Insecticides used to target the Anopheles mosquito vectors of the Plasmodium parasite have long been a cornerstone of global malaria control. The vast reduction of malaria transmission observed over the past decades is largely attributed to insecticide-based interventions (Bhatt et al., 2015; World Malaria Report 2023, 2023), but following increased use over the past two decades—most intensely in sub-Saharan Africa—resistance to these insecticides is now widespread (Ranson and Lissenden, 2016). There is recent evidence that increasing resistance may be impacting the efficacy of vector control tools, consistent with stalled progress in malaria control since 2019 (Alonso and Noor, 2017; World Malaria Report 2023, 2023). However, this has been difficult to quantify as insecticide resistance assays are not easily employable in all settings (Toé et al., 2014; Hemingway et al., 2016). In addition, there remain important gaps in the data on the intensity and spread of resistance in wild populations (Mnzava et al., 2015; Moyes et al., 2019, 2020). Practical, cost-effective techniques for routine resistance monitoring are urgently needed, both to understand the impact of insecticide resistance on malaria transmission and to signal the need for alternative interventions when resistance emerges (Kont et al., 2023).

Resistance to at least one class of insecticides is reported in the majority of malaria-endemic countries (World Malaria Report 2023, 2023). Insecticide resistance, broadly, is caused by four different mechanisms: target site resistance, metabolic resistance, cuticular resistance, and behavioral resistance (Suh et al., 2023). While there are individual molecular markers for some of these mechanisms, resistance from target site resistance, metabolic resistance, and cuticular resistance can all be measured using a bioassay, or simply assay, in which mosquitoes are exposed to insecticide and then assessed for survival.

In spite of this complexity, insecticide resistance is typically categorized as a binary measure (confirmed resistant or not) in WHO susceptibility tests, based on a threshold of 10% of mosquitoes surviving exposure to insecticide after 24 hours (World Health Organization, 2022). A secondary cutoff of 2% survival is used as a measure of “suspected resistance.” Monitoring insecticide resistance is one of the main pillars of the Global Plan for Insecticide Resistance Management in malaria vectors (GPIRM), and sentinel sites are directed to monitor populations at least annually (World Health Organization, 2012). Their recommendations for prolonging the efficacy of insecticide interventions are dependent on local insecticide resistance levels, although the evidence base for these recommendations is poor (Madgwick and Kanitz, 2022; Hobbs et al., 2023).

Systematic, long-term, and spatially representative monitoring, however, requires increased sampling of mosquito populations beyond the current system. The WHO’s ‘Test procedures for insecticide resistance monitoring in malaria vector mosquitoes’ advises that control programs select sentinel sites that will be representative of both the eco-epidemiological zones and malaria intensities in their region (World Health Organization, 2016). However, neither of these variables are static over time—malaria prevalence changes in response to malaria interventions, and eco-epidemiological zones change in response to interventions as well as climate and land use change (Ryan et al., 2015, 2020; Rodó et al., 2021). Thus, sentinel sites are likely to become less representative of the insecticide resistance landscape over time. In addition, the GPIRM emphasizes the importance of monitoring and responding to insecticide resistance at the local level (World Health Organization, 2012) because Anopheles populations cluster at a relatively small geographic scale (Carter et al., 2000; World Health Organization, 2012). High resolution monitoring of insecticide resistance is important in order to allow for local-level decision making, as well as to understand the changing distribution of insecticide resistance. Although the importance of insecticide resistance monitoring is well outlined, there remain large gaps in data in many malaria-endemic regions, both geographically and temporally (Mnzava et al., 2015; Moyes et al., 2019, 2020). Recent work used spatial modeling methods to try and fill in the gaps in resistance data across sub-Saharan Africa, constructing annual maps of predicted resistance level by district (Moyes et al., 2020). However, many locations still have insufficient data, leaving inferred results unable to capture geographic heterogeneity in resistance levels.

The WHO’s current field protocol for testing resistance recommends using mosquitoes raised from collected larvae or the first generation (F1) progeny of collected adults, both of which require access to mosquito rearing facilities and take a week or longer to rear to the correct age for testing (World Health Organization, 2012, 2022). Larval collection is generally easier than adult collection to obtain a large number of individuals. However, resistance assays conducted on wild-caught adult mosquitoes–which we will refer to here as adult-capture assays–offer an alternative with some benefits, and are already in use for capturing mosquitoes that are difficult to find at the larval stage (Hargreaves et al., 2003; Rakotoson et al., 2017; Matowo et al., 2021).

First, insecticide bioassays are not necessarily a good predictor of the failure of vector control tools or of malaria infection risk (Thomas and Read, 2016; Kleinschmidt et al., 2018). While larval testing under standardized conditions is a good indicator of technical resistance—to borrow language from Namias et al.—that is the genetic change that leads to reduced susceptibility in lab-controlled bioassays—testing on wild adults may actually be a better indicator of practical resistance—the change in the mosquito that leads to mosquito control failure in field settings (Namias et al., 2021). In addition to allowing for collection of vector species that are not easily found at the larval stage, sampling adults may better represent the mosquitoes of interest: as they are the vectors that directly contribute to transmission, resistance in adult mosquitoes may be a better representation of “real world” resistance. Samples collected as adults may also have a lower likelihood of pseudo-replication due to relatedness within larval samples collected from the same breeding site. It is therefore possible that testing for insecticide susceptibility on mosquitoes in less standardized conditions could help shed light on the connection between measured resistance and vector control efficacy (Namias et al., 2021). A second advantage of adult-capture assays, though they are more labor intensive with respect to sampling, is that they may be conducted in the absence of rearing facilities and in a shorter time frame, as there is no need to wait for eggs or larvae to grow to adulthood, lending this method more flexibility. Adult-capture assays therefore have potential as a tool for increasing the spatial coverage of insecticide resistance monitoring, and developing more systematic monitoring programs.

Despite these potential advantages, assays on wild-caught adult mosquitoes are not currently recommended by the WHO. This is in part due to age and exposure heterogeneity in adult sampled mosquitoes, and in part due to the difficulty of collecting a large number of adult mosquitoes. Unlike mosquitoes raised from larvae, which are tested at a standardized age (usually 3-5 days), mosquitoes captured as adults have unknown age at the time of testing and are likely to have a variety of insecticide exposure histories, including the number of past exposures, the timing of exposures, and their duration. Laboratory experiments have shown that insecticide resistance declines with mosquito age (Rajatileka et al., 2011). Resistance also appears to decrease when mosquitoes were previously exposed to insecticides, weakening even resistant mosquitoes (Viana et al., 2016). These heterogeneities and dynamic aspects of resistance measured in adult mosquitoes must be considered to compare adult capture assay results with standard methods, and to adjust for real-world settings.

In the absence of studies directly comparing insecticide resistance from adult mosquito samples to that of larval samples, mathematical models offer transparent frameworks to account for the likely impact of aging and exposure heterogeneities among adult mosquitoes. Dynamical models have a long history in the context of malaria interventions, because they capture the nonlinearities inherent in biological populations that may not be intuitive, and they distill the most important dynamic factors likely to affect complex phenomena like patterns of resistance. Here, we developed a mathematical model of the mosquito life cycle to examine the impacts of heterogeneity in age and exposure history of wild adult mosquitoes on measured assay survival in simulated adult-capture samples. We modeled a mosquito population with inherited insecticide resistance, incorporating variations in insecticide susceptibility due to age and insecticide exposure history. Using this model, we have quantified how sampling from heterogeneous adult mosquito populations could impact the categorical results of insecticide resistance assays, including the positive and negative predictive values. We also compared adult- and larval-capture assays as measures of the true resistance frequency in the modeled population. We find that adult mosquito sampling offers a practical complementary approach for monitoring insecticide resistance in malaria-endemic regions.

We ran the population model for 10 years starting with 1% resistance allele frequency in 5% increments of insecticide coverage, from 0% to 95%, and in 10% increments of the heterogeneity parameter, from 0% to 100%. Despite always starting with an almost entirely susceptible population, the population always reached either a fully susceptible (0% resistant allele frequency) or fully resistant (100% resistant allele frequency) equilibrium during this time ( Figure 2 ). The population became fully resistant when insecticide coverage was 70% or higher, and fully susceptible when insecticide coverage was 25% or lower. The precise coverage at which the population became fully resistant depends on parameter choices, and biological interpretation requires precise knowledge of a particular setting. The amount of time it took for the population to become fully resistant also depended on the coverage and degree of heterogeneity, with higher coverage and lower heterogeneity each leading to faster emergence of resistance ( Supplementary Figure S5 ).

The assumption of greater heterogeneity in the population led to a higher coverage required to reach the fully resistant equilibrium. When the heterogeneity parameter was equal to one, meaning that whether or not a mosquito was exposed on its first blood feed determined exposure on all of its future blood feeds, the population became fully resistant when insecticide coverage was 70% or higher. When the heterogeneity parameter is equal to zero, the population became fully resistant when insecticide coverage was as low as 30% or greater ( Figure 2 ).

To understand the implications of using adult-capture samples rather than larval-capture samples when the population is either fully resistant or fully susceptible, we simulated assays when the population was at equilibrium using each method and directly compared their results ( Figure 2 ). The difference between modeled bioassay survival using adult-capture vs larval-capture depended on the level of resistance in the population. In a population that is completely susceptible, adult-capture simulations overestimated resistance by approximately 3-5%. In a population that is entirely resistant, adult-capture assays underestimated resistance by 5-15%.

This difference is driven by the relative exposure history of the population at different levels of population resistance. In a predominantly resistant population, many resistant mosquitoes have survived at least one insecticide exposure, and prior exposure makes resistant adults more susceptible to insecticide exposure (Viana et al., 2016). A sample of the adult population will include many mosquitoes that will die even though they are homozygous resistant because they already have been exposed previously, making the population appear less resistant than it truly is.

The magnitude of this difference in the resistant population depends on coverage (i.e. varies horizontally in Figure 2 ), with higher levels of coverage leading to a larger difference between adult and larval-capture assay results. It does not, however, appear to differ by the degree of exposure heterogeneity (i.e., vertically in Figure 2 ). We interpret this to mean that the bias in an adult-capture assay is driven primarily on the level of resistance in the population, with coverage as a secondary determining factor.

While some populations may be either fully susceptible or fully resistant, it is more likely that a population in the field will have an intermediate level of resistance. To investigate the difference between these two sampling methods in a non-equilibrium population, we sampled from the population at varying levels of resistance allele frequency. We also sampled at varying levels of the heterogeneity parameter and insecticide coverage. When we aggregated across exposure heterogeneity and coverage, assays based on larval- or adult-capture both measured true resistance relatively well, but with an observed bias in the adult-capture sample tests ( Figure 3A ). Assays conducted with 50 mosquitoes in each of the exposed and control groups, rather than 100, performed similarly, although individual test results had a wider variance ( Supplementary Figure S6 ).

As was observed in the equilibrium populations, there is a bias in adult capture tests, where resistance is overestimated at lower levels of resistance allele frequency and underestimated at higher levels of resistance allele frequency. This is also driven predominantly by the exposure history of the population: in a mostly resistant population, resistant mosquitoes are likely to have a history of exposure to insecticides, and will therefore be more likely to die in a resistance assay than an unexposed resistant mosquito in a larval assay. On the other hand, in a mostly (but not entirely) susceptible population, resistant mosquitoes are disproportionately overrepresented in a sample of wild adults because they are able to survive insecticide exposure (whereas susceptible mosquitoes of the same age will have died when exposed). At intermediate levels of resistance allele frequency (~50%), these two effects counter each other, and both bioassays will measure the same level of survival ( Figure 3A ).

To “correct” for this difference between these two sampling methods, we took the mean of 100 adult-capture samples and the mean of 100 larval-capture samples for a given day (associated with a resistance allele frequency). We then subtracted to find the difference in mean bioassay result at each value of resistance allele frequency ( Figure 3B ). Depending on the level of resistance, the bias in the adult-capture model ranged from a mean 3% overestimate of resistance at very low resistance allele frequencies to a mean 11% underestimate of resistance at very high resistance allele frequencies. These results are similar to when the population was at equilibrium ( Figure 2 ). We then modeled this survival difference against adult assay survival using a quadratic function, generating a correction model function (black line in Figure 3B ).

We tested this correction model by applying it to adult-capture assays and, alongside the results of unadjusted adult-capture and larval-capture assay results, compared the results of this correction to the true resistance allele frequency. Using this adjustment ( Figure 3B ), the bias in the adult-capture assays was reduced: the mean square error (MSE) of the adjusted adult-capture (0.0038) was lower than in the unadjusted adult-capture (0.0059), but was still greater than in the larval-capture (0.0018) ( Table 2 ).

Finally, to evaluate the performance of adult-capture assays compared to larval-captured assays in a field context, we estimated the sensitivity, specificity, and negative and positive predictive values of adult-capture and larval-capture assays simulated in our model. We generated these for both the 10% “confirmed resistance” threshold and the 2% “suspected resistance” threshold used by the WHO. Although continuous measures of resistance are important for understanding the strength of resistance in a population, these categorical cutoffs remain a standard part of field malaria programs.

Assays done with adult-capture samples correctly identified “confirmed resistance” (10% or higher) 99.5% of the time with samples of 100 tested and 100 controls ( Table 3 ), and 98.5% of the time with samples of 50 tested and 50 controls ( Supplementary Table S4 ). This was higher than those done with larval samples, which had a sensitivity of 98.4% (n =100) and 97.3% (n = 50), respectively ( Table 3 ; Supplementary Table S4 ). Larval-capture samples, however, correctly identified insecticide susceptibility (<10%) 91.1% (n=100) and 89.9% (n=50) of the time, which was higher than adult-capture samples (73.1% with samples of n = 100, and 72.7% with samples of n = 50).

In field contexts, positive predictive values (PPV) and negative predictive values (NPV) are particularly relevant, as the true resistance allele frequency is unknown. We evaluated PPV and NPV across the entire dataset, which ranged from 0 to 100% resistance allele frequency. With adult-capture samples, results showing “confirmed resistance” (i.e. >10% resistance allele frequency) were truly >10% resistant 94% of the time when n=100 and 93.8% of the time when n=50. With larval-capture samples, the corresponding PPVs were slightly higher (97.9% and 97.6%, respectively). However, the NPV was much higher using adult-capture samples than larval-capture: with 100 samples each for test and control groups, NPV was 97% for adult-capture samples and 19.2% with larval-capture samples. The NPV was similar with n=50 (92.1% and 19.2%, respectively).

The results for suspected resistance (>2%) followed similar trends to those for confirmed resistance, although specificity and NPV were lower for both sampling methods and specificity in particular was much lower for adult-capture ( Table 4 ). Sensitivity was 99.8% (n=100) and 98.9% (n=50) for adult-capture samples, compared to 98.3% (n=100) and 96.5% (n=50) using larval capture ( Table 4 ; Supplementary Table S5 . Specificity, however, was 19.4% (n=100) and 5.1% (n=100) using adult capture compared to 60.2% (n=100) and 48.7% (n=50) using larval capture. Using adult-capture samples, PPV was 96.0% (n = 100) and 96.5% (n=50) compared to 97.7% (n=100) and 98.2% (n=50) using larval-capture samples. NPV for this category of resistance was particularly low for larval-capture samples. Using larval-capture samples, NPV was 4.2% for both sample sizes compared to 54.6% (n=100) and 44.3% (n=50) for adult-capture.

Insecticide resistance represents a major threat to the gains in malaria control achieved over the past two decades. Monitoring the spread and strength of insecticide resistance is a key component of surveillance for malaria control programs, and current approaches are lacking particularly with respect to spatial representation. This is, in part, connected to the limitations of current resistance assay protocols. Given the importance of insecticide resistance in limiting disease control, it is important to fully consider how monitoring insecticide resistance can be improved. Here, we used a mathematical model of a mosquito population to compare the results of simulated adult- and larval-capture assays. We found that, despite bias in assays using adult-capture, they may still be a good measure of resistance, particularly around the WHO categorical resistance cutoffs.

Although capturing adult mosquitoes is a more labor-intensive process than capturing larvae, there are settings in which adult-capture may be a more practical strategy for insecticide resistance monitoring than larval-capture ( Supplementary Table S6 ). Adult-capture can be done without the need for raising mosquitoes from larvae, and is therefore more portable, which may allow for testing in more remote and varied locations. In addition, they require much less time following collection—tests can be conducted immediately after collection, provided there are sufficient mosquitoes sampled, as opposed to the week or more required to raise larvae or F1 progeny to the correct age in the insectary. These advantages could allow for a shift beyond the current sentinel site model, towards more systematic monitoring over geographic regions. However, adult-capture does have its challenges. Collection must be done carefully, to avoid damaging the mosquitoes, and taxonomic identification without damaging specimens requires expertise. While adult-capture does not require the same insectary capacity as raising mosquitoes from larvae, it still requires some infrastructure so that mosquitoes can be kept alive for testing. These challenges must be considered locally before promoting the use of adult sampling over larval sampling.

Despite these challenges, adult-capture may in fact be a more representative way of monitoring the mosquitoes that are driving malaria transmission (World Health Organization, 2012). In theory, also supported by our model, larval-capture results are a very good measure of true resistance in the full mosquito population. However, this depends upon the assumption that the larvae are drawn from the same population that we are interested in measuring, i.e. the mosquitoes that are actively biting humans and, thus, likely contributing to malaria transmission. This may not be the case in the field: larval collections, in particular, may overrepresent a small portion of the mosquito population (due to egg laying behavior) and adult-capture can more easily target mosquitoes that are actively seeking out humans for blood feeding. The results of this modeling study show that even when there is bias in adult-capture assay results, it has a relatively comparable sensitivity and positive predictive value and therefore may be an improvement over larval-capture if the logistical challenges can be overcome.

One limitation of this model is that it relies on a single data set to parameterize the mortality of resistant mosquitoes following multiple exposures. For this reason, we are not able to model multiple Anopheles species, which often overlap in a single ecological setting. In addition, some field experiments have demonstrated greater survival in adult-captured mosquitoes than measured in the birth cohorts, which appears counter to the results of the lab work in Viana et al. (Hughes et al., 2020). However, from our simulated bioassays we show that adult-capture can yield either higher or lower mortality than larval-capture, depending on true population resistance.

Field experiments that directly compare larval- and adult-captured assays in regions with varying levels of resistance are necessary in order to understand whether adult-capture could be used to replace larval-capture assays. We hope that these modeling results, showing that the theoretical potential for using adult-capture samples for resistance assays, is a step towards motivating additional future investigation into this question.

The model framework is not intended to perfectly capture every detail of real-world contexts; rather, it examines the impact of assumptions about the most important drivers of heterogeneities in resistance patterns in the context of variable mosquito age and exposure. Our results demonstrate that adult-capture, in addition to the current standard larval-capture, could be a useful and consistent measure of resistance. Field experiments should be conducted to validate our results, which could allow adult-capture to be prioritized moving forward as an alternative strategy for wide scale testing to the methods currently recommended.