Ribavirin for treating Crimean Congo haemorrhagic fever.

BACKGROUND: Crimean Congo haemorrhagic fever (CCHF) is a tick-borne disease that occurs in parts of Asia, Europe and Africa. Since 2000 the infection has caused epidemics in Turkey, Iran, Russia, Uganda and Pakistan. Good-quality general supportive medical care helps reduce mortality. There is uncertainty and controversy about treating CCHF with the antiviral drug ribavirin.
OBJECTIVES: To assess the effects of ribavirin for treating people with Crimean Congo haemorrhagic fever. SEARCH
METHODS: We searched the Cochrane Infectious Diseases Group Specialized Register; the Central Register of Controlled Trials (CENTRAL); MEDLINE (PubMed); Embase (OVID); Science Citation Index-Expanded, Social Sciences Citation index, conference proceedings (Web of Science); and CINAHL (EBSCOHost). We also searched the WHO International Clinical Trials Registry Platform (ICTRP) and ClinicalTrials.gov for trials in progress. We conducted all searches up to 16 October 2017. We also contacted experts in the field and obtained further studies from these sources. SELECTION CRITERIA: We evaluated studies assessing the use of ribavirin in people with suspected or confirmed Crimean Congo haemorrhagic fever. We included randomised control trials (RCTs); non-randomised studies (NRSs) that included more than 10 participants designed as cohort studies with comparators; and case-control studies. DATA COLLECTION AND ANALYSIS: Two review authors assessed eligibility, risk of bias, and extracted data. For non-randomized studies we used the ROBINS-I tool to assess risk of bias. The main effects analysis included all studies where we judged the risk of bias to be low, moderate or high. We summarized dichotomous outcomes using risk ratios (RRs) and continuous outcomes using mean differences (MDs), and used meta-analyses where appropriate. We carried out a subsidiary appraisal and analysis of studies with critical risk of bias for the primary outcome, as these are often cited to support using ribavirin. MAIN
RESULTS: For the main effects analysis, five studies met our inclusion criteria: one RCT with 136 participants and four non-randomized studies with 612 participants. We excluded 18 non-randomized studies with critical risk of bias, where none had attempted to control for confounding.We do not know if ribavirin reduces mortality (1 RCT; RR 1.13, 95% confidence interval (CI) 0.29 to 4.32; 136 participants; very low-certainty evidence; 3 non-randomized studies; RR 0.72, 95% CI 0.41 to 1.28; 549 participants; very low-certainty evidence). We do not know if ribavirin reduces the length of stay in hospital (1 RCT: mean difference (MD) 0.70 days, 95% CI -0.39 to 1.79; 136 participants; and 1 non-randomized study: MD -0.80, 95% CI -2.70 to 1.10; 50 participants; very low-certainty evidence). We do not know if it reduces the risk of patients needing platelet transfusions (1 RCT: RR 1.23, 95% CI 0.77 to 1.96; 136 participants; very low-certainty evidence). For adverse effects (including haemolytic anaemia and a need to discontinue treatment), we do not know whether there is an increased risk with ribavirin in people with CCHF as data are insufficient.We do not know if adding ribavirin to early supportive care improves outcomes. One non-randomized study assessed mortality in people receiving ribavirin and supportive care within four days or less from symptom onset compared to after four days since symptom onset: mortality was lower in the group receiving early supportive care and ribavirin, but it is not possible to distinguish between the effects of ribavirin and early supportive medical care alone.In the subsidiary analysis, 18 studies compared people receiving ribavirin with those not receiving ribavirin. All had a critical risk of bias due to confounding, reflected in the mortality point estimates favouring ribavirin. AUTHORS'
CONCLUSIONS: We do not know if ribavirin is effective for treating Crimean Congo haemorrhagic fever. Non-randomized studies are often cited as evidence of an effect, but the risk of bias in these studies is high.

Summary of findings

Background

Description of the condition

Crimean Congo haemorrhagic fever (CCHF) is a tick‐borne viral disease. The virus that causes CCHF is a Nairovirus, a member of the Bunyaviridiae family. The most common vectors of the disease are Hyalomma ticks, which spread the disease and also act as a disease reservoir. CCHF is found in Africa, Eastern Europe, the Middle East, and Asia, with further occasional cases in other European countries such as Spain and Greece (Hoogstraal 1979; Zapata 2014; García Rada 2016). In recent years, CCHF incidence has been increasing in several areas worldwide (Zapata 2014).

The disease starts with a headache, fever, abdominal pain, musculoskeletal pain, and nausea. Over the next few days this is followed by gastro‐intestinal symptoms, including vomiting, diarrhoea, and haemorrhagic rash. After three to five days, a few patients progress to severe microvascular instability and the haemorrhagic phase of the illness, which is usually manifested first by a petechial rash followed by ecchymosis and bleeding. As the disease progresses into the second week, bleeding can worsen and become more severe, resulting most commonly in haemorrhage under the skin and within the abdomen. Death rates in people infected can reach up to 50% (Hoogstraal 1979). In endemic areas where high‐quality supportive care and access to diagnostics are offered, death rates can be as low as 5% (Leblebicioglu 2016b).

Many infections occur without symptoms and some estimates suggest this occurs in most infections (Bodur 2012). CCHF severity varies in people who are clinically unwell. Different scores to assess severity are used but it remains unclear what proportion of all infections are severe (Swanepoel 1987; Dokuzoguz 2013).

Infection in people is usually due to a tick bite or by contact with infected bodily fluids from humans or animals (Ergönül 2006b). Those at highest risk of contracting the virus are people who work outdoors in CCHF‐endemic areas, those who work with large domestic animals, and healthcare workers (Whitehouse 2004). CCHF has been linked to reservoirs such as sheep, goats, hedgehogs, and hares (Causey 1970; Saluzzo 1985; Yen 1985; Shepherd 1987). Human‐to‐human transmission occurs within families and in healthcare settings, including nosocomial outbreaks. The greatest risk of nosocomial exposure is from splash exposures and needle stick injuries (Conger 2015; Leblebicioglu 2016a). Case series studies also suggest that in rare cases airborne transmission from ventilated patients may also occur (Pshenichnaya 2015). Case reports suggest possible sexual transmission, although there is no published evidence of the virus being present in seminal or vaginal fluid (Pshenichnaya 2016). The virus is also transmitted from person to person by infected bodily fluids, and is highly infectious.

The disease may become more important in future years because of changes to the habitat of the Hyalomma tick vector, which is due in part to changes in the rural landscape from large diffuse habitats to smaller habitats. This is shown to lead to densely‐populated habitats for the tick vector, which is associated with increasing incidence of the disease (Estrada‐Peña 2007). Given the high mortality of patients, the lack of a widely‐available viable vaccine (Dowall 2016), and an emerging pattern of spread with multiple countries reporting re‐emergence of epidemics or new cases (Messina 2015), CCHF should be considered a potential threat to public health.

Description of the intervention

Supportive medical care underpins CCHF treatment, and use of fluids, good nursing care, and blood products in response to changes in the blood's ability to clot are key components (Leblecioglu 2012). Previous attempts at therapeutic regimens have explored intravenous (IV) immunoglobulin (Ig) isolated from horses (Hoogstraal 1979), and from recovered patients (Vassilenko 1990), but these are not currently widely used.

Ribavirin is commonly used with interferon to treat people who have hepatitis C, and is used alone in treating people who have Lassa fever (Debing 2013). Ribavirin is also used in healthcare settings as a form of post‐exposure prophylaxis for those exposed to CCHF (Leblebicioglu 2016a). Non‐randomized studies show that it could be effective in treating cases of CCHF (Fisher‐Hoch 1995; Mardani 2003; Dokuzoguz 2013), although this has been debated (Kalin 2014; Leblebicioglu 2016a). One such idea is that administration of ribavirin early in the disease, when it appears to be at its most effective, may be a promising approach (Dokuzoguz 2013; Ozbey 2014). This fits with the known course of the disease, where the virus is most commonly only present in the blood within the first week following onset of CCHF symptoms (Bente 2013).

Ribavirin has adverse effects, and, as well as the questions about its efficacy, clinicians debate whether or not to risk using the drug (Ceylan 2013; Oflaz 2015). Some of the adverse effects include risks of haemolysis, arrhythmia, bone marrow suppression, and deranged liver function (EMA 2015). Two previous systematic reviews have shown no clear benefit of ribavirin in people with CCHF, although the available evidence is limited mainly to confounded non‐randomized data (Soares‐Weiser 2010; Ascioglu 2011).

No alternative therapy has been proposed as the mainstay of therapeutic treatment. Although newer drugs, such as favipiravir, have shown promise in vitro (Oestereich 2014), widespread adoption of new therapies is years away. Current treatment guidelines and case definitions vary from region to region and from country to country (DoH South Africa 2014; Kalin 2014).

How the intervention might work

Ribavirin is a synthetic nucleoside that is active against a broad spectrum of DNA and RNA viruses (Sidwell 1972). It is one of few drugs shown to be active against CCHF in vitro (Watts 1989). Ribavirin can be given in hospital settings either intravenously or orally, according to World Health Organization (WHO) recommendations (WHO 2015). National guidelines from countries such as South Africa (DoH South Africa 2014), India (NCDC 2011), and Pakistan (NIH 2013) recommend prompt treatment with ribavirin following diagnosis of CCHF. However, these recommendations for management are not based on a robust evidence base (Soares‐Weiser 2010). CCHF can be mild or more severe, and it is often not deemed necessary to treat mild cases of the disease (Ergönül 2004). Questions remain about the overall benefits of ribavirin, how long after the onset of symptoms it is most effective, and whether it is more or less effective in severe cases (Ergönül 2006b; Dokuzoguz 2013; Leblebicioglu 2016b).

Why it is important to do this review

The controversy surrounding ribavirin use and the benefits of a widely‐available treatment for CCHF mean an up‐to‐date review of the existing evidence is required. There are mixed views on whether to treat CCHF with ribavirin, given the uncertainty about the balance between potential but unproven benefit and known risks of the drug (Kalin 2014). It is therefore important to use the data available to address whether ribavirin reduces the number of deaths from a lethal disease, whilst assessing the possibility of harm from serious, life‐threatening adverse effects.

Objectives

To assess the effects of ribavirin for treating people with Crimean Congo haemorrhagic fever (CCHF).

Methods

Criteria for considering studies for this review

Types of studies

We included the following types of study:

Randomized controlled trials (RCTs), quasi‐RCTs, and non‐randomized controlled studies of ribavirin compared to no ribavirin; also studies comparing early versus late administration of ribavirin

Cohort studies with ribavirin compared to no ribavirin, and studies comparing early versus late administration of ribavirin (prospective and retrospective, with more than 10 participants).

Case‐control studies with ribavirin compared to no ribavirin, and studies comparing early versus late administration of ribavirin

Types of participants

Children or adults of any age with CCHF confirmed with a laboratory test (immunoglobulin (Ig) or polymerase chain reaction (PCR))

Types of interventions

Intervention

Ribavirin (intravenous (IV) or oral)

Early ribavirin (as defined in identified studies)

Control

Supportive care only

Late ribavirin (as defined by study authors)

We accepted co‐interventions as long as the indication for the co‐intervention was consistent between groups, for example, administration of platelets according to homeostatic need.

Types of outcome measures

Primary outcomes

Death (in hospital or 28 days post‐admission)

Secondary outcomes

Length of hospital stay (days)

Requirement for transfusion (any blood products, including platelets, fresh frozen plasma, packed red cells, or whole blood)

Withdrawal of treatment due to serious adverse events

Serious adverse events, as defined according to the accepted US Food and Drug Administration (FDA) definition of: "if in the view of the investigator or sponsor, the event results in any of the following outcomes: death, life threatening adverse event, inpatient hospitalizations, prolongation of existing hospitalisation, disability or permanent damage, congenital abnormality, required intervention to prevent permanent impairment or other serious medical events" (FDA 2016).

Search methods for identification of studies

We tried to identify all relevant trials, regardless of language or publication status (published, unpublished, in press, and in progress).

Electronic searches

We searched the following databases using the search terms and strategy described in Appendix 1: the Cochrane Infectious Diseases Group Specialized Register (16 October 2017); the Cochrane Central Register of Controlled Trials (CENTRAL, Issue 9 of 12, September 2017), published in the Cochrane Library; MEDLINE (PubMed, 1966 to 16 October 2017); Embase (OVID, 1947 to 16 October 2017); Science Citation Index‐Expanded, Social Sciences Citation index, conference proceedings (Web of Science, 1900 to 16 October 2017); and CINAHL (EBSCOHost (1982 to 16 October 2017). We also searched the WHO International Clinical Trials Registry Platform (ICTRP; www.who.int/ictrp/en/) and ClinicalTrials.gov (clinicaltrials.gov/ct2/home) for trials in progress, up to 16 October 2017, using "ribavirin" and "Crimean Congo haemorrhagic fever" or "CCHF" as search terms.

Searching other resources

We searched the reference lists of any relevant systematic reviews. We contacted researchers in the field, requested information about grey literature and ongoing studies from the WHO, and checked reference lists of included studies.

Data collection and analysis

Selection of studies

Two review authors independently screened all citations and abstracts identified in the search according to predefined inclusion criteria. We obtained the full‐text reports of all potentially eligible studies or studies we were unclear about. Two review authors independently screened these full‐text articles, resolving any disagreements through discussion and if necessary consulting a third review author. We listed all studies excluded after full‐text assessment and their reasons for exclusion in the ‘Characteristics of excluded studies' table.

We included all unique studies in analyses; however, if there were any studies at critical risk of bias we excluded them from the main effects analyses. We included studies at critical risk of bias in a subsidiary descriptive analysis, using non‐overlapping samples as described in Appendix 2.

Data extraction and management

One review author extracted data using pre tested data extraction forms. A second review author cross‐checked the extracted data. We resolved any disagreements about data extraction by referring to the study report and through discussion. We attempted to contact the study authors where data were insufficient or missing.

We extracted data using a tool tailored for the inclusion criteria described above, including the following information:

Dose and method of administration (oral or IV)

Adult or child populations

Location

Setting

Design

Study size

Dates

Death

Length of hospital stay (days)

Transfusion of blood products

Serious adverse events

Time since onset of symptoms (days) to treatment with ribavirin or supportive care only

Assessment of risk of bias in included studies

Two review authors assessed the risks of bias of each included study, resolving any disagreements through discussion and consulting a third review author if necessary. For RCTs or quasi‐RCTs, we used the Cochrane ‘Risk of bias' tool for RCTs (Higgins 2011). For non‐randomized studies, we used the ROBINS‐I tool (Risk Of Bias In Non‐randomized Studies ‐ of Interventions) (Sterne 2016).

We assessed risks of bias through a hierarchy of domains, starting with critical then serious, moderate, and low. If any domain reached critical risk of bias we did not continue with the assessment, as further evaluation would not influence how we assess the certainty of the evidence.

As the risk of bias in the effect of an intervention may be different for different outcomes, we made a ‘Risk of bias' assessment for each outcome.

Our full methods for using ROBINS‐I are set out in Appendix 3. For assessment of confounding we considered length of time from onset of symptoms to receiving medical care or ribavirin, severity of disease, historical controls rather than contemporary controls, and quality of supportive care to be confounding domains. We made the decision to define these as confounding factors based on extensive debate in the literature (Ergonul 2009; Soares‐Weiser 2010; Kalin 2014), alongside consultation with clinicians with experience of treating viral haemorrhagic fever and CCHF. We listed co‐interventions that could differ between intervention groups impacting on outcomes as ‘quality of supportive care'.

The ROBINS‐I tool recommends only including non‐randomized studies that are not classified as having critical risk of bias. For our main effects analysis, we followed this approach. In addition, there was a further set of studies which met the inclusion criteria but which we classified as having critical risk of bias. As some of these studies have traditionally been used as part of the evidence base, we carried out a subsidiary descriptive analysis describing these studies and their estimates of effect. We established a non‐overlapping sample and performed meta‐analysis to describe the effect of confounding.

Measures of treatment effect

We analysed data using Review Manager 5 (RevMan 5) (RevMan 2014). For dichotomous outcomes, we presented analyses using risk ratios (RRs) with their 95% confidence intervals (CIs). For continuous data we used mean differences (MDs) with their 95% CIs.

Unit of analysis issues

We did not identify any studies that used a cluster‐randomized design or multiple interventions. For our subsidiary descriptive analysis of studies at critical risk of bias we established a non‐overlapping sample of studies using decision rules and methods set out in Appendix 2.

If we had identified studies of a cluster design we would have only used adjusted measures of effect. If the included study had not performed any adjustment for clustering, we would have adjusted the raw data ourselves, using an intracluster correlation coefficient (ICC).

We did not identify any studies with multiple intervention arms, but if we had we would have included data from these studies by either combining treatment arms, or by splitting the control group so that we only include participants once in the meta‐analysis.

Dealing with missing data

We attempted to contact the study authors to obtain missing data when the lack of reporting of necessary data restricted the use of the study.

We applied no imputation measures for missing data.

In one study there was an unclear amount of missing data from an analysis looking at the added benefit of corticosteroid use as well as ribavirin (Dokuzoguz 2013). This occurred because the number of participants included in the analysis did not tally with the explanation of how data were analysed in the text. These missing data may have affected the adjusted odds ratio (OR) presented in the study. We took this into account in the ‘Risk of bias' assessment, and it affected our decision not to present the adjusted estimate of effect; instead we presented a forest plot with the results stratified by severity of diease. As the missing data did not affect results relating to ribavirin and mortality, we did not classify the study as being at critical risk of bias.

Assessment of heterogeneity

We examined the included studies to determine whether there was heterogeneity in terms of co‐intervention, level of supportive care available, and risks of bias in the included studies.

We inspected forest plots to assess whether statistical heterogeneity was present. We deemed CIs that did not overlap as an indication of statistical heterogeneity. We also performed the Chi2 test using a cut‐off point of P < 0.10 to indicate statistical heterogeneity, and we used the I2 statistic to quantify the heterogeneity. We interpreted the I2 statistic value according to guidance in the Cochrane Handbook of Systematic Reviews of Interventions (Higgins 2011).

Assessment of reporting biases

If applicable, we intended to use funnel plot analysis or statistical tests (such as the Egger regression test), or both, to assess for publication bias. We planned to perform funnel plot analysis if there were more than 10 studies in any meta‐analysis. As there were fewer than 10 studies included in any of the effects analyses we did not perform this test.

Data synthesis

In order to deal with non‐standard study designs we have presented our main effects analysis by study design. We separated study designs into different subgroups and did not pool results across randomised and non‐randomized subgroups. This included RCTs, retrospective cohort studies, matched cohort studies and historically‐controlled cohort studies.

We stratified studies by their risk of bias in the descriptive analysis. We did not include studies which we assessed as having critical risk of bias in the main effects analysis. We performed a meta‐analysis of studies using the random‐effects model; this was due to varying study type, differences in populations and supportive therapy available.

We examined those studies classified as being at critical risk of bias in a subsidiary descriptive analysis that assessed the degree of confounding. We assembled a non‐overlapping sample, as set out in Appendix 2.

One study (Dokuzoguz 2013) used a model to adjust for confounding of effect due to severity of disease. This resulted in an adjusted OR of 0.04 (95% CI 0.004 to 0.48). The small sample size, the size of the adjusted effect, missing data from the corticosteroid analysis, concerns about residual confounding due to time from onset of symptoms to presentation to hospital/administration of ribavirin, and the fact that the study analysed severely‐ill patients with gastro‐intestinal haemorrhage "per protocol" and not by intention‐to‐treat, meant that we took a conservative approach to synthesis. We presented non‐adjusted data from the study in the main effects analysis, and the stratified data in a separate analysis (Analysis 1.2).

Analysis 1.2

Comparison 1 Ribavirin versus no ribavirin, Outcome 2 Mortality stratified by severity of disease (Dokuzoguz 2013).

We used the GRADE approach to assess the certainty of the evidence, and created ‘Summary of findings' tables and Evidence Profiles (GRADEpro 2015). Data from observational studies started as low quality, but we intended to upgrade this to moderate or high quality if the pooled estimates revealed a large effect size, negligible concerns about confounders, or a strong dose‐response gradient.

Subgroup analysis and investigation of heterogeneity

If unexplained heterogeneity occurred we intended to perform subgroup analyses of the results, to assess whether the effect of ribavirin was influenced by any of the following factors:

Severity of symptoms: severe, moderate, mild

Duration of treatment, presence of severe gastro‐intestinal symptoms, and route of administration

Age (children versus adults). Children are defined as under 16 years of age.

We did not conduct subgroup analyses, because there were insufficient data to apply the prespecified subgroups in the primary effects analysis.

Sensitivity analysis

If we had estimated an ICC to adjust the results from cluster trials, we would have performed sensitivity analyses to investigate the robustness of our findings. We had intended to perform a sensitivity analysis, to consider excluding studies that were at high risk of bias according to the Cochrane ‘Risk of bias' tool for RCTs (Higgins 2011), and serious risk of bias according to ROBINS‐I tool for observational studies (Sterne 2016).

We performed no sensitivity analyses because we identified no cluster‐controlled studies, and because all studies included in the main effects analysis were at high risk of bias according to the Cochrane tool or at serious risk of bias according to ROBINS‐I.

Results

Description of studies

Results of the search

See PRISMA flow diagram Figure 1.

Figure 1

Flow of studies

We identified 1161 records, plus a further three references through contacting experts in the field. From these we identified 611 unique references after removing duplicates. We considered 372 references to be irrelevant for our review, and we were unable to obtain the articles for 13 of the references. We considered 226 full‐text articles for inclusion, of which we excluded 203. Two did not report on CCHF, and two reported on outcomes not included in our review. Fifty‐one studies did not report on ribavirin for treating CCHF, nine had no comparator arm and 27 were single‐case reports. Fifty were cohort studies with fewer than 10 participants, 27 were narrative reviews and 35 studies were editorial letters or comments on other studies. Twenty‐three studies met our inclusion criteria and are included in the review. No prospectively‐registered ongoing studies met the inclusion criteria.

Included studies

See Characteristics of included studies.

We include 23 studies that tested the use of ribavirin in people with CCHF, with the outcomes of mortality, length of hospital stay, and requirement for transfusion.

Main effects analysis

For the main effects analysis we include five studies that were not at critical risk of bias; one RCT of 136 participants (Koksal 2010) and four non‐randomized studies of 612 participants (Elaldi 2009; Izadi 2009a; Bodur 2011; Dokuzoguz 2013).

Study design

Of the five studies included in our main effects analysis, one study was an RCT (Koksal 2010), one was a matched cohort study (Bodur 2011), one was a cohort with a historical control (Elaldi 2009), one was a mixed prospective and retrospective cohort study (Dokuzoguz 2013), and one was a retrospective cohort (Izadi 2009a).

The design of the studies included in our primary analysis and how controls were selected varied. Elaldi 2009 used historical controls from a period when ribavirin was unavailable. Dokuzoguz 2013 selected controls based on clinical criteria including time from onset of symptoms to diagnosis or contraindication to oral ribavirin. Izadi 2009a compared administration of ribavirin given early in the disease to late in the disease. Bodur 2011 used a retrospective design that matched 10 participants who received ribavirin to 40 controls that did not, using various clinical and physiological parameters.

Setting

Four out of the five studies included in the main effects analysis were conducted in Turkey, and one was conducted in Iran (Izadi 2009a).

Participants

Most participants described in Dokuzoguz 2013 were adults or adolescents, with the youngest participant aged 16 years. Izadi 2009a described an age range of 11 to 75, with a median age of 29.2 years. It was not possible to ascertain the exact numbers of adolescents, as they were not described in the included studies. All of the studies included in our main effects analysis included confirmed cases only, using either Ig enzyme‐linked immunosorbent assay (ELISA) or PCR to verify.

Intervention

Doses of ribavirin differed between studies. Elaldi 2009 described weight‐based prescribing (30 mg/kg initial loading dose; 15 mg/kg 4 times daily for 4 days; 7.5 mg/kg 3 times daily for 6 days). Dokuzoguz 2013 and Bodur 2011 described standard doses in adults (4 g daily for 4 days, followed by 2.4 g daily for 6 days). Doses were broadly the same and full details are described in the ‘Characteristics of included studies' table. All studies in this analysis administered oral ribavirin.

Comparators

None of the participants in the comparator groups received ribavirin for the comparison of ribavirin versus no ribavirin. One study (Izadi 2009a) was included in the main effects analysis that offered a comparison of early versus late ribavirin; all those in the comparator arm received ribavirin after four days.

Length of follow‐up

No studies specified a length of follow‐up. Instead, they relied upon discharge from hospital or clinical care as the sole measure of follow‐up time.

Subsidiary descriptive analysis

For the subsidiary descriptive analysis we included 18 studies rated at critical risk of bias. These studies are frequently cited as evidence of benefit, so we appraised them against the primary outcome of mortality.

The rationale for ‘Risk of bias' assessments is set out in Table 7. They all failed to control for confounding due to severity of disease, time from the onset of symptoms to receiving medical care, or all of these. This critically affected the reliability of data collected for these studies. ROBINS‐I recommends studies at critical risk of bias are excluded from the review.

Table 1

Table of studies at critical risk of bias: disease‐related outcomes

Studies at critical risk of bias outcomes: Death, timing of administration, length of stay in hospital, requirement for transfusion
Study	Bias due to Confounding	Comment
Alavi‐Nani 2006	Critical	Confounders not controlled for. No information reported on care received in hospital. Variation in disease severity between ribavirin and control groups not measured. No discussion of potential confounding by severity of disease in paper. No control for time from onset of symptoms to administration of ribavirin. Small size of control group suggests clinical contraindication to ribavirin, a factor in selection into control group (although this is not expressly commented on)
Belet 2014	Critical	Confounders not controlled for. Although criteria for administration of ribavirin reported, it is not clear whether recipients must fulfil all of these or only someParticipants receiving ribavirin were more severe at baseline. There is no adjustment for severity on admission, and length of time between symptom onset and admission/ribavirin treatment
Cevik 2008	Critical	Confounders not controlled for. Severe patients only included in case‐control study. No discussion of potential confounding in paper. Care provided during hospitalisation not described
Ergönül 2004	Critical	Confounders not controlled for. Severe patients only included in retrospective cohort. Baseline severity of disease not established. Classification of severe disease is at any time point for 22 participants. Time from onset of symptoms not controlled for. No method for dealing with potential confounders. Patients were given preparations of erythrocytes, fresh frozen plasma, and total blood, depending on their homeostatic state ‐ disentangling the effect of this supportive care from that of ribavirin is not considered. Oral ribavirin was given to severe CCHF patients
Ergonul 2006	Critical	Confounders not controlled for. Paper focuses on developing severity scoring system. Baseline characteristics not established between ribavirin and non‐ribavirin groups. Criteria for selection into control arm included clinical contraindication due to haematemesis. Time from onset of symptoms not controlled for. The authors developed specific criteria to identify severe cases
Ertugrul 2009	Critical	Confounders not controlled for. No methods for controlling potential confounders are discussed. Authors stated in Discussion that no information was available to them on severity of cases. No information reported on care received by participants
Ertem 2016	Critical	Confounders not controlled for. Controls for “time from onset of symptoms” for a comparison of early versus late ribavirin. However, not for the comparison of ribavirin versus no ribavirin. Rather than just comparing means, the authors should control for the confounders when comparing the groups. Mortality not reported
Gayretli Aydin 2015	Critical	Confounders not controlled for. Paper focuses on bradycardia in paediatric patientsNo discussion of potential confounding in paper and no controlling for confounding factors such as severity of illness or time from onset of symptoms to administration of ribavirin
Kalin 2014	Critical	Confounders not controlled for. Significant differences in baseline severity of disease and time from onset of symptoms. These confounders were measured but not controlled for by stratification or other method. Ribavirin group had more severe disease; confounding would reduce effect of ribavirin seen
Mardani 2003	Critical	Confounders not controlled for. Baseline characteristics not established between ribavirin and no‐ribavirin groups. No method for dealing with potential confounders. Significant differences in arms of study ‐ suggests heterogeneous samples with no controlling for severity of disease. Time from disease onset to presentation/treatment not assessed. Historical control arm used supportive treatment likely to have differed substantially between intervention and control arms
Metanat 2005	Critical	Confounders not controlled for. Conference abstract ‐ insufficient information reported by study authors about possible confounders such as severity of disease. No information provided on care received in hospital
Ozkurt 2006	Critical	Confounders not controlled for. No methods for controlling potential confounders are discussed. Timing of administration of ribavirin is documented but severity of infection is not considered. Baseline characteristics not established between ribavirin and no‐ribavirin groups
Sannikova 2009	Critical	Historical control group used. Study conducted from 1999‐2008, quality of supportive care likely to have changed significantly over this period of time. Control group originated during period before ribavirin was available. Substantial period of time from the start of follow up in historical control group to start of follow up in intervention group.
Sharifi‐Mood 2006	Critical	Confounders not controlled for. No information reported on care received in hospital. Baseline characteristics not established. Variation in disease severity expected, although influence of this across the two groups unclear. No method of controlling for confounding by severity of disease. No discussion of potential confounding in paper. Timing of administration investigated, raw data not presented.
Sharifi‐Mood 2013a	Critical	Confounders not controlled for. Time from onset of symptoms adjusted for by stratification into early/late ribavirin. Baseline characteristics and severity of disease not assessed, measured or controlled for. No information provided on care received in hospital
Tasdelen Fisgin 2009	Critical	Timing of administration of ribavirin controlled for by stratification. Baseline confounding due to severity of disease measured and not controlled for participants in no‐ribavirin group and late‐ribavirin group having more severe disease based on baseline biochemistry and haematology. At least one participant was included in no‐ribavirin group due to gastrointestinal haemorrhage and severe disease. Criteria for use of ribavirin changed during period and largely historical controls were used
Tezer 2016	Critical	Confounders not controlled for. No methods for controlling potential confounders such as severity of disease and time since onset of symptoms are discussed. Authors recognize highly‐confounded data as limitation of their study
Tulek 2012	Critical	Confounders not controlled for. Case‐control study with no information in abstract about how the controls were selected. Supportive care protocol was similar in both departments. No method for dealing with potential confounding by time since onset of symptoms. No matching for severity or time since onset of symptoms
Tuygun 2012	Critical	Confounders not controlled for. No method for dealing with potential confounding by time since onset of symptoms, baseline characteristics were not established, no method for controlling for severity. The patients were given erythrocyte suspension, thrombocyte suspension and/or fresh frozen plasma based on their haemostasis status, and other supportive care when necessary ‐ disentangling the effect of this care from that of ribavirin is not considered. Oral ribavirin was given to the patients who were evaluated as severe or had bleeding symptoms, or both.Entirely unclear how the authors selected the 50 participants included from 202 confirmed cases. Also at critical risk of bias on selection of participants into the study

All 18 studies at critical risk of bias were retrospective cohort studies. One study used a cohort of patients treated before the availability of ribavirin as a control arm. (Sannikova 2009).

In those studies included in the descriptive analysis 12 were conducted in Turkey, five were conducted in Iran and one study was set in Russia and was translated from Russian (Sannikova 2009). There were a number of studies that reported on populations that overlapped with each other. Our decisions on overlapping studies are outlined in Appendix 2.

In the subsidiary descriptive analysis for the comparison of ribavirin versus no ribavirin an additional 1214 participants in 10 studies at critical risk of bias were analysed.

In the subsidiary descriptive analysis for the comparison of early versus late ribavirin an additional 431 participants in 4 studies received either early or late ribavirin.

For the comparison of ribavirin versus no ribavirin doses were broadly the same; we give full details in the ‘Characteristics of included studies' table. Most studies in this analysis administered oral ribavirin.

For the comparison of early versus late ribavirin, participants received ribavirin according to the study author's definitions of early versus late. Studies used different cut‐off time points for the definition of early care with ribavirin, either less than three days since onset of symptoms (Sharifi‐Mood 2006; Sharifi‐Mood 2013a), less than four days since onset of symptoms (Izadi 2009a) or less than five days since the onset of symptoms (Metanat 2005; Tasdelen Fisgin 2009). 114 participants received ribavirin less than three days from the onset of symptoms, 97 received ribavirin after 3 days since the onset of symptoms. One hundred and thirty participants received ribavirin less than five days from onset of symptoms with 90 participants receiving ribavirin after this time point.

Of those studies covered by the descriptive analysis, four studies included children only (Sharifi‐Mood 2006; Tuygun 2012; Gayretli Aydin 2015; Tezer 2016). Three studies did not report the method used to confirm cases of CCHF (Metanat 2005; Tulek 2012; Sharifi‐Mood 2013a), all other studies used either Ig ELISA or PCR.

No studies specified a length of follow‐up. Instead they relied upon discharge from hospital or clinical care as the sole measure of follow‐up time.

Excluded studies

We excluded 203 studies at the full‐text screening stage because they did not study CCHF, did not relate to a relevant CCHF topic, had fewer than 10 participants, or they were narrative reviews or commentaries. See the Excluded studies tables.

Risk of bias in included studies

Main effects analysis

Randomized controlled trials

We identified one randomised control trial (Koksal 2010), which we assessed using the Cochrane ‘Risk of bias' tool for RCTs (Higgins 2011). Methods for random sequence generation and allocation concealment were unclear in the single RCT. The trial authors did not report methods for this in the text. We judged the methods for blinding of participants and outcome assessments to be unclear; we identified no missing data.

There was no protocol available to assess selective reporting. We judged mortality and length of hospital stay as unlikely to be subject to reporting bias (Figure 2).

Figure 2

‘Risk of bias' assessment for all included trials

Non‐randomized studies

We identified 22 non‐randomized studies, which we assessed using the ROBINS‐I tool. Of these we classified 18 studies as being at critical risk of bias, and four studies at serious risk of bias.

Confounding

We have presented comprehensive ‘Risk of bias' assessments for non‐randomized studies included in the main effects analysis in Table 8; Table 9; Table 10; Table 11.

Table 2

ROBINS‐I assessment: Bodur 2011

ROBINS‐I assessment
Reference: Bodur 2011
Risk of bias domain	Assessments by outcome	Comment	Conclusion
Bias due to confounding	Mortality, Length of hospital stay, transfusion, withdrawal of treatment: serious risk of bias due to baseline confounding	Matching controls were included in the study in order to increase the study's power. Baseline characteristics established and are similar between groups for severity of disease and time from onset of symptoms to admission.No significant differences in baseline laboratory findings. Differences occur between groups in rates of splenomegaly (1 case in each arm), petechiae, haematemesis (2/10 in ribavirin group, 3/40 in control), melena. However, limited information reported on how the controls or baseline characteristics were selected.Given the differences in clinical symptoms serious risk of bias was attributed	Serious
Bias in selection of participants into the study	Mortality, length of hospital stay, transfusion, withdrawal of treatment: serious risk of bias.Direction: would show increased effect of ribavirin	Selection into study did not appear to be related to intervention, outcome or any prognostic factor.Limited information reported on how the controls were selected.Controls selected “at random” that matched baseline characteristics	Moderate
Bias in classification of interventions	All outcomes	Interventions well defined	Low
Bias due to deviations from intended interventions	All outcomes	No information on deviation from intended intervention, as would be the case in usual practice	Low
Bias due to missing data	All outcomes: serious	All data appear to be reported	Low
Bias in selection of the reported result	All outcomes	No outcomes of interest to study authors are specified. No protocol available, no prespecified outcomes in Methods section	Serious

Table 3

ROBINS‐I assessment Dokuzoguz 2013

ROBINS‐I assessment
Reference:Dokuzoguz 2013
Domain	Assessments by outcome	Comment	Conclusion
Bias due to confounding	Mortality, length of hospital stay, transfusion, withdrawal of treatment: Serious risk of bias due to baseline confounding	Time from onset of symptoms not adequately controlled or adjusted for. All participants with time from onset of symptoms to diagnosis < 7 days received ribavirin unless contraindicatedBoth time from onset of symptoms and clinical contraindication are prognostic factors that predict whether the individual receives the intervention	Serious
Bias in selection of participants into the study	Mortality, length of hospital stay, transfusion, withdrawal of treatment: serious risk of bias	Control group selected by including patients with time from onset of symptoms to diagnosis > 7 days and clinical contraindication to ribavirinBoth time from onset of symptoms and clinical contraindication are prognostic factors that predict whether the individual receives the interventionAnalysis was performed per protocol (2 participants in control group were intended to be treated with ribavirin but due to gastrointestinal bleeding were unable to receive oral medication)	Serious
Bias in classification of interventions	All outcomes	Interventions well‐defined in Methods section	Low
Bias due to deviations from intended interventions	All outcomes	No deviation from intervention not expected in normal practice	Low
Bias due to missing data	All outcomes	Some missing outcome data not dealt with in text. This is related to numbers of participants receiving co‐administration of corticosteroids with ribavirin Unbalanced across groups	Serious
Bias in selection of the reported result	All outcomes	Analysis was performed per protocol. (2 participants in control group were intended to be treated with ribavirin but due to gastrointestinal bleeding were unable to receive oral medication). Effect of ribavirin as measured will be overestimated compared to intention‐to‐treat analysis	Serious

Table 4

ROBINS‐I assessment: Elaldi 2009

ROBINS‐I assessment
Reference: Elaldi 2009
Risk of bias domain	Assessments by outcome	Comment	Conclusion
Bias due to confounding	Mortality, length of hospital stay, transfusion, withdrawal of treatment: Serious risk of bias due to baseline confounding	Baseline characteristics established and are similar between groups for severity of disease and time from onset of symptoms to admissionDifferences occur between groups in rates of maculopapular rash, hepatomegaly and lactate dehydrogenase. None of these are markers of disease severity unless petechiae were misclassified as maculopapular rashUse of historical control arm at the onset of an epidemic establishes a difference in the quality of supportive care between groups. As the time elapsed was only one year we classified this as serious and not critical confoundingNo participants diagnosed received ribavirin in the historical control group All participants diagnosed received ribavirin in the intervention groupNo method to adjust for potential confounders reported	Serious
Bias in selection of participants into the study	Mortality, length of hospital stay, transfusion, withdrawal of treatment	Selection into study was not related to intervention, outcome or any prognostic factorHistorical control group may confound results as set out aboveIt appears that all potential participants for the specified study years have been included in the studies for the particular treatment groups	Serious
Bias in classification of interventions	All outcomes	Interventions well‐defined in Methods section	Low
Bias due to deviations from intended interventions	All outcomes	No information reported on adherence of participants to ribavirin treatment schedule. For supportive care: "Same proportions of patients received ES (12%) and FFP (39%) in treated and untreated groups. On the other hand, more patients in the treated group were infused with PS (52%) than those in the untreated group (42%)." No other information provided about co‐interventions	Low
Bias due to missing data	All outcomes: serious	All data appear to be reported	Low
Bias in selection of the reported result	All outcomes	Unclear selection criteria for establishing baseline similarities between groups. PT/APTT may be missing from baseline characteristicsReported results are in keeping with those specified in the study methods	Low

Table 5

ROBINS‐I assessment: Izadi 2009

ROBINS‐I assessment
Reference: Izadi 2009a
Risk of bias domain	Assessments by outcome	Comment	Conclusion
Bias due to confounding	Mortality, length of hospital stay, transfusion, withdrawal of treatment: serious risk of bias due to baseline confounding	Time from onset of symptoms adjusted for by stratification into early/late ribavirin Baseline characteristics and severity of disease are not assessed, measured or controlled forMultiple regression models used to identify factors predictive of mortality. Regression does not adjust for severity and prognostic factors for the efficacy of ribavirinAdjusted estimates of effect not included in analysis due to linear regression not outlined clearly, although it is unlikely to control for confounding of the effect of ribavirin	Serious
Bias in selection of participants into the study	Mortality, length of hospital stay, transfusion, withdrawal of treatment: low risk of bias	Unclear if selection into the study was based on participant's characteristics observed after the start of the study; retrospective design	Moderate
Bias in classification of interventions	All outcomes	Not recorded	Moderate
Bias due to deviations from intended interventions	All outcomes	No deviation from intended intervention, as would be the case in usual practice. Most participants received a transfusion ‐ Table 3 shows the proportions of participants who received a transfusion of platelet concentrates, and in some cases fresh frozen plasma and packed erythrocytes. No other aspects of care or co‐interventions are discussed No information on adhering to ribavirin treatment	Low
Bias due to missing data	All outcomes: serious	Outcome data (mortality or cured) reported for all 63 participants according to treatment group	Low
Bias in selection of the reported result	All outcomes	Although no protocol available or prespecified outcomes, the authors state that they attempted to assess the effect of ribavirin in reducing mortality	Moderate

One study (Bodur 2011) controlled for confounding by matching baseline characteristics of 10 cases with 40 controls (Table 8). One mixed retrospective/prospective cohort study (Dokuzoguz 2013) established a severity scoring index and stratified results using this as a way of controlling for confounding by severity. Time from onset of symptoms to diagnosis was addressed by not prescribing ribavirin to anyone with more than seven days history of symptoms (Table 9). One historically‐controlled cohort study (Elaldi 2009) established similar baseline characteristics between cohorts. The use of a historical control arm at the onset of an epidemic establishes a difference in the quality of supportive care between groups. The time period elapsed was only one year, so we classified this as serious and not critical confounding (Table 10). One retrospective cohort study (Izadi 2009a) stratified mortality outcome by time from onset of symptoms to administration of ribavirin (Table 11). This study performed a regression analysis, although this was designed to identify predictive factors for mortality and not to control for confounding. We classified all four non‐randomized studies included in the main effects analysis as being at serious risk of bias for the domain of confounding.

Bias in selection of participants into the study

In Bodur 2011 (Table 8), participants were matched by their baseline severity according to clinical presentation and laboratory values. The trial authors did not adequately describe the matching process, although stated the controls were selected "at random". The lack of clarity meant that we classified this study as being at serious risk of bias in this domain.

In Dokuzoguz 2013 (Table 9), participants in the control group were selected based on time from onset of symptoms (more than seven days) and clinical contraindication. Both are prognostic factors that predict whether the individual receives the intervention. We therefore judged this as being at serious risk of bias for this domain. We did not judge this domain as critical, because baseline severity was established, measured, and sufficiently comparable to garner useful data from the study.

In Elaldi 2009 (Table 10), selection of participants was not related to the intervention, outcome, or any prognostic factor; the historical control group may have confounded results. We therefore judged this domain to be at serious risk of bias. We did not judge this domain as critical, because baseline severity was established, measured, and sufficiently comparable to garner useful data from the study.

In Izadi 2009a (Table 11), it was unclear if selection into the study was based on participants' characteristics observed after the start of the retrospective study design.

Bias in classification of interventions

We judged all studies to be at low risk of bias, as the doses and methods of administration of ribavirin were well‐defined.

Bias due to deviations from intended interventions

We judged all studies to be at low risk of bias. In none of the studies were there deviations from the intended intervention other than what would be expected in normal practice.

Bias due to missing data

In Dokuzoguz 2013 there were some missing outcome data not described in the text. The missing data were related to numbers of participants receiving co‐administration of corticosteroids with ribavirin. We judged the balance of missing data across groups as unclear, as there was insufficient documentation explaining this. We therefore classed this study as being at serious risk of bias in this domain.

Bias in measurement of outcomes

We judged all studies to be at low risk of bias. Whilst investigators will have been aware of the intervention status of the participants (if they received ribavirin or not), none of the measured outcomes were subjective and thus prone to bias.

Bias in the selection of reported result

Bodur 2011 and Elaldi 2009 used unclear criteria to establish similar baseline characteristics between arms. Most of the expected clinical and laboratory criteria were included, but a severity score would be more comprehensive. Whilst severity indices have been developed since the publication of these papers, an accepted severity index was available at the time (Swanepoel 1987). We therefore classed both studies as being at moderate risk of bias in this domain.

Subsidiary descriptive analysis

We classified all 18 studies included in the descriptive analysis as being at critical risk of bias due to confounding, as described in Table 7. All of these studies were retrospective cohorts by design. The main reason for this was the failure to control for baseline confounding due to severity of disease. Most studies did not describe important baseline characteristics in intervention and control groups.

Effects of interventions

See: Table 1; Table 2

Summary of findings for the main comparison

Ribavirin versus no ribavirin for Crimean Congo haemorrhagic fever

Ribavirin compared to no ribavirin for Crimean Congo haemorrhagic fever
Patient or population: people diagnosed with suspected or confirmed Crimean Congo haemorrhagic fever Setting: global Intervention: ribavirin Comparison: no ribavirin
Outcomes	Anticipated absolute effects* (95% CI)		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with no ribavirin	Risk with ribavirin
Mortality	56 per 1000	63 per 1000 (16 to 240)	RR 1.13 (0.29 to 4.32)	136 (1 RCT)1	⊕⊝⊝⊝ VERY LOW2,3	‐
Length of hospital stay (days)	The mean length of hospital stay in 1 RCT was 0.7 days longer in the experimental group (0.39 days fewer to 1.79 days longer)		‐	136 (1 RCT)4	⊕⊝⊝⊝ VERY LOW2,3	‐
Requirement for transfusion (platelets)	306 per 1000	376 per 1000 (235 to 599)	RR 1.23 (0.77 to 1.96)	136 (1 RCT)	⊕⊝⊝⊝ VERY LOW2,3	‐
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). Abbreviations: CI: confidence interval; RCT: randomised controlled trial; RR: risk ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect.

1In addition there were three non‐randomized studies (mixed retrospective and prospective cohort; single arm cohort with historical control; matched case series) with serious risk of bias (ROBINS‐I), providing an estimate of RR 0.59, 95% CI 0.27 to 1.27; 549 participants; very low‐certainty evidence). 2Downgraded one level for risk of bias: one RCT with no description of randomisation or concealment of allocation. 3Downgraded two levels for imprecision. Few events and wide CI containing appreciable benefit and harm. 4In addition one non‐randomized study (matched case series) with serious risk of bias (ROBINS‐I) providing an estimate of 0.8 days fewer in the experimental group (2.7 days fewer to 1.1 days longer); very low‐certainty evidence.

Summary of findings 2

Early versus late supportive care plus ribavirin for Crimean Congo haemorrhagic fever

Early versus late supportive care plus ribavirin for Crimean Congo haemorrhagic fever
Patient or population: people diagnosed with suspected or confirmed Crimean Congo haemorrhagic fever Setting: global Intervention: early supportive care plus ribavirin1 Comparison: late supportive care plus ribavirin
Outcomes	Anticipated absolute effects* (95% CI)		Relative effect (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with Late ribavirin	Risk with Early ribavirin
Mortality in early versus late supportive care plus ribavirin	400 per 1000	156 per 1000 (64 to 380)	RR 0.39 (0.16 to 0.95)	63 (1 non‐randomised study)	⊕⊝⊝⊝ VERY LOW2,3	‐
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). Abbreviations: CI: confidence interval; RR: risk ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

1Early defined according to that reported in the included study (< 4 days since onset of symptoms) 2Downgraded one level for risk of bias: all studies at serious risk of bias. 3Downgraded two levels for imprecision: few events and wide CIs.

Our main effects analysis included one RCT and four non‐randomized studies. The remaining 18 studies, which we assessed as being at critical risk of bias, are used in a subsidiary descriptive analysis for our primary outcome of mortality.

Ribavirin versus no ribavirin

Mortality

One RCT and three non‐randomized studies were included that compared the effect on mortality of ribavirin and no ribavirin in participants with CCHF (Figure 3).

Figure 3

Forest plot of Ribavirin versus no ribavirin, outcome: mortality.

RCT

One RCT of 136 participants (Koksal 2010) found no statistically significant effect in favour of either ribavirin or no ribavirin (RR 1.13, 95% CI 0.29 to 4.32; Analysis 1.1).

Analysis 1.1

Comparison 1 Ribavirin versus no ribavirin, Outcome 1 Mortality.

Non‐randomized studies

One mixed retrospective and prospective cohort study of 281 participants stratified risk of death by severity of disease (Dokuzoguz 2013). No deaths occurred in 103 mild cases and risk ratios were therefore not calculable. In 152 moderate cases (subgroup 2) ribavirin reduced mortality (RR 0.09, 95% CI 0.02 to 0.50). In 26 severe patients no effect of ribavirin on mortality was seen (RR 0.79, 95% CI 0.44 to 1.41; Analysis 1.2, Figure 4 ). The two participants in the severe disease strata control group were unable to take oral ribavirin due to gastro‐intestinal bleeding, despite an intention to treat them with ribavirin.

Figure 4

Forest plot of comparison: 1 Ribavirin versus no ribavirin, outcome: 1.2 Mortality stratified by severity of disease (Dokuzoguz 2013).

One cohort study with a historical control arm of 218 participants had similar baseline characteristics in terms of severity of disease and time from onset of symptoms (Elaldi 2009). This study showed no statistically significant benefit of ribavirin on mortality (RR 0.60, 95% CI 0.26 to 1.38; Analysis 1.1)

One retrospective matched cohort study of 50 participants used a matched design where those who received ribavirin were randomly matched to a control group with similar baseline characteristics (Bodur 2011). In this study no statistically significant effect was seen (RR 1.33, 95% CI 0.32 to 5.64; Analysis 1.1).

In a pooled analysis of these three non‐randomized studies we found no statistically significant effect (RR 0.72, 95% CI 0.41 to 1.28; 549 participants; Analysis 1.1; Figure 3). With few events and wide CIs containing clinically appreciable benefit and harm, it is not possible to draw a conclusion of benefit or of no effect from the available evidence. Given the concerns over the internal validity of the studies, this further decreases our confidence in the effect estimate.

In summary, it is uncertain whether ribavirin reduces mortality, because the certainty of the evidence is very low from both the RCT and the non‐randomized studies (Table 1).

Length of hospital stay

One non‐randomized study (retrospective matched cohort design) and one RCT met our inclusion criteria and evaluated the effect of ribavirin on length of hospital stay in participants with CCHF receiving ribavirin or not (Koksal 2010; Bodur 2011; Analysis 1.3; Figure 5).

Analysis 1.3

Comparison 1 Ribavirin versus no ribavirin, Outcome 3 Length of hospital stay (days).

Figure 5

Forest plot of ribavirin versus no ribavirin, outcome: length of hospital stay (days).

Koksal 2010 showed no effect of ribavirin on the length of hospital stay in days (MD 0.70, 95% CI ‐0.39 to 1.79; 136 participants; Analysis 1.3).

Bodur 2011 showed no effect of ribavirin on the length of hospital stay in days (MD ‐0.80, 95% CI ‐2.70 to 1.10; 50 participants; Analysis 1.3).

In summary, we do not know if ribavirin reduces the length of stay in hospital, as the certainty of the evidence is very low (Table 1).

Requirement for transfusion

One included RCT compared the effect of ribavirin with no ribavirin on the need for transfusion of blood products in participants with CCHF (Koksal 2010). There was no statistically significant difference in requirement for transfusion of platelets between treated and untreated participants in the RCT (RR 1.23, 95% CI 0.77 to 1.96; 136 participants; Analysis 1.4).

Analysis 1.4

Comparison 1 Ribavirin versus no ribavirin, Outcome 4 Requirement for transfusion (platelets).

Withdrawal of treatment due to adverse events

One study included in the primary analysis reported on adverse events leading to discontinuation of treatment. One participant among 44 who received ribavirin and corticosteroids discontinued ribavirin due to elevated amylase levels (Dokuzoguz 2013).

Serious adverse events

No studies in the primary analysis reported on adverse events.

Timing of administration of ribavirin: early versus late ribavirin

One non‐randomized study (retrospective cohort) was included that addressed the timing of administration of ribavirin alongside supportive care and mortality (Izadi 2009a).

Izadi 2009a outlined an association between reduced mortality in those who received supportive care and ribavirin less than four days since the onset of any symptoms compared to those receiving supportive care and ribavirin after four days (RR 0.39, 95% CI 0.16 to 0.95; 63 participants; Analysis 2.1; Figure 6).

Analysis 2.1

Comparison 2 Early versus late supportive care with ribavirin, Outcome 1 Mortality.

Figure 6

Forest plot early versus late ribavirin, outcome: mortality in early versus late ribavirin.

Whilst an association was seen between early supportive care and ribavirin and reduced mortality in one included study at serious risk of bias, we are uncertain if early ribavirin is more effective than late ribavirin in treating CCHF. Separating the effect of early presentation to hospital, early diagnosis and early supportive care from the effect of early ribavirin treatment is very difficult without an adequately‐powered randomised study.

Subsidiary descriptive analyses

Ribavirin versus no ribavirin

In the subsidiary descriptive analysis we explored the effect of confounding on the effect estimates for ribavirin versus no ribavirin. We included 10 studies at critical risk of bias that reported mortality outcomes. We established a non‐overlapping sample using the methods described in Appendix 2 and present these in a forest plot alongside the single RCT and cohort studies at serious risk of bias (Analysis 3.1; Figure 7). In these studies with a critical risk of bias, the point estimates shows an effect skewed towards benefit for ribavirin (1 RCT; RR 1.13, 95% CI 0.29 to 4.32; 136 participants; 3 non‐randomized studies at serious risk of bias; RR 0.72, 95% CI 0.42 to 1.28; 549 participants; 10 non‐randomized studies at critical risk of bias RR 0.43, 95% CI 0.22 to 0.86; 1214 participants). There was also increasing heterogeneity (NRS serious risk of bias I2 statistic = 0%; NRS critical risk of bias I2 statistic = 58%). This supports the conclusions of a previous meta‐analysis (Soares‐Weiser 2010) that the effect seen is likely to be attributable to confounding and that no evidence of benefit could be drawn. Secondly, our descriptive analysis demonstrates that a critical failure to control for confounding is associated with an increase in heterogeneity and inconsistency between studies.

Analysis 3.1

Comparison 3 Subsidiary descriptive analysis ‐ Ribavirin versus no ribavirin, Outcome 1 Mortality stratified study type.

Figure 7

Forest plot of subsidiary descriptive analysis: ribavirin versus no ribavirin, outcome: mortality.

Early versus late supportive care with ribavirin

In the subsidiary descriptive analysis of early versus late ribavirin, we explored the effect of confounding of the effect estimates. We included four studies at critical risk of bias and present these in a forest plot alongside the single non‐randomized study at serious risk of bias (Analysis 4.1; Figure 8).

Analysis 4.1

Comparison 4 Subsidiary descriptive analysis: early versus late supportive care with ribavirin, Outcome 1 Mortality stratified by study type.

Figure 8

Forest plot of comparison: 4 Subsidiary descriptive analysis: early versus late supportive care with ribavirin, outcome: 4.1 Mortality stratified by study type.

Our subsidiary descriptive analysis showed an association between early supportive care with ribavirin and a reduction in mortality in studies with critical risk of bias (4 NRS; RR 0.57, 95% CI 0.38 to 0.85; 431 participants), there was no difference in effect compared to the study at serious risk of bias (1 NRS; RR 0.39, 95% CI 0.16 to 0.95; 63 participants; I2 statistic = 0%).

Discussion

Summary of main results

See Table 1; Table 2.

Five studies met the inclusion criteria for our main effects analysis. These was one RCT with 136 participants and four non‐randomized studies with 612 participants. We judged all four non‐randomized studies to have serious risk of bias by ROBINS‐I. There were a further 18 non‐randomized studies classified at critical risk of bias which we included in a subsidiary descriptive analysis. None of these studies attempted to control for confounding.

We do not know if ribavirin reduces mortality (very low‐certainty evidence).

We do not know if ribavirin is more effective when given early with supportive care rather than late with supportive care (very low‐certainty evidence), and we do not know if ribavirin reduces the length of stay in hospital (very low‐certainty evidence).

In terms of possible adverse effects, we do not know if it reduces the risk of patients needing platelet transfusions (very low‐certainty evidence), and we do not know what the adverse effects of treating CCHF with ribavirin are, because there is a lack of data for this outcome.

In the subsidiary descriptive analysis of studies with a critical risk of bias, the point estimates show an effect skewed towards benefit for ribavirin, as well as increasing heterogeneity.

Overall completeness and applicability of evidence

This review includes a single RCT and 22 non‐randomized studies from multiple countries in Europe and Asia. We found no studies from Africa, where CCHF is also endemic.

There is insufficient reliable evidence to be confident of the effects of ribavirin on mortality, length of hospital stay or the need for transfusion of blood products. There is insufficient high‐quality evidence to draw conclusions about the likelihood of serious adverse events occurring when administering ribavirin to people infected with CCHF. Ribavirin is frequently used in the treatment of hepatitis C and the side‐effect profile is well established (Brok 2009). However, given different dosing schedules and the differences in the length of use of ribavirin, we do not think this evidence is sufficiently generalizable to CCHF.

We wondered whether the non‐randomized studies would be sufficient to show a benefit for ribavirin if indeed they had a very large effect on mortality and were of sufficient quality. However, all but four of the studies were at critical risk of bias, which means drawing inferences from these studies is not possible (Sterne 2016). In those non‐randomized studies not at critical risk of bias, the evidence base is not of sufficient robustness to draw conclusions about benefit or harm, given our concerns about the internal validity of the studies and imprecision of the effect estimates.

Certainty of the evidence

The overall certainty of the evidence for all outcomes was very low. Any estimate of effect is highly uncertain and is likely to change with further research on the treatment of CCHF. Most research done in this area is of non‐randomized designs and is critically compromised by uncontrolled confounding and small sample sizes. Because of this, we are unable to reach any conclusions on the efficacy of ribavirin for treating CCHF.

For mortality, the single RCT, which was the study with the most reliable internal validity and which we felt provided the most reliable effect estimate, was at high risk of bias and underpowered to show an effect, with few events. As a result we downgraded it to very low‐certainty evidence for the outcome of mortality.

For mortality in early versus late ribavirin, all studies were of a non‐randomized design at serious risk of bias. The pooled effect estimate included few events and broad CIs, which meant we downgraded the evidence to very low certainty.

For length of hospital stay, the single RCT was at high risk of bias and underpowered to show an effect, with few events. We therefore downgraded it to very low‐certainty evidence for this outcome.

For the requirement for transfusion of blood products, the single RCT was at high risk of bias and underpowered to show an effect, with few events. We therefore downgraded it to very low certainty evidence for this outcome.

Potential biases in the review process

We have minimized the effect of confounding bias on the effect estimates in the non‐randomized studies by only presenting those at serious, moderate, high, low or unclear risk of bias in the main analysis. To describe the effect of confounding we conducted a subsidiary analysis only including those studies at critical risk of bias. We used the latest tools in assessing risk of bias in non‐randomized studies. We sought guidance from specialist methodologists developing the ROBINS‐I tool to aid our processes. Despite these efforts, we included no studies in this review with a low risk of bias, which means that confounding is still likely to bias any estimates in the main effects analysis.

The included studies populations largely came from Turkey and Iran, with little evidence available from other countries, although we attempted to include a broad range of geographic locations by searching extensively for literature and by including a PhD thesis from Russia (Sannikova 2009).

Agreements and disagreements with other studies or reviews

A previous systematic review (Soares‐Weiser 2010) concluded that there was no clear evidence of benefit from the data then available, as non‐randomized studies were heavily confounded. In our review we have tried to stratify analysis by different degrees of confounding in the studies. This analysis agrees with the opinion of the authors of the Soares‐Weiser review that the effect seen in their meta‐analysis was likely to have been due to confounding in non‐randomized studies.

Soares‐Weiser 2010 included two further studies not included in our review. We excluded these studies because of a sample size of less than 10 participants (Jamil 2005), and the lack of a comparator arm (Nadeem 2003). See the Excluded studies section.

We agree with the assessment of the authors of the Ascioglu 2011 review about the internal validity of the included studies and the effect of systematic bias on the effect estimate. We further agree that the results of a meta‐analysis of flawed studies cannot be used as evidence of an effect, and that a randomised controlled trial is needed and ethically justified, given the ambiguity of observational studies.

All studies included in the Ascioglu 2011 systematic review are included in our review.

We agree with the two previous systematic reviews on this topic (Soares‐Weiser 2010; Ascioglu 2011). We cannot draw conclusions about the efficacy of ribavirin for treating Crimean Congo haemorrhagic fever using the data currently available. This is largely attributable to too few studies that adequately control for confounding and the lack of a reliable RCT. Any estimate of effect based on currently available data is very uncertain.

Research in outbreaks

In a broader sense, the current status of the evidence for ribavirin in CCHF highlights the difficulties when non‐randomized studies or consensus is used to establish a treatment in the absence of reliable evidence. Once established as standard practice, clinicians feel uneasy about the ethics of conducting a placebo controlled trial whether reliable evidence of efficacy exists or not. This is made more acute because of a previous lack of preparedness for experimental research therapeutics in outbreak situations. In 2016 WHO issued guidance on managing ethical issues in infectious disease outbreaks which highlights the need to learn as much as possible as quickly as possible and that in such situations where no proven treatment exists research should be conducted using rigorous methodology that is capable of providing valid results (WHO 2016).

Whilst monitored use of experimental or unproven therapies can be ethically justifiable in outbreak situations provided; 1) no proven effective treatment exists; 2) it is not possible to initiate clinical studies immediately; 3) data providing preliminary support of the intervention’s efficacy and safety are available; 4) the relevant country authorities, as well as an appropriately qualified ethics committee, have approved such use; 5) adequate resources are available to ensure that risks can be minimized; 6) the patient’s informed consent is obtained; and 7) the emergency use of the intervention is monitored and the results are documented and shared in a timely manner with the wider medical and scientific community (WHO 2016). It is important always to be clear that no harm is likely, considering the potential for causing harm is important as sometimes there is a perception that any intervention will help because of the high mortality, however, a harmful intervention could push case fatality rates even higher as well as potentially costing valuable time and resources.

As this review demonstrates, establishing efficacy of a therapeutic in acute infectious diseases using observational or non‐randomised data is difficult and results can be unreliable. As such the ability to conduct methodologically rigorous research that is able to demonstrate efficacy should be a requirement of any use of unproven therapeutics in outbreak situations.

Authors' conclusions

We do not know from the current literature if ribavirin is an effective treatment for CCHF. Most research on this question is of a non‐randomized design and is critically confounded. Any estimates of effect based on the existing literature is highly uncertain and likely to change with further methodologically rigorous research.

This review improves on the previous systematic reviews by including all the relevant information from observational studies and assessing them systematically. We have used the latest methods examining confounding and other important methodological aspects important in assessing the findings of non randomised studies looking at the effects of ribavirin in CCHF.

There remains considerable controversy on the effects of ribavirin in CCHF and whether to use it, reflecting true uncertainty in the field, with some strong advocates (Ergonul 2006); and others recommend supportive care only (Kalin 2014). These clear variations in practice and viewpoints and the lack of any clear message from this independent systematic review of the of the evidence point us in the direction of a randomised clinical trial to establish or disprove the efficacy of ribavirin, as has been suggested previously (Soares‐Weiser 2010; Ascioglu 2011).

Study	Reason for exclusion
Abuova 2012	This was a case series with fewer than 10 participants
Ajazaj 2013	This was a case series with fewer than 10 participants
Alavi‐Naini 2004	This is a single case report
Ali 2010	This was a case series with fewer than 10 participants
Anon 1984	This was a survey and therefore was a different study design from our inclusion criteria
Ardalan 2006	This was a single case report
Athar 2003	This was a case series with fewer than 10 participants
Athar 2005	This was a case series with fewer than 10 participants
Barr 2013	This was a single case report
Canpolat 2011	This was a single case report
Caylan 2010	This was a single case report
Ceri 2013	This was a single case report
Chinikar 2013	This was a case series with fewer than 10 participants
Dilber 2010	This was a case series with fewer than 10 participants.
Drosten 2002	This was a single case report and discussion
El Bahnasawy 2015	This was a survey and therefore did not meet inclusion criteria; it was based on a different study design
Elata 2011	This was a case report of a single nosocomial transmission.
Ergonul 2009	This was a commentary on an included study
Ergonul 2014	This was a case series with fewer than 10 participants.
Ergonul 2017	This study compared individuals with CCHF to healthy individuals
Fazlalipour 2016	This was a case series with fewer than 10 participants.
Gonen 2014	This was a case series with fewer than 10 cases
Gozel 2013	This was a case series with fewer than 10 cases
Guner 2014	This was a case series with fewer than 10 participants
Gursoy 2014	Translated from Turkish. This was an editorial letter that reported a single case
Hasan 2013	This was a single case report
Izadi 2009b	This is an editorial letter written as a reply to comments on an included study
Jabbari 2006	This was a case series with fewer than 10 participants
Jamil 2005	This was a case series with fewer than 10 participants.
Joubert 1985	This was a case series with fewer than 10 participants.
Kadanali 2012	This was a cohort study that did not compare ribavirin to supportive care only ‐ no comparator arm
Kader 2011	This was a cohort study that did not compare ribavirin to supportive care only ‐ no comparator arm
Kleib 2016	This was a single case report
Kubar 2011	This reported the effects of administration of hyperimmunoglobulin, not ribavirin.
Kunchev 2008	This was a case series with fewer than 10 participants.
Leblebicioglu 2016a	This reported on the use of ribavirin for prophylaxis but did not report on ribavirin used as treatment for disease
Makwana 2015	This was a single case report
Mardani 2009	This was case series of fewer than 10 participants
Mardani 2013	This was a case series with fewer than 10 participants
Midilli 2007	This was a cohort study that did not compare ribavirin to supportive care only ‐ no comparator arm
Mishra 2011	This was a single case report
MMWR 1984	This was a case series with fewer than 10 participants
Mohamed 2016	This was a single case report
Nabeth 2004	This was a single case report
Naderi 2011	This was a case series with fewer than 10 participants.
Naderi 2013	This was a case series with fewer than 10 participants.
NCT00992693	This ongoing study did not include a comparator group where no ribavirin is given.
Oflaz 2013	This was a cohort study that did not compare ribavirin to supportive care only ‐ no comparator arm
Ozbey 2014	This did not compare use of ribavirin to supportive care only. It did not report mortality as an outcome in a useable way, reporting only a case fatality ratio in those who were transferred to tertiary centres or not transferred. As such this study did not meet our inclusion criteria.
Ozsoy 2015	This was a case series with fewer than 10 participants
Papa 2008	This was a single case report
Pourahmad 2011	This was a single case report
Pshenichnaya 2015	This was a case series with fewer than 10 participants
Raoofi 2012	This was a case series with fewer than 10 participants
Richards 2015	This was a case series with fewer than 10 participants
Sahin 2016	This was a single case report
Saluzzo 1985b	This was a single case report
Schwarz 1995	This was a single case report
Scrimgeour 1996	This was a case series with fewer than 10 participants
Sefikotullari 2013	This case series reported different outcomes from those in our review
Sharifi‐Mood 2008	This was a cohort study that did not compare ribavirin to supportive care only ‐ no comparator arm
Sharifi‐Mood 2009	This was an overlapping study reporting the same data as an included study (Sharifi‐Mood 2006)
Sharifi‐Mood 2013b	This was a quasi‐RCT that did not report on ribavirin compared to supportive care only ‐ all participants received ribavirin with or without corticosteroids
Sheikh 2005	This was a cohort study that did not compare ribavirin to supportive care only ‐ no comparator arm
Sheikh, 2004	This was a case series with fewer than 10 participants
Smego 2004	This was a case series with fewer than 10 participants
Suleiman 1980	This was a case series with fewer than 10 participants
Sunbul 2016	This was a single case report
Tall 2009a	This was a single case report
Tall 2009b	This was a case series with fewer than 10 participants
Tatar 2005	This was a case series with fewer than 10 participants
Tezer 2014	This was a case series with fewer than 10 participants.
Tulek 2010	This was a case series with fewer than 10 participants
Tutuncu 2009	This was a case series with fewer than 10 participants
Ugurlu 2013	This was a single case report
Unlusoy 2014	This was a single case report
Uysal 2012	This was a single case report
Van Eeden 1985a	This was a case series with fewer than 10 participants
Van Eeden 1985b	This was a case series with fewer than 10 participants
Weber 2001	This was a case series with fewer than 10 participants
Yadav 2013	This did not report on ribavirin use in CCHF
Yadav 2016	This did not report on ribavirin use in CCHF
Yesilyurt 2011	This did not report on ribavirin use in CCHF
Yildirmak 2016	This was a case series with fewer than 10 participants
Yilmaz 2009a	This did not report on ribavirin use in CCHF
Yilmaz 2009b	This was a case series with fewer than 10 participants
Yolcu 2014	This was a survey that did not report on ribavirin use for treatment of CCHF
Zakhashvili 2010	This was a single case report
Öztürk 2012	This was a case series with fewer than 10 participants

RCT: randomized controlled trial

Table 6

ROBINS‐I Interpretation of domain level and overall risk of bias judgements

Judgement	Within each domain	Across domains	Criterion
Low risk of bias	The study is comparable to a well‐performed randomised trial with regard to this domain	The study is comparable to a well‐performed randomised trial	The study is judged to be at low risk of bias for all domains
Moderate risk of bias	The study is sound for a non‐randomized study with regard to this domain but cannot be considered comparable to a well‐performed randomised trial	The study provides sound evidence for a non‐randomized study but cannot be considered comparable to a well‐performed randomised trial	The study is judged to be at low or moderate risk of bias for all domains
Serious risk of bias	the study has some important problems in this domain	The study has some important problems	The study is judged to be at serious risk of bias in at least one domain, but not at critical risk of bias in any domain
Critical risk of bias	the study is too problematic in this domain to provide any useful evidence on the effects of intervention	The study is too problematic to provide any useful evidence and should not be included in any synthesis	The study is judged to be at critical risk of bias in at least one domain
No information	No information on which to base a judgement about risk of bias for this domain	No information on which to base a judgement about risk of bias	There is no clear indication that the study is at serious or critical risk of bias and there is a lack of information in one or more key domains of bias (a judgement is required for this)

Reproduced from Sterne 2016.

Comparison 1

Ribavirin versus no ribavirin

Comparison 1 Ribavirin versus no ribavirin, Outcome 1 Mortality.

Comparison 1 Ribavirin versus no ribavirin, Outcome 2 Mortality stratified by severity of disease (Dokuzoguz 2013).

Comparison 1 Ribavirin versus no ribavirin, Outcome 3 Length of hospital stay (days).

Comparison 1 Ribavirin versus no ribavirin, Outcome 4 Requirement for transfusion (platelets).

Comparison 1 Ribavirin versus no ribavirin, Outcome 1 Mortality.

Comparison 1 Ribavirin versus no ribavirin, Outcome 2 Mortality stratified by severity of disease (Dokuzoguz 2013).

Comparison 1 Ribavirin versus no ribavirin, Outcome 3 Length of hospital stay (days).

Comparison 1 Ribavirin versus no ribavirin, Outcome 4 Requirement for transfusion (platelets).

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Mortality Analysis 1.1
4		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
1.1 RCT	1	136	Risk Ratio (M‐H, Random, 95% CI)	1.13 [0.29, 4.32]
1.2 Non‐randomized studies	3	549	Risk Ratio (M‐H, Random, 95% CI)	0.72 [0.41, 1.28]
2 Mortality stratified by severity of disease (Dokuzoguz 2013) Analysis 1.2
1		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
3 Length of hospital stay (days) Analysis 1.3
2		Mean Difference (IV, Random, 95% CI)	Totals not selected
3.1 RCT	1		Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
3.2 Non‐randomized studies	1		Mean Difference (IV, Random, 95% CI)	0.0 [0.0, 0.0]
4 Requirement for transfusion (platelets) Analysis 1.4
1		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected

Comparison 2

Early versus late supportive care with ribavirin

Comparison 2 Early versus late supportive care with ribavirin, Outcome 1 Mortality.

Comparison 2 Early versus late supportive care with ribavirin, Outcome 1 Mortality.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Mortality Analysis 2.1
1		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected

Comparison 3

Subsidiary descriptive analysis ‐ Ribavirin versus no ribavirin

Comparison 3 Subsidiary descriptive analysis ‐ Ribavirin versus no ribavirin, Outcome 1 Mortality stratified study type.

Comparison 3 Subsidiary descriptive analysis ‐ Ribavirin versus no ribavirin, Outcome 1 Mortality stratified study type.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Mortality stratified study type Analysis 3.1
14		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
1.1 RCT	1	136	Risk Ratio (M‐H, Random, 95% CI)	1.13 [0.29, 4.32]
1.2 Non‐randomized studies (serious risk of bias)	3	549	Risk Ratio (M‐H, Random, 95% CI)	0.72 [0.41, 1.28]
1.3 Non‐randomized studies (critical risk of bias)	10	1214	Risk Ratio (M‐H, Random, 95% CI)	0.43 [0.22, 0.86]

Comparison 4

Subsidiary descriptive analysis: early versus late supportive care with ribavirin

Comparison 4 Subsidiary descriptive analysis: early versus late supportive care with ribavirin, Outcome 1 Mortality stratified by study type.

Comparison 4 Subsidiary descriptive analysis: early versus late supportive care with ribavirin, Outcome 1 Mortality stratified by study type.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Mortality stratified by study type Analysis 4.1
5		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
1.1 Serious risk of bias	1	63	Risk Ratio (M‐H, Random, 95% CI)	0.39 [0.16, 0.95]
1.2 Critical risk of bias	4	431	Risk Ratio (M‐H, Random, 95% CI)	0.57 [0.38, 0.85]