Site-specific biases in phase-III clinical trials underestimate the effect of radical cure against Plasmodium vivax hypnozoites

Plasmodium vivax relapses caused by reactivating hypnozoites are a major barrier for elimination and control of this form of malaria. Radical cure is a form of therapy capable of addressing this problem. Recent clinical trials of radical cure have yielded efficacy estimates ranging from 65% to 94%, with substantial variation across trial sites. We performed an analysis of simulated trial data using a transmission model to demonstrate that variation in efficacy estimates across trial sites can arise from differences in the conditions under which trials are conducted. Our analysis revealed that differences in transmission intensity, heterogeneous exposure, and relapse rate can yield efficacy estimates ranging as wide as 11-82%, despite simulating trial data under the uniform assumption that treatment had a 75% chance of clearing hypnozoites. We show that vector control and parasite genotyping offer two potential means to yield more standardized efficacy estimates that better reflect protection against relapse.


INTRODUCTION
Plasmodium vivax is the most geographically widespread cause of human malaria, and its burden 24 in 2017 was estimated at 14.3 million clinical cases globally (1). Control of P. vivax is 25 challenging due to a unique life stage of the parasite, known as the hypnozoite, which latently 26 infects the livers of individuals with recent P. vivax blood-stage infections (2). Hypnozoites 27 activate to cause successive relapsing infections following the initial blood-stage infection, and 28 relapses comprise an estimated 79 -96% of all P. vivax infections (3,4). The prevention of 29 relapses is therefore an ongoing priority for P. vivax control (5), and clearance of hypnozoites 30 can be achieved through radical cure treatment with an 8-aminoquinoline, such as primaquine 31 (PQ) or tafenoquine (TFQ). 88 We first examined how transmission intensity and the level of heterogeneity in mosquito biting 89 patterns in the trial location interact to affect efficacy estimates. We varied transmission intensity CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint calculated based upon the proportion at risk using the complete record of recurrent infections 110 (Fig. S1). 111 In addition to the effect of transmission intensity itself, the effect of heterogeneous biting  At an EIR of 100, efficacy estimates had a non-monotonic relationship with the extent of 124 heterogeneous biting ( Fig. 2A). At high transmission intensities, greater heterogeneity in biting 125 contributed to greater variation in anti-parasite immunity among trial participants. Trial 126 participants that were more frequently exposed to mosquito biting were expected to have greater 127 levels of anti-parasite immunity, thereby reducing the detectability of each P. vivax infection 128 during follow-up. Therefore, although the percentage of participants reinfected during follow-up 129 decreased monotonically with greater heterogeneity in mosquito biting, the percentage of 130 participants with a detectable reinfection (e.g., clinical or subclinical/LM-detectable) varied non-131 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint monotonically and was 36% (1.0%), 38% (1.0%), 34% (1.0%), and 23% (1.0%) when the 132 respective variance in exposure to biting was 0, 1, 2, and 3 ( Fig. 2B-E). Effect of the Rate of Relapse and Duration of Follow-up 146 We next examined how the rate of relapse of P. vivax hypnozoites and the duration of follow-up 147 of trial participants interact to affect efficacy estimates. We varied the mean time to relapse from 148 30 to 180 days and simulated phase-III clinical trials in which participants were followed for 90, 149 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint 180, 365, or 730 days. To assess the effect of transmission intensity, all trials were simulated at 150 EIRs of 0.1, 1, 10, and 100, assuming homogeneous biting.

151
For a given duration of follow-up, efficacy estimates decreased with a longer mean time 152 to first relapse (Fig. 3A). In simulated trials in which participants were followed for 180 days at 153 an EIR of 1, we estimated that the efficacy of an 8-aminoquinoline with a clearance probability follow-up. The downward bias occurred because a longer duration of follow-up ensured that 168 more trial participants were reinfected by mosquito biting (Fig. 3C), similarly causing the 169 distribution of times to first recurrent infection to appear more similar across the trial arms. By 170 90 days of follow-up, 12% and 71% of trial participants were expected to have been reinfected 171 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint when the respective EIRs were 1 and 10. By 730 days of follow-up, 63% and 100% of trial 172 participants were now expected to have been reinfected at respective EIRs of 1 and 10.  CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted January 31, 2021. Higher transmission intensity biased our efficacy estimates downward, because trial participants 190 in both the treatment and control arms were frequently reinfected ( Fig. 2A). We assessed the 191 potential of long-lasting insecticidal nets (LLINs) and indoor residual spraying (IRS) to protect 192 trial participants from reinfection and therefore reduce the downward bias due to transmission 193 intensity. Because the effects of LLINs and IRS depend upon vector bionomics, we varied the 194 proportion of bites that occurred indoors and the proportion of bites that occurred while trial 195 participants were in bed.

196
Vector control interventions were effective in reducing the downward bias due to 197 transmission intensity in trial locations where the vector was predominantly endophagic (i.e.,

205
Across all mosquito biting behaviors simulated, the combination of LLINs and IRS was 206 generally more effective than either intervention in isolation in reducing the downward bias in 207 efficacy estimates (Fig. S2). When considering interventions in isolation, IRS was more effective 208 than LLINs in reducing the downward bias, and the distribution of LLINs most improved 209 efficacy estimates under scenarios in which an appreciable number of mosquito bites were taken 210 at night while trial participants were in bed.

211
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted January 31, 2021. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint this simulation study, we defined sensitivity as the probability of correctly identifying relapses 231 caused by hypnozoite batches acquired prior to treatment and specificity as the probability of 232 correctly identifying all other recurrent infections (e.g., reinfections and relapses caused by 233 hypnozoite batches acquired after treatment).

286
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint  For all three efficacy metrics considered, increasing transmission intensity caused a 301 downward bias in efficacy estimates, because, at higher transmission intensities, most 302 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint participants in both the treatment and control arms experienced a reinfection or relapse during 303 follow-up (Fig. 7). For PCR-detectable infections, estimated efficacy was highest if calculated 304 using the Cox proportional hazards model and lowest if calculated using the proportion at risk. 305 This suggested that, in the absence of parasite genotyping (Fig. S1), the magnitude of bias caused 306 by reinfection from mosquito biting was more severe when basing efficacy upon incidence rates 307 or the proportion at risk than upon proportional hazards.

308
The sensitivity of the assay by which trial participants were assessed for recurrent 309 infections also affected our efficacy estimates, though the direction of the effect depended upon 310 the chosen efficacy metric. For efficacy estimates based upon incidence rates or the proportion at 311 risk, shifting from a more sensitive assay (i.e., PCR-based detection) to a less sensitive assay 312 (i.e., monitoring for clinical infections) reduced the downward bias due to reinfection, 313 particularly at higher transmission intensities. This was due to the fact that the fraction of 314 recurrent infections ascertained increased with a more sensitive assay, causing us to detect more  infection events, such as reinfections by mosquito biting, that do not reflect the effect of radical 336 cure on P. vivax hypnozoites. In this simulation study, we identified features of the trial setting, 337 including transmission intensity, heterogeneous feeding patterns, and the relapse rate of the P. 338 vivax parasite, that affected our estimates of efficacy and the utility of clinical trial data to assess 339 the extent to which radical cure prevents relapse. We demonstrated that the use of vector control 340 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint and genotyping methods are two approaches that can reduce and, in some cases, even correct 341 these site-specific biases and yield more standardized estimates of efficacy against relapse.

342
In the recent GATHER, DETECTIVE, and IMPROV trials (6-9), efficacy estimates 343 varied across trial sites. Interpreting site-specific differences in efficacy is important for past and CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint data to distinguish relapses from reinfections have been successfully developed and could be 386 readily integrated into the analysis of future clinical trial data (12,33). 387 Beyond the factors that make resolving differences in efficacy challenging within the 388 context of a single trial, we identified features of trial design that could limit the comparability of 389 efficacy estimates across trials. Specifically, the details of how the efficacy endpoint was 390 calculated (e.g., time to first infection vs. incidence ratio) and the sensitivity of the assay used to  Trial Overview 456 We constructed phase-III clinical trials for a generic 8-aminoquinoline (e.g., primaquine or 457 tafenoquine). We simulated trials in order to quantify the recurrence-free efficacy as a measure   analyses revealed that the choice of intervention action did not substantially affect our efficacy 495 estimates generated using the Cox proportional hazards model (Fig. S4). Therefore, we assumed 496 by default that each 8-aminoquinoline had an "all-or-none" action.

497
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021.

520
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint Individuals that met the criteria for enrollment were then randomly assigned to the 521 treatment or control arm of the clinical trial. Allocation of participants to each trial arm occurred 522 with equal probability, provided that the current number of participants in each arm was less than 523 the desired sample size of 1,000.  CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint of mosquito bites that occur while individuals are indoors (Φ ! ) and in bed (Φ " ), the duration of 544 usage of the LLIN, and the duration of insecticidal activity. We assumed that the mean duration 545 of usage of LLINs was 3 years and that the mean duration of insecticidal activity was 2.5 years. 546 We also considered whether the use of indoor residual spraying (IRS) in each 547 participant's house administered in isolation or in combination with LLINs decreased the 548 potential bias in our efficacy estimates. As with LLINs, we assumed that the use of IRS 549 decreases the probabilities of mosquito biting and successful feeding and increases the 550 probabilities of repellency and death (10,36). To account for the waning effect of IRS over time, 551 we assumed that the mean duration of insecticidal activity was 6 months. infection was unobserved, all trial participants were interval censored. 580 We calculated efficacy using incidence rates computed from the recurrent infections that 581 occurred within the duration of follow-up. We calculated efficacy based on incidence rates as

588
The magnitude of the measured efficacy also depends upon the action of the radical cure 589 therapeutic. In absence of other sources of bias, we will overestimate efficacy for 8-590 aminoquinolines that elicit an "all-or-none" response if we calculate efficacy based on incidence   CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint  The authors have no competing interests to declare.

650
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. Rate of Liver-Stage Hypnozoite Clearance 1/383 day -1 (10)

658
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint To ensure that our simulated clinical trials were appropriately powered, we performed a where . and $/ ! " are the z-scores, -is the expected incidence rate in the control arm, and ) is 693 the expected incidence rate in the treatment arm.

694
Based on proportion at risk, we computed the number of participants in each arm as . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint In eq. (S3), . and $/ ! " are the z-scores, ) andare the expected proportion of participants at 699 risk in the treatment and control arms, and is the average of ) and -(52).    CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint Although vector control may decrease the bias by reducing the number of reinfection events, a 766 method capable of distinguishing relapses caused by hypnozoites acquired prior to treatment 767 from all other recurrent infections is needed to completely correct the bias in efficacy estimates. 768 Genotyping methods have been successfully used to determine the cause of each P. vivax 769 recurrent infection (12,33) and therefore could correct biases in phase-III clinical trials. 770 We evaluated the potential for a genotyping method at varying sensitivities and 771 specificities to correct the bias in the efficacy estimates. We considered a genotyping method  and, therefore, may bias upwards our efficacy estimates (6).

788
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

819
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint   to be 75%, the type-I error rate was set to 0.05, and the allocation ratio between trial arms was 855 one.

856
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted January 31, 2021. ; https://doi.org/10.1101/2021.01.28.21250689 doi: medRxiv preprint