Resolving the cause of recurrent Plasmodium vivax malaria probabilistically

Abstract

Relapses arising from dormant liver-stage Plasmodium vivax parasites (hypnozoites) are a major cause of vivax malaria. However, in endemic areas, a recurrent blood-stage infection following treatment can be hypnozoite-derived (relapse), a blood-stage treatment failure (recrudescence), or a newly acquired infection (reinfection). Each of these requires a different prevention strategy, but it was not previously possible to distinguish between them reliably. We show that individual vivax malaria recurrences can be characterised probabilistically by combined modelling of time-to-event and genetic data within a framework incorporating identity-by-descent. Analysis of pooled patient data on 1441 recurrent P. vivax infections in 1299 patients on the Thailand-Myanmar border observed over 1000 patient follow-up years shows that, without primaquine radical curative treatment, 3 in 4 patients relapse. In contrast, after supervised high-dose primaquine only 1 in 40 relapse. In this region of frequent relapsing P. vivax, failure rates after supervised high-dose primaquine are significantly lower (∼3%) than estimated previously.

Conflict of interest statement

The authors declare no competing interests.

Figures

Fig. 1. Dynamics of P. vivax relapse, recrudescence, and reinfection.

Estimates generated under the time-to-event model of the probabilities (log10 scale) of relapse (a), recrudescence (b), and reinfection (c) for all observed recurrences (n=1441) are shown as a function of the interval since the last episode of vivax malaria (VHX trial: dots; BPD trial: diamonds). Colours correspond to the treatment used in the previous episode, where Primaquine+ refers to high-dose primaquine with a partner drug (either chloroquine or dihydroartemisinin-piperaquine). For each treatment (colour), the population mean probabilities (linear scale) for the three recurrence states (solid, dotted and dashed lines) are shown as a function of time-to-recurrence (d).

Fig. 2. Probabilities of P. vivax relapse estimated under the time-to-event model.

Top panels: per-recurrence mean probability of relapse together with 95% credible intervals for 1309 recurrences following either artesunate or chloroquine monotherapies (No PMQ, a) and for 130 recurrences following high-dose primaquine with partner drug (PMQ+, b). The recurrences are ranked by their mean probabilities of relapse. The zone of uncertainty (same as in Fig. 3) is highlighted in blue. The upper and lower bounds are arbitrary. c The relationship between time since last episode and the uncertainty of the posterior estimates (width of the 95% credible interval on the log10 scale), coloured by treatment received. The dashed lines represent fitted LOESS curves to highlight trends in the No PMQ and PMQ+ groups, respectively.

Fig. 3. Probabilities of relapse for 487 genotyped P. vivax recurrences.

a All recurrences ranked by their probabilities of relapse coloured by treatment drug (orange: blood-stage treatment only, No PMQ; green: high-dose primaquine plus partner drug, PMQ+). In all, 95% credible intervals are shown by the vertical lines. b The same posterior probabilities as a function of the time since the last episode of P. vivax with the same colour coding. The uncertainty zone (same as in Fig. 2, and used to classify recurrences in Fig. 4) is shown by the blue zone (the upper and lower bounds are arbitrary).

Fig. 4. Classification of 487 genotyped P. vivax recurrences.

Each line represents one individual (n=208). Duration of active follow-up is shown by the span of the horizontal lines (green: high-dose primaquine given, PMQ+; orange: no primaquine given, No PMQ). Recurrences classified as failures are black triangles (point up), reinfections are hollow triangles (point down), recurrences with uncertain classification are blue squares since they fall in the blue zone of uncertainty in Fig. 3. Since there was little evidence of recrudescence, all failures are essentially relapses. The delayed failures (circa ten months after treatment of the previous episode) are circled with their follow-up duration shown by a black dotted line. A histogram summary of these recurrence classifications is shown in Supplementary Fig. 1.

Fig. 5. Probabilities of P. vivax recurrence states with simulated marker count.

Each plot corresponds to a different relationship simulation scenario: a sibling scenario; b stranger scenario; and c clonal scenario. Each coloured bar shows the median of 250 posterior probabilities (dots) with error bars extending ±two standard deviations. The effective cardinality of each marker was set equal to 13 in these simulations. The complexity of infection was one in both first and second infection. The prior probability of each recurrence state was 13. As the number of microsatellites genotyped is increased, the following are expected: under the sibling scenario (a), the probabilities should converge to 1 for relapse and 0 otherwise; under the stranger scenario (b), the probability of reinfection should converge to a probability greater than the prior and the complement of that of relapse, meanwhile the probability of recrudescence should converge to 0; under the clonal scenario (c), the probability of recrudescence should converge to a probability greater than the prior and the complement of that of relapse, meanwhile the probability of reinfection should converge to 0.