Windows of Selection for Anti-Malarial Drug Treatments
Windows of Selection for Anti-Malarial Drug Treatments
The aim of this paper was to investigate how accurately genotyping data obtained during anti-malarial drug clinical trials can estimate WoS. The results show that the current method of estimating WoS from the clinically observed patency of different genotypes is a poor surrogate and tends to underestimate the true WoS (Method 1, Table 1). For example, the duration of the true WoS following a default to 15-fold IC50 increase was 17 and 37 days for AR-LF and AS-MQ, respectively; the corresponding clinical WoS estimates were eight to 9 days or 4–6 days, respectively (Method 1, Table 1). Including 'censored' individuals in the analysis (Method 2, Table 1) greatly improved clinical WoS estimates, particularly when using the 25th centile cut-off. Practically, Method 2 requires a significant number (ideally 10–25% of patients) of new infections with sensitive parasites to occur during the 63-day follow-up period to prevent estimates falling into the censored BF period which would preclude estimating WoS. For example, it was impossible to determine a WoS for MQ-sensitive parasites as new infections rarely occurred during the follow-up period; Fig. 2 shows there is still only a 30% chance of acquiring a new infection with MQ-sensitive parasites 50 days after the initial treatment.
A large discrepancy did occur for PPQ when IC50 was increased from two to fivefold; the clinical WoS estimates of the true WoS were greatly overestimated regardless of the method or centile used. Lines A and B on Fig. 1 provide one possible explanation for this. The true WoS (Table 1) shows infections with a PPQ IC50 increased twofold can emerge 1 day after infections with a fivefold IC50 increase. If these infections grow at different rates as a result of their differing IC50s then the infections would become patent several days apart and so the clinical WoS would be expected to overestimate the 'true' WoS.
The choice of model parameters was previously validated for these ACT and the results generated herein are again highly consistent with field data. Fully effective ACT was able to clear infections with either sensitive or mildly resistant parasites when treated directly. Subsequent re-infections with LF-sensitive parasites typically become patent 25–35 days post-treatment in these simulations (central panel, Fig. 3), which closely match those reported by Sisowath et al. who found re-infections with LF-sensitive parasites occurring 24–30 days after treatment. The simulated results also show AS-MQ was able to prevent new infections by sensitive and mildly resistant parasites for longer than AR-LF. This implies MQ has the longer post-treatment prophylactic period, consistent with the data of Sagara et al. who reported re-infections in Peru occur less frequently in their follow-up period of 28 days following AS-MQ treatment than AR-LF.
A previous study used a similar PK/PD approach to elucidate the general principles underlying antimalarial drug windows of selection. This manuscript extends this seminal work in three important ways. Firstly, by incorporating recent methodological advances in PK/PD modelling of antimalarial dosing: this allows multiple dosing, drug absorption and conversion processes, and the co-administration of artemisinin as in the current generation of ACTs. It also allows the analysis of PPQ, a currently widely used antimalarial, which has more complicated (two or three compartmental) PK (e.g.). It is unlikely that these inclusions will have a large qualitative impact on the basic properties of WoS but it is important to repeat the analyses using the current best practice in PK/PD modelling. Interestingly, there were some notable quantitative differences: unlike the Stepniewska and White paper, the new analyses did not find WoS that lasted "several months". The estimates of WoS presented here last approximately 1–2 weeks; this has important policy implications as shorter WoS suggest a slower spread of drug resistance. The second advance is the simultaneous inclusion of the substantial variance that occurs in all the PK/PD parameters (Additional file 1: Table S1 http://www.malariajournal.com/content/14/1/292/additional); this reflects the reality in vivo and means that WoS become probabilistic (Fig. 2) rather an fixed entity, and requires that WoS be re-defined in this more probabilistic context. Finally, simulations included re-infection of patients to show how WoS may be estimated from clinical data. It is now considered mandatory in malaria drug clinical trials to genotype recurrent malaria to distinguish treatment failures from new infections. Recent advances in identifying candidate genes encoding resistance mean that these samples can later be re-genotyped at candidate genes to obtain clinical evidence of their impact on resistance. The new method of analysis (Method 2) will allow these data to be analyzed in the appropriate, accurate manner.
One objective of this study was to contribute to the theory describing the dynamics underlying the spread of anti-malarial drug resistance. Three processes drive drug resistance: the ability to survive treatment, the WoS and intra-host dynamics. The ability of a mutation to survive treatment can already be quantified by clinical trials. Quantifying intra-host dynamics is less easy although some progress has been made in mouse models and through inference from clinical data. The results presented here on WoS suggest that it is now possible to accurately quantify all three of the forces driving resistance. Table 1 suggests that the current method of estimating clinical WoS by 'visual inspection' or its more formalized Method 1 described above, underestimate the 'true' WoS for AR-LF, AS-MQ. The new Method 2 includes all patients when estimating the clinical WoS and obtains estimates much more consistent with the 'true' WoS. It is however important to consider that methodologically, Method 2 requires that estimates of WoS be undertaken in areas of high transmission to ensure a reasonable proportion of the population acquire a new infection during the follow-up period (see recommendations below). Quantifying this impact for the current generation of anti-malarials can only be achieved through accurate estimation of their true WoS, hence the importance of verifying whether clinical WoS accurately reflect the true WoS.
The results presented here are only applicable to the three drug regimens investigated. Researchers are urged to use the methodology contained within this and previous papers to estimate the likely clinical WoS for other drugs and/or drug regimens. By reproducing the plots in Fig. 2 for their chosen drug regimens, researchers can quickly and cheaply assess the likely success of estimating WoS from field data. For example, it would not be advisable to estimate the WoS for drugs where the true WoS is likely to be very short (such as PPQ; Fig. 2c) but it would be suitable to use the new method, Method 2, for drugs that have longer true WoS (Fig. 2a, b, respectively). This new methodology does also require that clinical trials be carried out in areas of moderate to high transmission in which both, or all, alleles are well represented so patients have an adequate chance of acquiring a new infection.
The study also has operational implications for anti-malarial drug deployment policies. The long half-lives characteristic of most malaria partner drugs have always been regarded as potent drivers of drug resistance but the results presented here suggest they are more potent drivers than implied by previous estimates of WoS based on 'visual-inspection' of re-infections (or the more formal Method 1). There is currently interest in the improved diagnostics of malaria treatment, primarily to ensure treatment is restricted to confirmed malaria cases, so treatment is appropriate to the underlying clinical condition. A secondary objective is to reduce the use of expensive ACT and, it is argued, reduce "selection pressure for drug resistance". For example, if a drug had a WoS of 15 days and a patient was given six treatments throughout the year, there would be 15 × 6 = 90 days or approximately 25% of the year in which the resistant parasites were being preferentially selected; cutting treatment down to two doses per year would obviously decrease this selection by two-thirds, i.e., to 15 × 2 = 30 days. The results suggest the decreased selection for resistance caused by improved diagnostics may be much greater than anticipated. This is because the drugs' true WoS are longer and potentially much more important drivers of resistance than suggested by analysis of clinical data using Method 1. The results also suggest that proposed drug deployment policies based on mass distribution of ACT, such as MDA and IPTs, which inevitably increases drug use in the population, have the potential to be more potent drivers of resistance than has been previously thought because the WoS are longer than anticipated. A detailed investigation of the advantages and disadvantages of such interventions is outside the remit of this paper, but it does emphasise that accurately quantifying the magnitude of WoS has both a broad application in quantifying the drivers of drug resistance, and also has a more specific role in the evaluating the likely impacts of interventions such as IPT and MDA.
Discussion
The aim of this paper was to investigate how accurately genotyping data obtained during anti-malarial drug clinical trials can estimate WoS. The results show that the current method of estimating WoS from the clinically observed patency of different genotypes is a poor surrogate and tends to underestimate the true WoS (Method 1, Table 1). For example, the duration of the true WoS following a default to 15-fold IC50 increase was 17 and 37 days for AR-LF and AS-MQ, respectively; the corresponding clinical WoS estimates were eight to 9 days or 4–6 days, respectively (Method 1, Table 1). Including 'censored' individuals in the analysis (Method 2, Table 1) greatly improved clinical WoS estimates, particularly when using the 25th centile cut-off. Practically, Method 2 requires a significant number (ideally 10–25% of patients) of new infections with sensitive parasites to occur during the 63-day follow-up period to prevent estimates falling into the censored BF period which would preclude estimating WoS. For example, it was impossible to determine a WoS for MQ-sensitive parasites as new infections rarely occurred during the follow-up period; Fig. 2 shows there is still only a 30% chance of acquiring a new infection with MQ-sensitive parasites 50 days after the initial treatment.
A large discrepancy did occur for PPQ when IC50 was increased from two to fivefold; the clinical WoS estimates of the true WoS were greatly overestimated regardless of the method or centile used. Lines A and B on Fig. 1 provide one possible explanation for this. The true WoS (Table 1) shows infections with a PPQ IC50 increased twofold can emerge 1 day after infections with a fivefold IC50 increase. If these infections grow at different rates as a result of their differing IC50s then the infections would become patent several days apart and so the clinical WoS would be expected to overestimate the 'true' WoS.
The choice of model parameters was previously validated for these ACT and the results generated herein are again highly consistent with field data. Fully effective ACT was able to clear infections with either sensitive or mildly resistant parasites when treated directly. Subsequent re-infections with LF-sensitive parasites typically become patent 25–35 days post-treatment in these simulations (central panel, Fig. 3), which closely match those reported by Sisowath et al. who found re-infections with LF-sensitive parasites occurring 24–30 days after treatment. The simulated results also show AS-MQ was able to prevent new infections by sensitive and mildly resistant parasites for longer than AR-LF. This implies MQ has the longer post-treatment prophylactic period, consistent with the data of Sagara et al. who reported re-infections in Peru occur less frequently in their follow-up period of 28 days following AS-MQ treatment than AR-LF.
A previous study used a similar PK/PD approach to elucidate the general principles underlying antimalarial drug windows of selection. This manuscript extends this seminal work in three important ways. Firstly, by incorporating recent methodological advances in PK/PD modelling of antimalarial dosing: this allows multiple dosing, drug absorption and conversion processes, and the co-administration of artemisinin as in the current generation of ACTs. It also allows the analysis of PPQ, a currently widely used antimalarial, which has more complicated (two or three compartmental) PK (e.g.). It is unlikely that these inclusions will have a large qualitative impact on the basic properties of WoS but it is important to repeat the analyses using the current best practice in PK/PD modelling. Interestingly, there were some notable quantitative differences: unlike the Stepniewska and White paper, the new analyses did not find WoS that lasted "several months". The estimates of WoS presented here last approximately 1–2 weeks; this has important policy implications as shorter WoS suggest a slower spread of drug resistance. The second advance is the simultaneous inclusion of the substantial variance that occurs in all the PK/PD parameters (Additional file 1: Table S1 http://www.malariajournal.com/content/14/1/292/additional); this reflects the reality in vivo and means that WoS become probabilistic (Fig. 2) rather an fixed entity, and requires that WoS be re-defined in this more probabilistic context. Finally, simulations included re-infection of patients to show how WoS may be estimated from clinical data. It is now considered mandatory in malaria drug clinical trials to genotype recurrent malaria to distinguish treatment failures from new infections. Recent advances in identifying candidate genes encoding resistance mean that these samples can later be re-genotyped at candidate genes to obtain clinical evidence of their impact on resistance. The new method of analysis (Method 2) will allow these data to be analyzed in the appropriate, accurate manner.
One objective of this study was to contribute to the theory describing the dynamics underlying the spread of anti-malarial drug resistance. Three processes drive drug resistance: the ability to survive treatment, the WoS and intra-host dynamics. The ability of a mutation to survive treatment can already be quantified by clinical trials. Quantifying intra-host dynamics is less easy although some progress has been made in mouse models and through inference from clinical data. The results presented here on WoS suggest that it is now possible to accurately quantify all three of the forces driving resistance. Table 1 suggests that the current method of estimating clinical WoS by 'visual inspection' or its more formalized Method 1 described above, underestimate the 'true' WoS for AR-LF, AS-MQ. The new Method 2 includes all patients when estimating the clinical WoS and obtains estimates much more consistent with the 'true' WoS. It is however important to consider that methodologically, Method 2 requires that estimates of WoS be undertaken in areas of high transmission to ensure a reasonable proportion of the population acquire a new infection during the follow-up period (see recommendations below). Quantifying this impact for the current generation of anti-malarials can only be achieved through accurate estimation of their true WoS, hence the importance of verifying whether clinical WoS accurately reflect the true WoS.
The results presented here are only applicable to the three drug regimens investigated. Researchers are urged to use the methodology contained within this and previous papers to estimate the likely clinical WoS for other drugs and/or drug regimens. By reproducing the plots in Fig. 2 for their chosen drug regimens, researchers can quickly and cheaply assess the likely success of estimating WoS from field data. For example, it would not be advisable to estimate the WoS for drugs where the true WoS is likely to be very short (such as PPQ; Fig. 2c) but it would be suitable to use the new method, Method 2, for drugs that have longer true WoS (Fig. 2a, b, respectively). This new methodology does also require that clinical trials be carried out in areas of moderate to high transmission in which both, or all, alleles are well represented so patients have an adequate chance of acquiring a new infection.
The study also has operational implications for anti-malarial drug deployment policies. The long half-lives characteristic of most malaria partner drugs have always been regarded as potent drivers of drug resistance but the results presented here suggest they are more potent drivers than implied by previous estimates of WoS based on 'visual-inspection' of re-infections (or the more formal Method 1). There is currently interest in the improved diagnostics of malaria treatment, primarily to ensure treatment is restricted to confirmed malaria cases, so treatment is appropriate to the underlying clinical condition. A secondary objective is to reduce the use of expensive ACT and, it is argued, reduce "selection pressure for drug resistance". For example, if a drug had a WoS of 15 days and a patient was given six treatments throughout the year, there would be 15 × 6 = 90 days or approximately 25% of the year in which the resistant parasites were being preferentially selected; cutting treatment down to two doses per year would obviously decrease this selection by two-thirds, i.e., to 15 × 2 = 30 days. The results suggest the decreased selection for resistance caused by improved diagnostics may be much greater than anticipated. This is because the drugs' true WoS are longer and potentially much more important drivers of resistance than suggested by analysis of clinical data using Method 1. The results also suggest that proposed drug deployment policies based on mass distribution of ACT, such as MDA and IPTs, which inevitably increases drug use in the population, have the potential to be more potent drivers of resistance than has been previously thought because the WoS are longer than anticipated. A detailed investigation of the advantages and disadvantages of such interventions is outside the remit of this paper, but it does emphasise that accurately quantifying the magnitude of WoS has both a broad application in quantifying the drivers of drug resistance, and also has a more specific role in the evaluating the likely impacts of interventions such as IPT and MDA.
Source...