The association between hepatocellular carcinoma and direct‐acting anti‐viral treatment in patients with decompensated cirrhosis

Direct‐acting anti‐viral therapy (DAA) has transformed hepatitis C virus (HCV) care, particularly in patients with decompensated cirrhosis. However, their impact on hepatocellular carcinoma (HCC) remains unclear.


| INTRODUC TI ON
Hepatitis C virus (HCV) infection is a leading cause of liver cirrhosis and hepatocellular carcinoma (HCC), the second most frequent malignant cause of death worldwide. 1 With the advent of direct-acting anti-viral (DAA) therapy for HCV, treatment options and curative rates have been transformed with high rates of sustained virological response (SVR). 2,3 These agents have also facilitated the treatment and cure of patients with advanced liver disease who remain at risk of HCC 4 and are therefore recommended to continue lifelong surveillance. 5,6 There is controversy around patients with cirrhosis who have cleared virus (ie achieved an SVR) on DAAs and their ongoing risk of developing HCC. Conti et al reported an increased incidence of HCC following DAA treatment with 3.16% (95% CI 1.45-5.90) of 285 patients developing an HCC within 24 weeks of therapy. 7 Supporting this Ravi et al found an unusually high risk (9%) of patients developing de novo HCC following DAA treatment. 8 Conversely, multiple studies have shown no increase in HCC occurrence 9 following viral clearance and a large American cohort of 62 354 patients with and without cirrhosis showed that although patients with cirrhosis who had cleared virus with DAA therapy did develop malignancy, the frequency was not increased. 10 These studies have suggested that alcohol consumption, diabetes mellitus, lower platelet count and higher aspartate aminotransferase (AST)/alanine aminotransferase (ALT) ratio 11 are baseline characteristics that predict HCC development.
In addition to the impact of HCV clearance on HCC development, there is controversy regarding the impact of HCC on HCV treatment outcome.
Prenner et al showed a greatly reduced SVR rate of 58% for patients with an HCC present on treatment initiation, with this rising to 97% in patients with a previous history of treated HCC prior to DAA commencement. 12 This implies that the presence of HCC may reduce the response to treatment though this study includes patients post-liver transplantation.
The prognosis following the diagnosis of HCC in patients with HCV and cirrhosis is poor with a median survival as low as 0.7-0.9 years. 13 In the SHARP trial of sorafenib in patients with advanced HCC, time to progression on imaging regardless of the initial cause was 2.8 months in the placebo group. 14 It is still not known whether clearance of HCV impacts tumour progression, but anecdotal evidence has suggested that it may slow evolution.
In light of these uncertainties, we examined the NHS England early access programme (EAP), which provided access to 12 weeks of alloral DAA therapy for patients with advanced liver disease. Patients in this programme remain on surveillance, and here, we report the incidence and factors predictive of de novo malignancy in patients developing HCC early (within 6 months) or late (after 6 months) after the onset of DAA therapy, the impact of HCC on DAA treatment response and the progression of cancers in viraemic and nonviraemic patients.

| Patients
All patients enrolled in the NHS England early access programme (EAP) were encouraged to enrol in HCV Research UK (HCVRUK) with written informed consent. Details of the treatment (June 2014-September 2015) and management of the early access programme cohort have been published previously. 15 In brief, patients with decompensated cirrhosis were offered 12 weeks therapy with either sofosbuvir/ledipasvir or sofosbuvir plus daclatasvir, with or without ribavirin at the clinician's discretion. Entry to the English early access programme specified that all patients had to have either a diagnosis of hepatic decompensation in the past or have current evidence of CTP score B or C.

| Data collection
Baseline data included age, gender, ethnicity, alcohol usage, smoking status, diabetes mellitus, HIV status and use of proton pump inhibitors or statins. Data were also available for HCV (route of infection, genotype), date of cirrhosis diagnosis and decompensation diagnosis, previous HCV treatment and Child-Turcotte-Pugh score within the year preceding treatment. The Child-Turcotte-Pugh score was converted to a stage centrally for interpretation purposes (stage A-score 5-6, stage B-7-9, stage C-10-15). Local accredited laboratory measurements for the preceding year were collected with the highest serum HCV RNA, lowest serum sodium, lowest creatinine, highest alanine aminotransferase (ALT), aspartate transaminase (AST), highest bilirubin, lowest albumin, highest alpha-foetoprotein (AFP), highest clotting studies and lowest full blood count measurements used. The model for end-stage liver disease (MELD) score, AST to platelet ratio index (APRI) score and albumin to bilirubin (ALBI) grade were calculated centrally. Length of follow-up was defined as the date of onset of DAA treatment until the date of death, date of transplantation or date of survey, whichever occurred first.
DAA treatment type and commencement date were noted.
Sustained virological response (SVR) was defined as negative for serum HCV RNA at 12 weeks (SVR12) following the completion of treatment. Patients with incomplete HCV treatment outcome data either due to death prior to SVR12 tests or those lost to follow-up were removed from the analysis.
All patients were subject to national guidelines recommending an ultrasound scan every 6 months with further cross-sectional imaging if indicated. All local imaging and multidisciplinary team (MDT) reports were collected centrally by the study team for the year prior to therapy and following therapy up until the study endpoint. Tumour size measurements were taken from radiological reports and Barcelona clinic liver cancer (BCLC) scores, 16 Liver Reporting & Data System (Li-RADS) grading, 17 Milan criteria 18 and response evaluation criteria in solid tumours (RECIST) criteria 19 were applied and appropriate scores/ grades generated by the study team. RECIST criteria, which take into account the size and progression of the primary lesion, secondary lesions, nodal, vascular and metastatic disease to give an overall definition for complete resolution, partial resolution, stable or progressive disease, were used to assess tumour progression with the date of cross-sectional imaging being used to define the observation period.

| Statistics
Baseline characteristic data are presented as the median and interquartile range (IQR) for continuous variables or as frequencies and percentages for categorical ones. Mann-Whitney U and chisquare tests were used for baseline characteristic and subsequent comparisons.
Count data for two-group comparisons were analysed with two proportions tests using the normal approximation method to calculate the P-values. We have also performed odds ratio analyses using the z-score calculated as ln(OR)/SE{ln(OR)}. The odds ratio (OR), standard error and 95% confidence intervals were calculated according to Altman, 1991. 20 To analyse the association of HCC development with several variables in our dataset and investigate potential confounding factors, we have used multiple logistic and Cox regression models. The binomial logistic model was built to explain the HCC status (Yes/No) with the inclusion of important predictors from an initial univariate analysis in respect to both deviance and Hosmer-Lemeshow goodness-of-fit tests while maintaining the variance inflation factor to the minimum. We also investigated potential interactions that were included as interaction terms in the model. The Cox proportional hazards regression analysis was used for a time-dependent outcome (time to develop HCC) and produced hazard rates allowing the quantification of the effect (risk) per group or unit change depending on the nature of each predictor. The effect of each variable is presented with hazard rates and 95% confidence intervals. For continuous variables, the hazard rate was calculated for a clinically meaningful increment of change.
Time-to-event analyses were performed using the nonparametric Kaplan-Meier method. 21 The survival distributions were compared for equality for two groups at each comparison. All lost to follow-up cases were censored up to the most recent time point with available information. For each comparison, the log-rank test results are presented, but the Breslow and Tarone-Ware tests were also considered. P < 0.05 was considered statistically significant.
Data analyses were performed using IBM SPSS version 25 (Armonk, NY) and GraphPad Prism version 6.0 (San Diego, CA).

| Baseline demographics
We identified 81 patients in the early access programme within the HCV Research UK database treated with DAA therapy between June 2014 and September 2015 who developed HCC subsequent to the onset of therapy. These were frequency matched with 178 early access programme patients who were treated with DAAs but did not develop HCC within the follow-up period. We excluded patients lost to follow up or who died before SVR outcome became known (1 HCC patient, 13 non-HCC patients). HCC was diagnosed by MRI in 45 patients, CT scan in 26, while eight patients had incidental HCC diagnosed within their explanted liver. One patient had a date of diagnosis, but no mode of diagnosis was available.
The demographics of the cohort are shown in Table 1.  end-stage liver disease (MELD) score, a median of 11  15 of the 24 (63%) nonmalignant lesions were considered to have progressed to HCC, with six of these patients presenting with an early HCC and the remaining nine developing a late malignancy.
The breakdown for these baseline lesions is shown in Figure 1.
In univariate analysis comparing the 80 HCC patients with the matched population, factors associated with the development of HCC were diabetes, lower albumin, nonmalignant lesion seen on pretreatment ultrasound scan and a lower platelet count. These variables were entered into both logistic and Cox regression models for multivariate analysis, with both models returning all but albumin as statistically significant predictors. The effect size of albumin was reduced in the multivariate models due to its strong correlation with platelets (Spearman rho P-value = 0.007). In Table 2, we present the results from the Cox regression analysis in order to fully incorporate the time-dependent nature of the outcome (time from the start of treatment to HCC development).

| Progression of liver cancers arising early after starting DAA compared to later cancers
We compared cancers that developed soon after therapy with those developing later to test the hypothesis that the elimination of the virus-associated inflammatory response leads to a more aggressive tumour. Figure 2 shows that there was no significant difference in either the progression of the tumour (Figure 2A) or overall survival ( Figure 2B) between these two groups. Indeed, patients with HCC developing soon after viral elimination appeared to fare slightly better, although this was not statistically significant.

| Progression of liver cancer following viral clearance
To examine the hypothesis that malignancy developing in an uninfected liver (ie post-SVR) may be more aggressive than cancers that develop in an HCV-infected liver, we examined HCC prognosis by Kaplan-Meier estimation. Figure 3A,B shows that the time from cancer diagnosis to progression (P = 0.17) and death (P = 0.7), respectively, were similar in patients who did, or did not, achieve viral clearance.   Table 3 which shows no statistically significant differences between viraemic and nonviraemic patients.  At present, we would recommend more intensive HCC surveillance in patients with these characteristics to allow early identification of lesions at a stage where they may be amenable to therapy.

| D ISCUSS I ON
The significance of pretreatment nonmalignant lesions presents a challenge for hepatologists. The LI-RADS criteria were developed to try and overcome this, but diagnostic uncertainty remains. 17,29 We have shown that patients with apparently nonmalignant lesions on scans taken within 12 months of the onset of DAA therapy are more likely to go on to develop HCC. This is in keeping with the notion that many HCCs diagnosed after the onset of DAA therapy were already present beforehand, a phenomenon previously noted by others. 30 Nahon et al found that 5/15 patients had a nonmalignant nodule observed within 6 months prior to starting DAA treatment and subsequently developing HCC with this shown as a statistically significant risk factor for HCC development. 24 Alternate to this, Toyoda et al recently found no effect of previously identified nonhypervascular hypointense nodules (NHHNs) on HCC incidence; however, these were all compensated cirrhotic patients with all nodules found on contrast-enhanced MRI scans as opposed to the less sensitive ultrasound scanning, which most of our patients received. 31  follow-up period is a year longer, this suggests that the progression of these nodules occurs early following DAA initiation. 32 Vigilance is clearly indicated in patients with pre-existing liver lesions.
We found that patients diagnosed with HCC within 6 months of the onset of DAA therapy are less likely to achieve SVR12. Prenner et al reported that in a cohort of 137 patients with pre-existing HCC treated with regimens incorporating sofosbuvir, ledipasvir, simeprevir, ombitasvir/paritaprevir/ritonavir and ribavirin, 21% failed to achieve SVR, significantly more than those patients without HCC at baseline (P = 0.009). 12 These data may be interpreted as indicating a difficulty for DAAs to penetrate a small pre-existing liver cancer effectively. Alternatively, a strain of HCV which has a higher oncogenic effect may be present which renders DAAs less effective when coupled with the above. However, in our study, we also detected a lower SVR12 rate (65%) in patients who were diagnosed with HCC more than 6 months after the onset of therapy.
This suggests that either virus-infected premalignant/malignant cells that are treatment resistant are present for a very long time before presenting as overt malignancy or viral or host factors that predispose to malignancy are also involved in treatment failure. Whatever the mechanism of tumour development, physicians should be aware that patients who fail DAA therapy may be at increased risk of HCC development to allow early detection of malignancy. We have adopted a local, albeit none evidence-based protocol involving 3 monthly scanning of such patients for the first 12 months following completion of anti-viral therapy.
The important question of whether liver cancer is more common and/or more aggressive following viral clearance is difficult to answer.
This would necessitate randomising patients with cirrhosis to treatment or observation and is unlikely to be popular with patients or, in our view, ethical. The use of historical controls is, to some extent, flawed as changes to treatment regimens and surveillance introduce time-dependent differences that are difficult to reconcile. We have previously shown that, in the English early access programme, there is no difference in the frequency of liver cancer in treated or untreated patients 2,15 , and here, we address the question of whether cancers in a 'virus free' environment are more aggressive than those in patients with persisting virus. Given the uncertainty about the delay from cancer initiation to presentation (it is unknown whether small, invisible, lesions are present for months or weeks prior to detection), we studied all cancers that developed in patients who did, or did not, respond to therapy as well as examining HCC developing 6 months after therapy. We chose 6 months as an arbitrary, convenient time period that was likely to exclude cancers present before treatment was initiated although we accept that other periods could have been selected. We found no difference in outcomes in either of the groups between HCC in infected or noninfected livers leading us to conclude that viral clearance does not alter cancer behaviour. We accept that the ideal study would have involved untreated patients with comparable degrees of cirrhosis, but we do not believe such a study to be ethical.
Our study is a nationwide prospectively collected real-world study of decompensated cirrhotic patients. The standard of data collection was high throughout the study and carried out to a clinical trial standard, although not formally audited. In our opinion, the results of this study are readily translatable to everyday patient care.
Although our study is one of the larger studies examining HCC in the post-DAA era, we nevertheless had only 80 HCC patients treated with DAAs. This may limit our ability to detect small yet significant differences in populations and is compounded by the relatively short period of follow-up. Another limitation of our study is the selection of controls which although frequency matched to remove bias for age, gender, stage of disease and length of follow-up, were not otherwise All (n = 72)

Nonviraemic (n = 47) Viraemic (n = 25)
Size of primary lesion, mm 9.5-120 9.5-120 14-100 More than 1 lesion, n (%) 20 (28)  TA B L E 3 Description of tumours split by viraemic and nonviraemic, excluding those found on explant matched. However, as liver function has the greatest impact on the development of hepatocellular carcinoma, we felt that these measures would be most sensitive for this. We removed all patients without data for SVR and this may have led to missing of ultra-aggressive cancers in the very early stages of follow-up. We chose to use the worst value for the blood tests in the year prior to treatment to provide an assessment of 'baseline, most severe' liver function. We accept that other approaches are possible but as liver function values are often modified by specific treatments (eg albumin infusions), we believe that it is most appropriate to use the worst value within a reasonable time period to avoid potentially artificially adjusted values. As this is a realworld observational study, some data were unavailable due to patient engagement or ability to gain this from the records; nevertheless the clear outcomes from the majority patients where data were available provide us with confidence that the conclusions are robust. Finally, the question of whether the presence of HCC hinders SVR is difficult to answer without a randomised controlled trial which would be unethical.
In conclusion, we have shown the presence of baseline nonmalignant lesions in addition to diabetes and a lower platelet count, to be indicative of HCC development. An absence of effect of DAA treatment on HCC progression as well as an absence of effect of viraemia on patient survival was evident.

ACK N OWLED G EM ENTS
Declaration of personal interests: The authors thank the patients, their families and all participating studies for contributing data to this study.