001] [HR=0.67, 95% CI (0.54-0.82), P<0.001].
PNPLA3 genotype The role of the PNPLA3 genotype was well recognized as a modifier of hepatic triacylglycerol accumulation and NAFLD progression40. A total of 4 articles suggested that the PNPLA3 genotype was a risk factor for HCC in NAFLD, especially the GG genotype. The multivariate analysis results of Grimaudo et al30 and Danti et al31 suggested that the PNPLA3 genotype increased the risk of HCC in patients with NAFLD by 168% and 61% respectively. Seko et al. showed a 6.36-fold increased risk for the PNPLA3 genotype GG [HR=6.36, 95%CI (1.36-29.8), P=0.019]20. The research by Liu et al21 which include 2 single-center studies, concluded that that carrying each G allele was associated with a doubled risk of HCC [OR=2.26, 95%CI (1.23-4.14), P=0.0082].
Dyslipidemia 7 studies investigated lipid levels and their prognostic value for HCC22,23,24,26,27,28,32. Low high-density lipoprotein, high triglycerides, combined lipid abnormalities, and hypercholesterolemia were exposures of interest. Our meta-analysis showed the results no significance (Figure 2). The relationship between dyslipidemia and HCC progression needed to be further investigated. It has been suggested that dyslipidemia is associated with NAFLD progression and that statins may have a protective effect on HCC progression [HR=0.29, 95% CI (0.12-0.68), P=0.005] 22. Dyslipidemia remained a risk factor of concern.
Overweight 5 studies reported overweight and HCC. All studies used body mass index (BMI) to measure overweight. Some studies analyzed the effect of BMI>25 on the development of HCC in NAFLD, with the other analyzed BMI>30. Being overweight was an independent risk factor of HCC. But our meta-analysis results seem to be negative and not statistically significant (Figure 2). The relationship between overweight and HCC in NAFLD needs to be further explored.
Abnormal liver enzymes 6 studies investigated the predictive value of liver function abnormalities, studying high AST, high ALT, or high ALT/AST. High AST was mentioned as a risk factor in 4 studies 23,26,28,30. A large retrospective study in Japan suggested an 8-fold increase in HCC risk with AST≥40IU26. 4 studies mentioned high ALT as a risk factor17,22,26,28. Lee's large study suggested a 6-fold increase in risk with high ALT22. One study mentioned AST/ALT>12.83 increased nearly 5-fold in HCC risk34. We should pay attention to liver enzyme abnormalities in patients with NAFLD, which may be associated with the development of HCC.