This study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guideline [8].
2.1 Search strategy
Searches for studies were performed in PubMed, Web of Science, Scopus, Embase, Google Scholar (first 100 hits), the website ClinicalTrials.gov, and the preprint server medRxiv from January 1, 2020 to May 23, 2022. The search was limited to studies published in full-text versions, without language restriction. In ClinicalTrials.gov, only completed studies with results were analyzed. The reference lists of all eligible studies and reviews were evaluated to identify additional studies for inclusion.
We used the following structured search strategy for each electronic database and other sources: (nitazoxanide) AND (COVID-19 OR “2019-nCoV Infection” OR “Coronavirus Disease-19” OR “2019-nCoV Disease” OR SARS-CoV-2). To expand the number of eligible studies, specific filters for RCTs were not used.
2.2 Study selection and eligibility criteria
Two reviewers (P.R.M.-F. and E.M.N.-J.) independently screened the search results and identified studies that were potentially relevant based on their title and abstract. Relevant studies were read in full and selected according to eligibility criteria. Disagreements between the two reviewers were resolved by consensus.
The following elements were used to define eligibility criteria: (1) Population: individuals with COVID-19; (2) Intervention: nitazoxanide; (3) Comparison: placebo; (4) Outcomes: viral load, positive RT-PCR status, composite measure of disease progression (ICU admission or invasive mechanical ventilation), death, serum biomarkers of inflammation (white blood cells [WBC], neutrophils, lymphocytes, C-reactive protein [CRP], D-dimer, lactate dehydrogenase [LDH], IL-6, IL-8, TNF-α), and any adverse events; (5) Study type: blinded, placebo-controlled, RCTs. Eligible studies must report at least 1 of the outcomes of interest. Potential overlapping populations, open-label trials, and observational studies were excluded. Trials testing drug associations were also excluded.
2.3 Data extraction
Two authors (P.R.M.-F. and E.M.N.-J.) extracted the data from included studies and crosschecked them for accuracy. Using a standardized data extraction sheet, the following information were extracted from the studies: registry of the study protocol, demographic characteristics of study participants, pre-existing medical conditions, treatment arms, nitazoxanide protocol, concomitant medications, follow-up duration, and outcome data.
2.4 Risk of bias assessment
Risk of bias was judged according to the Cochrane guidelines for RCTs [9]. The following domains were evaluated: sequence generation and allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessment (detection bias), incomplete outcome data (attrition bias), selective outcome reporting (reporting bias), sample size calculation, power analysis, and early stopping for futility (operational bias), outcome measurements (information bias), and the authors' financial or non-financial conflicts of interest that could appear to affect the judgment of research team when designing, conducting, or reporting study. Studies using real-time reverse transcription-polymerase chain reaction (RT-PCR) to detect SARS-CoV-2 or, if testing was limited, provided a clinical diagnosis based on COVID-19-related symptoms and epidemiological data were considered as having a low risk of bias.
2.5 Data synthesis
Treatment effects were reported as relative risk (RR) for dichotomous variables (positive RT-PCR status, composite measure of disease progression, death, and any adverse events) and standardized mean difference (SMD) for continuous variables (viral load and serum biomarkers of inflammation) with 95% confidence intervals (CI). To calculate the RR, the number of events and individuals in each treatment group were extracted. To calculate SMD, means and standard deviations (SD) were obtained for each study group. If the means and SD were not directly reported in the publication, indirect methods of extracting estimates were used [10]. For viral load, we analyzed the results based on changes from baseline. A negative effect size indicated that nitazoxanide decreased viral load and levels of inflammatory biomarkers in patients with COVID-19.
We used either a fixed or random-effects model to pool the results of individual studies depending on the presence of heterogeneity. Statistical heterogeneity was quantified by the I2 index using the following interpretation: 0%, no between-study heterogeneity; <50%, low heterogeneity; 50–75%, moderate heterogeneity; >75%, high heterogeneity [11]. In the case of heterogeneity, we used the random-effects model, otherwise, the fixed-effects model was used.
Although funnel plots may be useful tools in investigating small study effects in meta-analyses, they have limited power to detect such effects when there are few studies [12]. Therefore, because we had only a small number of included studies, we did not perform a funnel plot analysis. Forest plots were used to present the effect sizes and the 95% CI, and a 2-tailed p < 0.05 was used to determine significance. Analyses were conducted using Review Manager, version 5.3 (Cochrane IMS).
2.6 Grading the strength of evidence
We graded the strength of evidence for the association between the use of nitazoxanide and the outcomes of interest as high, moderate, low, or very-low using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) rating system [13, 14]. In the GRADE system, RCTs begin as high-quality evidence but may be downrated according to the risk of bias assessment, inconsistency, indirectness, imprecision in the results, and publication bias [15]. Certainty is uprated for estimates with large (RR > 2.0 or RR < 0.5; SMD > 0.8) magnitude of effect.
Although the funnel plot asymmetry was not evaluated, we reduced the potential for publication bias by planning a comprehensive search including grey-literature without restrictions. In this criterion, we analyzed discrepancies in findings between studies and the influence of small trials (< 100 patients per arm) on estimated treatment effects. The influence of small trials on the pooled estimates was analyzed using a “leave-one-out” sensitivity approach [16].