Baseline data of parasite clearance in patients with falciparum malaria treated with an artemisinin derivative: an individual patient data meta-analysis

Background Artemisinin resistance in Plasmodium falciparum manifests as slow parasite clearance but this measure is also influenced by host immunity, initial parasite biomass and partner drug efficacy. This study collated data from clinical trials of artemisinin derivatives in falciparum malaria with frequent parasite counts to provide reference parasite clearance estimates stratified by location, treatment and time, to examine host factors affecting parasite clearance, and to assess the relationships between parasite clearance and risk of recrudescence during follow-up. Methods Data from 24 studies, conducted from 1996 to 2013, with frequent parasite counts were pooled. Parasite clearance half-life (PC1/2) was estimated using the WWARN Parasite Clearance Estimator. Random effects regression models accounting for study and site heterogeneity were used to explore factors affecting PC1/2 and risk of recrudescence within areas with reported delayed parasite clearance (western Cambodia, western Thailand after 2000, southern Vietnam, southern Myanmar) and in all other areas where parasite populations are artemisinin sensitive. Results PC1/2 was estimated in 6975 patients, 3288 of whom also had treatment outcomes evaluate d during 28–63 days follow-up, with 93 (2.8 %) PCR-confirmed recrudescences. In areas with artemisinin-sensitive parasites, the median PC1/2 following three-day artesunate treatment (4 mg/kg/day) ranged from 1.8 to 3.0 h and the proportion of patients with PC1/2 >5 h from 0 to 10 %. Artesunate doses of 4 mg/kg/day decreased PC1/2 by 8.1 % (95 % CI 3.2–12.6) compared to 2 mg/kg/day, except in populations with delayed parasite clearance. PC1/2 was longer in children and in patients with fever or anaemia at enrolment. Long PC1/2 (HR = 2.91, 95 % CI 1.95–4.34 for twofold increase, p < 0.001) and high initial parasitaemia (HR = 2.23, 95 % CI 1.44–3.45 for tenfold increase, p < 0.001) were associated independently with an increased risk of recrudescence. In western Cambodia, the region with the highest prevalence of artemisinin resistance, there was no evidence for increasing PC1/2 since 2007. Conclusions Several factors affect PC1/2. As substantial heterogeneity in parasite clearance exists between locations, early detection of artemisinin resistance requires reference PC1/2 data. Studies with frequent parasite count measurements to characterize PC1/2 should be encouraged. In western Cambodia, where PC1/2 values are longest, there is no evidence for recent emergence of higher levels of artemisinin resistance. Electronic supplementary material The online version of this article (doi:10.1186/s12936-015-0874-1) contains supplementary material, which is available to authorized users.


Background
Parasite clearance is a robust measure of the efficacy of anti-malarial drugs, which has been used particularly to measure the pharmacodynamic effects of artemisinin derivatives [1]. Initial studies conducted in patients with severe malaria employed frequent parasite counting to characterize clearance profiles, and these demonstrated that artemisinin derivatives cleared parasitaemia more rapidly than quinine [2]. More recently, frequent parasite counting has been used to characterize artemisinin susceptibility in vivo [3], and to validate molecular markers [4,5] and in vitro assays for detection of artemisinin resistance [6]. Parasite clearance following artemisinin treatment is influenced by a number of factors other than parasite susceptibility, including host immunity, initial parasite biomass and partner drug efficacy. It is therefore essential to control for such potential confounding factors in order to identify temporal changes in parasite clearance resulting from reduced anti-malarial drug susceptibility.
The WorldWide Antimalarial Resistance Network (WWARN) Parasite Clearance Estimator (PCE) [7] was developed to automate and standardize analysis of frequent parasite count data. This tool is freely available online [8] and provides an automated report for each patient. The derived measure, parasite clearance half-life (PC 1/2 ), generated by the PCE reflects the extent to which ring-stage parasites are killed and removed from the circulation, and is currently considered the most reliable measure of parasitological responses to treatment with artemisinin or its derivatives [5,[9][10][11][12][13][14][15][16]. This standardized approach to PC 1/2 measurement allows comparison in space and time of artemisinin resistance, which manifests as a slow parasite clearance rate in patients. Within the WWARN framework, investigators who obtained frequent parasite count data have joined several study groups [17] to evaluate this metric. This pooled analysis presents reference parasite clearance estimates stratified by geographic location, treatment and study population, and explores the relationship between parasite clearance measures and the risk of recrudescent infection (treatment failure). The effects of different sampling strategies on clearance estimates have been published separately [18].

Data acquisition
Any study involving patients with uncomplicated falciparum malaria, treated with either artemisinin combination therapy (ACT) or oral artesunate monotherapy, in which peripheral parasitaemia was measured at least twice daily in the first 3 days after starting treatment, was eligible for inclusion in this pooled analysis. In addition, the minimum data required were enrolment date, patient age, drug treatment, study location and characteristics, and details of the parasite counting method. Studies with frequent parasite counts were identified using literature reviews and existing collaborations within WWARN. Principal investigators were subsequently approached to participate in this study group [19]. The datasets uploaded to the WWARN repository were standardized using the WWARN Data Management and Statistical Analysis Plans for clinical data [20] and pooled into a single database of quality-assured individual patient data.
Parasite inclusion criteria, counting methods and blood sampling schedules were different among studies; for a detailed description see Additional file 1: Table S1.

Statistical analysis Definitions
As a measure of transmission intensity, malaria endemicity estimates were obtained for study sites and year from the Malaria Atlas Project [21]. Anaemia was defined according to WHO guidelines [22], (i.e., haemoglobin concentration cut-offs for moderate anaemia were 10 g/dL in children <5 years of age and 11 g/dL in older patients, and for severe anaemia were 7 and 8 g/dL, respectively). For studies where haematocrit only was measured, the following relationship was used to estimate haemoglobin: haematocrit (%) = 5.62 + 2.60 × haemoglobin (g/ dL) [23]. Nutritional status of children aged <5 years was assessed by the weight-for-age indicator using the igrowup package developed by the WHO [24].

Analysis of parasite counts
PC 1/2 was estimated only for patients with sufficient parasite counts defined as sampling at least 12-h in the first 48 h (a maximum of a 16-h gap between any two measurements, as a 2-h window on each side was allowed) and at least 24-h sampling (maximum 28-h gap) after 48 h until parasite clearance [18]. The following deviations from this rule were accepted as they were deemed not to have substantial effects on the PC 1/2 estimate [18]: sampling was not performed until parasite clearance but the last recorded parasitaemia was <100 or <1000 parasites/ µL with at least five positive parasite counts available; a longer gap was observed between a set of measurements but there were at least two positive parasite counts directly after the gap, or a zero count was recorded after the gap and the last recorded parasitaemia before the gap was either <100 or <1000/µL and at least five positive parasite count measurements were available before the gap. PC 1/2 was calculated by the PCE [7] for each patient (variable called slope_half_life in the output files), based on the linear segment of the decline in the log-transformed parasitaemia-time profile. A lag-phase (an initial, flat part of the parasitaemia-time profile which precedes the log-linear decline) and a tail (a levelling out in the parasitaemia-time profile which follows the log-linear decline), if present, are identified by the PCE automatically.
Reliability of PC 1/2 estimates was assessed by (a) the standard deviation of residuals from the final linear model used to estimate PC 1/2 ; (b) the duration of the lag phase (as a long lag phase is very unlikely if an artemisinin derivative is given and absorption is adequate); (c) the number of positive parasite counts used in the estimation; (d) pseudo-R 2 statistics; and, (e) the width of the 95 % confidence interval around the PC 1/2 .
Pseudo-R 2 is a measure of goodness of fit of the final model and is provided by the PCE tool. Low values of pseudo-R 2 indicate that the predicted values from the polynomial model are far from the measured parasitaemias. Pseudo-R 2 is calculated from the fitted values of the final linear model used to estimate the PC 1/2 (after exclusion of the lag phase and tail) and the observed log-parasitaemias, excluding zero counts.

Parasite clearance and clinical covariates
Factors affecting PC 1/2 were investigated in the random effects regression model (to account for study site heterogeneity) with PC 1/2 being modelled after log transformation. Separate analyses were performed in artemisinin-resistant and artemisinin-sensitive areas. The resistant areas were defined as locations in which delayed parasite clearance had been reported previously [3, 9-11, 14-16, 25-28] (i.e., western Cambodia, western Thailand after 2000, southern Vietnam, southern Myanmar), while the sensitive areas were defined as all other locations.
In studies which randomized treatment arms to 2 and 4 mg/kg/day artesunate doses, meta-analysis of the differences in mean log-transformed PC 1/2 between treatment arms was performed using a fixed effects model using the inverse variance method. Heterogeneity was evaluated by I 2 [29].

Analysis of treatment outcome
The risk of recrudescence was assessed by survival analysis using WHO definitions of therapeutic efficacy outcome [30]. Patients with no PCR results were excluded from the treatment outcome analysis. Cox regression model with random effects in the form of frailty parameters were used to adjust for study site effects [31]. The proportional hazard assumption was tested based on Schoenfeld residuals [32]. PC 1/2 , presence of a lag phase, duration of lag phase and presence of a tail were evaluated as possible predictors of outcome, together with all other baseline clinical and treatment characteristics.
Covariates for the final regression models (for treatment outcome and PC 1/2 ) were selected on the basis of the likelihood ratio test and examination of residuals.
Relationship between the independent variable and continuous covariates such as age and parasitaemia was examined using fractional polynomials. All statistical analyses were performed using Stata 13.0.

Data summary
Data from 9318 patients enrolled from 1996 to 2013 in 24 studies [3, 9-15, 26, 33-43] (Additional file 2: Table  S2 Among hyperparasitaemic patients (Study ID 1), 882 of 3393 (26 %) patients were excluded from analysis because of one or more of the following: severe malaria (n = 108), slow parasite clearance and administration of rescue treatment with intravenous or intramuscular artesunate (n = 642), blood transfusion before clearance of parasites (n = 215), or incomplete treatment information (n = 19). Table 1 shows the demographic and clinical parameters of patients with uncomplicated Plasmodium falciparum malaria who were included in this analysis.

Estimates of parasite clearance
Among 8536 patients, 6975 (82 %) had sufficient parasite counts taken for PCE estimation of PC 1/2 . The majority  IDs 4,20,21) and from the study with hyperparasitaemic patients (36 %, Study ID 1), which for 10 years routinely recorded parasitaemia every 6 h until clearance.
Only two positive parasite counts were used to estimate PC 1/2 in 878 patients, either because only two positive measurements were available (n = 844) or measurements were excluded as being part of the lag or tail phases (n = 34). For these profiles with only two positive parasite counts available, the PCE replaces the first zero count with the detection limit [7] and the resulting PC 1/2 estimate clearly overestimates the true PC 1/2 . However, the estimated PC 1/2 was still considered informative as 25 % (214/844) of these profiles had estimated values <2 h and 73 % (618/844) had estimated values <3 h, indicating that parasite clearance in these patients was rapid and thus provided no evidence for artemisinin resistance. Of the remaining patients, 73 % (165/226) with an estimated PC 1/2 >3 h had an initial parasitaemia <10,000 parasites/µL and 77 % (175/226) had parasite counts measured using one of the twice-daily schemes. For 21 % (1489/6975) of profiles, a non-zero lag phase was estimated with median (range) duration of 6 (1.5-60) hours, with 6 % (90/1489) having a lag phase duration >12 h.
The median (range) goodness of fit statistic, pseudo-R 2 , was 0.938 (−198 to 0.999), with 89 % (6197/6975) of profiles having a pseudo-R 2 >0.8. Only 0.9 % (65/6975) of profiles had a negative pseudo-R 2 , indicating that the model was not a good representation of the data.
The 95 % confidence interval (CI) for the estimated PC 1/2 was wide for 11 % (740/6975) of profiles; the 95 % CI either included negative values or the upper limit was greater than twice the PC 1/2 estimate. Of these, 70 % (519/740) were for patients with only two positive parasite counts available.
For the distribution of PC 1/2 by location, treatment and study year see Additional file 3. See Additional file 4: Table S3 for summaries of PC 1/2 and other parasitological measures by location and treatment and Additional file 5: Table S4 for proportion of profiles with PC 1/2 longer than 3, 4, 5 and 6 h.

Areas with slow parasite clearance
Delayed parasite clearance was observed at sites in Cambodia, Thailand, Myanmar, and Vietnam. For all treatments and locations, the longest PC 1/2 were observed in three western Cambodian sites: Pailin, Tasanh and Pursat where data from 2007 to 2012 were available; study median PC 1/2 ranged from 5.6 to 6.7 h, and the proportion of PC 1/2 >5 h ranged from 61 to 80 %. Importantly, no significant trend of increasing PC 1/2 was observed at these sites over that time interval. At the two other Cambodian sites, Ratanakiri and Preah Vihear, parasite clearance was significantly faster (p < 0.001; median PC 1/2 of 3.0 and 3.8 h, and proportion of PC 1/2 >5 h of 4 and 22 %, respectively) and also different between these two sites (p = 0.011).

Areas with rapid parasite clearance
Among studies in areas with artemisinin-sensitive parasites (n = 3208), patients who received AS 2 mg/kg (with or without partner drugs) or the standard six-dose AL regimen had longer PC 1/2 values compared to patients who received AS 4 mg/kg (with or without partner drugs) by 7.3 % (95 % CI 1.9-12.9, p = 0.007) and 7.4 % (95 % CI 3.8-11.2, p < 0.001), respectively (Fig. 3). These comparisons are adjusted for study site and study design characteristics which affect PC 1/2 estimates: (a) patients with twice-daily sampling (Study IDs 4, 13, 20, 21) had 16.2 % (95 % CI 7.6-25.6) longer PC 1/2 compared to those with more frequent schedules (p < 0.001); (b) patients with insufficient number of data points to estimate lag phases had 31 % (95 % CI 26-37) longer PC 1/2 than those with sufficient data (p < 0.001). Since patients with very low initial parasitaemias and short PC 1/2 are excluded from the analysis because of insufficient data, this creates a negative association between the initial parasitaemia and PC 1/2 (Fig. 4). Mean PC 1/2 was estimated to decrease by 16 % (95 %CI 15-18) per tenfold increase in parasitaemia.
In artemisinin-sensitive areas, profiles with a lag phase had 6.4 % (95 % CI 3.6-9.1) shorter PC 1/2 compared to profiles without a lag phase (p < 0.001, adjusted for all the above factors). No association was observed between the duration of the lag phase and PC 1/2 among 687 patients with a non-zero lag phase (p = 0.220, adjusted for the above factors).

Initial parasitaemia
Studies differed in their admission parasitaemia inclusion criteria. It was assumed that the log-transformed initial parasitaemias followed a truncated normal distribution with lower and upper truncation consistent with the inclusion criteria. In all but five studies, there was no evidence against this assumption of truncated normality (p values ranged from 0.17 to 0.98); the exceptions were four multi-centre and/or multi-country studies (Study IDs 4, 21, 22, 23) and a study with two distinct age groups (Study ID 20; young children weighing 12-20 kg and older children weighing 20-40 kg). For these five studies, there was clear evidence of a multimodal distribution of initial parasitaemias.

Areas with slow parasite clearance
In areas with previously documented slow parasite clearance rates, no significant association between PC 1/2 and patient age was observed. After adjusting for study design factors, admission gametocytaemia was associated with an 11.1 % (95 % CI 5.5-16.9, p < 0.001, n = 3574) increase in PC 1/2 , and temperature >37.5 °C was associated with a 7.3 % (95 % CI 1.3-13.8, p = 0.017, n = 1491) increase in PC 1/2 . The relationship between PC 1/2 and initial parasitaemia was the opposite of that in the artemisininsensitive population: a tenfold increase in parasitaemia was associated with a 5.2 % (95 % CI 0.7-9.9, p = 0.024, n = 3574) increase in PC 1/2 when adjusted for study site.

Artemisinin dose and PC 1/2
Six studies at 15 locations had randomized AS treatment arms of 2 and 4 mg/kg. The higher dose was associated with an 8. After adjusting for the initial parasitaemia, PC 1/2 and parasite sensitivity status, the recrudescence rates varied (2) predicted relationship from multivariate model using fractional polynomials (c); adjusted for treatment group, region, initial parasitaemia, presence of lag phase and study design characteristics between regimens with different partner drugs or time of their administration. Recrudescence rates were significantly higher in patients receiving artesunate-chlorproguanil-dapsone than any other ACT (HR = 3.62, 95 %CI 1.74-7.52, p = 0.001) Recrudescence rates were significantly lower in patients receiving AS for 3 days followed by a standard ACT at 72 h (HR = 0.28, 95 % CI 0.11-0.74, p = 0.010) than in all other patients. Other baseline covariates, as well as the presence or duration of lag and tail phases in the parasite clearance curve, were not associated with treatment outcome.

Discussion
The rate at which asexual P. falciparum parasites are cleared from the blood following treatment is the best measure of the anti-malarial effect of artemisinin and its derivatives. This is assessed from the linear component of the log-linear decline in parasite densities and is expressed conveniently as PC 1/2 [1]. Resistance to artemisinins results in prolongation of the PC 1/2 . This pooled analysis combines the largest set of data, collected in 24 studies over 18 years, from nearly 7000 patients with uncomplicated falciparum malaria in whom frequent measurements of parasitaemia were made. The reference PC 1/2 estimates provided for 46 locations across Africa and Asia are essential comparators for the early recognition of emerging resistance, and so will be updated continuously as others join the WWARN [44] collaborative effort and provide relevant data sets. An important output of this analysis is that there was no evidence for worsening of artemisinin resistance in western Cambodia. There is substantial concern that failure to eliminate falciparum malaria in this area, the 'cradle of antimalarial drug resistance' , will lead to higher levels of artemisinin resistance, rendering ACTs progressively less effective. While further worsening of the degree of artemisinin resistance fortunately has not happened, at least until 2012, continued monitoring is vital. This large dataset allowed estimation of the additional contributions of patient characteristics and study design to parasite clearance estimates, information that is crucial in interpreting and monitoring changes in these estimates, and attributing them to true artemisinin resistance rather than the effects of partner drugs, study design or patient characteristics. The recent discovery [4] and validation [15] of the molecular marker kelch13 in the Greater Mekong area and the development of suitable in vitro sensitivity tests [6,45] provide important information. Data from in vitro ring-stage survival assays do reflect artemisinin resistance in vivo, but their use is likely to be limited to few resourced laboratories and thus unlikely to provide comprehensive surveillance information across endemic countries. Mutations in kelch13 Estimation of PC 1/2 requires sufficient quality-assured serial parasite blood counts for analysis. In this very large series, the most common problem encountered (13 %) was that only two positive counts were available because of rapid parasite clearance and low initial parasitaemias. Other problems encountered (10 %) were a very long lag phase, large variations in parasite counts resulting in poor fits or large confidence intervals around the estimate. These were most likely a consequence of inaccurate microscopy counts. The initial parasitaemia and frequency of sampling had the greatest effects on the PC 1/2 estimates, which accords with results of a previously reported simulation study [18]. Ideally, only patients with initial parasitaemia >10,000 parasites/µL should be included in PC 1/2 assessments. In patients with only two positive parasite counts, estimated PC 1/2 should be interpreted with caution as it is likely to be overestimated. This is because the lag phase cannot be evaluated and the first recorded zero parasitaemia is treated as a parasite density at the detection limit (so the worst case scenario is assumed). Profiles for which the lag or tail phases were identified, and after their exclusion only two data points were left for the PC 1/2 estimation, should be excluded from analysis as they likely represent limitations in microscopy-based parasite counting.
A lag phase was detected more frequently in patients presenting with fever, possibly because of the association of fever with synchronous schizont rupture. The more frequent lag phase with AL treatment may result from the initial lower dose and slower absorption and conversion to DHA of oral artemether compared to oral artesunate [46]. Patients with profiles beginning with a lag phase may have had more rapid clearance in the log-linear part of the parasitaemia-time curve (lower PC 1/2 ); however, the difference was rather small (6.4 %, 95 % CI 3.6-9.1). This is an artefact of the way the model is fitted-as the lag phase is defined only if the initial clearance is slower and the ratio of the clearance rates between this initial period and the rest of the parasitaemia profile reaches a pre-specified cut-off. Some of the observed differences in slopes are caused by random variation of the microscopy measurement. Excluding this randomly occurring slower (but not faster) part of the profile will result in the overestimation of PC 1/2 in profiles with a detected lag phase. This phenomenon was observed in 3-10 % of simulated parasite profiles (using previously described methodology [18]), from distributions of PC 1/2 with mean of two to 6 h and standard deviation (log scale) from 0.05 to 0.3.
The treatment, clinical and demographic variables studied had modest effects on PC 1/2 estimates, all resulting in less than 20 % change in PC 1/2 , and none associating with an increased risk of PC 1/2 being >5 h in the rapid-clearing parasite populations.
In areas with artemisinin-sensitive parasite populations, parasite clearance was faster in patients receiving the 4 mg/kg dose of AS than in those receiving the 2 mg/ kg dose, which was a robust finding confirmed in metaanalysis performed in a subset of randomized studies as well as in a multivariate analysis of studies with either of the doses administered. It is therefore expected that there will be marked differences between the various currently available ACTs, including AL, ASAQ, DHA-PQP and ASMQ, depending on the dose of artemisinin derivative. However, after adjusting for the sampling scheme, the proportions of patients with PC 1/2 estimates >5 h were not significantly different between treatments in this study and ranged from 0 to 10 % for studies with sixhourly sampling, and from 0 to 7 % after exclusion of profiles with pseudo R 2 statistic <0.8.
Therapeutic responses in malaria are enhanced by immunity [1]. As expected from previous work [47,48], young children had slower parasite clearance rates compared to older patients. However, this was observed only in artemisinin-sensitive parasite populations, with most data coming from Africa. Resistant parasite populations, present only in Southeast Asia, did not demonstrate an age effect. The lack of an age effect on PC 1/2 could be due to one or more of the following factors: lower background immunity in those patients from low transmission settings, different age distributions studied with 70 % of patients being older than 12 years, nonlinear negative age effect on PC 1/2 (Fig. 5), or a qualitative pharmacodynamic difference in that whereas most of the clearance of artemisinin-sensitive parasites results from clearance of ring-stage parasites in low transmission settings, in artemisinin-resistant infections cytoadherence becomes a more important contributor to the initial decline in parasitaemia (as it is following quinine treatment) [1]. In both populations, the presence of fever on admission was associated with longer PC 1/2 . This has also been reported in other studies from Kenya [49] and Uganda [47] and may be a surrogate marker of a less effective host immune response. Also, fever in malaria is thought to be caused partly by TNF and other pyrogenic cytokines released as part of the human immune response to products of schizont rupture [50,51].
The relationship between PC 1/2 and parasitaemia was different between the sensitive and resistant parasite populations. In sensitive areas, high parasitaemia was associated with shorter PC 1/2 largely because patients with low initial parasitaemias and rapid clearance are not included in the analysis. In resistant populations with longer PC 1/2 , high parasitaemias were associated with slightly longer PC 1/2 (by 5.2 % per tenfold increase).
The main limitation of this analysis is the heterogeneity in study designs and treatments which did not permit a more detailed examination of treatment and dose effects, as they were confounded by the exclusion of patients with relatively low initial parasitaemias, different partner drugs, and different timings and frequencies of sampling.

Conclusion
This pooled analysis showed that the main factor affecting estimates of parasite clearance is the study designrelatively low initial parasitaemia resulting in too few data points to estimate the clearance accurately, and too infrequent sampling. Additionally, in artemisinin-sensitive parasite populations, PC 1/2 is affected by artemisinin dose, patient age and the presence of fever as likely surrogates of acquired immunity. Therefore, it is important to consider these factors in early surveillance of changes in parasite sensitivity. This pooled analysis provides critical baseline information to monitor future evolution of PC 1/2 in malaria endemic countries.