Main text

The compartmentalization of liver cirrhosis

Both epidemiologically and clinically, liver cirrhosis and other chronic liver diseases are generally compartmentalized into separate categories based on their etiology. Thus, the Global Health Estimates [1] or the Global Burden of Disease Study [2] give prevalence, incidence, and mortality rates for them in different categories, separating cirrhosis and other chronic liver diseases by etiology: alcohol use, hepatitis B virus infection, hepatitis C virus (HCV) infection, non-alcoholic steatohepatitis, and other causes. Similar differentiation can be found clinically and in the International Classification of Diseases [3], even though there have been some calls to change this system [4, 5]. This contribution will not focus on the logic of current classifications, but will look at the intersection between two of these seemingly separate categories, i.e. liver diseases due to alcohol use and due to HCV infection.

Aims of the current contribution

Based on a systematic literature search, we examined the role of heavy alcohol use—as operationalized via alcohol use disorders (AUDs) [6, 7]—on the progression of liver disease in people with chronic HCV infection. We hypothesized, based on a recent large-scale retrospective cohort study on all hospitalizations in France [8], that a large proportion of complications arising over the course of liver disease in people with HCV infection is attributable to AUD. The main outcome was “decompensated liver cirrhosis”, defined as an acute deterioration in liver function in a patient with cirrhosis, and characterized by jaundice, ascites, hepatic encephalopathy, hepatorenal syndrome and/or variceal hemorrhage [9, 10]. We will summarize the link between AUDs and decompensated liver cirrhosis and other complications of liver disease in people with HCV infection, by pooling relevant studies using meta-analytical techniques [11].

Methods

Systematic search and inclusion/exclusion

As a first step, we conducted a systematic search of epidemiological studies on the relationship between alcohol use and progression of liver disease due to HCV, using Ovid Medline databases, and applying the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) criteria [12]. The exact search terms can be found in the Supplementary Materials (Table S1), but we looked for cohort or case-control studies (for definition, see [13]) of people with chronic HCV infections; with at least two different forms of alcohol use (e.g., alcohol use yes/no; AUD yes/no); and a verified indicator of progression of liver disease (e.g., progression of fibrosis, progression of cirrhosis to decompensated cirrhosis; liver death). This search was initially conducted on July 28, 2019 and updated on December 22, 2019, with the updated search yielding 467 references (see Fig. 1).

Fig. 1
figure 1

Systematic search results and selection process of studies for the meta-analyses

Full size image

In addition, we conducted a search on systematic reviews and meta-analyses on this theme and searched for relevant articles this way (in particular [14,15,16], see the latter for further detail on the search strategy). As the second step, all articles which included AUDs as exposure were selected as the basis of the analyses of this paper.

For measurement of AUDs, we included the following: hospital record, other record in the healthcare or government database (e.g., registration), physician judgement, standardized measurement with a validated instrument such as the World Health Organization (WHO) Composite International Diagnostic Interview [17], self-report about major criteria of AUDs (for a general discussion of criteria, see [6, 18]). While chronic heavy drinking above 80 g pure alcohol per day before or at some point over the course of the disease could be part of the inclusion criteria (e.g., [19]), it alone did not suffice for inclusion in the study, as no generally accepted quantitative threshold for AUDs based on level of drinking has yet been established [7, 20]. We also excluded studies solely based on screening instruments, such as the CAGE [21] or the AUDIT [22].

All articles were screened by two of the authors (either LLF, MG, or JR). An overview of the selection process can be found in Fig. 1 [23]. All included studies were extracted by two authors (LLF and JR) for key information (study characteristics: title, authors, year published, country, study design, year of study; study population: total number of patients with HCV infection; participant details: mean age, sex, HIV coinfection, number of people with AUD; measurement of AUD; risk relations: relative risk indicator, confidence intervals, p value, adjustments (other covariates); and outcomes).

Statistical methods

Relative Risk estimates (either Odds Ratios, Relative Risks or Hazard Ratios [13, 24]) were taken directly from the respective article or calculated based on a 2*2 table [25] or, in one case, using the methodology specified by Hamling et al. [26].

In the main analysis, only decompensated liver cirrhosis or its main constituents (defined above [9, 10]) as confirmed from medical records, hospitalization or death was chosen as the endpoint (n = 7 estimates from 5 studies). A random effect meta-analysis [27], accounting for the hierarchical structure of the data (three estimates from one study) was conducted [28].

In a secondary analysis, we repeated the random-effect meta-analyses for all estimates of progression of liver disease, i.e. applied a looser inclusion criterion. Here, the following endpoints were including as well (see also Table 1: advanced fibrosis, liver cirrhosis, liver deaths). For both models, we tested if adjustment for important covariates has an impact on the overall effect. Cochran’s Q and the I-squared statistic were used as indicators for heterogeneity [38, 39].

Table 1 Characteristics of studies included in the meta-analyses
Full size table

Lastly, we calculated the population-attributable fraction (PAF), i.e. the proportion of cases with liver disease progression that could have been avoided if no person with a chronic HCV infection had an AUD. The PAF was calculated using Formula 1 by combining data on the prevalence of AUD (P) with corresponding RRs [40]. All analyses were performed with R version 3.6.1 [41].

$$ \boldsymbol{PAF}=\frac{\boldsymbol{P}\left(\boldsymbol{RR}-\mathbf{1}\right)}{\mathbf{1}+\boldsymbol{P}\left(\boldsymbol{RR}-\mathbf{1}\right)} $$
(1)

Results

Table 1 gives an overview of the studies and their characteristics. In total, studies including 286,641 people with chronic HCV infection fulfilled the inclusion criteria, of whom 63,931 (22.3%) qualified for an AUD.

In the main analysis, 268,114 people with chronic HCV were included, of whom 21,882 had decompensated liver cirrhosis (8.2%). A total of 63,335 people, or 23.6% of this sample, were identified with AUD, a proportion much higher than seen in the general population [2, 42]. Using decompensated liver cirrhosis as the outcome for the main analysis (based on n = 7 estimates), an AUD diagnosis was associated with a 3.3-fold risk for progression of liver disease among people with a chronic HCV infection (95% Confidence Interval (CI): 1.8–4.8), see Fig. 2). There was no significant difference between studies that were adjusted for important covariates and those that were not (p-value = 0.878). In terms of population attributable fractions, slightly less than 4 out ten cases of decompensated liver cirrhosis cases were attributable to AUD: 35.2% (95% CI: 16.2–47.1%).

Fig. 2
figure 2

Forest plot for risk of decompensated liver cirrhosis associated with alcohol use disorder

Full size image

As a secondary analysis, all estimates of liver disease progression were pooled, which yielded a similar overall effect of AUD as compared to the main analysis (n = 13 estimates; OR = 3.7; 95% CI: 2.2–5.3; see Supplementary Materials Figure S1). The risk difference between decompensated liver cirrhosis and the other indicators of liver disease progression was not significant (relative risk ratio: 0.6; 95% CI: 0.3–1.3). Again, adjustment for important covariates was not related to substantial reductions of the effect (p-value = 0.969). Again, this would be equivalent to about 40% of cases with liver progression being attributable to AUDs (attributable fraction: 39.3%; 95% CI: 21.9–50.4%).

In both analyses, substantial heterogeneity was identified using Cochran’s Q statistic ( [38]; main analysis for n = 7 estimates: Q (df = 6) = 507.3, p < .001; secondary analysis for n = 13 estimates: Q (df = 12) = 1184.5, p < .001), presumably associated with the large variation in sample sizes in the included studies (minimum = 77, maximum = 97,347). The I-square statistics also indicated substantial heterogeneity.

Discussion

Before discussing the results and implications of our findings further, we would like to highlight the potential limitations.

Limitations

One limitation to this review and meta-analysis is the reliance on aggregate data, which relies on the qualities of the underlying published studies, based on heterogeneous populations, different study designs and different statistical models, and in different historical periods of time [11]. Even though the populations were heterogeneous, almost all of the studies included are from high-income countries. Given the global load of alcohol-attributable liver cirrhosis burden [43, 44], we urgently need data from other regions of the world, especially from regions with a high prevalence of HCV infections such as Africa and Central Asia [2, 45], but also from countries in Eastern Europe where prevalence of HCV infections and of AUDs are high (e.g., Moldova, Georgia [46]).

Also, the largest studies [8, 29] relied on medical hospital records of AUD, which likely underestimated the true prevalence, as this disorder is highly stigmatized [47] and neither necessarily disclosed nor recorded in hospitals or healthcare settings (for a wider discussion, see [48, 49]), even for 100% alcohol-attributable disorders [50, 51]. However, the bias introduced by underestimating the prevalence of AUDs is conservative; the attributable fractions would likely be higher with higher prevalence (for formulas, see [40]). Additionally, relying on hospital records for the largest studies removes potential biases due to self-report of AUDs [52].

Another potential limitation involves the exclusion of studies where AUDs could only have been inferred by a mention of chronic heavy drinking or other drinking behaviours closely related to AUDs. On the one hand, heavy drinking is a key characteristic of AUDs [20]. To give one example, there is a high likelihood that lifetime drinkers with more than 175 g pure alcohol consumed daily—such as in the study of Corrao and colleagues [53]—would qualify for AUDs had this condition been measured with validated instruments. One the other hand, it is hard to draw a threshold. In the same study by Corrao and colleagues, the following thresholds were used to indicate the drinking level: 50 g, 75 g, 100 g, 125 g and 150 g pure alcohol per day. It is not clear which of these drinking-level categories would indicate AUD. Thus, while AUDs constitute a common medical diagnosis, the use of this diagnosis—which is clinically relevant and can be used in health services research—in epidemiological research may lead to biases, as the active ingredient in disease progression—ethanol—is only indirectly assessed (see also [7, 54]). Another aspect of patterns of drinking deserve mentioning. These patterns – especially the prevalence of heavy episodic drinking – differ vastly between the countries examined here [55]. It has been shown that a pattern of daily heavy drinking is most detrimental for worsening of liver disease [56, 57], for daily drinking is less prevalent among heavy drinkers in countries like Poland or Scotland, compared countries like France, to mention just three of the countries in our sample. Without measuring patterns of drinking at the individual level, variation is introduced into our results. Future research should not only rely on wide categories such as AUD [6], but should measure drinking level and patterns.

As we wanted to conduct a meta-analysis with a narrow outcome – decompensated liver cirrhosis, we defined our search terms excluding wider definitions such as hepatocellular carcinoma. Indeed, we achieved this goal, and only in one study [35], a minority of cases included hepatocellular carcinoma. This does not mean, however, that AUDs do not causally impact on hepatocellular carcinoma in people with chronic HCV infections. As Schwarzinger and colleagues demonstrated in the national French hospital cohort comprising 6404 patients with hepatocellular carcinoma [8], AUDs were associated with a fourfold-increased risk (4.23; 95% CI: 3.99–4.49).

Alcohol use disorders as a key determinant of liver disease progression

Our results show that AUDs are quite common among people with chronic HCV infections, and that they are a key determinant for worsening of liver disease. Our design did not allow us to answer the question of whether alcohol use or AUDs were the only factor in disease progression (see [15]); however, other research seems to indicate that HCV in people with AUDs also showed increased disease progression (e.g., [6, 58, 59]) and, thus, there seems to be an interaction effect of alcohol use and HCV infection. There are also plausible biological pathways, such as increased viral replication and altered immune response [60].

Our design also does not answer the question regarding a dose-response relationship for alcohol use, i.e., whether all levels of alcohol use are detrimental for liver disease progression (see [16, 61]). However, we can clearly state that AUDs, with their high levels of alcohol consumption, produce a markedly worsened progression for liver diseases, and were responsible for about 40% of all these complications in the large cohorts underlying our study (see [62, 63], for further discussion). Such a high attributable fraction also calls into question the compartmentalization of liver cirrhosis into subtypes/categories [4, 5].

Conclusions

AUDs are relatively frequent in people with chronic HCV infections and contributed markedly to progression of liver disease. Two main conclusions result: despite the current clinical compartmentalization, alcohol use and AUDs should be assessed in patients with liver cirrhosis of any etiology (see also [64] for the dose-response relationship for alcohol use on any kind of liver cirrhosis), and irrespective, if the HCV infection has been treated successfully of not. This is even true for so-called “non-alcoholic” liver disease categories, where AUDs do not play a role by definition, but alcohol use still may [65]. But assessment is not sufficient, AUDs need to be treated by either achieving full abstinence (the best outcome of alcohol interventions for any kind of liver disease [66]) or, if this is not possible, reducing consumption to the highest degree possible [67,68,69].