Temporal distribution of Plasmodium falciparum recrudescence following artemisinin-based combination therapy: an individual participant data meta-analysis

Abstract

Background: The duration of trial follow-up affects the ability to detect recrudescent infections following anti-malarial treatment. The aim of this study was to explore the proportions of recrudescent parasitaemia as ascribed by genotyping captured at various follow-up time-points in treatment efficacy trials for uncomplicated Plasmodium falciparum malaria.

Methods: Individual patient data from 83 anti-malarial efficacy studies collated in the WorldWide Antimalarial Resistance Network (WWARN) repository with at least 28 days follow-up were available. The temporal and cumulative distributions of recrudescence were characterized using a Cox regression model with shared frailty on study-sites. Fractional polynomials were used to capture non-linear instantaneous hazard. The area under the density curve (AUC) of the constructed distribution was used to estimate the optimal follow-up period for capturing a P. falciparum malaria recrudescence. Simulation studies were conducted based on the constructed distributions to quantify the absolute overestimation in efficacy due to sub-optimal follow-up.

Results: Overall, 3703 recurrent infections were detected in 60 studies conducted in Africa (15,512 children aged < 5 years) and 23 studies conducted in Asia and South America (5272 patients of all ages). Using molecular genotyping, 519 (14.0%) recurrences were ascribed as recrudescent infections. A 28 day artemether-lumefantrine (AL) efficacy trial would not have detected 58% [95% confidence interval (CI) 47-74%] of recrudescences in African children and 32% [95% CI 15-45%] in patients of all ages in Asia/South America. The corresponding estimate following a 42 day dihydroartemisinin-piperaquine (DP) efficacy trial in Africa was 47% [95% CI 19-90%] in children under 5 years old treated with > 48 mg/kg total piperaquine (PIP) dose and 9% [95% CI 0-22%] in those treated with ≤ 48 mg/kg PIP dose. In absolute terms, the simulation study found that trials limited to 28 days follow-up following AL underestimated the risk of recrudescence by a median of 2.8 percentage points compared to day 63 estimates and those limited to 42 days following DP underestimated the risk of recrudescence by a median of 2.0 percentage points compared to day 42 estimates. The analysis was limited by few clinical trials following patients for longer than 42 days (9 out of 83 trials) and the imprecision of PCR genotyping which overcalls recrudescence in areas of higher transmission biasing the later distribution.

Conclusions: Restricting follow-up of clinical efficacy trials to day 28 for AL and day 42 for DP will miss a proportion of late recrudescent treatment failures but will have a modest impact in derived efficacy. The results highlight that as genotyping methods improve consideration should be given for trials with longer duration of follow-up to detect early indications of emerging drug resistance.

Keywords: Distribution; Efficacy; Follow-up; Malaria; Plasmodium falciparum; Recrudescence.

Fig. 1

Flow chart showing selection of clinical trials and participants in the study. AL = artemether-lumefantrine; DP = dihydroartemisinin-piperaquine; ASMQ = artesunate-mefloquine; ASAQ = artesunate-amodiaquine, and n = number of recrudescences; PCR = Polymerase Chain Reaction; WWARN = WorldWide Antimalarial Resistance Network; ACT = artemisinin-based combination therapy

Fig. 2

Observed time-to-recrudescence for ACT stratified by region and duration of follow-up. The y-axis depicts the time for the recrudescent infection to reach microscopic limit of detection (50 parasites/µL). In Africa, the distribution is shown for children < 5 years age, whereas in Asia/ S. America, data are shown for patients of all ages. Each dot represents an observed recrudescence. AL = artemether-lumefantrine; DP = dihydroartemisinin-piperaquine; ASMQ = artesunate-mefloquine; ASAQ = artesunate-amodiaquine, and n = number of recrudescences. Only data from studies with at least three molecular markers are shown in the graph. For DP (Asia), the graph depicts recrudescences observed in studies from Asia and S. America combined

Fig. 3

Fractional polynomial estimates of cumulative and instantaneous hazard for AL and DP. The cumulative baseline hazard estimated from Cox model (adjusted for age, baseline parasitaemia and mg/kg dosage of partner drug) together with fractional polynomial smoother (left panel). The temporal trend of observing recrudescence during the follow-up period, estimated by the instantaneous baseline hazard function (right panel). Data in Africa was restricted to children < 5 years whereas patients of all ages were included in Asia and S. America. Data from S. America were grouped with Asia. All studies used at least 3 locus genotypes (the usage of microsatellites was considered as a separate locus). Under-dosed was defined as total piperaquine dose ≤ 48 mg/kg (those receiving > 48 mg/kg defined as not under-dosed). AL = artemether-lumefantrine; DP = dihydroartemisinin-piperaquine. The equations for the cumulative baseline hazard functions derived using the fractional polynomial smoothing are presented in the Additional file 1

Fig. 4

Distribution of recrudescent infection for artemether-lumefantrine and dihydroartemisinin-piperaquine. Dotted line represents distribution derived from all data. Solid line represents the distribution derived by including data up to day 42 (observations beyond day 42 were censored on day 42). For Africa, the data included children < 5 years old whereas for Asia, no age restriction was applied

Fig. 5

Absolute overestimation in efficacy due to sub-optimal follow-up duration (compared to day 63 estimates). Overestimation of efficacy relative to day 63 estimates for AL and DP regimen. In areas of high transmission, there was a median of 15% new infections in the DP arm and 30% new infections in AL arm representing. In areas of low transmission, there was a median of 8% new infections in the DP arm and 15% new infections in the AL arm. Simulation assumed 500 patients per treatment arm and was repeated for 1,000 runs. AL = artemether-lumefantrine; DP = dihydroartemisinin-piperaquine (See Additional file 1 for further details on the functions used for simulation)

Fig. 6

Simulation study comparing hazards ratio of recrudescence for AL against DP in children < 5 years in Africa. A There were a median (across 1000 simulation runs) of 8% new infections in DP arm and 15% in AL arm representing areas of low transmission. B There were a median of 15% new infections in DP arm and 30% in AL arm representing areas of high transmission. In both simulation settings, approximately 4% recrudescence was observed by day 63 on AL and DP arm. Data for AL was simulated based on the estimated hazard function of recrudescence in children < 5 years Africa, and for DP regimen the data was simulated based on the hazard function for those who received piperaquine dose greater than 48 mg/kg (See Additional file 1 for the functions used for these simulations). The dotted horizontal line shows the line of no effect (hazard ratio = 1). The simulation assumed 500 patients per treatment arms. AL = artemether-lumefantrine; DP = dihydroartemisinin-piperaquine