Electronic searches of the National Library of Medicine's MEDLINE database, Current Contents database and manual searches of selected specialty journals were performed to identify all the pertinent literature. MEDLINE database engines (Ovid, PubMed and GratefulMed) were used with the keywords "rectal", "artemisinin", and "treatment". The search was further refined with the words "artesunate", "artemether", "artemisinin", "dihydroartemisinin", "suppositories", and "rectocaps" from 1980 to March 2006. Reference lists from qualitative topic reviews and published clinical trials in English were also searched. We attempted to obtain the original individual patient data from all studies, regardless of publication status.

Selection criteria for inclusion of studies were clinical trials that assessed the efficacy of a rectal artemisinin-based preparation where individual patient data (inclusion/exclusion criteria, age, parasitological status at baseline and parasitology post treatment) were available. Comparative and non-comparative studies were included, regardless of study design, geographical area, patient age group, disease severity or the artemisinin derivative used. Safety data and information on dosage was specifically requested from the individual study investigators if this was not provided or evident from a publication. Data extraction was conducted by two investigators (MP, IR) for efficacy data and two investigators (MW, MG) for safety.

The main focus of the review was on the assessment of parasite reduction and clinical response of patients during the first 24 hours following treatment. This was considered most appropriate, given the intended indication as emergency pre-referral treatment, where ameliorating disease progression within 24 hours, i.e. while the patient is being transported to a clinic or hospital, is likely to be most important.

All efficacy definitions used are consistent with the scientific literature. Parasite reduction ratio (PRR) at 12 and 24 hours was assessed as the percent reduction of parasitaemia at 12 and 24 hours from baseline parasitaemia. Parasitological success was defined as the absence of clinical deterioration from baseline and a PRR at 24 hours of ≥ 90%. In the analysis (see further below) both the continuous variable PRR and the binary variable of ≥ 90% parasite reduction were used, the former to define efficacy at 12 hours, and the latter to define efficacy at 24 hours.

Because definitions of parasite and fever clearance times differed from study to study, these time-to-event variables were re-derived from the serial parasite density estimations and temperatures for individual patients. Parasite clearance and fever clearance time were defined as the time at which the first negative blood smear or normal temperature (<37.5°C axillary or <38.0°C oral) was recorded. The effect of consolidation treatment on recrudescence of the infection during follow-up was evaluated from baseline to the reappearance of parasitaemia.

Except in two trials, laboratory markers of safety (including haematological and biochemical indices) were not available. Safety analyses were consequently restricted to clinical descriptions of reported adverse events. In the absence of prospective standardised methods for defining, assessing, reporting and classifying adverse events across all trials, and due to inherent difficulties in clinically distinguishing drug side-effects from manifestations of malarial infection, principal investigators who contributed data were asked to re-review individual patient data retrospectively and reassess all reported adverse events. Ideally this was performed directly from case record forms where archived data were accessible. Each reported event was thus re-classified by the clinical investigator as being either "unlikely", "possibly", "probably" or "definitely" due to the treatment. Those events considered possibly, probably or definitely drug-related were thereafter re-classified as "potentially drug-related" for the purposes of the pooled analysis.

Original data from different studies were merged into a master data set. In the analysis mainly hierarchical models were used, with treatment arm as the second level of clustering. In some cases within-study comparisons of different treatment arms was possible and methods for meta-analysis were then accordingly applied in the analysis of PRR. A hierarchical mixed model applying the DerSimonian & Laird method was used with the estimate of heterogeneity being taken from the inverse variance fixed-effect model 19.

Most of the analyses, however, were based on comparisons of treatment arms from different studies, here called pooled analyses, and conducted as follows: For PRR, linear mixed effect models with random intercepts were applied. In the analysis at 12 hours, an identity link (normal distribution) was used. At 24 hours a large proportion of the PRR values for the artemisinins were close to 100% at 24 hours, and these were then categorised into binary observations (PRR >90% versus <90%). A logistic link was then used in the analysis. It should be noted that crude mean values of PRR at 24 hours using the identity link are also given for description. We systematically examined the effect (stepwise backward elimination of covariates not reaching a significance level of p = 0.05) of the following covariates within all analyses: baseline parasitaemia, age, region, total treatment dose provided within the initial 12 and 24 hours, and severity of disease as defined by the evaluating clinician.

Time-to-event analyses, including time to parasite and fever clearance and the efficacy of the consolidated treatment in suppressing parasitaemia post treatment (reappearance of parasites), were represented using crude Kaplan-Meier plots (these are provided as graphs only where the difference in estimates was significant). Hazard-ratios were estimated using a non-hierarchical Cox regression model applying the same stepwise backward elimination of covariates as above. Intra-rectal treatments were compared to parenteral treatment when followed by the same consolidation treatment in one analysis and intra-rectal treatments followed by different consolidation treatments were compared in a second analysis. Time to parasite clearance, age, region, disease severity, parasitaemia at 72 hours (often the start of consolidation treatment) were assessed as covariates through stepwise backward elimination of covariates.

All statistical computations were performed with Stata for Windows (version 8 and 9.2, Stata Corporation, College Station, Texas-USA).

A total of 27 studies were identified in patients using the search criteria; 4 studies were in healthy normal volunteers 11151618. Twelve studies were not included in the analysis: 5 trials were on pharmacokinetics – 8 to 15 patients per trial, total of 59 patients 1417202122 with no efficacy data available. This left 7 trials, 242 patients with uncomplicated malaria23242526 and 185 patients with severe and complicated malaria treated with rectal artemisinins23272829 that were not available to be included in our analyses.

Individual patient data were available from 15 clinical trials (Table 1). There were 4 randomised, controlled, clinical trials with parenteral quinine as comparator: Molyneux1997–8 30, Barnes1998 30, Phuong1992–5 31, Aceng2002–3 32, and 6 trials in which the comparator was another artemisinin given parenterally or orally in moderately severe, hyperparasitaemic and severe patients: Krishna1996 10, Looareesuwan1996 (unpublished), VanVugt1997–9 (unpublished), Vinh 1997–9 33, Hien1998 (unpublished), Karunajeewa2003–4 34. The remaining trials used a different dosage or treatment regimen with the same artemisinin-based rectal preparation as a comparator: Looareesuwan1995 35, Looareesuwan2000 (unpublished), Than1998 (unpublished). There were 2 non- comparative trials – Bhatt1994–5 36, Karunajeewa2001 37. We were provided with an additional 10 individual patient observations not included in one published trial (Phuong1992–5) 31 which completed with fewer patients than planned due to recruitment difficulties, but where the team continued to collect clinical descriptions of severe malaria in children, using rectal artemisinins which the hospital preferred to quinine. The data from these 10 cases were added to the pooled analyses.

Altogether the studies included in this analysis enrolled a total of 1167 patients in 37 separate treatment arms. Five patients who simultaneously received rectal artesunate and quinine were excluded, leaving a total of 1162 patients included in the analysis. Of these 786 had been treated with rectal administration of an artemisinin derivative and 376 with a comparator drug which was either a parenteral artemisinin (236) oral artesunate (17) or parenteral quinine (123). The majority of included patients treated with a rectal artemisinin were from mainland South East Asia (487) or Papua New Guinea (89), with relatively fewer patients from Africa (210). Thirty-one percent of patients were children under 5, 11.8% children aged 6–10, 21.3% adolescents aged 11–20 years and 36% adults over 20 years of age. Treatment exposure information is provided in Table 2.