Smoking and Risk of Fatty Liver Disease: A Meta-Analysis of Cohort Studies

1. Introduction

Fatty liver disease (FLD), which is categorized into two major types such as nonalcoholic FLD (NAFLD) or alcoholic FLD (AFLD), indicates a morphological spectrum consisting of hepatic steatosis and steatohepatitis [1]. It is a persistent liver condition marked by macrovesicular steatosis in liver cells, which has the potential to advance to hepatic cirrhosis, liver failure, and possibly hepatocellular carcinoma [1]. Ethanol consumption serves as a pivotal determinant in distinguishing between NAFLD and AFLD in the guidelines from the European Association for the Study of the Liver [2]. NAFLD diagnosis involves considering ethanol intake of 20 g/d or less in females and 30 g/d or less in males, following the thorough exclusion of alternative causes such as hepatitis virus infection and the use of steatogenic drugs [3]. Beyond alcohol-related impacts, the pathogenesis of FLD is intricately linked to various contributors, including insulin resistance (IR), oxidative stress, mitochondrial dysfunction, immune system deregulation, and the release of adipokines [3,4]. These factors collectively underscore the complexity of FLD, emphasizing the importance of a comprehensive understanding to address its progression and associated risks such as hepatic cirrhosis, liver failure, and hepatocellular carcinoma [5].

The overall prevalence of NAFLD worldwide also has increased from 25.5% in or before 2005 to 37.8% in 2016 or later based on the report from a recent meta-analysis published in 2022 [6]. Also, FLD has become a predominant chronic liver disorder in developed Western countries [7]. Its risk factors include a higher body mass index (BMI), consumption of saturated fat and fructose, type 2 diabetes, and known single nucleotide polymorphisms [8].

However, it remains unclear whether smoking is associated with an increased risk of FLD. An animal study has reported that smoking increases lipid accumulation in hepatocytes by modulating an activity of critical molecules related with lipid synthesis [9]. Also, another animal study has shown that the histological severity of NAFLD in obese rats was exacerbated by tobacco exposure [10].

In the meantime, observational epidemiological studies [11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30] have reported inconsistent findings. Several cohort studies [14,17,18,20,23,24,27,28,29] have reported that smoking was associated with an increased risk of FLD, whereas other cohort studies [11,12,13,15,16,19,22,26] have reported no significant association between them.

In 2018, the only meta-analysis of observational studies has reported that smoking was significantly associated with the increased risk of NAFLD [31]. However, it included a small number of cohort studies to confirm the association, and subsequent cohort studies have been published since then. Furthermore, to our knowledge, no meta-analysis of cohort studies has been published regarding the association between smoking and the risk of FLD encompassing NAFLD and AFLD.

This study aimed to explore the associations between smoking and the risk of FLD by using a comprehensive meta-analysis of cohort studies and subgroup meta-analyses by important factors.

2. Materials and Methods

2.1. Search Strategy

PubMed and EMBASE were searched in September 2023 with terms of the National Library of Medicine (NLM) Medical Subject Headings (MeSH) and commonly used keywords. We used a PICO framework to combine search terms: P for population is any type of population; I for intervention is ‘smoking; C for comparison is ‘non-smoker’; and O for outcome is ‘fatty liver disease’. Also, the study type was confined to cohort studies. Thus, the final search terms were ‘smoking’, ‘fatty liver disease’, and ‘cohort study’.

2.2. Study Selection and Data Extraction

We included a cohort study that explored the associations between smoking and FLD (NAFLD or AFLD) and presented risk estimates such as odds ratio (OR), relative risk (RR), or hazard ratio (HR) with their corresponding 95% confidence intervals (CI). Two independent authors (MH. Lee and SH. Lee) conducted a selection of relevant studies by reviewing titles and abstracts. Discrepancies between them were resolved through discussion. The extracted information included the last name of the first author, publication year, study region, study design (prospective or retrospective cohort study), gender, study participants, comparison of exposure, risk estimates (OR, RR, and HR with corresponding 95% CIs), type of outcomes, and adjusted variables.

2.3. Assessment of Methodological Quality

We used the Newcastle-Ottawa Scale (NOS) in order to assess the methodological quality of the cohort studies included in the current meta-analysis [32]. The NOS comprises eight items and provides a scoring system ranging between 0 and 9. We classified individual cohort studies as having high or low quality based on the mean score.

2.4. Main and Subgroup Analyses

We investigated the association between smoking and the risks of FLD for the main analysis. We also conducted subgroup meta-analyses by type of study (prospective or retrospective cohort study), region (Europe, Asia, or US), type of FLD, gender (male or female), follow-up period (<5 years or >5 years), and study quality (high or low quality).

2.5. Statistical Analysis

A combined OR, RR, or HR with its corresponding 95% CIs was calculated utilizing the adjusted OR, RR, or HR and their respective 95% CIs from each study that reported the association between smoking and the risk of FLD. The DerSimonian and Laird method [33] was employed, opting for a random-effects model due to the diverse populations across studies. Heterogeneity was evaluated using Higgins I² computed as follows:

I² = 100% × (Q – df)/Q,

where Q is Cochran’s statistic for heterogeneity, and df is degrees of freedom [32]. I² values range between 0% (no heterogeneity) and 100% (maximal heterogeneity) [32]. Publication bias was assessed using both Begg’s funnel plot and Egger’s test. We used the STATA SE version 15.1 software package (StataCorp, College Station, TX, USA) for statistical analyses.

3. Results

3.1. Study Selection

Figure 1 shows the diagram of identifying relevant studies. A total of 806 studies were identified by the initial search of PubMed and EMBASE databases by using keywords. After excluding duplicates, 622 articles were screened based on the review of each title and abstract. After excluding 546 articles according to the predetermined selection criteria, 76 full-text articles were reviewed and assessed for eligibility. Among them, 56 articles were excluded for the following reasons: not relevant (n = 46); not cohort studies (n = 4); identical population (n = 4); and insufficient data (n = 2). The final analysis included 20 cohort studies (Figure 1) [11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30].

3.2. General Characteristics of Included Studies

Table 1 shows general characteristics of the cohort studies included in the final analysis. Eight studies were conducted in Europe, 10 were conducted in Asia, and the remaining two studies were conducted in the US. Types of study designs are prospective cohort studies (n = 5), retrospective cohort studies (n = 6), and cross-sectional surveys based on cohort studies (n = 9). Types of outcomes are FLD (n = 6) and NAFLD (n = 14).

3.3. Methodological Quality of Studies

The average score of the 11 prospective and retrospective cohort studies assessed using the NOS was 7.5 (Table 2). Each study’s methodological quality was categorized as either high (a score of ≥8) or low (a score of <8). Out of the 11 studies, six were classified as high-quality studies, while the remaining 5 were categorized as low-quality studies.

3.4. Association Between Smoking and Risk of FLD

In the meta-analysis of all the included studies, smoking was associated with an increased risk of FLD (OR/RR/HR = 1.14; 95% CI, 1.05 – 1.24; n = 20) (Figure 2).

3.5. Subgroup Meta-Analyses

Table 3 shows the associations between smoking and the risk of FLD in the subgroup meta-analysis by various factors. In the subgroup meta-analysis by study design, smoking increased the risk of FLD in prospective cohort studies (HR = 1.15; 95% CI 1.05 – 1.18; n = 5), but not in retrospective cohort studies (OR/RR = 1.23; 95% CI 0.94 – 1.62) and cross-sectional surveys (OR/RR = 1.12; 95% CI 0.92-1.46; n = 9). In the subgroup meta-analysis by study region, smoking was significantly associated with an increased risk of FLD in in Europe (OR/RR/HR = 1.32; 95% CI 1.16 – 1.50; n = 8), but not in Asia (OR/RR/HR = 1.03; 95% CI 0.91 – 1.18; n = 10) and the US (OR/RR/HR = 0.75; 95% CI 0.28 – 2.06; n = 2). In the subgroup meta-analysis by gender, which data are available only for Asian studies, smoking significantly increased the risk of FLD in men (OR/RR/HR = 1.15; 95% CI 1.06 – 1.25; n = 4), while there was no significant association between them in women (OR/RR/HR = 1.12; 95% CI 0.94 – 1.34; n = 4).

Regardless of type of FLD, smoking consistently increased the risk of FLD. On the contrary, subgroup meta-analyses by follow-up period and study quality showed no significant association between smoking and the risk of FLD.

3.6. Publication Bias

Both the Begg’s funnel plots (Figure 3) and Egger’s test (P = 0.29) did not show publication bias.

4. Discussion

In this meta-analysis of cohort studies, we found that smoking was significantly associated with an increased risk of FLD. In the subgroup meta-analysis by study design, smoking increased the risk of FLD in prospective cohort studies, but not in retrospective cohort studies and cross-sectional surveys. Also, smoking was significantly associated with an increased risk of FLD in Europe, but not in Asia and the US. Interestingly, in the subgroup meta-analysis by gender in Asians, smoking significantly increased the risk of FLD in men, while there was no significant association between them in women.

There are several possible biological mechanisms that could explain the increased risk of FLD by smoking. First, Yuan et al.’s study demonstrated that smoking stimulated lipid accumulation in hepatocytes in mice and cultured hepatocytes [9]. When mice and cultured hepatocytes were exposed to sidestream whole smoke, lipid accumulation was increased by modulating the activity of AMP-activated protein kinase and sterol response element binding protein-1, which are critical molecules in lipid synthesis [9]. Second, smoking has the potential to induce insulin resistance, which leads to the development of NAFLD [34,35]. A study revealed that cigarette consumption is correlated with degree of insulin resistance in smokers [36]. Also, a study in biopsy-proven NAFLD patients and health control subjects showed that homeostasis model assessment of insulin resistance (HOMA-IR) levels were significantly higher in the NAFLD patients [35]. Insulin resistance induced by smoking could enhance hepatic fat accumulation through increasing free fatty acid delivery to the liver and through hyperinsulinemia to contribute to triacylglycerol accumulation in the liver [37]. Third, nicotine in tobacco smoke could increase the release of norepinephrine and epinephrine, which could affect thermogenesis in adipose tissue, leading to the increased lipolysis and the subsequent recycling of fatty acids into triglycerides [38]. Consequently, it may contribute to the development of NAFTD. Lastly, adiponectin could inhibit liver fat deposition, and gluthathione peroxidase (GPx) could reduce lipid and hydrogen peroxide [39]. Thus, it has been suggested that cigarette smoking in combination with single-nucleotide polymorphisms in the adiponectine gene and GPx1 gene mutation could contribute to the development of NAFLD [39].

Previously, a meta-analysis of 12 observational studies by Akhavan et al. has already reported that smoking moderately increased a risk NAFLD [31]. However, it included only two cohort studies, and the remaining studies were seven cross-sectional and three case-control studies. Although it is possible to combine different study designs such as cross-sectional, case-control, and cohort studies when conducting a meta-analysis, conducting subgroup meta-analyses by study designs is crucial and useful because there could be discrepancies in findings between different study designs. Also, based on the ‘levels of evidence pyramid’, cohort studies give us a higher level of evidence than cross-sectional and case-control studies [40]. Akhavan et al. performed subgroup meta-analyses by study designs and reported significant increased risks of NAFLD by smoking in all the subgroup meta-analyses by each study design such as cross-sectional, case-control, and cohort studies [31]. However, they included just two cohort studies, which are not enough to confirm the association between smoking and the risk of NAFLD. In the current meta-analysis, we included 11 cohort studies involving five prospective and six retrospective cohort studies and nine cross-sectional surveys based on cohort studies. Thus, our findings would provide more clear and convincing evidence on this topic.

A notable strength lies in the subgroup meta-analyses by various factors. We confirmed a significantly increased risk of FLD in prospective cohort studies, which are generally considered as having a higher level of evidence than retrospective cohort studies and cross-sectional studies although retrospective and cross-sectional studies in the current analysis showed no significant association between smoking and the risk of FLD.

Interestingly, a significantly increased risk of FLD by smoking was observed in the studies conducted only in Europe, but not Asia and the US. We included just two studies conducted in the US that are not enough to draw a definite conclusion. Thus, more studies are required to confirm the association for people in the US. On the contrary, the number of studies conducted in Asia is sufficient to draw a conclusion. We do not have exact reasons why there were discrepancies in findings on this topic between Europeans and Asians. However, we have a potential explanation. A misclassification of Asian women’s smoking status might lead to a non-significant association between smoking and the risk of FLD in the current analysis. Unlike adult men, smoking rates among adult women have been known to be very low (<10%) in South Korea, China, and Hong Kong [41,42]. One of the main reasons of the low smoking prevalence in Asian women is under-reporting. The accuracy of smoking rates self-reported by women in Asian countries has been doubted because social repression and disapproval of women’s smoking might make women reluctant to report their smoking status [43,44,45]. A study using the Korean National Health and Nutrition Examination Survey reported that among the cotinine-verified smokers, about 60% of women classified themselves as non-smokers in a self-report survey [42]. When we performed the subgroup analysis by gender, smoking significantly increased the risk of FLD in Asian men, although there was no significant association between them in Asian women. Thus, the discrepancy in findings between Europeans and Asians might be mainly attributable to a misclassification of Asian women’s smoking status, but not race or ethnicity.

Despite the strengths, there are several limitations in this study. First, we included only five prospective cohort studies. Out of 20 cohort studies included in the current analysis, most studies were retrospective cohort studies and cross-sectional surveys based on cohort studies. As mentioned above, because prospective cohort studies give us a higher level of evidence than retrospective cohort studies and cross-sectional studies, our findings should be confirmed by further prospective cohort studies. Second, most studies measured smoking status based on a self-report. As discussed earlier, self-reported measure of smoking status may lead to under-reporting and a misclassification of smoking status, which could result in biased conclusions. Thus, further studies with biochemical validation of smoking status such as urinary cotinine levels are warranted to confirm our findings. Last, we were unable to investigate the association between the number of cigarettes smoked and the risk of FLD due to a lack of data reported in each study.

In summary, we found that smoking increases the risk of FLD in the meta-analysis of cohort studies. In addition to well-known risk factors of FLD such as obesity and alcohol consumption, clinicians should recommend smoking cessation for the management of FLD.