Treatment optimisation for hepatitis C in the era of combination direct-acting antiviral therapy: a systematic review and meta-analysis

Background: Prior to direct-acting antiviral (DAA) therapy, personalised medicine played an important role in the treatment of hepatitis C virus (HCV). Whilst simplified treatment strategies are central to treatment scale-up, some patients will benefit from treatment optimisation. This systematic review and meta-analysis explores treatment optimisation strategies in the DAA era. Methods: We systematically searched Medline, Embase, and Web of Science for studies that adopted a stratified or personalised strategy using a licensed combination DAA regimen, alone or with additional agents. We performed a thematic analysis to classify optimisation strategies and a meta-analysis of sustained virologic response rates (SVR), exploring heterogeneity with subgroup analyses and meta-regression. Results: We included 64 studies (9450 participants). Thematic analysis found evidence of three approaches: duration, combination, and/or dose optimisation. We separated strategies into those aiming to maintain SVR in the absence of predictors of failure, and those aiming to improve SVR in the presence of predictors of failure. Shortened duration regimens achieve pooled SVR rates of 94.2% (92.3-95.9%) for 8 weeks, 81.1% (75.1-86.6%) for 6 weeks, and 63.1% (39.9-83.7%) for ≤4 weeks. Personalised strategies (100% vs 87.6%; p<0.001) and therapy shortened according to ≥3 host/viral factors (92.9% vs 81.4% or 87.2% for 1 or 2 host/viral factors, respectively; p=0.008) offer higher SVR rates when shortening therapy. Hard-to-treat HCV genotype 3 patients suffer lower SVR rates despite treatment optimisation (92.6% vs 98.2%; p=0.001). Conclusions: Treatment optimisation for individuals with multiple predictors of treatment failure can offer high SVR rates. More evidence is needed to identify with confidence those individuals in whom SVR can be achieved with shortened duration treatment.

1,2 1 1 Introduction Viral hepatitis is a leading cause of mortality globally 1 and an estimated 71 million individuals are infected with hepatitis C virus (HCV) 2 . The recent transformation in HCV treatment has led to ambitious World Health Organization (WHO) targets for its elimination as a public health threat by 2030; 3 meeting those targets will require at least 80% of infected individuals accessing treatment and in richer countries the ambition will be to exceed this.
Simplified treatment strategies will be essential for rapid treatment scale-up 4 . However, some subgroups experience worse cure rates with standard therapies and may benefit from a stratified or personalised approach. Stratified medicine refers to the separation of patients into subgroups based on factors known to be associated with outcome. Personalised medicine goes further, individualising therapy according to the unique host and/or viral context. Both are used for treatment optimisation. Such approaches may become increasingly relevant for patients who struggle to engage with care, and in some settings may be more cost-effective 5 . The balance between simplified and personalised care has been well described for tuberculosis 6 . An understanding of the risks and benefits of a stratified or personalised approach for HCV is required to inform treatment in different settings.
Treatment strategies centred on direct acting antiviral (DAA) combinations are evolving rapidly, taking into account new knowledge of efficacy, safety, simplicity, and cost. Recent guidelines advocate treatment optimisation for certain subgroups 7,8 . Both host (liver disease stage, treatment history) and viral (HCV genotype, baseline viral load) factors inform treatment decisions, with the duration and/or regimen being adapted to optimise sustained virologic response rates (SVR).
We have carried out the first detailed review of treatment optimisation strategies in the DAA era. We used a mixed methods approach to synthesise data on such strategies for the management of HCV within a theoretical framework. We aimed to initiate a discussion on the role of treatment optimisation within an increasingly simplified treatment landscape.

Search strategy
We performed this review in accordance with PRISMA 2009 guidelines 9 . We searched Medline, Embase, and Web of Science using a free text strategy (Extended data, Supplementary Table 1 10 ) combining hepatitis C; host/viral factors associated with treatment outcome; and DAAs used in combination regimens. The search period spans January 2013 (year of first publication of phase 2 trial of sofosbuvir) and July 2019 (end of review period). We searched the National Institutes of Health clinical trials registry 11 and the European (EASL) or American (AASLD) International Liver Congress websites to identify ongoing trials. We reviewed reference lists to identify additional studies.

Study selection criteria
One reviewer (CRJ) screened abstracts against pre-determined inclusion/exclusion criteria. A second reviewer (BFF) independently screened a 20% sample of all records. We resolved discrepancies through discussion between reviewers (CRJ, BFF) and the senior author (GSC). We included phase 2-4 interventional and observational studies of adults (>18 years old) with chronic HCV (>6 months duration) where SVR 12 weeks post-therapy was the primary outcome and at least one treatment arm met our definition of a stratified or personalised treatment strategy. We only included studies that assessed a licensed combination DAA regimen, alone or in combination with other DAAs, interferon (IFN), or ribavirin (RBV). We excluded studies assessing one DAA in combination with IFN or RBV, or unlicensed DAAs (unless used alongside a licensed regimen).
Stratified or personalised treatment strategy definition Evidence of a priori stratification of patients into different treatment approaches based on predictors of response to therapy 12-14 . Strategies were considered stratified if patients were separated into broad subgroups according to prognostic factors, e.g. treatment history, liver disease stage, baseline HCV viral load. Strategies were considered personalised if therapy was individualised e.g. viral resistance testing, on treatment response kinetics, host genetics. Included strategies contained an adaptation to standard treatment, either defined in guidelines 7,8 or by an equivalent 12 week standard of care. A priori or post hoc stratification for analysis only is not included in this definition.

Data extraction
The first reviewer extracted data into a pre-designed Excel spreadsheet v16 (Microsoft Excel, RRID:SCR_016137). Where trials included multiple treatment arms, we only extracted data from arms that used a stratified or personalised strategy. We identified the following data a priori for extraction: year; author; journal; title; location; design; population; sample size; DAA regimen(s); treatment optimisation approach; host and/or viral optimisation factors; baseline demographics; resistance data at point of treatment failure; and SVR rates from intention-to-treat and per-protocol analyses.

Quality assessment
We assessed included studies for methodological quality using the Cochrane Risk of Bias tool for randomised trials 15 within Review Manager v5.3 (RevMan, RRID:SCR_003581), or the Newcastle Ottawa Scale for non-randomised studies 16 . Since most trials were non-comparative, we modified the Newcastle Ottawa Scale by removing the comparability domain. We only considered virologic outcome. Since SVR is a highly objective outcome measure, we did not consider lack of blinding a threat to validity.

Thematic analysis
To characterise treatment strategies and guide quantitative analyses, we performed a thematic analysis as previously described 17 .
We examined the methods of each study in detail using a deductive theory-driven approach, coding data related to treatment strategy and predictive factors. We analysed codes in order to identify themes. We then re-analysed the data using an inductive approach, reviewing and refining themes before generating a conceptual framework.

Quantitative analysis
We calculated pooled SVR rates using a DerSimonian and Laird random effects model 18 with the Freeman-Tukey double arcsine transformation to stabilise variances 19 . We weighted effect sizes using an inverse variance approach and calculated 95% confidence intervals using the Clopper-Pearson exact method 20 . We assessed statistical heterogeneity using Higgin's I 2 test (0-25%, 25-75%, and >75% representing low, moderate, and high degrees of inter-study heterogeneity, respectively) 21 .
We conducted pre-defined subgroup analyses to explore clinical variables associated with SVR (stage of liver disease, treatment history, genotype 3) or strategic variables that we hypothesised are associated with SVR in the context of treatment optimisation (duration, RBV use, number of host/viral treatment optimisation factors, pangenotypic vs genotype-specific regimen, stratified vs personalised approach). We performed a random effects metaregression to explore clinical (stage of liver disease, treatment history, genotype 3, mean age, male proportion, mean body mass index (BMI), baseline HCV RNA, IL28B CC proportion) and methodological (duration, RBV use, number of host/viral treatment optimisation factors, randomisation, sample size, trial design, pangenotypic vs genotype-specific regimen, stratified vs personalised approach) sources of heterogeneity. We constructed a multivariable model using variables identified as significant (p≤0.1) in the univariable analysis. For studies not reporting the mean of continuous variables, we estimated mean and variance using the median and range 22 . We did not formally assess publication bias since funnel plots are inaccurate in meta-analysis of proportion studies 23 . We performed all analyses in Stata v14.1 (Stata, RRID:SCR_012763) using the metaprop 24 and metareg 25 commands. We considered all p values as significant if p≤0.05, unless otherwise stated.
To explore the prevalence of resistance-associated substitutions (RASs) at the point of virologic failure following a shortened treatment duration, we performed a separate pooled analysis of virologic failure resistance data, where available. We considered resistance to all three classes of DAAs, and both treatment-enriched and treatment-emergent substitutions. We stratified data by treatment duration and performed the analysis using a Chi-square test for trend in Prism v7 (GraphPad Prism, RRID:SCR_002798).

Role of the funding source
There was no funding source for this study. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Results
Thematic analysis Treatment optimisation strategies were categorised into three approaches: i) duration optimisation -shorten or lengthen therapy, with or without a response-guided element; ii) combination optimisation -add or remove DAAs, RBV, or IFN, or select a regimen with an advantageous characteristic e.g. higher genetic resistance barrier; and iii) dose optimisation -adjust dose or dosing schedule, either by increasing frequency or intermittent dosing. Personalised treatment strategies draw on ≥1 of these approaches (Extended data, Supplementary Figure 1 10 ). Host factors used to optimise treatment were stage of liver disease, treatment history, and weight/BMI. We did not find any studies using sex, ethnicity, or IL28B genotype. Viral factors were HCV genotype, baseline viral load, resistance-associated substitutions (RAS), and viral kinetics.
The intended purpose of treatment optimisation allows distinction of two groups: i) those with factors known to be negative predictors of SVR, in whom treatment optimisation aims to improve SVR rates compared to those expected with standard care -'improve SVR' group; and ii) those without negative predictors, in whom the intention is to simplify/shorten therapy whilst achieving SVR rates non-inferior to those expected with standard care -'maintain SVR' group.

Summary of studies
We identified 133 studies for full text review and 64 studies for final inclusion ( Figure 1) 26-89 . We extracted data from 104 treatment arms including 9450 participants (Table 1). We identified six (9.4%) randomised controlled trials (RCT). Three (4.7%) RCTs were designed and powered to test an optimised treatment strategy, demonstrating non-inferiority to a standard regimen 35,57,88 . Most treatment arms were quasi-experimental (74 arms -71.2%) and included <50 participants (72 arms -69.2%). Adjustment to the duration and combination of therapy was the most common optimisation strategy (52 arms -50%), followed by duration (39 arms -37.5%). Only one (1%) treatment arm adjusted the dosing schedule 83 .
In total, 63 arms (60.6%) fall within the 'maintain SVR' group and 37 arms (35.6%) within the 'improve SVR' group. Four (3.8%) real-world observational studies personalised treatment on a case-by-case basis. Most arms used treatment history (82.7%) and/or stage of liver disease (77.9%) when optimising therapy. Eight arms (7.7%) personalised according to baseline resistance testing. Individual studies are summarised in the Extended data, Supplementary Table 2 10 . Treatment optimisation strategies within the 'maintain SVR' and 'improve SVR' groups used opposing strategies to study participants with different prognostic characteristics. Consequently, we anticipated a difference in treatment outcome and performed separate meta-analyses. A narrative summary of the four real-world studies that personalised therapy on a case-by-case basis is provided separately.
Subgroup analysis ( Improve SVR group meta-analysis Figure 3 displays pooled intention-to-treat SVR rates for strategies that aim to improve SVR in the presence of negative predictors. Pooled SVR for 12 weeks duration was 97.7% (94.9-99.5%;   Pooled analysis of resistance following treatment failure Although short duration regimens resulted in lower SVR rates, there was a significantly lower prevalence of RASs at the point of treatment failure (p=0.0004 for trend) (Table 4). However, data were only available for less than a third of treatment arms included in this study. A post hoc analysis stratified by RBV use did not find an association between RBV and RAS prevalence at treatment failure (data not shown).
Real-world studies reporting case-by-case personalisation Four real-world studies described personalised therapy on a case-by-case basis, with treatment left to the physicians' discretion. All personalised according to HCV treatment history, liver disease stage, and baseline RASs, with two also considering baseline HCV viral load. Rates of SVR were high using various combination DAA regimens (range, 94.7-100%).

Assessment of quality and bias
Assessment of randomised studies using the Cochrane Risk of Bias tool found that the majority of studies are at a low risk of bias (Extended data, Supplementary Figures 2 and 3 10 ). Assessment of non-randomised studies using a modified Newcastle Ottawa Scale found that the majority of studies were of high quality (Extended data, Supplementary Table 5 10 ). However, no trials included an active comparator group.

Discussion
The recent transformation in HCV treatment has led to increasingly widespread access to several IFN-free, DAA combinations. The WHO has set ambitious targets to eliminate HCV as a public health threat. Simplified universal treatments made possible by the new DAA therapies will be important in reaching such goals. However, a proportion of patients will not achieve  cure with standard regimens. Furthermore, as countries approach elimination targets, the remaining pool of patients are likely to be harder to reach and/or harder to treat. For some patients, treatment optimisation may lead to better outcomes. The ideal balance between these approaches will vary in different settings.
This is the first detailed review of stratified and personalised treatment strategies in the combination DAA era. We identified a range of strategies, based on host and viral factors, with which the duration, combination, and/or dose of therapy may be adjusted to optimise rates of SVR. Such strategies are appropriate for a small but important proportion of hard-to-treat patients. Specialist infrastructure and expertise are often required, therefore these strategies are likely to be most relevant in wellresourced settings.
Individuals with HCV genotype 1a and NS5A RASs at baseline may experience worse outcomes when treated with standard durations of elbasvir/grazoprevir 90 or sofosbuvir/ledipasvir 91 . The growing recognition of sub-genotypes that are relatively  resistant to standard therapies (genotype 4r) 92 presents a challenge to simplified treatment strategies. A better understanding of regional sub-genotype differences and their impact on cure rates is required. High SVR rates (94.7-100%) from four real-world studies identified in this review support the use of baseline resistance testing within resistance-optimised treatment strategies. However, given the costs, technologies, and specialist input required, strategies based on resistance testing are likely to be confined to high-income settings at present. Importantly, we provide evidence that the higher failure rates seen with short duration therapies are associated with a lower prevalence of resistance at the point of treatment failure relative to standard or extended duration therapies. Without individual participant data, we were unable to evaluate the influence of baseline RASs on short duration treatment therapies.
The recent European HCV treatment guidelines 7 emphasised RBV-free regimens. This avoids RBV-related toxicity and allows simplification of treatment algorithms. In general, evidence from this review does not support additional RBV when personalising treatment. An ongoing RCT is exploring biomarkerstratified short course (4-7 weeks) versus fixed duration (8 weeks) therapy and includes a factorial randomisation to RBV for each arm (see Extended data, Supplementary Table 6 for a summary of ongoing trials 10 ). This will hopefully address uncertainty regarding the utility of additional RBV for shortened duration therapy.
Despite stratified or personalised treatment strategies, we found significantly lower rates of SVR for individuals with genotype 3 HCV and poor prognostic factors (92.6% vs 98.2%; p=0.001).
In the ASTRAL-3 study 96 , 89% of treatment experienced, cirrhotic individuals with HCV genotype 3 achieved SVR following 12 weeks of sofosbuvir/velpatasvir. Higher SVR rates were reported with sofosbuvir/velpatasvir/voxilaprevir 31 and glecaprevir/pibrentasvir 86 in this difficult to treat group. Whilst simplified, decentralised care will be the cornerstone of treatment scale up, HCV genotype testing may yet be important in settings with a substantial genotype 3 prevalence. This will allow an appropriate regimen to be selected. In general, we find no evidence to support the extension of treatment beyond 12 weeks.
We identified one study that altered the dosing schedule. Uemura et al. 83 studied a 2-week lead-in period of daily IFN-beta injections before 12 weeks' sofosbuvir/ledipasvir in DAA-experienced individuals. Only six participants were recruited (SVR 66.7%). Interferon-related toxicity and the poor response makes this an unattractive strategy. However, long-acting injectable DAA preparations, if developed, may allow reduced dosing schedules for difficult to access individuals.
We did not find any studies that used host genetics (e.g. IFNlambda 4 (IFNL4)) to personalise treatment. There is evidence suggesting that an unfavourable IFNL4 genotype is associated with virologic relapse following short duration therapy with sofosbuvir/ledipasvir 97 and sofosbuvir/velpatasvir/voxilaprevir 98 . This may be of relevance during the planning and analysis of future short duration trials.
There are limitations to this study. Firstly, we have metaanalysed heterogeneous studies that explore different DAA regimens, populations, and methodologies. We attempted to address this using a random effects model, which accounts for some heterogeneity. However, pooled SVR rates should be interpreted with caution. Secondly, we used a broad definition for stratified and personalised approaches to treatment optimisation. Consequently, we included several phase 2 exploratory stud-ies that primarily aimed to establish optimum treatment duration. However, we feel that their inclusion provides interesting conceptual data. Thirdly, most data were derived from studies of sofosbuvir/ledipasvir. In the pangenotypic era the relative importance of sofosbuvir/ledipasvir may decrease. Fourthly, we have not included adverse events as an outcome for pooled analysis. Implementation of any optimised strategy will require confirmation of a similar or better side-effect profile compared to standard therapy. Finally, we have not considered costeffectiveness. The cost of standard regimens varies widely and can be a barrier to access. In some settings, personalised regimens may be more cost-effective 5 . However, the increased complexity associated with monitoring and infrastructure required for such strategies may present a pragmatic barrier.
Stratified and personalised treatment strategies have the potential to complement elimination efforts in some settings. Although existing standards of care for DAA therapy offer high SVR rates, there is evidence that treatment optimisation can improve outcomes in those with a higher predicted risk of failure. Whilst emerging data summarised in this review are encouraging, more evidence is needed to identify with confidence those individuals in whom SVR can be maintained whilst shortening treatment.

Data availability
Underlying data All data underlying the results are available as part of the article and no additional source data are required.

Anne Lindebo Holm Øvrehus
Department of Infectious Diseases, Odense University Hospital, Odense, Denmark This paper presents a systematic review and meta-analysis on available data on chronic hepatitis treatment with direct acting antivirals from trials aiming at either shortening treatment "optimization" or enhancing response "maintaining" in difficult patients. The review thus includes both industry initiated phase 2 and 3 trials and investigator initiated more experimental trials. The authors have made an impressive effort and an important contribution in setting up a conceptual framework for the analysis and communication of studies on treatment optimization in Chronic Hepatitis C in the DAA era. The decision to make separate analysis for studies aiming at maintaining SVR in easy to treat and optimizing SVR in hard to treat patients is just. It might however have been a better strategy to make separate papers to overcome the large conceptual span of these somewhat opposing aims. That said the benefit of being able to compare RAS between the two strategies justifies this approach. The paper comprises a wealth of information and although the method is mainly that of a systemic review, the discussion is more based on the narrative or concept.

Aim
The stated aim is to initiate a discussion on the role of treatment optimization and in this the authors have succeeded.

Method
Well described and choices made in the process are supported. Please define "enriched" RAS (does it mean RAS present at baseline but sequence population proportion had increased at failure?). Table 1 Given that data are analyzed separately for the "Maintain" and "Optimize" studies (and that DAA used in the two groups are quite different) pooling them into one table is a little misleading. Table 4 Important point -Please consider if presenting data on treatment emergent RAS and "enriched" RAS should be separate.

Results
should be separate. Treatment emergent RAS is a major concern in treatment failure patients and it would be interesting to see these data.

Discussion
Opens with a focus on hard to treat patients and optimizing SVR rates. As the results sections starts with the maintain SVR in absence of negative predictors approach it is a little confusing.
That the majority of studies were conducted using LDV/SOF is rightly stated as a limitation, as is that safety was not included as a factor.
The important finding that short therapy at 4 weeks does not lead to development of RAS is an important finding in a public health perspective (fear of initiating therapy in vulnerable populations that might discontinue early) but the actual results are not mentioned.
Genotype 4r is singularly mentioned as genotype relatively resistant to standard DAA -there are many others -so maybe better to leave out?
The discussion does not explicitly include consideration on the target population demographic characteristics for treatment optimization. Would in general be considered to be populations where adhering to an extended course of treatment can be challenging -like in very active PWID or limited resource settings?

Is the statistical analysis and its interpretation appropriate? Yes
Are the conclusions drawn adequately supported by the results presented in the review? Yes

Kali Zhou
Department of Medicine, Division of GI/Liver, University of California, San Francisco, CA, USA This is a well-written and comprehensive systematic review and meta-analysis (with meta-regression) on treatment optimization for hepatitis C using direct-acting agents. This is an interesting topic and relevant as we push the boundaries of simplifying and improving HCV regimens for patients to ultimately decrease costs, improve adherence, and make elimination a reality. The majority of the significant findings are not surprising and have mostly been adopted as SOC. I commend the authors for completing this review.
The methodology was fairly complex, with several layers, but overall well-defined. The first is separation of predictor (optimization strategy) to stratified strategy (defined as subgroups related to prognostic factors such as severity of liver disease and prior treatment history) and personalized strategy (individual host and viral factors such as viral resistance). There were also two goals of treatment optimization that were separated in analysis: 1) improve SVR in those with negative prognostic factors and 2) maintain current SVR rates with easier or shorter regimens in those without negative prognostic factors. Each optimization strategy could have been done by changing duration, drug or both. Primary outcome of meta-analysis was % SVR. The purpose of meta-regression was to identify clinical and methodological contributors to heterogeneity in findings.
The authors were thorough in their review methodology and adhered to PRISMA guidelines. Selection of studies was performed by one reviewer while a 2 reviewer screened 20% of studies, reasonable given their initial search identified >8000 articles. However, only one reviewer performed data extraction and having a minimum of 2 is standard.
Reasons for exclusion in Figure 1 were reasonable. It is surprising that no additional records were identified (perhaps in reference lists?) although the broad initial search likely limited missed publications. I would consider excluding studies using IFN in meta-analysis due to anticipated heterogeneity (Uemura).
Bias was assessed with separate tools for RCTs and observational studies. Publication bias was not assessed although the authors could have considered Peter's test, which can be used with proportions.
The major finding of the study is that shortened duration of therapy (8 weeks) is an acceptable strategy to maintain SVR in those with no negative prognostic factors with pooled SVR of 94%, but SVR drops with 6 and 4 week regimens. The authors provide subgroup analysis by addition of RBV, number of personalization factors, strategy, regimen, and GT -however, what specific personalization factors were considered is not clearly evident for these patients with no negative prognostic factors (presumably, stage of disease and prior treatment history are not applicable).
For improving SVR in those with negative prognostic factors, duration did not significantly impact SVR (thus 12 as good as 24). In subgroup analysis, GT3 patients (not unexpected) had lower SVR rates. I would be interested in differences between arms with SVR<95 and >95% using duration of 12 weeks and hypothesized reasons for lower SVR reports in these arms of smaller sample size. Also, did SVR rates for 12-week regimens differ by negative prognostic factor (not so overall but not evaluated by duration). At present time, however, since 12-week regimens are standard of care for negative prognostic groups, this nd