Objective The natural course of chronic hepatitis C varies widely. To improve the profiling of patients at risk of developing advanced liver disease, we assessed the relative contribution of factors for liver fibrosis progression in hepatitis C.
Design We analysed 1461 patients with chronic hepatitis C with an estimated date of infection and at least one liver biopsy. Risk factors for accelerated fibrosis progression rate (FPR), defined as ≥0.13 Metavir fibrosis units per year, were identified by logistic regression. Examined factors included age at infection, sex, route of infection, HCV genotype, body mass index (BMI), significant alcohol drinking (≥20 g/day for ≥5 years), HIV coinfection and diabetes. In a subgroup of 575 patients, we assessed the impact of single nucleotide polymorphisms previously associated with fibrosis progression in genome-wide association studies. Results were expressed as attributable fraction (AF) of risk for accelerated FPR.
Results Age at infection (AF 28.7%), sex (AF 8.2%), route of infection (AF 16.5%) and HCV genotype (AF 7.9%) contributed to accelerated FPR in the Swiss Hepatitis C Cohort Study, whereas significant alcohol drinking, anti-HIV, diabetes and BMI did not. In genotyped patients, variants at rs9380516 (TULP1), rs738409 (PNPLA3), rs4374383 (MERTK) (AF 19.2%) and rs910049 (major histocompatibility complex region) significantly added to the risk of accelerated FPR. Results were replicated in three additional independent cohorts, and a meta-analysis confirmed the role of age at infection, sex, route of infection, HCV genotype, rs738409, rs4374383 and rs910049 in accelerating FPR.
Conclusions Most factors accelerating liver fibrosis progression in chronic hepatitis C are unmodifiable.
- HEPATITIS C
Statistics from Altmetric.com
Significance of this study
What is already known on this subject?
Hepatitis C progression towards cirrhosis varies in each single patient, depending on the occurrence of several factors affecting liver fibrosis progression.
Some factors influencing liver fibrosis progression are modifiable (alcohol drinking, obesity, diabetes, iron overload), whereas others are not (age at infection, sex, route of infection, HCV genotype and the host genetic background).
Managing modifiable factors positively affects liver disease progression in hepatitis C, although it is unclear what the relative importance of modifiable versus unmodifiable factors is.
What are the new findings?
The factors contributing to the highest extent to liver fibrosis progression in patients with hepatitis C cannot be modified: older age at infection, male sex, intravenous drug use (IVDU) as risk factor for HCV infection and HCV genotype 3 are the most important, together with genetic variants associated with PNPLA3, MERTK and the major histocompatibility complex region, with the potential exception of lifestyle associated with a past IVDU as risk factor for HCV acquisition.
Significant alcohol consumption contributes to fibrosis progression, but only marginally, whereas anti-HIV, diabetes and body mass index do not.
A significant proportion of the liver fibrosis progression in hepatitis C is explained by hitherto unidentified factors.
How might it impact on clinical practice in the foreseeable future?
These results may help improving patients’ profiling (including genetic testing) to prioritise currently expensive therapeutic interventions.
The HCV is a major pathogen infecting 130–170 million individuals worldwide.1 HCV causes persistent infection in ∼75% of cases, mostly associated with chronic hepatitis and characterised by a relentless deposition of fibrotic tissue, culminating in the development of cirrhosis. The complications of cirrhosis are a major cause of mortality in hepatitis C and a leading indication for liver transplantation.1
Hepatitis C progression towards cirrhosis is variable, as it may be influenced by several factors, such as older age at infection, male sex, excessive alcohol drinking, obesity, insulin resistance, diabetes, HCV genotype 3, iron overload and immunosuppression.2––4 Patients cumulating two or more of these factors may progress to cirrhosis very rapidly, that is, in 5–10 years, whereas those with a more favourable profile may develop cirrhosis only after several decades, if ever at all.5 To these well-characterised factors, others have been associated with a more rapid liver disease progression in hepatitis C, although the published evidence is less consistent, such as tobacco and cannabis smoking.6 Finally, recent candidate gene7––12 and genome-wide association (GWA) studies have identified single nucleotide polymorphisms (SNPs) associated with either fibrosis stage or fibrosis progression rate (FPR): six from our GWA study carried out in Caucasian populations (rs16851720 (RNF7), rs9380516 (TULP1), rs4374383 (MERTK), rs9976971 (IFNGR2), rs738409 (PNPLA3) and rs8099917 (IFNL3)),13 and four from a GWA study performed in Japan (rs910049, rs3817963, rs3135363, rs3129860 (all in the major histocompatibility complex (MHC) region)),14 including genes involved in the innate (IFNL3 and IFNGR2) and adaptive (MHC region) immune responses, lipid metabolism (PNPLA3) and the control of apoptosis (RNF7, TULP1 and MERTK).
Patient profiling is important for prioritising interventions, both pharmacological and behavioural, for example, implementing alcohol abstinence and/or increasing physical activity, although it is clear that only a part of the cofactors of hepatitis C progression can be modified. What is unclear, however, is what portion of the liver disease progression rate is indeed affected by these cofactors and what may be ascribed to hitherto unidentified factors. The purpose of the present work is to provide an estimate of the attributable fraction (AF) of different factors of liver FPR in a large, well-characterised cohort of patients with chronic hepatitis C.
Materials and methods
In the discovery cohort, 1461 Caucasian patients from the Swiss Hepatitis C Cohort Study (SCCS) were enrolled.15 Inclusion criteria were (1) having at least one liver biopsy with fibrosis or activity scoring performed prior to any antiviral treatment and (2) an estimated time of infection with HCV. Patients with concomitant liver diseases (hepatitis B, autoimmune hepatitis, α-1 antitrypsin deficiency, Wilson's disease and haemochromatosis) were excluded. Liver biopsy specimens were formalin-fixed and paraffin-embedded for histological evaluation and analysed by experienced pathologists.
The FPR was defined as [(F+1)/D], where D is the duration of the infection in years and F the fibrosis stage, corresponding to the Metavir fibrosis stages 0–4.5 As a threshold for accelerated FPR, we used the sample median of 0.13 fibrosis units per year.
Risk factors included age at infection, sex, route of infection, HCV genotype, body mass index (BMI), diabetes, significant alcohol consumption (defined as ≥20 g/day for ≥5 years), presence of anti-HIV and several SNPs associated with the risk of stage-specific fibrosis progression (rs16851720 (RNF7), rs9380516 (TULP1), rs4374383 (MERTK), rs9976971 (IFNGR2), rs738409 (PNPLA3), rs8099917 (IFNL3), rs910049, rs3817963, rs3135363 and rs3129860 (all in the MHC region)). With the exception of SNP rs9405098, all SNPs were available in our dataset. Steatosis was dropped from the model due to its insignificant effect. Markers of tissue or systemic inflammation, such as liver histology activity scores16 and ferritin,17 were not included, as these were considered as surrogate markers of other factors.
The route of infection was initially separated into ‘IVDU’, ‘blood transfusion’ and ‘accidently infected by needle-stick’ together with ‘having undergone an invasive medical procedure’. Blood transfusion and accidently infected by needle-stick/having undergone an invasive medical procedure were then pooled together under the common definition ‘iatrogenic’, as the logistic regression showed similar effect sizes (data not shown).
HCV genotypes were grouped into two classes: type 1, 2 or 4 (hereafter called ‘non-3’) and 3, where ‘3’ is assumed to have a greater risk for accelerated FPR.2 ,4 For the variable ‘BMI’, we used an average value calculated over the whole duration of infection, starting from the date when infected to the date of biopsy. Diabetes, significant alcohol drinking and anti-HIV were categorised as ‘absent’ or ‘present’. Missing data in binary variables were imputed based on the whole cohort prevalence, coding absent as 0 and present as 1. There were no missing data for the variable ‘diabetes’; there were 332 missing data for the variable ‘significant alcohol consumption’, which was replaced by 0.22 (our population prevalence), whereas the 227 missing values for the variable ‘anti-HIV’ were substituted with a prevalence of 0.081. SNPs genotypes were available in 575 patients.18 Here, the criteria for imputing missing data were the same as those used for the whole cohort.
For all computations we used the program R.19
Logistic regression and AF
We developed a logistic regression model with FPR as outcome and risk factors as explanatory variables to quantify the fraction attributable due to that risk. The AF of risk of accelerated FPR (ie, higher than the median) was calculated according to a simplified approach called ‘average AF’,20 which allows one to control for confounding factors. Briefly, the AF of each risk factor is the proportion of cases attributable to that risk factor. In other words, the AF is the fraction of cases which would be prevented if the risk factor could be eliminated. We used a similar approach as implemented in R, based on Greenland and Drescher,21 but we kept the variables continuous (and not dichotomised as proposed, except for age at infection, which was dichotomised using the median age of 20). The 95% CIs were calculated using a bootstrap method.22 ,23
In addition, a logistic regression was performed without age at infection, as this variable is often missing. Finally, we provided also the results of a linear regression (with log[FPR] as outcome) and a Cox proportional hazard model (using Metavir fibrosis stage >2 as outcome and the duration of infection as the time component). Both types of models were developed using Akaike information criterion (AIC) model selection and done for both variable sets (ie, models 1 and 2).
We performed two multivariate logistic regression analyses to obtain OR and AF estimates: the model 1 included all variables except for the SNPs (n=1461), whereas the model 2 included all variables (n=575). We selected logistic regression models according to the AIC. The models were developed according to three procedures: one starting from the full model and going stepwise backwards, one starting from an intercept model and going stepwise forward, and one starting from a model with a random selection of variables and going in both directions. For both models, the three procedures gave comparable results and we chose the one with the lowest AIC.
We applied Bonferroni correction (α/m, m=number of covariates) for the meta-analysed results to correct for multiple testing.
Model validation and calibration
An overall goodness of the logistic regression is given by the pseudo R2,24 accounting for the optimism correction,25 using bootstrap methods. Sensitivity, specificity, positive predictive and negative predictive values were calculated using the leave-one-out cross-validation approach.
We validated our results using data from three additional cohorts: a French (FR) cohort (from the ANRS Genoscan Study Group; n=327),12 a mixed cohort of French and US patients (FR–US) with European-descent patients from Marseille and New York (n=401 for model 1 and n=412 for model 2)12 and a Sydney cohort (n=219).26 For each replication cohort we fitted the same models (ie, one without and one with the SNPs) developed based on the results from the SCCS dataset. AF estimates were then meta-analysed using a fixed effect inverse variance weighting method 27 resulting in a combined estimate for each risk factor. We first provided unbiased AF estimates 28 coming from the three replication cohorts only, and then using all four cohorts. To assess the replication success we used the combined estimate of all four cohorts. Finally, we tested for heterogeneity across the four cohorts (Cochran's Q test).
Swiss patient population characteristics
The whole SCCS dataset included, at the time of the analysis, 3702 persons. Two thousand two hundred and forty-one persons were excluded due to missing data: age at infection in 634 (17.1%), fibrosis stage in 1848 (49.9%) and route of infection in 642 (17.3%). Patients reporting ‘living with an HCV-infected person’ as risk factor for infection were excluded because of their paucity (n=32). The proportion of missing variables within the complete dataset is reported in online supplementary table S1.
Table 1 shows the variable distribution in the enrolled study population (n=1461): 63.7% of patients were men, 53.3% reported a history of IVDU, 28.4% had HCV genotype 3, 8.4% had diabetes, 19.8% reported significant alcohol consumption and 7.2% were anti-HIV-positive. The median (IQR) age at infection was 20 years (9), the median duration of infection 21 years (13) and the median BMI 24 (5). The table shows also the distribution of patients according to the same variables and the FPR, dichotomised according to the median value of 0.13 fibrosis units per year.
Table 2 shows the characteristics of the 575 patients for whom the genotyping at the SNPs associated with accelerated FPR was available. All were anti-HIV negative (except 83 which were imputed), 62.4% were men, 49.6% had a history of IVDU, 29.2% had a HCV genotype 3, 9.2% had diabetes and 25.7% reported a history of significant alcohol drinking. The median (IQR) age at infection was 19 years (9), the median duration of infection 22 years (14), and the median BMI 24.8 kg/m2 (5.1). The table shows also the minor allele frequency of the 10 tested SNPs. We used additive genetic model coding genotypes as 0/1/2. Some SNPs were imputed, resulting in continuous allele dosages between 0 and 2.
AF of risk for accelerated FPR in the Swiss patient population
The results of the univariate and the multivariate analyses (carried out on the whole patients’ population (model 1) or on the 575 patients with SNP genotypes (model 2)) are presented in table 3. Figures 1A, B show the AFs (95% CI) for each risk factor. When considering the whole study population, age at infection, sex, route of infection, HCV genotype and significant alcohol drinking were independently associated with FPR ≥0.13, the logistic regression model having a pseudo R2 of 0.17 (optimism-corrected performance as per online supplementary table S2). Age at infection (in years) had the highest AF (28.7%), with an OR of 1.10 (95% CI 1.08 to 1.12) for each additional year. The risk for having an FPR ≥0.13 was significantly higher in male patients (OR 1.39; 95% CI 1.09 to 1.77), with a corresponding AF of 8.2%. IVDU as route of infection (OR 2.22; 95% CI 1.75 to 2.81; AF 16.5%) and HCV genotype 3 (OR 2.05; 95% CI 1.59 to 2.66; AF 7.9%) were also significantly associated with having an FPR ≥0.13. The lowest AF was found for significant alcohol drinking (1.8%), with an OR of 1.26 (95% CI 0.93 to 1.72). None of the remaining variables (anti-HIV, diabetes and BMI) were associated with an FPR ≥0.13. A subset analysis omitting, the patients who had to be imputed due to missing variables, provided similar results (data not shown).⇓
Considering the model 2, age at infection, route of infection and HCV genotype were independently associated with FPR ≥0.13, whereas excess alcohol drinking, diabetes and BMI were not. Sex did not reach statistical significance. In addition, the four SNPs rs9380516 (TULP1), rs738409 (PNPLA3), rs4374383 (MERTK) and rs910049 (MHC region) were associated with higher FPR. Anti-HIV was not included in the model because all genotyped cases were anti-HIV negative. Age at infection showed the strongest impact on FPR, with an AF of 31.5%, the OR being 1.11 (95% CI 1.08 to 1.14), followed by IVDU as route of infection (AF 18.3%), with an OR of 2.55 (95% CI 1.74 to 3.78), HCV genotype 3 (AF 6.6%, OR 2.06; 95% CI 1.35 to 3.14), rs4374383 (MERTK; AF 21.1%, OR 1.49; 95 CI 1.14 to 1.95, effect allele=G), rs910049 (MHC region; AF 9.4%, OR 1.68; 95% CI 1.23 to 1.31, effect allele=C), rs738409 (PNPLA3; AF 7.6%, OR 1.40; 95% CI 1.04 to 1.90, effect allele=C) and rs9380516 (TULP1; AF 6.2%, OR 1.57; 95% CI 1.12 to 2.20, effect allele=C). The multivariate logistic regression allowing for a larger set of variables had a pseudo-R2 of 0.21 (optimism-corrected performance as per online supplementary table S2).
Online supplementary tables S3–S5 show, respectively, the results of the logistic regression as in models 1 and 2, but without including age at infection in the variable sets, of the linear regression with log (FPR) as outcome and of the Cox proportional hazards regression model with fibrosis stage >2 as outcome and duration of infection as follow-up time. Finally, the specificity, sensitivity, positive predictive and negative predictive values were 71.4%, 63.6%, 66.9% and 68.3 for model 1, and 70.3%, 72.2%, 70.0% and 72.5% for model 2, respectively.
AF of risk for FPR in the replication cohorts
Online supplementary tables S6, S7A,B and S8 show the features of the three replication datasets. Not all previously examined variables were available for these cohorts: the Sydney cohort included only cases with HCV genotype 1 and had no information about alcohol drinking or route of infection; for the FR–US cohort, only SNPs rs4374383 (MERTK) and rs9380516 (TULP1) were available (and only in 412 patients), whereas the remaining two cohorts were fully genotyped. For each cohort we performed a logistic regression, a first time including only sex, age at infection, route of infection, HCV genotype and alcohol consumption (table 4), and then adding the SNPs identified in the discovery analysis (table 5).
Finally, we performed two meta-analyses (table 6).
When considering model 1 (age at infection, sex, route of infection, HCV genotype and alcohol consumption), using only the three replication cohorts—to obtain unbiased estimates—we confirmed the SCCS results only as far as sex and age at infection were concerned. HCV genotype was not supported by the unbiased estimate (p=0.25) neither did significant alcohol consumption (p=0.16). In model 2, the AF for rs4374383 (MERTK) was 10.3% (p=0.048) and for rs738409 (PNPLA3) was 7.6% (p=0.024), confirming the SCCS results, whereas all other SNPs did not replicate.
In the meta-analysis including all four cohorts (model 1), the AF of age at infection, sex, route of infection, HCV genotype and significant alcohol consumption were all confirmed. In model 2, rs9380516 (TULP1) failed to show a strong effect over all four cohorts (p=0.1), whereas all the remaining SNPs did so. Taking multiple testing into account, the Cochran's Q test showed homogeneity among the cohorts.
This is the first study to analyse the AF—related to a comprehensive range of factors—of risk for accelerated FPR in a large, well-characterised population of patients with chronic hepatitis C. Our major finding consists of the observation that FPR in chronic hepatitis C seems mostly attributable to factors, such as age at infection, sex, route of infection, HCV genotype and some host SNPs—that cannot be modified by targeted interventions. The contribution from the variable ‘route of infection’ may, however, represent a theoretical exception, assuming that this factor is an indirect measure of a liver ‘unfriendly’ lifestyle, possibly characterised by regular tobacco and/or cannabis smoking, known contributors to liver disease severity.29––31 Alternatively, patients who acquired HCV in the IVDU setting may have been repeatedly exposed to unidentified bloodborne pathogens contributing to liver damage. Both hypotheses need to be further explored in more detailed datasets: if smoking were confirmed to possess an elevated AF of risk for accelerated FPR, it should clearly become the target of appropriate lifestyle-modifying interventions.
Interestingly, some common factors of disease progression, such as diabetes, excess body weight and HIV infection, failed to confirm their reputed role as major players in driving the fibrotic process. Although the lack of association with the HIV infection may be explained by good harm reduction programmes and high level of adherence to antiretroviral therapy, the surprising finding is that the contribution of significant alcohol drinking appeared to be small. Finally, a significant portion of the FPR could not be explained by common cofactors. Hitherto unidentified viral and/or host and/or environmental cofactors may participate in the liver disease progression process. It is unlikely that these include known factors that were not documented in our population, as these are usually either confined to small subgroups (such as iron overload and immunosuppression) or may play but a controversial role in the fibrotic process (like tobacco and/or cannabis smoking). We performed a post hoc power calculation to estimate what sample would be needed to detect the observed effects. Assuming 90% power and a type 1 error rate=0.05, and taking into consideration the actual frequencies observed, we estimated a minimal sample size of n=248 for the HCV genotype 3 versus non-3 or n=503 per group when comparing sex. Because we used the actual data to determine the ‘hypothesised’ effect size, the sample size should have been much higher for non-significant variables, such as diabetes (n=16 141) or HIV (n=2728).
Although somewhat surprising, we think that our results are reliable for several reasons. First, we did not rely solely on the static assessment of liver fibrosis by a single liver biopsy, but rather calculated the rate of fibrosis progression, because all patients had an estimated date of infection. Second, the data obtained in our cohort were corroborated by the replication analysis on three additional cohorts—with slightly different variables acquisition and geographical background—and finally confirmed by a meta-analysis. These observations may have a significant impact for the management of hepatitis C, ranging from accrued surveillance of patients at risk of liver disease progression to setting priorities in antiviral therapy allotment in resource-constrained settings. In particular, these data may prove useful to carry out cost utility analyses, especially when assessing the cost-effectiveness of antiviral therapy in patients before reaching the cirrhotic stage.
Our findings are especially unexpected with regard to the marginal effect of significant alcohol misuse. Although our data do not contradict the notion that significant alcohol consumption bears negative consequences—in terms of morbidity and mortality—in chronic hepatitis C, the observation that the AF of alcohol misuse was as low as 1.8% is surprising, although consistent with a previous study.32 This is in contrast to a recent work from Scotland, highlighting the substantial role of heavy alcohol use in promoting the development of liver cirrhosis in persons with chronic HCV infection, with an AF exceeding 50%.33 However, the Scottish model may not necessarily apply to other countries where alcohol misuse is not so widespread and the drinking pattern may be different. The appreciation of significant alcohol drinking was also very different: in the Scottish study this was defined as 50 units/week in most weeks for at least 6 months at any point prior to study enrolment and found in as many as 34% of patients.33 In our study, we chose the lower threshold of 20 g/day (ie, 14 units/week) for 5 years of significant alcohol drinking, and yet this was reported by as few as 19.8% of participants. Furthermore, when we run a similar analysis (limited to the Swiss cohort) using the higher threshold of 40 g/day (corresponding to 28 units/week), the impact of alcohol drinking lost statistical significance, with an OR of 1.07 (95% CI 0.98 to 1.16; p=0.13) by univariate, suggesting some degree of underestimation of the alcohol consumption in some heavy drinkers. Indeed, we cannot rule out false-negative data concerning self-declared alcohol consumption: this limitation may be overcome by measuring carbohydrate-deficient transferrin in future studies, as suggested.34 Finally, although our previous analysis from the same data set showed that the standardised mortality ratio in case of heavy alcohol use was 3.66 in non-cirrhotic patients, reaching a staggering 15.62 in cirrhotic patients,35 the increased mortality reported in alcohol misusers is not necessarily accounted for by liver-related events, but rather by heart disease, stroke, a variety of cancers and accidents.36
The impact of HCV genotype 3 on liver disease progression was confirmed in both models, suggesting a substantial role (AF 5.4% in the meta-analysis) of this variable in promoting fibrosis progression. The mechanism of this effect is unknown. Viral steatosis, typically associated with this genotype37 does not seem to contribute to fibrogenesis, as shown previously.2 Oddly, patients with HCV genotype 3 infection do not have increased necroinflammatory activity,37 the most important mediator of fibrosis progression.16 We hypothesise that HCV-infected hepatocytes secrete profibrogenic cytokines, directly stimulating hepatic stellate cells to synthesise collagen bypassing the classical wound-healing process. This observation requires further research, particularly in view of the lukewarm results of therapy for genotype 3-infected patients with the novel antivirals.37
Among the various SNPs previously reported to be associated with fibrosis progression by GWA studies,13 ,14 three were confirmed as playing a significant, independent role in liver fibrosis progression by our meta-analysis. It has to be emphasised that the SNPs reported by Patin et al13 were strongly correlated with other measures of fibrosis, such as non-dichotomised FPR, F0 to F4 or other transitions regardless of fibrosis duration, thus we do not expect all of them to replicate in our set up. The potential mechanistic link between these genetic variants and fibrosis has been discussed previously.13 ,14 The confirmed association between rs4374383 and fibrosis deserves attention. This SNP is located in MERTK, a tyrosine protein kinase expressed in macrophages, where it controls the phagocytosis of apoptotic cells.38 Interestingly, it has been shown that the host immune response towards apoptotic autoantigens, mediated by the activation of polyfunctional CD8+ T-cell responses, may be a driver of liver disease progression.39 Genetic testing for these SNPs may prove useful in patient profiling and become part of the diagnostic algorithm to identify patients with hepatitis C at risk of progression, for example, when prioritising costly treatments.
Our study has various limitations. The date of infection in patients with a history of intravenous drug injection was assumed to correspond to the first year of drug misuse, in keeping with previous work, although more recent studies suggest that there may be a median lag of 3 years between first injection and HCV infection.40 However, given the fact that the median duration of infection in patients with a history of IVDU was 20 years, assuming that drug users would acquire HCV at the start of their misuse would prolong the estimated duration of their infection by ∼15%, modestly affecting the FPR estimate. Second, missing data on HIV infection status and alcohol drinking compelled us to impute them in a small proportion of cases. Then, because our main study aim was to assess AF of risk factors, we explored only a dichotomised FPR measure. Analysis of the continuous FPR variable, however, led to similar conclusions. In model 1, age at infection, sex, route of infection and HCV genotype remained significant (adjusted R2=0.26). In model 2, the SNPs were rs16851720 and rs910049, additionally to the SNP rs16851720 (RNF7) and age at infection, sex, route of infection and HCV genotype (adjusted R2=0.31, optimism corrected).
Notably, the low pseudo-R2s show that fibrosis progression remains poorly understood. In particular, it is difficult to test for interactions with relatively low number of patients. As the progression of fibrosis is an extremely complex process, various phenotype definitions describe different aspects of it, each of which may have slightly different underlying risk factors with a diverse strength of impact. We believe that there are many risk factors (and interactions) we do not know and that the FPR is probably a more complex process which is not yet fully understood.
We also concede that the AF, as discussed elsewhere,25 ,41 ,42 can be calculated according to many distinct approaches in a multivariate setting, and it is unclear which is the best. Innes et al33 used logistic regression models with multilevel risk factors. Additionally, using some approaches, the sum of the AFs of all risk factors can be higher than one, and some AFs are even estimated to be negative. However, dichotomising risk factors there might be a loss of information of the variable (risk factors are not necessarily binary). We calculated the AF under the assumption that the AF stands for each risk factor itself, and thus can add up to more than one.
Note that the assumptions for the calculation of valid population AF require the factor to be uncorrelated to other outcome-related factors and causal for the outcome. The first assumption cannot be guaranteed because many possible unobserved factors affecting fibrosis progression speed are unknown, thus not testable. The question of causality is very complex and also hard to address in a cross-sectional study, where most frequently propensity score or instrumental variable analyses are applied. In our situation, propensity scores are not applicable as clearly not all confounders are measured.43 However, we could not identify a sufficiently strong instrumental variables, which renders this approach (given our low sample size) underpowered. It has been discussed in more detail why AFs do not necessarily imply a causal relationship.44 For these reasons, our estimated AFs must be interpreted with care. Importantly, however, most of the factors we identified (such as age at infection, sex, route of infection, HCV genotype and certain host genotypes) are established chronologically well before liver fibrosis starts, thus these are bound to be—even if not causes—at least early predictors rather than just correlates of FPR. We have conducted further analyses (linear regression on FPR, logistic regression for binary outcome, Cox regression) to examine the effect direction and ranking of the identified risk factors. This analysis showed that no matter which measure of risk effect we apply, the results are comparable and robust.
There are several possible sources of bias in this study:45 selection bias (if the study population does not represent the target population: eg, IVDU have only limited healthcare access and therefore some information is missing for certain samples, eg, HIV) and information bias (individuals cannot remember the date of infection or other medical history details). In addition, referral and recruitment bias may be due to the fact that patients with chronic hepatitis C are included into cohorts only if they are referred to the study centre, or when they develop symptoms and/or complications due to the liver disease. Further intervention studies with better characterised patients’ populations may address all these issues.
In conclusion, accelerated fibrosis progression in hepatitis C is primarily accounted for by unmodifiable cofactors, with the potential exception of lifestyle associated with a past IVDU as risk factor for HCV acquisition. A large portion of its interindividual variation remains unexplained and may be either due to unknown viral, host and/or environmental factors, although new, potentially actionable knowledge of fibrosis progression may follow from functional studies on the molecular mechanisms of the host genetic risk factors. These results may form the basis of algorithms for antiviral treatment allocation, especially at a time when it becomes mandatory to prioritise the use of expensive interferon-free regimens (especially in some parts of the world where access to new regimens will be limited by cost issues). For example, patients with three or more unfavorable and yet unmodifiable factors of accelerated FPR—for example, older age at infection, male sex and at least one gene variant—may benefit from immediate treatment. However, those with profiles less prone to accelerate FPR may be safely deferred from therapy until cheaper but equally effective combinations become available. We propose that the differential impact of the factors of accelerated FPR be incorporated in decision-making algorithms to increase the cost effectiveness of hepatitis C treatment interventions.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online tables
SR, P-YB, ZK and FN contributed equally.
Contributors SR, P-YB and ZK analysed the data and wrote the manuscript; J-FD, BM, DS, MHH, DM, AC, RM, DB, VS, JG, LA, PH, MB, AHT, IMJ, EP, BN, TP, SP, LA and FN collected the data and wrote the manuscript; all authors accepted the final version of the manuscript.
Funding The SCCS is supported by grants from the Swiss National Science Foundation (3347C0-108782/1 and 33CS30_148417/1). The French cohort is supported and sponsored by The National Agency for Research on AIDS and Viral Hepatitis (ANRS; ANRS Study HC EP 26 Genoscan). FN is supported by the Swiss National Science Foundation (314730-146991), ZK by the Swiss National Science Foundation (31003A-143914) and the Leenaards Foundation and P-YB by the Swiss National Foundation (32003B-127613), the Leenaards foundation and the Santos-Suarez foundation.
Competing interests None.
Patient consent Obtained.
Ethics approval IRB from all local centres participating to the various cohort studies.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.