Comparison of Efficacy and Safety of Artemisinin-Based Combination Therapies for Treating Uncomplicated Falciparum Malaria in Sub-Saharan African Countries: An Update on the Changes in Efficacy Using Network Meta-Analysis

Background: Several artemisinin-based combination therapies (ACT) are available to treat uncomplicated malaria in Africa. The present study aimed to assess the ranking of their efficacy and tolerance. Methods: A database of randomized controlled trials was retrieved from published papers. Network meta-analysis was used to compare efficacy on day 28 and day 42 after initia-tion of treatment. Age covariate effect on treatment outcome was assessed, and a modeling approach to reduce heterogeneity among trials was evaluated under the hypothesis of consistency in a meta-regression. Safety and adverse events were compared among different ACTs. A Bayesian analysis was performed to implement the consistency models using WinBUGS software. The results were compared to those of the frequentist approach using the R soft-ware. Results: Eighty-one articles, in which a total of 15 different ACTs were tested in more than 36,000 patients, were included. On day 28, dihydroartemi-sinin-piperaquine (DHPP) was more effective than artemether-lumefantrine (AL) before (odds ratio [OR], diarrhoea or headache post-treatment was significantly lower with DHPP than AL. Artesunate-mefloquine (ASMQ) was associated with a significantly lower prevalence of vomiting or nausea (OR, 0.80; 95% CI, 0.48 - 1.30) and headache (OR, 0.53; 95% CI, 0.40 - 0.68) compared to AL. On the contrary, vomiting and nausea occurred more frequently after fixed-dose artesu-nate-amodiaquine formulation (ASAQf) than with AL (OR, 1.45; 95% CI, 1.18 - 1.78). The risk of anaemia was higher with ASAQf and co-blistered ar-tesunate-amodiaquine (ASAQc) than with AL. There was no significant difference in risk of anaemia (P > 0.05) between AL and different formulations of ASAQ. Conclusions: Based on the available evidence, this study demon-strated the superiority of DHPP, followed by AL, among currently recommended ACTs in terms of efficacy and tolerance. Network meta-analysis could be an alternative analytical tool but needs more data input from therapeutic efficacy studies. The determination of the best available therapy requires data triangulation and data science.


Background
Since the World Health Organization (WHO) recommended artemisinin-based combination therapy (ACT) as the first-line treatment of uncomplicated Plasmodium falciparum malaria, there has been a growing interest in field-based randomized controlled trials. Most African countries adopted this drug policy during the 2000s, resulting in a large amount of published evidence-based data on the efficacy, tolerability, and safety of these antimalarial drugs. Antimalarial drug efficacy is usually assessed in therapeutic efficacy studies based on the standardized WHO protocol, initially published in 1994 and updated in 1996, 2003, and 2009 [1] [2] [3] [4]. The amount of data being generated in the field is ever-increasing, and the results of many, but not all studies, have been published. Because of this growth of evidence-based data, novel methods to collect, analyze, and summarize clinical data to identify the best available therapy have become indispensable.
Systematic reviews and meta-analyses have become a common approach to extracting relevant information from therapeutic efficacy studies based on either individual patient data or study level data (i.e., aggregate data) [5] [6] [7] [8].
The doses of ACT are variable, and for ASAQ, several different doses and formulations have been used. Most ACTs are administered as a single daily dose for three days. The fixed-dose AL is an outstanding exception, requiring twice daily doses for three days (a total of six doses administered over three days, with an 8-hr interval between the first and second dose). ASAQ is administered as a once daily fixed-dose combination (FDC) or twice daily doses for three days [18] [19]. ASAQ dose varies due to the availability of three formulations: non-fixed dose combinations (NFDC), which may be either loose NFDC (ASAQl) or co-blistered NFDC (ASAQc), and fixed-dose combinations (ASAQf) [20]. The dose effect should be taken into consideration in the comparison of antimalarial drugs. AL, ASAQ, and DHPP have been found to be safe for use in children, but a non-negligible number of adverse events have been associated with the intake of ACT [5]. In an earlier work [21], DHPP was identified as the best available ACT to treat uncomplicated falciparum malaria in terms of efficacy and safety. At the point where the public health community needs a comprehensive, reliable and timely information to slow the emergence of antimalarial drug resistance and maintain the efficacy of antimalarial treatments, there is a need for an update on the efficacy, safety, and tolerability of available ACTs and a strategy to reduce heterogeneity in the evaluation of treatment by considering baseline characteristics of the study. Given an increasing number of ACTs and published studies on their safety profile in children, the aim of the present study was to investigate the modifiers of treatment effect by analyzing treatment by covariate interactions and adverse events associated with each ACT.

Data Collection and Extraction of Effect Modifiers
Data were extracted from a recent database of randomized clinical trials involv-

Antimalarial Drugs
Drugs were assigned numbers from 1 to 13 as in our previous study [21]. With two additional formulations of ASAQ found in the database, the numbering for drug combinations was extended to 15, i.e., AL was assigned treatment number

Measures of Treatment Effect and Statistical Analysis
The primary outcome, i.e., the proportion of ACPR, was combined and pre-  [22]. In addition to that model, a meta-regression analysis was developed using the existing code, with age covariate to assess the change in efficacy of the different interventions. All analyses were based on a random effect model to account for different end-point evaluations. Heterogeneity was explored using I 2 statistics and test of inconsistency. To account for heterogeneity in the patient populations, a dummy random variable was defined as 1 if the patient was a child less than 15 years and 0 for other patient populations. In addition, meta-regression was used to assess the changes in treatment effect with the same heterogeneity variance assumed for every comparison and the effect of covariates on treatment effect. The age covariate was used to fit an NMA-regression model. Studies with missing covariates were excluded. AL was the main comparator drug.

Choice of Prior Probability Distribution
All treatment effect was given uniform priors between −10 and 10. This strategy avoids numerical "traps" encountered when running the model with a uniform prior between 0 and 20 [23]. Prior age covariate effect distribution was given a flat normal distribution N (0, 10 −5 ). Vague prior between-trial standard deviation was a uniform distribution between 0 and 20.

Treatment Ranking
To measure how the treatment was comparatively better than another treatment, P-scores were used and averaged over different treatments [24]. The method was found to be comparable with the surface under the cumulative ranking curve (SUCRA) [25]. All analysis was carried out using WinBUGS and the R package net meta [26].

Analysis of the Efficacy Trials on Day 28
The updated malaria network evidence illustrated in Figure 1   In the present study, AL and DHPP, followed by AL and ASAQf, were compared most frequently. *AQSP, a non-ACT combination, was included for comparison.

Clinical Efficacy on Day 42
A total of 20 studies assessed clinical efficacy on day 42 with two or more of the following combinations: AQSP, ASAQ, AL, ASMQ, ASPY, ASSP, and DHPP (Table 3). ASCD was excluded from analysis because this ACT is not recommended, and its production had ceased due to serious adverse effects associated with dapsone in African patients with glucose-6-phosphate dehydrogenase deficiency. By setting AL as the comparator, DHPP was 1.6-fold more efficacious than AL (OR = 1.65; 95% CI, 1.06 -2.56; P < 0.05) ( Table 3). On day 42, the difference in therapeutic efficacy between AL and ACTs other than DHPP was not statistically significant (P > 0.05).

Secondary Outcome Results
The safety of AL, as compared with other ACTs, was assessed. Adverse events were extracted from 43 studies. Thirty-eight of 81 studies did not report drug tolerance. The frequency of adverse effects and studies in which they were assessed are presented in Supplementary Table 3. The adverse events reported in different studies are summarized in Table 4 and Table 5. Adverse events were heterogeneous among studies. The common adverse events (≥1/100 and <1/10) among patients receiving AL included the following: anorexia, vomiting, anaemia, diarrhoea, vomiting, and abdominal pain (Table 4). Cough was the only very commonly reported adverse events (≥1/10) of AL and DHPP. The respiratory and gastrointestinal tracts were the most commonly affected organs, constituting 35% and 33% of all reported adverse events in AL-treated children, respectively. The number of events was very low with new alternative drugs (Table  4), while adverse events were more prevalent with ASSMP (Table 5).  For studies comparing AL with other ACTs, the risk of adverse event was compared (Table 6). Descriptive statistics for all studies involved in this comparison can be found in Supplementary Table 2. Comparisons showed that the risk of having a headache (relative risk [RR], 0.80; 95% CI, 0.67 -0.96), diarrhoea (RR 0.85; 95% CI, 0.74 -0.97), and cough (RR, 0.91; 95% CI, 0.85 -0.98) was significantly lower (P < 0.05) with DHPP than with AL. However, treatment with DHPP resulted in an increased risk of anaemia than with AL (raw RR, 2.23; 95% CI, 1.64 -3.02). On the other hand, ASMQ was associated with less vomiting and nausea (OR, 0.80; 95% CI, 0.48 -1.30), anaemia (OR 0.8; 95% CI, 0.74 -0.99), and headache (OR 0.53; 95% CI, 0.40 -0.68), compared to AL (Table 6).

Discussion
The efficacy and tolerance of ACTs were updated using the existing database.
The primary outcome, i.e., the rate of ACPR, was extended beyond 28 days and the mixed effect model was used to account for the variability of children's responses to different ACTs. Different doses of ASAQ were also accounted for. For instance, in the selected studies, the efficacy of ASAQ FDC was assessed as it was found that this combination ensures optimal dosing and provides higher treatment efficacy.  Each treatment was taken as the reference and compared to AL in the same trials that compared both treatments. Data on adverse events occurring at any time during the follow-up period from 43 published studies. *AQSP, a non-WHO-recommended, non-ACT combination, was included for comparison since several African countries have adopted its use during the transition period in the 2000s before full implementation of ACT-based antimalarial drug policy.
Age covariate was used in a meta-regression analysis to assess the benefit of each drug compared to AL while accounting for variation among trials in terms of protocol, patients, and follow-up period. The rate of adverse events was compared between ACTs to assess the safety and tolerability of different pairs of treatment arms.
The inclusion of age covariate did not modify the differences between treatments, as found in previous works [21] [27]. Compared to AL, treatment with DHPP and ASMQ resulted in a statistically significant increase (P < 0.05) in the chance of recovery, i.e. ACPR, as well as a lower probability (P < 0.05) of encountering adverse events in African children. The present analysis showed similar conclusions to a recent study that compared three combinations, DHPP, AL, and ASMQ, in which it was found that the treatment success rate was higher with DHPP compared to AL [28]. In addition, children who received one of these two combinations (DHPP or ASMQ) experienced a lower prevalence (P < 0.05) of cough, weakness, abdominal pain, or loss of appetite compared with  [29]. However, AL efficacy is still very high, with the rate of ACPR above 95% in most clinical studies conducted in Africa [30]. Indeed, despite the observed mild to moderate adverse events reported in the eligible studies, AL has a good tolerability [31] [32].
In this work, ASSMP and ASPY were the less studied novel ACTs. Adverse events were reported more frequently with ASSMP than with other ACTs. There is another novel non-WHO-approved ACT, such as AMPP, which, compared to AL, showed high cure rates in the ITT-analysis on day 28 (96.6% vs. 95.0%) and day 42 (94.4% vs. 93.1%), respectively [33]. However, more evidence-based data on the new alternative ACT are required in sub-Saharan Africa to evaluate their efficacy and safety in comparison to the currently WHO-recommended ACTs.
This study has several limitations. First, the study did not re-assess the selection bias for the evidence since this procedure was performed in our previous work [21]. Secondly, the performance of NMA proved to be difficult for one study reporting the outcome on day 42 due to inconsistency in treatment effect and inconsistent variances in a multi-arm study [34]. This methodological problem related to the difficulty in performing estimation with meta-regression analysis of a small number of studies led to an unreliable ranking of treatment efficacy. To overcome this problem, different treatments were ranked using a frequentist approach. Thirdly, because only a small number of studies were available, the rates of adverse events were compared using a fixed effect model. Moreover, a limited amount of evidence with new alternative ACTs did not allow a more precise estimate of treatment effect.

Conclusion
This study suggests that continued use of ACTs for treating uncomplicated malaria in Africa is warranted, but more attention should be paid to mild to moderate adverse events that have been reported frequently in different clinical studies and to potential ACT resistance. Indeed, it is essential to understand the mechanisms involved in the acquisition of artemisinin resistance by P. falciparum to adapt malaria treatment policies and propose new therapeutic strategies.
Understanding the mechanisms of artemisinin resistance, regular updates on the epidemiology of drug-resistant malaria, and surveillance are critical components of the overall strategy to prevent the expansion of artemisinin-resistant P. falciparum. Country-specific meta-analyses are also needed to ensure that ACTs remain effective in Africa.