Background
The World Health Organization (WHO) has set a goal to eliminate malaria in the Greater Mekong Subregion (GMS) by 2030, an objective that Viet Nam agreed to in 2011 [
1]. Since 2000 there has been a marked reduction in malaria cases in Viet Nam [
2], with an almost 50% reported decrease in indigenous cases from 9331 in 2015 to 4813 in 2018 [
3]. Over the same period, malaria transmission has become increasingly focal in Viet Nam [
4], with pockets of high incidence within areas of ongoing transmission, particularly in Binh Phuoc province, and parts of the central highlands, which could be sources for the spread of malaria to other regions. It appears from a review of the most recent publicly available data that progress towards malaria elimination has stalled.
From April to September in 2019 there was a 18% increase in confirmed cases nationwide and a 39% increase in
Plasmodium falciparum compared to the same period in 2018 [
5]. As the causes of this increase are not clear, with possible contributors including changes in climate or forest activities (personal communication from National Institute of Malariology, Parasitemia and Entomology (NIMPE)), a detailed study of which groups are now most at risk for malaria transmission is critically needed.
A major challenge for malaria elimination programmes (MEPs) in the developing world is access to funds for national programmes. For many countries the sources of funds for MEPs are external, primarily from the Global Fund to Fight AIDS, Tuberculosis and Malaria [
2]. With the decreasing external and domestic funding for malaria programmes in recent years [
3] developing countries must necessarily learn to optimize the use of limited resources. Optimizing the use of these resources will require, among other things, targeting of interventions and health services to those groups most at risk of infection, localities of active transmission, and those where malaria is at-risk of being reintroduced.
Gaps remain in the understanding of the epidemiology of malaria in Viet Nam, particularly in the sparse information about groups most at risk, risk behaviours and patterns of population movement which may associate with malaria transmission. These knowledge-gaps hamper efforts to effectively and comprehensively target resources at high-risk populations to accelerate elimination.
To provide evidence-based information for planning of the national malaria strategy, a prospective case–control study was conducted aiming to identify risk factors for malaria and patterns of population movement associated with malaria risk in the highest incidence area of Viet Nam.
Methods
A prospective, observational case–control study in health facilities in the two highest incidence provinces, Gia Lai and Binh Phuoc, and neighboring Dak Nong province, in central Viet Nam was done from March 2018 to September 2019. Gia Lai and Binh Phuoc provinces have accounted for around half of the national total reported cases in recent years (personal communication from NIMPE). Eligible subjects were patients of at least 6 months of age, living or working in these provinces, and self-presenting to health services in which the treating clinician prescribed a test for malaria. Study sites were the commune health stations (CHS) in the three provinces. CHS are the main source of care for malaria in the public sector [
6]. All 117 CHS in the 10 malaria endemic districts of the 3 provinces were included as study sites (Additional file
1: Table S1). Of these, 61 CHS were able to enroll both malaria cases and controls. The planned sample size was 1000 patients with confirmed malaria of any species (“cases”), plus an equal number of malaria negative individuals matched for age and gender. All subjects were tested at the health facility for malaria, as per routine practice, by rapid diagnostic test (SD BIOLINE Malaria Ag Pf/Pv) and/or peripheral blood microscopy (Giemsa-stained slides).
Blood was taken from all patients by finger prick or venipuncture for real-time PCR for Plasmodium. Blood samples were sent to the NIMPE laboratories in Hanoi, Viet Nam where real-time PCR was performed to detect parasites and determine Plasmodium species. Briefly, Plasmodium genomic DNA was purified from dried blood spots (DBS) using QIAamp DNA 96 Blood kit (Qiagen, Cat No./ID: 51161).
Real-time PCR was performed on the Applied Biosystems 7500 Fast Real-Time PCR system (Foster City, CA, USA) and targeted the 18S rRNA gene of
Plasmodium spp. following a previously published method [
7] with a self-designed primer set for detection of
P. falciparum and
Plasmodium vivax. The primers and probe used are listed in Additional file
1: Table S2. The reaction mixture (25 µL) included master mix (10 µL) (Quantinova probe PCR kit), forward primer (25 µM; 0. µL), reverse primer (25 µM; 0.35 µL), probe (10 µM; 0.75 µL), purified DNA (5 µL) and nuclease-free water (2.1 µL). The PCR program included 3 min of denaturation at 95 °C followed by 45 cycles of 05-s denaturation at 95 °C, 30-s annealing at 58 °C and 30-s elongation at 72 °C. All results with threshold cycle values (Ct) ≤ 40 were considered positive. The limits of detection of the assay were: 0.5 parasite/µL for
Plasmodium; 0.5 parasite/µL for
P. falciparum and 0.8 parasite/µL for
P. vivax [
8].
All enrolled subjects provided written, informed consent. Patients who presented with fever a second-time after an interval of ≥ 2 months were enrolled again.
Subjects were interviewed by trained staff at the CHS about possible risk factors including demographics, place of residence, employment, travel patterns, housing type, use of protective measures against malaria, previous malaria episodes and treatment. Their malaria test result was recorded and, where relevant, a microscope slide saved for later re-examination and parasite quantification. All malaria positive patients were treated in accordance with the Ministry of Health guidelines.
Data analysis
Study questionnaire data were compared between cases and controls to identify risk factors for malaria. Demographic data from both groups was compared to data from the most recent population census in Viet Nam to identify age and gender risk groups. Potential risk factors between cases and controls was assessed using multiple logistic regression with the use of dummy variables for categorical data. Block-wise forward selection by thematic group was used to choose predictor variables. Contingency tables were analysed by the Chi Squared method. Time in the forest was assessed by pairwise Wilcoxon test. Median age was compared between two groups using Mann–Whitney test and multiple groups using one-way ANOVA with the Kruskal Wallis test and Dunn’s correction for multiple comparisons. Due to space limitations, the analysis of travel data was confined to identifying risk factors for malaria. More detailed results by demographic and occupation group and extended analyses of the travel data will be published separately.
Ethics, consent and permissions
This study was approved by the Viet Nam Ministry of Health Ethics Committee and reviewed by the U.S. Department of the Navy Human Research Protection Program in compliance with all applicable federal regulations governing the protection of human subjects (HRPO.NMRCA.2018.0004 and HRPO.NMRCA.2018.0010). All participants provided written, informed consent to participate.
Discussion
This large case–control study found 38 risk factors associated with being infected with, or protected from, malaria in Viet Nam. Of these, when combined in a model, 10 independent risk factors, excluding age and gender, were able to correctly classify 78.8% of participants as cases or controls. This study was designed to include matching of age and gender between cases and controls as a strong association of malaria risk with male gender and being a young adult was evident from independent analysis of routine malaria programme data (personal communication from NIMPE) and prior research [
10,
11]. Matching of cases and controls in this study was intended to increase the power to detect other risk factors that are age or gender specific. The association of malaria risk with male gender and young adulthood has been shown elsewhere in the GMS in recent studies, particularly those measuring parasite prevalence [
12‐
15]. This correlation is thought to be due to young males spending more time travelling and working in the forest which are focal points of malaria transmission. In some locations in the GMS, clinical episodes of malaria are more common in children [
16].
This study investigated a large number of possible risk factors using a block design for the regression analysis to identify which risk factor within each category were most discriminatory. This analytic method has the advantage over the alternative of a stepwise analysis of all individual risk factors in that the range of questions within each theme can first be narrowed, allowing multiple different risk categories to be identified. The number of possible combinations of the 50 variables is also very large with a high chance of interactions among the variables, which would render analysis across all variables infeasible. The limitations of stepwise regression for large datasets has been described in the literature [
17]. Instead, a stepwise analysis was conducted within each theme using expert knowledge to decide which variables make sense to combine followed a stepwise analysis of the discriminatory variables from each theme. It is possible that this method may have missed some individual variables which could have improved the final multiple regression model whilst not appearing significant in the analysis of each block. It is likely that any additional contribution of these variables to the model would be very marginal, however.
The final output from the multiple logistic regression cannot be directly related to causality, as the analysis yields purely statistical associations. This is because variables that dropped out of the multivariable model could be more closely related to causality whilst being less strongly associated with the outcome than risk factors which remained in the model. Variables that are positively associated with being a case in a univariate or block-wise analysis can be associated with being a control when combined with additional variables. An example is the number of days spent in the forest, which was slightly higher among cases, but predicted in the opposite direction in the final model. This is because other variables in the model accounted for all of the positive association of this variable with being a case. For this reason, the output of a multivariable model like this, although of academic interest, is of limited value for guiding national program planning as it only works if all variables are included and it does not give information about the relative importance of individual risk factors in isolation. It is, however, of value in identifying from particular sets of variables from different domains which combination are the best predictors of malaria risk and this could be used to help guide what information to collect in future studies.
Block-wise and univariate analyses are more informative to identify individual risk factors and risk groups. In this study, visiting the forest was strongly associated with being a case, with a higher number of nights in the forest increasing that risk. Forest workers were the occupation group most strongly associated with malaria, however, these comprised only 7.8% of all cases. An additional sixty percent of cases visited the forest, but did not work there. Of note, there was no association between the different activities in the forest and risk of being a case.
Approximately a third of cases had not been to the forest in this study during the 2 months prior to diagnosis, during which almost all new infections should have occurred [
18,
19]. One possibility is that some of these cases were recurrences of
P. vivax malaria from hypnozoites from a previous malaria episode. However, there was no difference in the proportion with
P. vivax between those who had and those who had not been to the forest suggesting this not to be the case. Another more likely possibility is that these cases were infected elsewhere, either in their village of residence, village transmission is a possibility due to the observation of cases in children under 10 years of age (Fig.
1) who are not likely to visit the forests. It is also possible these cases were infected in a location they do not consider to be forest or they failed to recall having actually been to the forest. Previous studies have found a strong association between forest visits and malaria in Viet Nam [
10,
20] and have highlighted the challenges of reducing transmission among forest-goers. These challenges include that people visiting the forest are less likely to use a bed net when doing so and that much of the biting occurs before sleeping time [
21].
Other major risk factors for malaria from this study included duration of illness, confirmed fever and prior treatment for malaria, as well as having previously attended a government health centre or private sector facility for this illness. These may be an indication of healthcare seeking behaviour; people may be shopping around other healthcare providers before visiting the CHS, hence presenting late in their infection course. A plausible explanation is that their malaria infection is being missed on the initial encounter, or that they are failing treatment due either to inappropriate or insufficient anti-malarials, or they are infected with anti-malarial resistant strains. Of the cases who had received anti-malarial treatment within the preceding two months, almost all were treated at a government health facility and almost all took the full course of medication as prescribed. Over 80% of those who could remember their treatment were managed in accordance with the national or WHO guidelines. This suggests insufficient or inappropriate anti-malarials were not a major problem. This relies, of course, on accurate recall of details of treatments received by individuals but the information was volunteered by participants, without prompting, so is likely to be accurate. Around a quarter of enrolled people had attended a pharmacy for this illness and just over 3% a private sector health facility. However, none reported definitely receiving an anti-malarial from these sources. This may be due to a reported misconception that pharmacies are not authorized to provide anti-malarials, whereas they are able to provide medication for those with a prescription [
6].
There are a range of implications of these findings for NIMPE malaria management and elimination. This study confirms that males of working age were at higher risk for malaria than other demographic groups; over 90% of cases were male in this study. Secondly, independent of age and gender, people who spent time in the forest in the two months preceding presentation to a clinic, for any reason, were at higher risk of being a malaria case. The malaria risk was even higher for those who spent more nights away from their home and for those who worked in the forest. It should be noted that over 80% of the subjects who visited the forest did not work in the forest. Of the reasons given for visiting the forest, no factor further increased the risk of malaria above that of forest-goers in general. This implies that NIMPE should target interventions at anyone who visits the forest as opposed to specific occupations. Particular attention should be paid to males of working age, people who travel overnight for longer periods and those who work in the forest.
People who had malaria in the past year were also at a higher risk for infection, as were those with household members who had malaria in the past year and those who had previously received anti-malarial treatment at any time. This suggests there are particular groups of people who are being repeatedly infected through high risk behaviours. The number of previous episodes of malaria was independent of whether they had visited the forest in the previous 2 months. This could be because these people had visited the forest previously or because they were being infected elsewhere. It would be informative to explore in more detail about these multiply infected people to identify if their previous diagnoses were confirmed and whether they cluster geographically which could indicate potential transmission hotspots. It may also be worthwhile, targeting these people and households to educate them about malaria risk and ensure they are using healthcare services and personal protection measures optimally.
The identified risk factors are helpful at a population level to identify which groups to target with more intensive interventions and/or to help guide allocation of limited resources e.g. ensuring access to diagnostic and treatment services, LLIN distribution, personal protection measures or targeting of indoor residual spraying. This is particularly pertinent as numbers of cases decline with consequent reductions in funding for malaria elimination [
3]. Knowledge about who is at risk can also help to guide delivery of audience-specific public health messaging about malaria prevention and working with these groups to design and implement context appropriate measures, such as hammock bed nets [
22] or mobile outreach teams (MOTs).
Another potential application of these findings is to identify modifiable risk factors for malaria. This study found a range of such risk factors, which could be addressed by specific interventions. These could include reducing forest visits and avoiding or minimizing overnight stays in the forest, particularly in areas where the proportion of cases was highest; for example, Bu Gia Map National Park. This has been attempted elsewhere with varying success e.g. through enforcing logging and timber export bans [
23]. Just under half the people in this study who visited the forest cited exploitation of timber, minerals or animals as the reason for doing so. As this study was not able to quantify what proportion was for logging, particularly illegal logging, the potential impact of a strictly enforced logging ban on malaria in Viet Nam cannot be determined. Although use of LLIN was high, this study did not collect information on use of personal protection measures specifically in the forest and further work would be required to explore this in detail.
Other potentially modifiable risk factors identified included longer duration of illness and the presence of recorded fever. This is of concern as longer duration of malaria infection could lead to increased transmission before treatment clears parasites from the blood. Early diagnosis and treatment of malaria (EDTM) has been a key component of national malaria strategy in Viet Nam for many years [
24].
Community engagement and education to re-emphasize the importance of EDTM and encourage and support people with a fever to seek medical care as early as possible would be beneficial to address this as well as control for other emerging infectious diseases, such as novel coronavirus.
LLIN coverage and usage in the study population was very high, most LLIN were less than 1 year old and very few had holes. This suggests the LLIN distribution programme is functioning well in Viet Nam which is reassuring as it accounts for the majority of external funding for malaria elimination. Smaller numbers of people used other bite prevention methods and tended to use multiple methods or none at all. Repellents and plug-ins were protective against malaria, but coils were not. Unfortunately, due to space limitations, data on use of LLIN or other measures whilst in the forest was not collected.
This information is of limited use for clinical management of individuals as all possible malaria cases are required to have a confirmatory diagnostic test before treatment. Having a pre-test probability of malaria based on risk factors would not change this for people living in endemic areas as testing should be based on the presence or absence of fever without an alternative diagnosis. However, for locations where malaria transmission is very low, it becomes inefficient to test everyone with a fever for malaria. The WHO recommends testing should be based on whether a person may have been exposed to malaria, using history of, for example, travel to a malaria endemic area without protective measures together with a fever or history of fever with no other obvious cause [
25]. A more nuanced approach could be to include some of the more strongly associated risk factors identified in this study to develop a screening questionnaire, either as a checklist to derive a score or to assign an individual level pre-test probability of malaria to guide testing. This could be used to help guide who should undergo a diagnostic test as part of passive case detection or even as part of active case detection strategy to optimize use of limited testing resources. Scoring systems based on symptoms are not recommended by the WHO, as they can be complicated to implement and supervise and the key features may be different in different locations [
25]. Similarly, risk factors are likely to be different in different locations. Even if it was possible to implement, such a system may only be of use in areas where testing rates are low as it could increase appropriate testing of higher risk individuals thus improving sensitivity of the surveillance system. In areas where testing rates are already high, there is a risk that it could reduce sensitivity by excluding some individuals from testing with the benefit of increased specificity.
The rate of submicroscopic infection was low in this study, with 31 out of 1031 PCR positive individuals being microscopy negative. This differs from other studies in Viet Nam and other low transmission settings which have found the majority of infections to be submicroscopic and/or asymptomatic in cross-sectional surveys [
26]. This is probably because all individuals in this study were symptomatic with likely higher parasite burdens than those who are asymptomatic, and were thus more likely to be detectable by microscopy than the general infected population. Differences in sensitivity of PCR and/or microscopy are less likely explanations as the PCR in both studies was done by NIMPE (although previously done by semi-nested PCR, the sensitivity was the same or lower than nested PCR—personal communication) and microscopy was the routine diagnostic methodology used in Viet Nam.
The methodology used had several strengths worth highlighting. Diagnostic testing to identify cases and controls was robust, being done by microscopy and confirmed by PCR in all cases. Microscopy performed very well against PCR with minimal discrepancy. All interviews were done by healthcare workers with relatively little research experience and training. This was supported by academic partners without needing to employ large numbers of field staff. The study materials were developed and optimized by the study partners prior to roll-out to ensure maximal clarity and acceptability of the data capture form and high quality of the information recorded. The data capture form itself was kept as short as possible to minimize workload whilst still collecting sufficient detail. The results suggest relevant questions were properly selected as the model using data from the shorter data capture form was able to correctly identify nearly 80% of cases and controls. The most resource intensive part of the data management and analysis was of the travel data due to the range of different types of travel and amount of geographic information collected. This was made more efficient by use of a standardized travel survey format, which had already been used in other studies across the GMS. This was then entered into a previously developed data processing pipeline and analysis framework [
27]. Taken together, these also confer the advantage that this study is relatively easy and cost efficient to repeat and scale up in future in this or other locations. A final strength is that the study was run in close collaboration with NIMPE who are the main beneficiary of the results. Thus very early discussions could be had about the implications as results were generated.
Limitations of this study include that the quality of the data relied on the participant’s accuracy of recall and willingness to disclose information. This is one reason this study focused on travel within the previous 2 months as recall over longer periods can be notoriously unreliable. It is likely that there are items that people were not willing to discuss, such as illegal motivations for visiting the forest, unofficial employment or travel to sensitive locations or illicit crossing of international borders. In some cases, study participants may have adjusted their responses to fit with what they thought the interviewer (a government health worker) wanted to hear, for example they may not want to disclose about use of health services in the private or informal sector. The study sites were all CHS so the study would have missed people who only attended elsewhere including the private sector and larger hospitals. This is particularly a concern for migrant populations whose access to health services is restricted by the Law on Residence which limits health insurance cover to those with permanent registration status [
28]. However, it is thought that the vast majority of people with malaria in Viet Nam access health services via CHSs [
6]. The quality of the data also relied on the training of the interviewer. For most variables, the data was of extremely high quality and complete. The level of detail for a few variables was constrained, however, by grouping responses into categories at the time of data collection. One example is the cited reasons for visiting the forest where a breakdown of the terms “exploitation” and “foraging” was not available. A further limitation is that age and gender were identified as risk factors by comparison with census. It may be that people living in the malaria endemic areas have different age and gender profiles from the census. However, it is unlikely that they matched the profile of malaria cases of being almost all male and of working age.
This study collected detailed information on travel and it was only possible to provide a summary here, the focus being on identifying risk factors for malaria. A follow-on paper with a more detailed descriptive and modelling analysis of travel and the impact on malaria distribution in Viet Nam will be published separately.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.