Summary of main results

In this systematic review we have summarised the available evidence from randomised clinical trials comparing peginterferon alpha‐2a versus peginterferon alpha‐2b, both given in combination with orally administered ribavirin administered in doses in accordance with the weight of the patient. Our results suggest that only one trial reported on mortality (McHutchison 2009); only one trial reported on liver‐related morbidity (Silva 2006); and no trial reported on liver‐related mortality. Therefore, our present knowledge regarding patient‐important outcomes is very sparse. Our results also suggest that the two peginterferons are comparable in regard to adverse events leading to treatment discontinuation and serious and non‐serious adverse events. However, evidence on adverse events is also sparse and the meta‐analysis on adverse events is likely to be underpowered to detect any significant differences. Likewise, evidence on quality of life was sparse. The combination of peginterferon alpha‐2a and weight‐based ribavirin may achieve a significantly higher sustained virological response than the combination of peginterferon alpha‐2b and weight‐based ribavirin.

Overall completeness and applicability of evidence

The GRADE (Guyatt 2008) summary of findings table reveals that, in general, we can have moderate confidence in the current evidence on treatment benefits measured as sustained virological response and we can only have low confidence in the current evidence on harms measured as all‐cause mortality, liver‐related morbidity, serious adverse events, and adverse events leading to treatment discontinuation (summary of findings Table for the main comparison). Information to assess the risk of bias was incomplete in a few trials with small numbers of participants. However, our sensitivity analyses did not show any important change in the intervention effects when we focused on trials with lower risk of bias. In our study, the trials that adequately reported on the trial methodology are large trials and they dominate the estimates of intervention effects from the meta‐analyses. Therefore, it is less likely that the pooled estimates are biased. In the meta‐analysis for sustained virological response, there were no serious inconsistencies across trials and the meta‐analysis crossed the Lan‐DeMets monitoring boundary (trial sequential monitoring boundary) leaving any random error less likely. The comparison of the largest trial (McHutchison 2009) with the second and third largest trials (Ascione 2010; Rumi 2010) yielded discrepancies however. The largest trial, which was funded by the manufacturer of peginterferon alpha‐2b, observed the smallest benefit of peginterferon alpha‐2a whereas the second and third largest trials both reported peginterferon alpha‐2a to be significantly superior to peginterferon alpha‐2b, and these trials were not funded by either of the two manufacturers. It is well known that industry bias may affect the outcomes and interpretation of trials (Als‐Nielsen 2003; Lexchin 2003; Lundh 2012).

Quality of the evidence

Because the meta‐analysis for sustained virological response included 12 trials, we drew a funnel plot to explore bias (Figure 6). There was no significant asymmetry. The meta‐analysis included a seemingly reasonable mix of small and large trials yielding fairly consistent results, which gives little concern about the presence of publication bias and other biases. When we included only trials without risk of bias due to other factors than lack of blinding, the results remained the same regarding sustained virological response. This effect did not pass the test of trial sequential analysis (Figure 11). However, we have some concerns in regard to indirectness. In the identified trials, virological response was the predominant measure of benefit. Many of the trials measured sustained virological response, which is currently the commonly used non‐validated surrogate outcome (Gluud 2007). Recent large cohort studies show a correlation between the presence of viraemia and mortality (Adeel 2009; Hirofumi 2009). However, it is important to remember that sustained virological response (and early virological response and end of treatment virological response) are still only non‐validated surrogate outcomes for antiviral intervention effects (Gluud 2007; Koretz 2013; Gurusamy 2014). We do not know the effects of the interventions on patient‐relevant outcomes (Gluud 2007). Because randomised clinical trials need to inform clinical practice, patient‐relevant outcomes such as risk of liver failure, hepatocellular carcinoma, and mortality would be of greater interest to patients and clinicians. Nevertheless, to be able to report on these outcomes, a much larger sample size and a follow‐up of several years would be required. Currently, no randomised clinical trials comparing the two peginterferons are of such a size or duration.

There were serious discrepancies across trials in the meta‐analysis on adverse events. The proportions of observed adverse events differed greatly across the trials, and the direction of effect was also heterogeneous. It is noteworthy that the IDEAL trial (McHutchison 2009) included three intervention groups, one for peginterferon alpha‐2a and two for peginterferon alpha‐2b. The two peginterferon alpha‐2b groups consisted of the usual 1.5 µg/kg/week dose and a low 1.0 µg/kg/week dose. The usual dose group yielded a similar proportion of adverse events as the peginterferon alpha‐2a group, whereas the low dose peginterferon alpha‐2b group yielded a lower proportion of the group with adverse events. Including or excluding the low dose peginterferon alpha‐2b group from the meta‐analysis had no visible impact on the estimated adverse events however. Furthermore, the meta‐analysis on adverse events leading to treatment discontinuation had low precision. The frequency of all other adverse events varied greatly among the included trials. In the majority of the included trials each patient had at least four to six adverse events during the treatment (Di Bisceglie 2007; Laguno 2009; Ascione 2010; Rumi 2010; Miyase 2012), but there are some trials where patients experienced no or only one adverse event (Scotto 2008; McHutchison 2009; Kamal 2011). Such a discrepancy in reporting adverse events imposes the necessity of uniform reporting of the adverse events in future trials. A post hoc calculation of the required information size to detect a minimally important difference of 10% relative risk reduction, based on the assumption of an average population risk of 10% and employing a 5% maximum type I error and 80% power, suggested that a minimum of 27,000 patients should be randomised to obtain a conclusive meta‐analysis on adverse events. The current number of patients in the meta‐analysis on adverse events is approximately 5000 (that is, less than 20% of what is required).

There have been some concerns regarding the non‐standardisation of the ribavirin dose across trials. The weight‐based dose of ribavirin ranged from 800 mg to 1400 mg. However, the weight cut‐off varied among trials as well as within the same trial. In the largest included trial (McHutchison 2009), patients weighing from 40 kg to 65 kg received a lower dose of ribavirin (800 mg) in the peginterferon alpha‐2b group compared with a higher dose of ribavirin (1000 mg) in the peginterferon alpha‐2a group. Patients in the peginterferon alpha‐2b group achieved a lower sustained virological response compared with patients in the peginterferon alpha‐2a group (39% versus 41%). Patients weighing more than 105 kg received a higher dose of ribavirin in the peginterferon alpha‐2b group (1400 mg) compared with a lower dose of ribavirin (1200 mg) in the peginterferon alpha‐2a group. Patients in the peginterferon alpha‐2b group achieved a slightly higher sustained virological response compared with patients in the peginterferon alpha‐2a group (43% versus 39%) (McHutchison 2009). It is also interesting that in the same trial, patients who developed anaemia and thus required ribavirin dose reduction achieved a higher sustained virological response than the patients who did not require a ribavirin dose reduction (McHutchison 2009). Accordingly, we do not think that the varying doses of ribavirin have any major confounding influence on our observations regarding the effects of the type of peginterferon. More research needs to be done to explore the optimal ribavirin dose. A ribavirin dose reduction due to adverse events was reported in only seven trials (Yenice 2006; Scotto 2008; McHutchison 2009; Ascione 2010; Rumi 2010; Kamal 2011; Miyase 2012). Six of these trials applied one and the same dose reduction to all trial groups (Yenice 2006; Scotto 2008; Ascione 2010; Rumi 2010; Kamal 2011; Miyase 2012). Only one trial applied a different ribavirin dose reduction to the intervention groups (McHutchison 2009). Our estimate did not change noticeably when we excluded the latter trial from our meta‐analysis for the outcome sustained virological response. 

Selective outcome reporting was difficult to assess in this review. Most of the included trials were not adequately registered or did not have their protocols publicly available prior to the trial completion. The risk of bias from selective reporting was considered low if the trial protocol was available and all of the pre‐specified outcomes that were of interest in the review were reported. It turned out that protocols were not available for all trials but two (McHutchison 2009; Kamal 2011). The outcomes reported in the protocols (sustained virological response and adverse events) matched the outcomes that were reported (McHutchison 2009; Kamal 2011). We also considered that low risk of reporting bias was present if the trial reported both the sustained virological response and adverse events. However, since the other primary outcomes that were of interest in this review related to morbidity and mortality and were rarely reported in any of the included trials, we could have chosen to assess all the trials as having high risk of reporting bias due to the lack of reporting of important, patient‐relevant outcomes. Hopefully, the initiation of the World Health Organization (WHO) International Clinical Trials Registry Platform coupled with timely and correct registrations of trials will facilitate such assessments for future trials (WHO 2009). Another limitation in this review was insufficient reporting of the included trials. Investigators of future trials are, therefore, well advised to adhere to the Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) guidelines (Chan 2013) and the Consolidated Standards for Reporting of Trials (CONSORT) (Moher 2012) in order to improve the quality of trial reports.

Potential biases in the review process

The strengths of this Cochrane Hepato‐Biliary Group systematic review are that it builds on a peer‐reviewed published protocol (Awad 2010), uses extensive searches until October 2013, considers the risks of systematic errors (bias, that is overestimation of benefits and underestimation of harms), and considers risks of random errors (play of chance) by adjusting the threshold for statistical significance according to the information and strength of evidence in the cumulative meta‐analysis.

A possible limitation is the unavailability of full reports of all the included trials. Two out of the 12 trials that provided data on sustained virological response for the meta‐analysis are only available as abstracts. However, we were able to successfully retrieve the necessary data for one of the two abstracts via e‐mail correspondence with the authors, and thus the risk of bias assessment of the included trial was carried out satisfactorily. Our sensitivity analyses did not show any important changes in our results whether including or excluding the trials. In our review, the trials that were published as a full paper are large and dominated the meta‐analysed estimates of effects. Moreover, empirical evidence suggests that trials that fail to refute the null hypothesis have lower odds of getting published, especially those not funded by industry (Als‐Nielsen 2003; Lee 2008; Hopewell 2009). Thus, many of the included abstracts may have a low probability of getting published as full articles. In fact, including these abstracts in our systematic review may likely be a strength rather than a limitation. By including abstracts we consider the complete available body of evidence. By excluding abstracts we would only obtain a subset that is defined through present‐day biased publication mechanisms. This would considerably increase the likelihood of publication bias. These potential limitations and concerns may lower our confidence in the estimates of intervention effect. However, in our meta‐analysis of sustained virological response there is no apparent heterogeneity (I² = 0%) and the direction of the treatment effect is the same across all included trials. Further research is unlikely to change our confidence in the estimate of the effect. It is a common misconception that large randomised trials are generally more reliable than meta‐analyses. The reason this misconception has prevailed is due to a number of highly cited papers that compared large trials with low risk of bias to meta‐analyses of small trials with high risk of bias. This is of course an unfair comparison. In empirical studies where large trials with low risk of bias are compared to a meta‐analysis of small trials with low risk of bias, the results from the two are typically non‐discrepant. In the case of the IDEAL trial (McHutchison 2009), the results still show an effect, albeit small, in favour of peginterferon alpha‐2a. There are many examples of large trials that, merely by the play of chance, underestimate (or overestimate) the true treatment effect.

Agreements and disagreements with other studies or reviews

There are several meta‐analyses that have been published recently which compare the efficacy and safety of the two pegylated interferons (Alavian 2010; Barros 2010c; Cheinquer 2010; Zhao 2010; Coppola 2011; Singal 2011; Druyits 2012; Romero‐Gomez 2012; Flori 2013; Yang 2013). Four of them have been published as abstracts only (Barros 2010c; Cheinquer 2010; Coppola 2011; Romero‐Gomez 2012) and, despite the different numbers of included studies, the common conclusion is that pegylated interferon alpha‐2a has advances over pegylated interferon alpha‐2b. Six meta‐analyses that were published as full papers (Alavian 2010; Zhao 2010; Singal 2011; Druyits 2012; Flori 2013; Yang 2013) also uniformly report the superior efficacy of pegylated interferon alpha‐2a over pegylated interferon alpha‐2b while the safety profile remains similar for both treatments. Two meta‐analyses are not comparable to ours because of the different search strategy and inclusion criteria in one (Druyits 2012) and because non‐randomised trials were included for assessment of benefits in the other (Flori 2013). The authors did not assess the risk of bias of the included trials (Druyits 2012). The other four meta‐analyses have similar search strategies and outcomes as ours but included both fewer trials and smaller number of patients (Alavian 2010; Zhao 2010; Singal 2011; Yang 2013). They excluded conference abstracts or limited the participants to only naive patients. The conclusions of those three meta‐analyses are in concordance with our conclusions on the higher efficacy of pegylated interferon alpha‐2a on sustained virological response and comparable safety profile with pegylated interferon alpha‐2b.

Current evidence suggests that peginterferon alpha‐2a is significantly better than peginterferon alpha‐2b regarding sustained virological response, which is the clearance of the virus from the blood. However, there is insufficient evidence to detect any differences regarding effects on clinical outcomes (liver complications and mortality) as well as harms (adverse events). Future trials need to study the correlation between sustained virological response and risk of cirrhosis, hepatocellular carcinoma, and mortality.