Malaria is a significant global public health problem and the greatest burden of disease is found in the world’s poorest countries [1]. The majority of malaria morbidity and mortality is caused by two of the five species of Plasmodium that naturally infect humans, Plasmodium falciparum and Plasmodium vivax. The broader global distribution of P. vivax relative to P. falciparum puts an estimated 2.5 billion people at risk for endemic vivax malaria [2, 3]. An increasing body of evidence has shown that P. vivax should no longer be thought of as a benign and rarely fatal disease [4–9], but instead as being capable of causing severe disease and death, particularly in pregnant women and small children [9–12].

Plasmodium vivax is epidemiologically and biologically different to P. falciparum and it is not, therefore, appropriate to assume that control methods developed for falciparum malaria are directly transferable to P. vivax[13–16]. Biological features of P. vivax that distinguish it from P. falciparum also present unique challenges to the control of the parasite [17–19]; in elimination settings, P. vivax is often the “last parasite standing” following P. falciparum elimination [20, 21]. Plasmodium vivax gametocytes are present earlier in the progression of a primary or recrudescent infection compared to P. falciparum[17, 22], such that the majority of patients have sufficient gametocytaemia to allow for transmission before diagnosis or treatment may occur [23–25]. In addition, P. vivax gametocytes are transmitted more efficiently to Anopheles mosquito vectors than those of P. falciparum and are transmissible at lower parasite densities [18, 26, 27]. Within the mosquito, P. vivax sporozoites develop faster than P. falciparum at equivalent temperatures, which contributes to its exploitation of a wider geographic range [28].

Perhaps the most epidemiologically important feature of P. vivax biology is its ability to relapse in the weeks and months following a primary parasitaemia via a dormant liver stage known as the hypnozoite [29–31]. This potential for long-term latency provides the obvious advantage of safe harbour during cold winter months when circulation in blood creates potential host immune system dangers without the benefit or opportunity for onward transmission. Therefore the term “infection” has various meanings for P. vivax. Infection may refer to the introduction and presence of parasites in the body, but with P. vivax, unlike P. falciparum, this can also refer to a symptom-less latent infection. The origin of renewed parasitaemia following a primary vivax infection or a “recurrence” is also ambiguous; it could be due to a hypnozoite-triggered relapse, a resurgence of erythrocytic parasites as a recrudescence, or an entirely new reinfection. See Figure 1 and Table 1 for a description of the pathways between types of infection and attack, and distinctions in terminology. The hypnozoite fundamentally distinguishes P. vivax from P. falciparum in almost every important biological, epidemiological, clinical, and public health respect.

The hypnozoite stage in the life cycle of P. vivax and the potential for relapse makes chemical therapies that target only the blood stage of infection ineffective as a radical cure. The 8-aminoquinolines are the only class of drugs known to have activity on the hypnozoite parasite [32–34]. Primaquine therapy, the only currently licensed radical cure, comes with caveats that add to the challenge of controlling the parasite to the point of elimination. Primaquine is associated with potentially fatal haemolysis in individuals with glucose-6-phosphate dehydrogenase (G6PD) deficiency [32, 35, 36] and is contra-indicated in pregnant women because of the risk of acute haemolytic anaemia in the foetus of unknown G6PD status [37]. The hypnozoite stage and the paucity of therapy for safe and effective treatment render vivax malaria an exceedingly difficult challenge for clinicians and those responsible for the control of endemic malaria. Relapse also has critical implications for understanding epidemiological metrics such as the basic reproduction number and force of infection, obtained from prevalence rates derived from malariometric surveys and cartographic studies that form a central part in elimination scenario planning [38, 39].

It has long been known that there is significant geographical variation in the rate at which a “strain” of P. vivax relapses [40–42]. Temperate and subtropical strains often exhibit either a long incubation or latent period (Figure 1) of around eight to ten months. Tropical strains are characterized by short incubation times and short latency (approximately three to six weeks) [43]. Incubation period refers to the time from sporozoite inoculation (the mosquito bite) to the primary blood-stage infection. The latent period describes the time from the primary attack to relapse. How hypnozoite relapse is triggered, and the source of this phenotypic variation, is unresolved [44]. One theory is that the mechanism is an adaptive trait of the parasite to sequester or “hibernate” during times when climatic conditions would be inhospitable to the parasite’s anopheline mosquito vectors [45–47]. Another is that latent hypnozoites are activated by a systemic febrile illness, explaining the large number of P. vivax relapses that follow P. falciparum infections [47–49]. These hypotheses need not be mutually exclusive.

Regardless of the triggering mechanism and aetiology of relapse, evidence from both controlled experimental and natural settings indicate considerable geographical variation in the timing of relapse. Although the historical perception has been that frequent relapsing strains originate from the tropics and long-latency strains from temperate regions [31], it does not sufficiently describe the observed variation in relapsing phenotypes. This binary classification conflicts with evidence of long-latency strains in tropical regions in the Americas, for instance. Coatney et al. [43] described in 1971 that there were three patterns of relapse. These included the Chesson strain of New Guinea-South Pacific which exhibits a short incubation period (seven to fourteen days), followed by regular re-invasions of the bloodstream within approximately three weeks after the primary infection and may continue to relapse for more than 18 months without a radical cure of the hypnozoite stage. The St Elizabeth strain from southern USA has a similar incubation period to the Chesson strain, but hypnozoites remain quiescent for several months following the primary infection before relapsing at regular intervals of three to four weeks for up to two years [43]. A third variety, such as the strain once found in the Netherlands [42], has a long period of incubation (around eight months) before the primary clinical episode followed by frequent relapses (the lower arm of Figure 1). This three-type classification likely oversimplifies the degree of variation in P. vivax relapse periodicity and offers limited information regarding the geographic origin of the described phenotypes. Furthermore, the majority of P. vivax strains have short incubation periods and the greatest difference lies in the latency period from primary attack to first relapse.

Geographic zones have been proposed for distinguishing areas with similar timing and frequency of P. vivax relapse following sporozoite inoculation. Figure 2A, adapted from White [47], and modified by the boundaries of the malaria endemicity map proposed by Lysenko and Semashko (for the maximum range of malaria circa 1900) [50], illustrates a proposed distribution of relapse phenotypes. White noted the historical perception that strains that relapse quickly originate from Southeast Asia. Temperate and subtropical areas are characterized by long-latency strains. The Indian subcontinent and South America are thought to contain both long-latency and frequent relapsing strains. A second geographic description of variation in time to relapse is described in a recent study by Lover and Coker [51]. The authors analysed the time to relapse in experimentally infected individuals relative to the geographic origin of the strain. They found that, overall, temperate strains relapse more slowly than tropical ones. However, they also found that New World tropical strains relapse more slowly than Old World tropical strains, and Eurasian temperate strains relapse more slowly than temperate strains from the Western Hemisphere [51]. Figure 2B illustrates this implied classification system.

While the maps in Figure 2 were derived from observations of relapse phenotype, others have grouped regions based on similar ecological and epidemiological characteristics in the absence of relapse observations. These regions may reveal patterns of relapse and help elucidate determinants of the periodicity observed. The zones proposed by Macdonald [52] are shown in Figure 3. They are described by Macdonald as “zoogeographical” malaria regions and share commonalities with many historical biogeographical zonations [53–55]. The approximate boundaries of the zones are delineated by climatic variables that influence malaria transmission rates, such as temperature and rainfall, the intensity of transmission observed, as well as the abundance and behaviours of the locally dominant vector species [52].

A systematic review of P. vivax relapse events was conducted with the aim of revealing systematic geographical patterns of relapse frequency and a quantitative description of the potential time to relapse in different regions of the P. vivax-endemic world.

The aims of this paper were to review the timing and frequency of P. vivax relapse of known origin in patients not treated with a hypnozoitocide to characterize variance in these patterns within geographic dimension, to identify a system to classify the variation in relapse observed and to describe the pattern of relapse in each area. A modified classification of the zoogeographical zones of malaria transmission outlined by Macdonald [52] was found to best describe the observed variation in relapse incidence. The rate of relapse and mean time from primary infection to relapse was predicted in each of the nine zones. These quantitative estimates of the contribution of relapses to P. vivax case incidence are crucial in informing estimates of disease burden and the origin of acute attack, i.e., from biting mosquitoes or emergent hypnozoites. This understanding, in turn, informs critical decision-making in control strategies that effectively weigh the benefit of anti-mosquito versus hypnozoitocidal interventions. They also help identify regions in which strains have long-term latency and are therefore undetectable to standard diagnostic methods (rapid diagnostic tests and microscopy) for long periods of time.

The results presented here further refine historical interpretations and recent analyses of the geographic variation in relapse periodicity. As shown in Figure 2, tropical strains relapse more rapidly than temperate strains and New World strains vary from those in the Old World. White’s illustration of the variation in relapse phenotype shown in Figure 2A shows that Southeast Asia and Asia-Pacific is the only region having exclusively frequent relapse behaviour. The results also showed infections from this region to relapse quickly, with a few rare exceptions. White showed that both frequent relapsing and long-latency strains are present in India and South America. The data from India appear to affirm this, with relapse patterns from the subcontinent and surrounding areas so heterogeneous that it was impossible to generate logical model predictions of relapse incidence for the region. The low incidence in region 8 shown in Table 3 is not believed to be a reflection of the presence of long relapsing strains, but rather a result of natural infections that either did not result in a relapse or the resulting parasitaemia was too low to be detected by the study. There is not presently an explanation for the lack of relapses, but this phenomenon has also been observed in recent tafenoquine trials in India [75]. The variation in relapse timing in the raw data observed in India (Figure 5B) is likely a result of the wide variation in transmission settings found within this zone. There are tropical forest areas, similar to zone 12, dry habitats like those in zone 5 and highland areas that border zone 11. In addition, the presence of Anopheles stephensi, adapted to breeding in artificial water collections [76, 77], has extended transmission into urban areas. Therefore, in addition to issues of data availability and study design, the range of ecological settings in the zone, and likely some of the other zones, may also contribute to the variance in relapse behaviours observed.

The results presented for South America were different from what was shown in White’s phenotype map (Figure 2A). South America was predicted to have a high relapse incidence, comparable to Southeast Asia, and a two-month mean time from primary attack to relapse. The data available from this region were limited and determining the cause for the observed difference is therefore difficult. There has been renewed interest in the origin of the American strains of P. vivax, whether they originate from somewhere in Asia or were sent west from Africa by migration and the slave trade, as has been proposed for P. falciparum[78]. This could influence the nature of the relapse periodicity observed. Improved understanding of the phylogeny of P. vivax may reveal information about the pattern of relapse in this region.

The analysis by Lover and Coker [51] revealed that tropical regions relapse more quickly than temperate strains. However, Central America and sub-Saharan Africa had relapse patterns similar to the Mediterranean with moderate relapse incidence (around 250 relapses per 100,000 person days) and five to six months between primary infection and relapse. These regions seem to be better described as an “intermediate” relapsing phenotype between the frequent relapsing strains in Southeast Asia and South America and the long-latency temperate strains in North America and Europe/Asia.

The results of the northern temperate regions concur with the findings of Lover and Coker [51]. The authors noted that while in general, temperate strains relapse more slowly than tropical strains and that New World tropical strains were slower than Old World strains, the opposite was true of the temperate strains. Based on this analysis, the New World temperate strains relapsed more rapidly than the Old World temperate strains (Figure 7). The modelled results showed that the relapse incidence was 455 per 100,000 person days and mean time to relapse was six months in North America. However, again, this high incidence compared to that in northern Asia and Europe (134 relapses per 100,000 person days) could be due to a few exceptional experimental subjects who received large sporozoite inoculations [79–81].

The utility of the predictions made is limited by the nature of the data available. There are few contemporary data on P. vivax infections in patients not treated with a hypnozoitocide. Therefore, much of the data used were from drug trials on adult workers, military personnel and prison “volunteers”, as well as data from when malaria was used to treat neurosyphilis patients. The age and immunity of the patients would perhaps not be representative of relapse as it would occur among residents of the strain region of origin. This could be important because the children in many endemic settings likely carry the greatest burden of relapsing infections [82]. Some experimental challenge subjects were inoculated with relatively heavy sporozoite dosages, a factor that greatly influences the time from primary attack to relapse [47] and is likely to differ from relapses following more modest numbers of sporozoites acquired from wild anophelines. It is straightforward to attribute an infection as a relapse or re-infection in experimental settings, but this could not be distinguished for P. vivax infections acquired in the wild. Effort was made to obtain studies where the follow-up period was conducted in a non-endemic area (for example, in a hospital in a city). Of the 5731 records of relapse 88% (n = 5030) were obtained from experimental or non-local populations (such as military personnel). Lastly, the strains used in therapy and drug trials in the first half of the twentieth century were of “known” origin and infections were geopositioned to those sites. However, it is not certain if the strains used were of their named origin. For example, the “Madagascar” strain was obtained from an Indian seaman whose last port of call was in Madagascar [83] and may conceivably, therefore, originate from elsewhere.

A principal limitation of this work is the inability to conclusively identify a recurrent infection as a relapse, recrudesence or reinfection. This is particularly an issue given the rising rate of resistance to standard treatments such as chloroquine [84]. In cases of chloroquine failure, recurrence can occur one to two months after initial treatment [63], making differentiation between relapse and recrudescence a challenge. Chloroquine and chloroquine-combination therapies were by far the most common treatment regimens (82%, 24787/30049). Of those patients to receive chloroquine, 2519 patients experienced a relapse or recurrence, 523 of which occurred before 60 days. This is equivalent to 9% (523/5731) of the relapse records. In addition, the effect of resistance on recurrence is likely abated by the historical nature of the dataset. The first cases of chloroquine-resistant P. vivax were reported in 1989 [85]. Of the 5731 records of relapse, 2080 occurred before 60 days and 82% of those (n = 1701) were observed before 1989. Therefore, increased resistance is unlikely to have a large effect on the results and instances of recrudesence being classed as a relapse would have been rare. While the relapse signal represented in these data certainly contains some noise due to reinfection or recrudescence, we considered these other sources of recurrences improbable relative to relapse. Figures 5C and 6 seem consistent with this assumption because recurrence due to reinfection or recrudescence would have been far more stochastic than relapse, obscuring or effacing the patterns shown by the randomness of timing of those events relative to primary parasitaemia.

There are aspects of the data that were not incorporated into the analysis performed here that could be addressed in future work. First, it was difficult to account for strains with long incubation periods before primary infection (information only available for a subset of data from experimental settings) followed by relatively short time to relapse. This was occasionally exhibited by the North American St Elizabeth strain and P. vivax multinucleatum from China (see Additional file 3). The link between the sporozoite dose and latency, mentioned above, was shown in the literature to be an important factor in determining relapse patterns [46, 86, 87], but was not incorporated into the analysis as it cannot be known for wild infections. This may be a possible explanation why the incidence of relapse in North America was greater than that predicted for Central America. However, this was likely not a common problem with the experimental studies used. The majority of studies aimed to induce patency and the sporozoite inoculations were large, but not extreme. There were only a handful of studies included that used particularly large inoculations in order to study the effect of dosage on relapse pattern [79–81].

The type of patient varied among studies (prison volunteers, military personnel, malaria therapy patients, outpatients in an endemic area, etc.) and the drug type and dosage varied within and among studies. In some studies, primary attacks were treated with insufficient doses of 8-aminoquinolines or drugs that have long half-lives. Mepacrine has a half-life of up to a month and can delay relapses by about 30 days [88, 89] and chloroquine, the most common drug in the dataset, can delay parasite re-appearance by anywhere from two to six weeks [47]. Inclusion of the subject-type and treatment as explanatory variables was tested; however, the results were similar to the simpler model used (see Additional file 1). Finally, the analysis only addressed the periodicity between primary attack and first observed recurrence (relapse). The modelled estimates of incidence do not account for multiple relapses. Both the frequency and number of relapses will vary based on a variety of factors including inoculums, age of patient and origin of infection. While many studies did not follow patients long enough to report multiple relapses, further work in this area will be essential to obtain measures of the P. vivax force of infection.

The mechanism of hypnozoite activation to cause an acute attack (relapse) remains unknown. There is clearly variation in relapse “phenotype”. Based on the results presented here, timing of relapse appears to vary geographically in conjunction with areas of similar ecology and malaria transmission patterns. However, it is difficult to determine whether long latency occurs in regions of frequent relapse (tropical areas such as zones 9 + 10 and 12). A long-latency relapse may be thought to be another short-term relapse in a succession of rapid relapses, and the genotype cannot reveal if it is in fact a separate relapse “event” [47]. Nonetheless, understanding broad patterns of relapse is of use epidemiologically. There tends to be fewer overall relapses in the long-latency strains because hepatocytes that host the hypnozoites may die before the relapse event occurs. The resulting burden of hypnozoites from different strains or regions has implications for sensitivity to primaquine and therefore the dosage of primaquine that should be used [47]. This was observed in treatment of soldiers returning from Korea (fewer hypnozoites) [90] relative to those returning from the Pacific (high hypnozoite burden) [58] and will have implications for future control strategies.

Frequency of relapse varies geographically. The association between relapse rate and the geographic regions does not clarify causation. Geographic variation does not directly imply environmental cues as triggers for relapse, even if the revised Macdonald system resembles the distribution of transmission suitability. Relapse frequency may result from evolved responses to average transmission season duration or arise from proximate cues, such as triggers from other infections, correlated with P. vivax transmission and/or vector suitability periods. There is likely an interaction between activation of latent hypnozoites from infection and an evolved trait for strains from areas of seasonal transmission to remain dormant during periods of low mosquito abundance. Regardless of the cause, these patterns are important for the treatment of individual infections, measures of P. vivax burden and the prospects for control and eventual elimination of the disease from endemic areas.