PREreview of Efficacy of 60 personal mosquito repellents and perfumes against Aedes aegypti mosquitoes

This review was prepared in partial fulfillment of BIO 5201 as part of a peer review training exercise. All authors contributed equally and are affiliated with the Department of Biology, Baylor University.

Summary Statement

Ross and colleagues ask two practical questions: how well do personal mosquito repellents sold in Australia prevent Aedes aegypti landings over time, and do products deliver the protection claimed on their labels? The authors tested 60 products bought in stores and online in 2025: 55 marketed repellents and five non-repellent perfumes, deodorants and an antiperspirant included as comparators. The repellents comprised 43 topical formulations spanning 18 primary active ingredients, four wristbands, three clothing stickers, three ultrasonic devices, one fabric-applied repellent and one permethrin-treated garment. Each product was tested once on each of eight human subjects (480 trials) in an arm-in-cage assay adapted from WHO guidelines. Every test was paired with an untreated control arm, and testing was repeated hourly for up to 8 hours. Complete protection was defined as a reduction of at least 95% in landings relative to the control. The permethrin socks were assessed with a blood-feeding assay.

Only topical or fabric repellents containing DEET, picaridin, oil of lemon eucalyptus or IR3535 provided more than 2 hours of complete protection, with substantial variation among products sharing an active ingredient. Wristbands, stickers and ultrasonic devices never provided complete protection, and most essential-oil products lost protection within an hour. Nearly half of the non-exempt repellents (25/52) were not registered with the APVMA, and these tended to be less effective. Only 5 of 28 products met their advertised protection time, with no relationship between advertised and measured protection. Although relative product rankings were consistent across subjects, overall efficacy differed markedly between people. By pairing standardized testing with registration status and label claims, the study provides timely, consumer-relevant evidence on regulatory gaps and on which repellents work.

Overall, this is a large and useful dataset. The conclusions about the most effective active ingredients and the ineffectiveness of wearable devices are well supported and consistent with the literature. Our main concerns relate to observer blinding, the comparability of the protection endpoint with label claims, the stopping rule and the statistical treatment of protection time, and confounding in the market-level and registration comparisons. Addressing the points below would considerably strengthen the manuscript's comparative claims.

Major Comments

1. Blinding of the observer recording landings

Landings were scored by eye (a mosquito on the skin for more than 3 seconds), and the arm being scored was by design known to be either treated or untreated. The manuscript does not state who counted landings, whether that person was the subject or a second observer, or whether the counter knew which product was applied. Because the primary endpoint depends on a judgment call at the 3-second boundary, unblinded scoring could bias counts in the direction of expectation (for example, toward strong performance of DEET products and poor performance of wristbands). We recognize that full blinding is likely impossible for strongly scented products, and that this is itself worth disclosing. The authors could consider stating who recorded landings and whether they were blinded, which would help readers judge the risk of observer bias, adding this as a limitation if they were not, which would make the potential for expectation bias transparent, and, if any trials were video recorded, reporting agreement between the original counts and blinded re-scoring of a subset, which would show how robust the counts are to the 3-second judgment call.

2. Comparability of the study's protection endpoint with advertised protection times

The conclusion that only 5/28 products met their advertised protection time compares two quantities that may not measure the same thing. The authors define loss of complete protection as relative protection falling below 0.95, which permits some landings during the "complete protection" period, and measure it at 1-hour intervals to a ceiling of 8 hours. Manufacturers' claims may be based on time to first bite or landing, on different assays, mosquito species or doses, and two products claimed 48 hours, far beyond the observation window. The Introduction notes that information on how advertised times are derived is lacking, but the Results and Discussion treat the comparison as direct. The authors could consider (i) reporting the WHO-standard time to first landing alongside the 95% definition, which should be recoverable from the per-test data in Table S3, so readers can see whether the 5/28 figure is robust to the endpoint used; (ii) noting how the 1-hour sampling interval and 8-hour cap bound the measured values; and (iii) framing the claim as products not meeting advertised times "under these assay conditions". Together, these changes would clarify whether the 5/28 figure reflects a true shortfall or a difference in how protection was defined and measured.

3. Stopping rule and how protection time was assigned

The stopping rule is an absolute count (more than 5 landings on the treated arm in two consecutive tests), while loss of protection is a relative threshold (below 0.95). Control counts ranged from 30 to 361, so the two criteria diverge: with a control count of 300, six treated landings still represent 98% protection, so a trial could stop while protection is still "complete" by the authors' own definition. It is not stated how protection time was recorded in that case. The rule is also described inconsistently: "more than 5" in the Methods, "5 mosquitoes" in the Statistical analysis and "5 or more" in the Figure S4 legend. The authors could consider stating the rule used, which would resolve the inconsistent descriptions, explaining how protection time was assigned when the stopping criterion was met before relative protection fell below 0.95, and reporting how many trials this affected, which would show whether this issue could have influenced the results. In future experiments, the authors could consider a stopping rule scaled to each test's control count to align data collection with the analyzed endpoint.

4. Statistical treatment of time to loss of protection

Time to loss of protection is recorded at hourly intervals and is capped at 8 hours, so it is interval-censored and, for the longest-lasting products, right-censored; it is analyzed as a continuous Gaussian response in linear mixed models. The model description (category + product within category + human subject + interaction) is also unclear. With one trial per product and subject, a product-by-subject interaction cannot be separated from residual error, and the fabric-repellent category contains a single product, so its category estimate has no within-category variance. Several separate mixed models are reported (category, active ingredient, formulation type, registration) without stating whether any correction was applied across them. The authors could consider an interval-censored survival or time-to-event model (for example, a mixed-effects accelerated failure time or Cox model with subject as a random effect), which would account for the censoring and the 8-hour cap; clarifying exactly which terms were fitted and what "interaction" refers to, which would make the fitted model clear and reproducible; reporting the number of products per category, with n = 1 categories described rather than modeled, which would avoid estimates with no within-category variance; and stating the approach to multiple testing across analyses, which would clarify how reliable the reported P values are.

5. Representativeness of the product sample and market-level claims

Products were deliberately selected to include at least three per known-effective active ingredient (DEET, picaridin, oil of lemon eucalyptus, IR3535 and citronella), plus targeted searches for other ingredients. This purposive sampling suits comparing active ingredients but does not give a random or representative sample of the Australian market. However, the Abstract and Discussion present proportions such as 25/52 unregistered products and 5/28 meeting advertised times as statements about the market's prevalence of unregistered and ineffective products. The authors could consider describing the selection explicitly as purposive, stating how many candidate products were identified and how the final 60 were chosen, and wording market-level percentages as properties of the tested sample rather than estimates of market prevalence. Doing so would clarify the scope of the conclusions and keep readers from interpreting sample proportions as market prevalence.

6. Confounding in the APVMA registration analysis

The finding that registered products were more effective (Figure 3E) may reflect differences in what registered and unregistered products contain rather than registration itself. All unregistered products were bought online, and the product list suggests they are concentrated in essential-oil and botanical formulations, which the authors show are generally less effective. Without adjustment, registration status is confounded with active ingredients, concentration, and formulation. The authors could consider including primary active ingredient class, and ideally concentration, as a covariate in the registration model, or stratifying the comparison within ingredient classes that contain both registered and unregistered products, and tempering the causal language accordingly. These changes would help clarify whether the apparent advantage of registered products reflects registration itself or differences in formulation.

7. Unstandardized application dose

Topical products were applied "liberally" from wrist to elbow, and the amount applied was not recorded. The WHO guidelines from which the assay is adapted specify a standardized dose per unit area of skin. The authors acknowledge in the Discussion that formulations ranging from balms to wipes likely led to different amounts being applied, but acknowledging the problem does not control for it, and the dose is a plausible driver of the large within-ingredient variation in Figure 3B and the formulation effect in Figure 3D (for example, roll-ons outperforming wipes). If product mass was recorded (for example, by weighing containers before and after application), the authors could consider reporting it and including it in the models, which would help separate the effect of dose from the effect of the product. If not, the authors could consider stating this explicitly as a limitation in the Methods and qualifying the formulation-type comparison, which would help readers interpret the formulation effects appropriately. The authors might also consider a standardized-dose retest of a small subset, such as the DEET products, to show how much of the variation is attributable to dose.

8. Absence of a reference standard across an 11-month testing period

Trials ran from September 2025 to August 2026 using mosquitoes from generations 18 to 22, aged 5 to 25 days, in separate batches. The untreated control arm normalizes for mosquito activity within a test but does not show whether the assay's sensitivity to repellents stayed constant across cohorts and months. Including a fixed reference repellent at regular intervals (for example, a standard DEET formulation, as is common in arm-in-cage protocols) would have allowed drift to be detected and corrected. To be clear, the DEET products in this study were test articles, not a reference standard. The authors could acknowledge this limitation explicitly, which would help readers assess the comparability of results across the testing period. If testing date, mosquito cohort and mosquito age per trial are available, the authors could consider reporting them in Table S3 and testing them as covariates or random effects, which would show whether assay sensitivity drifted over time or across cohorts.

9. Order of control and treatment exposures

In every test, the untreated control arm was introduced into the cage before the treated arm. Control and treatment arms were alternated between left and right, but the order of exposure was not. After 3 minutes of host contact, mosquitoes may be partially depleted or habituated, or more activated, before the treated arm is introduced, which would systematically inflate or deflate relative protection. The authors could consider justifying the fixed order and, if possible, testing for order effects in a subset of trials with the order reversed, which would clarify whether exposure order influenced relative protection. Otherwise, adding this as a limitation would make the potential for bias transparent.

10. Carry-over between same-day tests and uncontrolled personal products

Subjects could test multiple products on the same day when a product failed to provide complete initial protection, with only a soap-and-water wash and a 1-hour wait between products. Residue from fragrances or oily products could carry over and affect the next product's result. Subjects were also not asked to avoid deodorant or other personal products elsewhere on the body, although the study itself shows that one deodorant (Lynx Black) provided more than an hour of complete protection. Such products could reduce mosquito attraction during control and treated tests and may contribute to the wide range of control counts (30 to 361) and between-subject differences. The authors could consider reporting how many same-day sequences occurred and whether results differed for products tested second on a given day, and stating whether subjects' personal product use was recorded, which would clarify whether carry-over or personal products contributed to the variation in results. Including that information as a covariate, or acknowledging it as a source of variation, would strengthen the subject-effects section.

11. Scope of ethics approval for the landing trials

The ethics statement (project ID 28583) is given in the context of blood-feeding mosquitoes on a single volunteer to obtain eggs, and the permethrin section notes that subjects consented to blood feeding. It is not stated explicitly that the 480 arm-in-cage landing trials, which repeatedly exposed eight subjects to 50 mosquitoes for up to 8 hours with products that were sometimes unregistered, were covered by the same or another approval. The authors could consider stating the approval covering the repellent trials and how informed consent was obtained for them, which would confirm that all human-subject testing was ethically approved and consented.

12. Interpretation of the correlation between control landing counts and protection time

The positive correlation between mean control landing count and mean protection time across subjects (Spearman ρ = 0.74, P = 0.046, n = 8) is described as "marginally significant" in the Results. The Discussion, however, treats it as a finding, stating that repellents tended to be more effective on subjects with higher control counts. With eight data points, no correction for the other subject-level tests, and P close to 0.05, this result is fragile. It may also partly be built into the metric: relative protection is (control − treated)/control, so the same number of treated landings produces higher relative protection when control counts are higher. The authors could consider presenting this result as exploratory and discussing the possible contribution of the ratio-based metric, which would help readers weigh how robust the finding is. A sensitivity analysis using absolute treated-arm counts or a model with control count as an offset would help clarify whether the correlation is independent of how relative protection is calculated.

13. Definition of "primary active ingredient"

The primary active ingredient is defined as the first-listed active ingredient, and 25 products contained multiple actives or synergists. Order on a label does not necessarily reflect concentration or contribution to repellency, so some products may be misclassified in Figure 3B and in the active-ingredient model. The Discussion acknowledges that multiple ingredients confound concentration effects, but does not justify the first-listed convention. The authors could consider explaining why it was chosen and running a sensitivity analysis that either classifies multi-active products by the ingredient at highest stated concentration or excludes them from the active-ingredient analysis. Either approach would clarify whether the active-ingredient results depend on the first-listed convention.

14. Integration of the permethrin-treated sock results

The permethrin socks were evaluated with a different assay (blood feeding over 15 minutes) at a single time point, because pilot trials showed that the garment did not prevent landings. This is a sensible decision. However, the resulting protection value is plotted on the same axis as landing-based relative protection in Figure 2B, and it is unclear whether the socks are included in summary counts and category analyses. A blood-feeding reduction and a landing reduction are not equivalent measures, and a single time point gives no information on duration. The authors could consider presenting the sock results separately, stating explicitly which counts and analyses exclude them, and either testing additional time points or noting that persistence of protection was not assessed. These changes would avoid implying that blood-feeding and landing reductions are directly comparable and would clarify how long protection lasts.

Minor Comments

1. Landings versus bites

The Abstract frames landings as a proxy for bites, but several Results sentences state that products protected "against mosquito bites" (for example, 17 products providing complete initial protection against bites) when landings were measured. The authors could consider using "landings" consistently for the arm-in-cage data and reserving "bites" or "blood feeding" for the sock assay, which would make clear what was actually measured in each assay.

2. Landing counts exceed the number of mosquitoes in the cage

Control counts of up to 361 landings in 3 minutes from 50 females mean that counts reflect repeated landings by the same individuals, not the number of individuals attracted. The authors could consider stating this explicitly in the Methods, which would help readers interpret both the 30-landing activity threshold and the relative-protection values.

3. Inconsistent product denominators and title wording

Different denominators are used without a reconciling statement: 60 products, 59 assessed by landing assay, 55 repellents, 54 in the Discussion ("17/54"), 52 non-exempt, 43 topical and 28 with advertised times. Comparing the Abstract and Results, 54 appears to be the 55 repellents minus the permethrin socks. The essential-oil figures also differ (7/30 in the Abstract, which includes perfumes, versus 6/25 in the Results). The Results report 26 products exceeding 1 hour, which appears to include Lynx Black deodorant, whereas the Abstract counts repellents only. The title describes "60 personal mosquito repellents and perfumes", although only 55 are repellents and the five comparators include deodorants and an antiperspirant. The authors could consider adding a product flow diagram or denominator table, which would resolve these inconsistencies and make clear which products underlie each figure.

4. Sample size per time point in Figure 2C–D

Because trials stopped once protection was lost, the set of products and subjects contributing to the 1 and 2 hour panels changes over time, and this change could be mistaken for declining protection. The authors could consider annotating n per product per time point, or adding it to the legend, which would help readers distinguish changes in sample composition from true declines in protection.

5. Concentration–efficacy relationship in Figure 3C

The Results invite readers to see the regression lines in Figure 3C, but no corresponding model or test is described in the Statistical analysis. The authors could consider stating how the lines were fitted and whether any test was performed, or labeling them explicitly as descriptive, which would clarify the strength of evidence for a concentration–efficacy relationship.

6. Negative relative protection values

Fifteen of 480 initial comparisons had relative protection below zero, meaning more landings occurred on the treated than on the control arm. This is not discussed. The authors could consider commenting on likely causes (for example, attraction to some fragrances, or variation in control counts between sequential tests) and stating how negative values were handled in the models, which would clarify whether these values reflect biology or measurement noise and whether they affected the estimates.

7. Handling of zero-landing comparisons in Figure 4

The Figure S5 legend states that comparisons with no treated-arm landings were included, but the Figure 4 legend does not. The authors could consider stating this consistently across legends, which would avoid ambiguity about which comparisons underlie Figure 4.

8. Figure 3F

Two products advertised at 48 hours fall outside the plotted axes. The authors could consider showing them with an axis break or inset, which would let readers see the full extent of the discrepancies, since these are among the largest in the dataset. The authors might also consider color-coding points by primary active ingredient, as in panels B and C, which would also let readers see whether failures to meet claims cluster by ingredient.

9. Figure 1 legend

The legend introduces panels "(B-F)" as examples for different product categories but then lists B through G within that clause, while also saying separately that G involves a blood-feeding assay. The authors could consider revising it to "(B–G)" or separating G clearly, which would remove ambiguity about which panels show landing data and which show the blood-feeding assay.

10. Formula for relative protection

Relative protection and the blood-feeding protection metric are described only in prose. The authors could consider adding displayed equations, for example, RP = (C − T)/C, which would remove ambiguity, particularly for the averaged repeated initial test.

11. Mosquito handling during exposure

The Methods state that mosquitoes were not allowed to ingest blood, but do not say how feeding was prevented during the 3-minute exposure (for example, whether landing mosquitoes were dislodged after 3 seconds). It is also unclear whether subjects' breath or exhaled CO₂ could reach the cage; Figure 1A suggests the subject's head is close to it. The authors could consider describing these procedures, which would help readers assess how mosquito activation was controlled and allow the assay to be reproduced.

12. "Wild-type" terminology

A population maintained for 18 to 22 laboratory generations and cleared of Wolbachia with tetracycline is reasonably called field-derived, but "wild-type" may suggest otherwise to general readers. The authors could consider defining the term as used here or replacing it, which would avoid misinterpretation by general readers.

13. Participant and mosquito details

The authors could consider reporting the age range of the eight subjects (sex is given) and, in Table S3, the age of mosquitoes in each trial rather than only the aggregate 5 to 25 day range, given the emphasis on between-subject and cohort variation, which would allow readers to evaluate whether subject or mosquito age contributes to that variation.

14. Victoria's Secret Bombshell formulation

The authors note that lilial may no longer be present in the tested formulation. The authors could consider checking the current ingredient list for the batch tested, or stating that it could not be verified, which would clarify whether the product still serves its stated purpose, since lilial was the reason for including it.

15. Unsupported statement in the Discussion

The statement that ineffective products, particularly certified ones, may give users a false sense of security and increase bite risk is plausible but uncited. The authors could consider adding supporting literature or softening the wording, which would ensure the claim is appropriately substantiated.

16. Supplementary data and code availability

Tables S1 to S3 are listed but were not available with the version reviewed. The authors could consider ensuring that they accompany the preprint, which would allow the reported results to be verified, and depositing the raw per-test data and R scripts in a public repository so the analyses can be reproduced.

17. Reference list formatting

Several DOIs are duplicated (for example, "https://doi.org/https://doi.org/…" in Banks et al. 2014, Barrera 2022 and Curtis et al. 1987, among others), and Fradin & Day (2002) uses "doi:" inside the URL. The WHO (2009) guideline is cited in-text as "Organization, 2009" and listed as "Organization, W. H.", rather than "World Health Organization". The emmeans package is cited as "Lenth (2018) Package 'lsmeans'" with journal details that do not correspond to that package, and the R stats package entry ("Team, R. C., Team, M. R. C., Suggests, M., & Matrix, S.") appears to be corrupted citation metadata; it could be cited as R Core Team. The authors could consider correcting these entries and moving the raw URLs in the Introduction and Discussion (APVMA, ACCC, CHOICE) into the reference list or footnotes, which would make the citations accurate, easy to retrieve and consistently formatted.

18. Proofreading

The authors could consider correcting several small errors, which would improve the clarity and polish of the text. "e.g." appears both as "e.g." and as "e. g.". A space is missing before the citation in "PMD)(Carroll & Loye, 2006)". "Permethrin treated sock" should be hyphenated as "permethrin-treated". "Mosquito-borne pathogens transmission" should read "mosquito-borne pathogen transmission". In "relative efficacy of the products were similar", the verb should be "was".

Competing interests

The authors declare that they have no competing interests.

Use of Artificial Intelligence (AI)

The authors declare that they did not use generative AI to come up with new ideas for their review.