A population pharmacokinetic model of piperaquine in pregnant and non-pregnant women with uncomplicated Plasmodium falciparum malaria in Sudan

Malaria is one of the most important infectious diseases with an estimated 216 million people infected worldwide in 2010 [1]. Pregnant women are at an increased risk of developing a malaria infection [2] and are at a higher risk of progressing to severe malaria [3‐5]. Plasmodium falciparum malaria is a major contributor to maternal mortality in Sudan; around 37% of all maternal deaths between 1985 and 1999 at the Medani Teaching Hospital in Medani City, Sudan, were attributed to malaria [6]. Malaria also has severe effects on the foetus causing both foetal loss and low birth weight.

Artemisinin-based combination therapy (ACT) is recommended as first-line treatment for P. falciparum malaria in all endemic areas. The artemisinin derivatives have a very rapid parasiticidal effect, which substantially reduces the parasite biomass during the first days of treatment. These drugs have a short terminal elimination half-life and are, therefore, used in combination with longer acting anti-malarials, with the aim of preventing recrudescence by killing residual parasites. Combination therapies consisting of drugs with different mechanisms of action also reduce the risk of the development of drug resistance [7, 8].

The oral fixed-dose combination of dihydroartemisinin and piperaquine has shown excellent efficacy in the treatment of P. falciparum malaria [9‐13]. Piperaquine is highly bound to plasma proteins (>99.9%), has a large apparent volume of distribution, (103–874 L/kg), a low apparent elimination clearance (0.6-1.3 L/h/kg) and, therefore, a long terminal elimination half-life (12–28 days) [14‐20].

Pregnancy has considerable effects on the pharmacokinetic properties of many drugs. Renal elimination, expression of metabolising enzymes, volume of body water and the degree of plasma protein binding all change during pregnancy [21‐23]. This could result in lower drug plasma concentrations [24]. Previously published studies have reported a decrease in drug exposure during the later stages of pregnancy for artesunate, artemether, dihydroartemisinin, lumefantrine, sulphadoxine, atovaquone, proguanil and cycloguanil [19, 25‐31]. Other anti-malarial drugs (e.g. quinine and amodiaquine/desethylamodiaquine) show no differences in pharmacokinetic properties in pregnant women compared to non-pregnant women [32‐34].

Only one previous study has investigated the impact of pregnancy on the pharmacokinetic properties of piperaquine in patients with uncomplicated P. falciparum malaria [19]. Pregnancy was found to affect the elimination clearance and the bioavailability of piperaquine, but with no change in total drug exposure. This was further supported by a non-compartmental analysis of the same study [35]. No published information is available on the population pharmacokinetic properties of piperaquine in pregnant or non-pregnant women in an African country.

The aim of this study was to describe the population pharmacokinetic properties of piperaquine in pregnant and non-pregnant women with uncomplicated P. falciparum malaria in Sudan.

Fourteen non-pregnant and twelve pregnant women were recruited into the study but two non-pregnant women withdrew their consent and were excluded from the analysis. Full demographics are given in Table 1. The treatment was well-tolerated, none of the patients vomited after treatment and no severe adverse events were reported during the study. One non-pregnant woman had a PCR-confirmed new infection on day 35. None of the women presented recrudescent malaria during the nine weeks of follow-up.

Five hundred sixty-four (564) post-dose plasma samples of piperaquine were used in the pharmacokinetic analysis. A total of 7 (1.2%) of these samples were measured to be below the limit of quantification and omitted. A three-compartmental disposition model resulted in a significantly better model fit compared with a two-compartment model (ΔOFV=−32.8). An additional peripheral compartment (four-compartment disposition model) did not further improve the fit (ΔOFV=0). A transit-compartment (n=3) absorption model was superior to all other absorption models (ΔOFV≥−44.5). In the final model the absorption rate constant (k_a) and the transit-compartment rate constant (k_tr) were set equal to increase the stability of the model. An additive residual error model was adequate in describing the residual random variability. BSV could be estimated reliably on elimination clearance, one inter-compartmental clearance parameter, and one peripheral volume parameter. The addition of BSV on the bioavailability resulted in a significant improvement in model fit (ΔOFV=−70.9). BOV had considerable impact on both mean transit absorption time and bioavailability (ΔOFV=−340). The final parameter estimates and a schematic picture of the final structural model is presented in Table 2 and Figure 1. Body weight was incorporated with an allometric function on all clearance and volume parameters (ΔOFV=−6.14).

Secondary parameters (i.e. total drug exposure, maximum concentration after first dose, time to maximum concentration, elimination half-life and day 7 concentrations) were analysed using the Mann–Whitney U-test to identify differences between pregnant and non-pregnant women (Table 3). There was a significant difference in terminal elimination half-life (p=0.0014), time to maximum concentration (p=0.0177) and maximum concentration (p=0.0205) [median (range) in pregnant vs. non-pregnant women: 22.1 (19.1-25.8) vs. 25.7 (20.9-33.3) days, 3.07 (1.65-4.64) vs. 1.48 (0.887-4.18) hours and 185 (109–363) vs. 102 (40.6-235) ng/mL, respectively]. However, no significant differences were found in day 7 concentrations (p=0.67), day 28 concentrations (p=0.84) or the total drug exposure (p=0.80) between the pregnant and non-pregnant women.

The MCMP analysis (5% significance level and 80% power) indicated that 8+8 and 13+13 women are needed to detect pregnancy as a covariate on elimination clearance and bioavailability, respectively (Figure 2). A formal stepwise covariate search was therefore not performed since it might result in a biased covariate selection in this small population sample.

Two full covariate models were constructed from the final model to investigate the clinical relevance of pregnancy and EGA separately. Pregnancy had a relatively large impact on mean transit absorption-time, volume of distributions and inter-compartment clearances but no significant effect on elimination clearance (Figure 3). The inclusion of EGA as a power function or a linear function produced similar results. EGA was therefore implemented as a linear function for the full covariate approach and resulted in similar results compared to pregnancy as a categorical covariate (Figure 4A). Bootstrap results for elimination clearance, stratified by trimester (i.e. non-pregnant, second trimester at 20 weeks, and third trimester at 32 weeks) were also investigated and resulted in a non-significant trend of increasing clearance with increasing EGA (Figure 4B).

The final model resulted in good model diagnostic performance and reliable parameter estimates (Figure 5 and Table 2). Calculated epsilon-shrinkage was low (13.0%) which indicates that model diagnostics can be assessed reliably. However, eta-shrinkage was relatively high for certain parameters (CL/F=18.9%, MTT=13.2-52.8%, F=12.4-46.6%) (Table 2) and empirical Bayes estimates should therefore be interpreted with caution (Table 3) [52]. The final model had good predictive performance (Figure 6) with 4.8% (95% CI. 1.4%-11%) of observed data below and 2.1% (95% CI. 1.4%-10%) of observed data above the simulated 90% prediction interval (the three observations at day 90 were excluded because too few patients were followed up to this time for reliable simulations).

In this study, the pharmacokinetic properties of piperaquine have been investigated using nonlinear mixed-effects modelling in pregnant and non-pregnant Sudanese women treated with piperaquine-dihydroartemisinin for uncomplicated P. falciparum malaria. Few studies have been performed to evaluate the effect of pregnancy on the population pharmacokinetics of piperaquine, and this is the first study conducted in an African population. The treatment was well-tolerated and none of the participating women had recrudescent malaria infections.

Previous studies of piperaquine pharmacokinetics have presented both two- and three-compartment disposition models depending on the amount of data included in the analysis [14, 15, 17, 19, 20, 53]. A three-compartment disposition model described the piperaquine concentration-time data adequately in this study. This supports the general finding that a three-compartment disposition is more appropriate than a two-compartment disposition when modelling data from patients followed for a sufficient period of time.

The absorption phase was best described with a transit compartment model with three transit compartments including random effects on bioavailability (BSV and BOV) and mean transit absorption time (BOV). The transit-compartment model provides a more physiological representation of the absorption process compared to the absorption models used in previous studies (i.e. first-order absorption and parallel first-order absorption with lag time) [15, 17]. Recently published studies have also implemented the transit-compartment model but with two- and five-transit compartments which support the absorption model presented here [19, 20]. Small variations in the number of transit compartments are to be expected when modelling different studies due to population differences, sampling schedules and study size. The inclusion of BSV and BOV in the absorption model improved the description of the data in the absorption phase and accommodated the large between-subject and between-occasion variability in these data. The data in the absorption phase was not rich enough to estimate separate absorption rates for k_a and k_tr, and they were therefore set to be equal. This is a common limitation and the same approach has been used in previous studies [19, 20]. Incorporating BOV resulted in an increasing median bioavailability (0.77, 1.19 and 1.40 at dose 1, 2 and 3, respectively) and mean transit absorption time (1.55, 1.95 and 2.05 at dose 1, 2 and 3, respectively) during the treatment regimen. Similar patterns have been identified in previously studies on piperaquine [19]. This might be an effect of disease recovery or differences in the food intake over the course of the dose regimen. This cannot be verified since parasite densities were not counted at each dose and food intake was not monitored in this study.

Pregnant women had a shorter terminal half-life compared to non-pregnant women, which is in agreement with the non-compartmental analysis [36], and higher maximum concentrations after the first dose. However, there were no differences in total piperaquine exposure, day 7 concentrations or day 28 concentrations, which supports previously published findings in an Asian pregnant population [19].

The main aim of this study was to investigate the pharmacokinetic differences between pregnant and non-pregnant women, but the sample size was not large enough to make a conventional covariate search. The MCMP analysis resulted in a minimum of 13 patients needed in each group in order to identify the previously described covariate relationships with 80% power and a significance level of 0.05 (Figure 2). However, this is under the assumption of perfect sampling since the MCMP analysis was based on simulated data using protocol sampling times. In the present study, some patients were not sampled for the complete follow-period and some samples were randomly missing which might increase the number of patients needed to identify the assumed covariate-relationships. Therefore the final model did not include any covariates except body weight, which has a strong biological prior [47, 48] and in addition gave a drop in OFV when included in the model. The full covariate approach suffered from identifiability issues when incorporating the pregnancy covariate simultaneously on clearance parameters, volume parameters and relative bioavailability. The full covariate approach was therefore used to investigate the net effect of a potential covariate on all parameters except relative bioavailability. This approach resulted in a model with reduced volume of distribution and inter-compartment clearance in pregnant women compared with non-pregnant women but no net-effect on apparent clearance. This is in agreement with previously published results where pregnancy affected both clearance and relative bioavailability but in different directions [19]. These covariate relationships would also explain the difference in terminal elimination half-life and the lack of difference in total drug exposure.

The model presented in this study was built on data from few patients and a single individual can therefore have a considerable impact on the results. Piperaquine population pharmacokinetic parameter estimates from the final model are in agreement with previous reports (Table 4). However, the elimination clearance presented in this study for an African population is lower compared to previous studies in non-pregnant and pregnant patients. This might suggest an ethnicity related effect on elimination clearance but this needs to be confirmed in a larger population.

In conclusion, this study presents the population pharmacokinetic properties of piperaquine in pregnant and non-pregnant women with uncomplicated P. falciparum malaria in Sudan. The terminal half-life was shorter in pregnant compared to non-pregnant women, but the total drug exposure was comparable between the two groups. This supports previous findings that no dose adjustments are needed on account of altered piperaquine pharmacokinetics in pregnancy.