Types of studies

We included randomised controlled trials (RCTs), including cluster‐RCTs. We excluded any trials of a cross‐over design as this is incompatible with our review question. For RCTs published after 2010, we considered only prospectively registered RCTs to be eligible for this 2021 update, in accordance with Cochrane Injuries Group policy.

Types of participants

Any person of any age, diagnosed with oral paraquat poisoning (although poisoning via inhalation or contact with skin is possible, these were not foci of the review). The review focuses on participants with moderate to severe poisoning, given that generally in trials (as in clinical practice) individuals with mild cases of paraquat poisoning (which tend to resolve on their own) and those with extremely severe intoxication (see definition of 'fulminant' in both Description of the condition and Included studies) may not be treated; or, if they are treated, may not be analysed with other groups of participants, due to prognosis.

Types of interventions

• Intervention: glucocorticoid with cyclophosphamide, in combination.

• Eligible comparators: placebo; standard care alone; or any alternative therapy in addition to standard care.

We excluded studies that focused on any single immunosuppressant (either a glucocorticoid drug or cyclophosphamide by themselves) or other combinations of therapies.

Types of outcome measures

• Mortality at 30 days following the ingestion of paraquat.

• All‐cause mortality at the end of the maximum follow‐up period recorded by investigators, as long as they are clinically compatible.

Otherwise we reported findings in the short term (typically at hospital discharge, which of necessity varied between participants), and longer term (three months after discharge in the one trial that reported any follow‐up), as appropriate.

• New infections diagnosed within one week after initiation of treatment.

Electronic searches

The Cochrane Injuries Group's Information Specialists searched:

• the Cochrane Central Register of Controlled Trials (CENTRAL) (which contains the Cochrane Injuries Trials Register) in the Cochrane Library (searched 21 September 2020);

• Ovid MEDLINE(R), Ovid MEDLINE In‐Process & Other Non‐Indexed Citations, Ovid MEDLINE Daily and Ovid OLDMEDLINE (1946 to 21 September 2020);

• Embase Classic + Embase (OvidSP) (1947 to 21 September 2020);

• ISI Web of Science: Science Citation Index Expanded (SCI‐EXPANDED) (1970 to 21 September 2020);

• ISI Web of Science: SCI‐EXPANDED, SSCI, A&HCI, CPCI‐S, CPCI‐SSH, ESCI (2017 to 22 September 2020)

• US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov (www.clinicaltrials.gov) (21 September 2020);

• Current Controlled Trials (www.controlled-trials.com/) (20 September 2020);

• World Health Organization International Clinical Trials Registry Platform (apps.who.int/trialsearch/) (21 September 2020).

The following Chinese databases were searched by one review author (LL):

• China National Knowledge Infrastructure (CNKI 数据库) (12 November 2020);

• Wanfang Data (万方数据库) (12 November 2020);

• VIP (维普数据库) (12 November 2020).

Searching other resources

We searched the internet through search engines google.com and baidu.com on 12 November 2020 using the term 'clinical trial & paraquat'. We also checked the reference lists of reports and systematic/literature reviews on paraquat poisoning for potentially relevant published or unpublished trials. We contacted the authors of the included trials for further information.

Selection of studies

For the present version of the review, three review authors (LL, JD and HW) independently screened the search results from the English language databases. LL, CY and BC independently screened the results from the Chinese databases. We obtained and assessed the full‐text versions of potentially relevant trials. Duplicate reports were identified and noted.

Review authors LL and BC disagreed about the inclusion of the Afzali 2008 study due to the use of alternate allocation as the method of sequence generation. Review author CY moderated the discussion on inclusion of this trial, and in the last published version of this review (Li 2014), it was agreed that the trial would be included, but classified as being at high risk of bias. Following statistical advice in 2021, the author team as a whole decided once again to retain the trial, whilst highlighting issues to do with its interpretation, on the basis not only of design, but also due to investigator‐acknowledged failure to assess participants via plasma concentration.

We identified one new study that met our eligibility criteria (Gawarammana 2017). A flow chart of the study selection process is shown in Figure 1.

Figure 1

---

---

Study flow diagram for 2021 update

Data extraction and management

Three review authors (LL, BC, and JD) independently extracted the following data from the four included trials:

• study design;

• study setting (country, city, number of centres);

• inclusion and exclusion criteria for studies;

• number, gender, and severity of intoxication of participants in the intervention and control groups;

• outcome data including number of deaths and infection cases assessed at each data collection point;

• loss to follow‐up/withdrawals from analysis;

• information on study conduct sufficient to assess risk of bias, including evidence of prospective trial registration for all studies published from 2010 onwards.

Assessment of risk of bias in included studies

Four review authors (LL, BC, JD, and HW) independently evaluated the risk of bias for each included trial based on the following domains: sequence generation, allocation concealment, blinding, incomplete outcome data, selective reporting, and any other sources of bias. We based our judgements on the criteria outlined in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We assessed each domain as at low or high risk, or unclear risk if the information provided was insufficient to permit a judgement on a particular area of bias, or if it was unclear in which direction a bias might lead. Risk of bias judgements are provided in the risk of bias tables in Characteristics of included studies, and summaries of the judgements are given in Figure 2 and Figure 3.

Figure 2

---

---

Methodological quality graph: review authors' judgements about each methodological quality item presented as percentages across all included studies

Figure 3

---

---

Methodological quality summary: review authors' judgements about each methodological quality item for each included study

Measures of treatment effect

Where feasible, we calculated the risk ratio (RR) and 95% confidence interval (CI). We summarised data for all‐cause mortality at final follow‐up in a meta‐analysis using a fixed‐effect model. When assuming a 75% and 65% mortality rate for control and treatment group respectively, with a level of significance of 5% and a power of 90%, we calculated that 439 participants would be required in each arm of a study to detect the effect sufficiently.

Unit of analysis issues

No unit of analysis issues arose within the current version of this review. We excluded cross‐over studies as inappropriate for our review question (given our focus on mortality). We identified no cluster‐RCTs and no trials with multiple eligible intervention (or comparator) arms.

Should the issue of clustering arise in future updates, we plan to use the methods detailed in the Cochrane Handbook for Systematic Reviews of Interventions as applicable (Higgins 2021), based on information provided by investigators, including: the number of clusters (or groups) randomised or the average (mean) size of each cluster; and an estimate of the intracluster (or intraclass) correlation coefficient (ICC).

Should the issue of multiple eligible arms or comparator groups arise, we will consider appropriate methods including pooling data from eligible arms, or conversely treating a multiple‐arm study as two or more studies. In the latter case, we will take care to reduce the number of participants in the arm(s) left intact, in order to preserve the independence of findings.

Dealing with missing data

The amount of missing data in each trial was assessed, outcome by outcome, and any potential effects on results informed our risk of bias assessment (see Risk of bias in included studies) and GRADE ratings.

Assessment of heterogeneity

We considered possible causes of clinical heterogeneity (e.g. how studies measured severity of poisoning, treatment regimen, variation in standard care, length of follow‐up) when making decisions about whether to pool data.

Where possible and appropriate, we examined statistical heterogeneity using the Chi² test and I² statistic. For the outcome of mortality, we considered an I² value of more than 50% as indicative of substantial heterogeneity, and acknowledged this in our interpretation of the data.

Assessment of reporting biases

Had we identified 10 or more eligible trials for this update, we would have investigated the possibility of reporting biases, including publication bias, by assessing funnel plots for asymmetry where 10 or more studies reported on the same outcome (Egger 1997). We acknowledge that asymmetry could be due to publication bias or to a genuine relationship between trial size and effect size. We also searched for evidence of prospective trial registration and trial protocols for all studies included in the review.

Data synthesis

Where feasible, we analysed data using Review Manager 5 (Review Manager 2020). Where this was not feasible, we reported results narratively.

Subgroup analysis and investigation of heterogeneity

Had there been sufficient studies, we would have performed subgroup analysis based on the treatment regimen used to investigate the impact on both mortality and adverse effects.

Sensitivity analysis

We performed post hoc sensitivity analyses to investigate the effect of reporting bias in the Lin 1999 study, as well as on the effect of possible selection bias in the Afzali 2008 study (see Effects of interventions). These two studies also merited exclusion from sensitivity analyses because they were subject to bias based on lack of appropriate assessment of plasma paraquat levels on entry to the study, leading ultimately to our decision to present a sensitivity analysis for the outcome 'all‐cause mortality at final follow‐up: at hospital discharge' in summary of findings Table 1.

Summary of findings and assessment of the certainty of the evidence

We summarised our findings for all outcomes in summary of findings Table 1.

We assessed the certainty of the evidence for each outcome measure using the GRADE approach (Yan 2016). GRADE is used to assess the certainty of evidence according to the following five domains: imprecision, indirectness, inconsistency, risk of bias, and publication bias. The certainty of evidence assessments are provided in summary of findings Table 1 and were generated within the online tool GRADEpro GDT (GRADEpro GDT 2020).