Small dried fish as an affordable source of key micronutrients in Madagascar: nutritional benefits and contamination risks

Article activity feed

1. 

   PREreview

   Sep 3, 2026

   This Zenodo record is a permanently preserved version of a PREreview. You can view the complete PREreview at https://prereview.org/reviews/22287424.

   Below is a detailed peer review based on the submitted preprint. I have treated the manuscript as a scientific article rather than simply proofreading it, with particular attention to study design, analytical validity, exposure assessment, statistical inference, and whether the conclusions are supported by the data.

   Peer Review

   Manuscript: Small dried fish as an affordable source of key micronutrients in Madagascar: nutritional benefits and contamination risks Recommendation: Major Revision

   Overall assessment

   This is a potentially valuable and timely study addressing an important intersection of nutrition, food security, fisheries, and food safety in Madagascar. The combination of …

   This Zenodo record is a permanently preserved version of a PREreview. You can view the complete PREreview at https://prereview.org/reviews/22287424.

   Below is a detailed peer review based on the submitted preprint. I have treated the manuscript as a scientific article rather than simply proofreading it, with particular attention to study design, analytical validity, exposure assessment, statistical inference, and whether the conclusions are supported by the data.

   Peer Review

   Manuscript: Small dried fish as an affordable source of key micronutrients in Madagascar: nutritional benefits and contamination risks Recommendation: Major Revision

   Overall assessment

   This is a potentially valuable and timely study addressing an important intersection of nutrition, food security, fisheries, and food safety in Madagascar. The combination of market surveys, taxonomic characterization, nutritional analysis, and contaminant assessment is a major strength. The manuscript also identifies an important practical tension: some dried-fish products appear exceptionally nutrient dense while simultaneously presenting potentially important arsenic or cadmium exposure concerns.

   However, in its present form, several methodological and inferential issues substantially limit the strength of the conclusions. The largest concerns are the representativeness of the chemical sampling, the use of only three analytical replicates per sample, the statistical treatment of these replicates, the assumptions used to convert total arsenic and mercury into iAs and MeHg exposure, the choice and interpretation of reference values, and the relatively strong claims about affordability and population-level nutritional impact. These issues appear potentially addressable without fundamentally changing the study, but they require substantial revision and, where possible, additional analyses.

   1. Title and Abstract

   Strengths

   The title clearly communicates the two principal dimensions of the study: nutritional benefit and contamination risk. The abstract is concise and gives the reader the principal findings, including nutrient contributions and contaminant concerns.

   Major concerns

   Issue Specific evidence Recommendation The title may overstate the evidence for affordability. The study reports prices per Kapoaka and converts these to cost/kg, but the abstract concludes that SDF are "affordable" without presenting a quantitative affordability metric. Either quantify affordability relative to household income/food expenditure or soften the title and conclusions to "relatively inexpensive"/"potentially affordable." The abstract presents RNI contributions as nutritional benefit without sufficiently emphasizing uncertainty. It states that several samples exceeded 100% of RNIs for selenium, calcium, iron, or manganese. Explicitly state that these are modeled contributions from standardized hypothetical portions, not observed nutrient intakes in Malagasy populations. The contamination findings deserve stronger prominence. Nine samples reportedly produced iAs exposures above the BMDL05 in infants/young children. Present this as a central finding rather than primarily as a qualification to the nutritional message. "Strong potential to alleviate micronutrient deficiencies" is stronger than the study design supports. The study measures food composition and modeled nutrient contributions, not changes in nutritional status. Replace with wording such as "could contribute substantially to dietary micronutrient intake."

   2. Introduction

   Strengths

   The Introduction establishes a convincing rationale linking micronutrient deficiency, limited access to animal-source foods, and the potential importance of dried fish. It also identifies a specific evidence gap in Madagascar. The four research questions are clearly stated.

   Major concerns

   2.1 Population-level claims are broader than the sampling frame

   The study surveyed markets along National Road 7, covering 16 markets in 12 cities.

   This is a useful transect, but it cannot straightforwardly establish that SDF are representative of Madagascar nationally.

   The manuscript later states that:

   "The widespread presence of SDF across all surveyed markets confirms their central role in local food systems."

   This is reasonable for the surveyed RN7 corridor, but not necessarily for Madagascar as a whole.

   Recommendation: Clearly distinguish:

   • availability along RN7;

   • availability in the sampled cities;

   • national availability.

   The Discussion should acknowledge that northern, eastern, and other coastal regions were not sampled.

   2.2 The rationale for the nine focal types needs greater methodological justification

   The authors selected the nine most common types, accounting for 275/389 batches.

   This is logical, but selecting common products may systematically exclude less common products that could have either unusually favorable or unfavorable nutrient/contaminant profiles.

   Recommendation: Explain why frequency of occurrence was considered an appropriate criterion for nutritional-risk characterization and provide the frequency distribution of all 389 batches.

   3. Materials and Methods

   This is the section requiring the greatest revision.

   3.1 Market sampling

   The authors surveyed 112 vendors and 389 batches, which is a substantial field dataset.

   However, several details are missing:

   1. How were the 16 markets selected?

   2. Were markets selected purposively or randomly?

   3. Why were only seven markets subjected to vendor randomization?

   4. Were the same vendors repeatedly sampled?

   5. How many respondents contributed to each appreciation estimate?

   6. Were vendors and consumers analyzed separately?

   The manuscript defines an appreciation rate as the proportion of positive responses, but the observational structure is unclear.

   Major recommendation: Provide a flow diagram or sampling table showing:

   markets → vendors → batches → respondents → focal SDF types → analytical samples.

   This would make the sampling hierarchy transparent.

   3.2 "Affordability" is insufficiently demonstrated

   The manuscript records the price of one Kapoaka and converts it to cost/kg.

   Cost/kg alone does not establish affordability.

   For food-security research, affordability generally requires comparison against:

   • household income;

   • food expenditure;

   • cost per nutrient delivered;

   • cost relative to alternative animal-source foods;

   • or a recognized affordability metric.

   The Discussion states that SDF are more accessible than meat, but this inference appears insufficiently quantified.

   Recommendation: Add a table showing price per Kapoaka, price/kg, price per 10-g serving, and—ideally—cost per unit of key nutrient (e.g., iron, calcium, selenium).

   If household expenditure data are unavailable, substantially soften the affordability claim.

   4. Taxonomic identification

   Strengths

   The combination of morphology and DNA barcoding is a strong feature. The manuscript explicitly recognizes that drying can compromise morphological identification and therefore uses 12S sequencing for selected specimens.

   Concerns

   Only 36 caudal-fin fragments from 16 morphospecies were DNA tested.

   This is useful for validation but does not establish that all individuals within a commercial type were correctly identified.

   The use of BLAST similarity thresholds of 99–100%, 96–99%, and 91–96% also needs justification. A percentage identity threshold alone may be inadequate when closely related species share highly similar 12S sequences.

   Furthermore, several samples remain unidentified at genus/species level, particularly the mixed coastal samples.

   Recommendation:

   • Report sequence accession numbers.

   • Provide alignment/sequence quality information.

   • State how many sequences successfully amplified.

   • Report best-hit identity and database accession for every barcode.

   • Clarify whether taxonomic assignments were independently confirmed against the local 12S database.

   • Avoid implying species-level nutritional attribution when the sample is taxonomically heterogeneous.

   5. Biochemical analyses

   Major methodological concern: pseudoreplication

   The manuscript states:

   "Six 100 g replicates from each sample were finely ground…"

   but then says three replicates were used for vitamin A and three for minerals/heavy metals.

   This raises a critical question: Are these independent biological samples or merely technical/analytical subsamples from the same bulk commercial sample?

   If the three replicates were aliquots of the same homogenized sample, they do not provide independent biological replication.

   Yet the manuscript uses ANOVA and Tukey tests to compare the ten samples.

   This distinction is crucial.

   Recommendation

   Explicitly distinguish:

   • biological/field replicates: independently purchased batches;

   • laboratory subsamples;

   • analytical replicates.

   If the n=3 observations are technical replicates from a single pooled sample, inferential statistics across SDF types should not be presented as though they represent population-level variability.

   This is arguably the most important statistical issue in the manuscript.

   6. Statistical analysis

   The manuscript states:

   "Element concentrations, their contributions to RNIs, and the risk of exposure to heavy metals were compared across the ten samples using a one-way ANOVA followed by post-hoc Tukey tests."

   Major concerns

   6.1 ANOVA assumptions are not documented

   There is no indication that the authors tested:

   • normality;

   • homogeneity of variance;

   • independence;

   • influential observations.

   With n=3 observations per sample, formal assumption testing is weak, but the assumptions still need consideration.

   Recommendation: Report diagnostic procedures and consider transformations or non-parametric methods where appropriate.

   6.2 The experimental unit is unclear

   If the three replicates originate from the same collected sample, then treating them as independent observations constitutes pseudoreplication.

   This could make p-values substantially too small.

   6.3 Multiple testing

   The study analyzes many nutrients, several contaminants, three population groups, and numerous pairwise comparisons.

   The manuscript therefore performs a very large number of implicit statistical comparisons.

   Tukey controls pairwise comparisons within a given ANOVA but does not necessarily address the broader multiplicity arising from dozens of separate outcomes.

   Recommendation: Clearly define primary outcomes and consider controlling false-discovery rate for exploratory analyses.

   6.4 Derived RNI percentages should not necessarily undergo ANOVA

   RNI contribution is a deterministic transformation of measured concentration and assumed portion size/reference intake. Testing it separately may add little beyond testing the underlying concentration.

   Recommendation: Focus inferential statistics on measured concentrations and present RNI calculations descriptively, preferably with uncertainty intervals.

   7. Nutrient intake calculations

   This is another major area for revision.

   The authors use standardized portions of 5 g for infants and 10 g for young children and women and describe these as "realistic" amounts.

   However, the manuscript does not appear to establish these portions from an observed consumption survey.

   Concern

   The conclusions may therefore be interpreted as actual dietary recommendations even though the portions are modeling assumptions.

   For infants especially, 5 g/day of dried fish could have important implications for complementary feeding, but the study does not measure actual intake, feeding frequency, age-specific dietary practices, or preparation losses.

   Recommendation:

   1. Present these explicitly as scenario-based exposure estimates.

   2. Provide sensitivity analyses for realistic observed intake distributions if available.

   3. Avoid wording implying that 5 or 10 g/day is an established Malagasy consumption norm.

   8. RNI versus bioavailability

   The authors use iron and zinc RNI values corresponding to assumed moderate bioavailability—10% for iron and 30% for zinc.

   This is reasonable as a scenario, but the manuscript should emphasize that RNI contribution does not equal absorbed nutrient contribution.

   The whole-fish consumption model may alter bioavailability, and the dietary matrix containing rice, cassava, vegetables, phytate, etc., could affect absorption.

   Recommendation: Add a limitation explicitly stating that the analysis estimates nutrient provision rather than physiological nutrient absorption.

   9. Mean Adequacy Ratio (MAR)

   The manuscript uses MAR to summarize adequacy across nutrients.

   This requires careful interpretation.

   MAR is usually applied to observed dietary intake and is sensitive to which nutrients are included. Here, it is calculated from a hypothetical single food portion.

   Thus, a high MAR should not be interpreted as showing that the food is nutritionally adequate as a diet.

   Recommendation: Rename this as something like a modeled nutrient adequacy ratio for the SDF portion, or make explicit that it is not equivalent to adequacy of the individual's total diet.

   10. Heavy-metal risk assessment

   This is the most important scientific issue in the manuscript.

   10.1 Total arsenic → inorganic arsenic conversion

   The authors did not measure iAs directly. Instead, they assume:

   • 30% of total As is iAs for freshwater fish;

   • 6% for marine fish.

   This is a very consequential assumption because iAs drives the principal risk conclusion.

   The manuscript itself acknowledges that the iAs exceedances are partly driven by the conservative BMDL05 value.

   Concern

   The resulting risk estimates may have substantial uncertainty that is not propagated into the exposure calculations.

   For example, if the actual iAs fraction differs materially from 6% or 30%, the risk classification could change substantially.

   Recommendation: Conduct sensitivity analyses using plausible ranges of iAs fractions rather than a single point estimate. Ideally, directly measure arsenic species in representative high-As samples.

   10.2 Total mercury → methylmercury assumption

   The authors assume:

   "100% of total Hg was present as MeHg."

   This is deliberately conservative, but again it directly determines exposure.

   Recommendation

   Perform sensitivity analysis using plausible MeHg fractions and clearly distinguish measured total Hg from modeled MeHg throughout the Results.

   The manuscript should not give the impression that MeHg was analytically measured.

   10.3 BMDL interpretation

   The manuscript uses the iAs BMDL05 of 0.06 µg/kg BW/day.

   The authors appropriately acknowledge that this is conservative and compare it with a higher value used by another study.

   However, exceeding a BMDL is not equivalent to exceeding a tolerable daily intake.

   A BMDL is a reference point used for risk characterization, not a safe/unsafe threshold in the same sense as a tolerable daily intake.

   Recommendation

   Replace language such as "exceeded acceptable limits" with more technically precise terminology such as:

   "the estimated exposure exceeded the EFSA BMDL05 reference point."

   This distinction is especially important because the manuscript is positioned in a public-health context.

   11. Body-weight assumptions

   For infants and young children, the authors use the highest reported male body weights to produce "conservative" exposure estimates.

   This is unusual because a larger body weight generally reduces exposure expressed per kg body weight for a fixed food intake.

   Thus, describing these as conservative exposure estimates requires reconsideration.

   For example:

   \[ Exposure = \frac{C \times Intake}{Body\ weight} \]

   A higher body weight produces a lower calculated dose per kg.

   Major recommendation

   Recalculate using:

   • representative median/mean body weight;

   • lower-percentile body weight;

   • ideally age-specific distributions.

   Then present a sensitivity analysis.

   This could materially alter the reported risk estimates.

   12. Results

   Strengths

   The Results are logically organized into appreciation/consumption, taxonomy, micronutrients, contamination, nutrient contribution, and exposure. The results reveal substantial heterogeneity among SDF types, which is scientifically interesting and practically relevant.

   Major concerns

   12.1 Nutritional and contamination rankings need uncertainty

   For example, Matsiroky has the highest reported total As and Pb, while Varilava and Tovy have particularly high Cd.

   However, because the analytical sample size is small, the reader needs confidence intervals or at least explicit replicate-level distributions.

   Recommendation: Provide raw replicate values in supplementary data and report 95% CIs where statistically meaningful.

   12.2 Results should distinguish sample-level findings from type-level findings

   The manuscript sometimes moves between:

   • "sample";

   • "type";

   • "species";

   • "category."

   For example, Kalatambo was sampled twice, producing M1 and M2, whereas other types have a single analytical sample.

   This makes the statistical unit uneven.

   Recommendation: Establish a consistent terminology:

   commercial type → analytical batch/sample → constituent taxa.

   Then explain exactly which level is being compared statistically.

   13. Discussion

   Strengths

   The Discussion does a good job of identifying the central trade-off between nutritional value and contamination. The type-specific nature of the findings is an important contribution.

   Major concerns

   13.1 Causal explanations are speculative

   The manuscript proposes that high iAs may reflect environmental contamination or bioaccumulation and discusses the Onilahy River and mining activities.

   However, the study does not measure:

   • water contamination;

   • sediment contamination;

   • fishing location-specific contaminant exposure;

   • tissue bioaccumulation pathways.

   Therefore, these explanations are hypotheses rather than demonstrated mechanisms.

   Recommendation: Use explicitly cautious language:

   "may reflect…"

   rather than implying that the Onilahy River or mining caused the observed contamination.

   13.2 Processing effects on vitamin A are overinterpreted

   The manuscript states that the absence of vitamin A in sun-dried fish "likely reflects degradation during drying."

   But other explanations include:

   • species differences;

   • initial vitamin A content;

   • storage conditions;

   • analytical detection limits;

   • differences in lipid content.

   Because only one smoked sample contained detectable vitamin A, the data do not permit separation of species and processing effects.

   Recommendation: Present this as a possible explanation rather than a demonstrated processing effect.

   14. Conclusion

   The conclusion is directionally consistent with the results, but some statements are too strong.

   For example:

   "SDF represent a promising aquatic food for improving access to essential micronutrients in LMICs."

   and

   "In Madagascar, they are … widespread, affordable…"

   The study supports potential nutritional value, but does not directly demonstrate improved nutritional status or access.

   Similarly, recommending:

   "promoting safer SDF types"

   requires caution because safety was inferred from a very small number of analytical samples.

   Recommendation

   Reframe the conclusion around:

   1. demonstrated nutrient density;

   2. observed contaminant variability;

   3. uncertainty in exposure estimates;

   4. need for larger geographically representative sampling.

   15. Figures and Tables

   Strengths

   The figures appear well aligned with the study objectives. Figure 4 in particular integrates nutrient concentrations/RNI contributions across the three population groups. Figure 6 addresses the public-health relevance of contamination.

   Recommendations

   Figure 2

   The appreciation-rate figure should show 95% confidence intervals rather than only percentages and sample sizes.

   Figure 4

   The RNI heatmap/plot is information-rich but potentially difficult to interpret because it combines many nutrients and three populations.

   Consider:

   • clearly separating measured versus modeled values;

   • highlighting uncertainty;

   • providing a concise legend explaining RNI, AI, and MAR.

   Table 3

   Because the nutrient concentrations are central to the paper, the table should clearly state whether:

   • values are wet-weight or dry-weight;

   • replicates are analytical or biological;

   • ± values represent SD or SE.

   The manuscript says that concentrations are reported as measured on samples as sold and gives a mean dry-matter content of approximately 90.2%.

   This distinction should be made prominent.

   Table 4

   This table is particularly important. It combines:

   • measured total As;

   • Pb;

   • Cd;

   • Hg;

   • estimated iAs;

   • estimated MeHg;

   • reference values.

   The table should clearly distinguish measured analytes from modeled species.

   16. Ethical considerations

   Strengths

   The manuscript reports a Research Ethics Committee advisory opinion (UM 2023-040bis) and a Madagascar research permit.

   Concern

   The market interviews involve vendors and consumers, yet the manuscript does not clearly explain:

   • informed consent;

   • whether participation was anonymous;

   • how interview data were stored;

   • whether verbal or written consent was obtained.

   Recommendation

   Add a short statement describing participant consent and confidentiality procedures.

   17. Reproducibility

   The analytical methods are reasonably well described, including instrumentation, digestion, ICP-MS/ICP-AES, HPLC, PCR, and sequencing.

   Nevertheless, reproducibility would be substantially improved by providing:

   • raw nutrient data;

   • raw contaminant data;

   • individual replicate values;

   • DNA sequences/accession numbers;

   • exact R code;

   • exact formulas used for RNI and exposure calculations;

   • the complete reference-value table;

   • sample collection locations;

   • analytical limits of detection/quantification for all analytes.

   This is especially important because several major conclusions depend on calculated rather than directly measured quantities.

   18. Major issues requiring resolution before publication

   Priority Issue Severity 1 Clarify whether the n=3 laboratory replicates are independent biological samples or technical replicates; address pseudoreplication Critical 2 Reconsider the statistical framework and ANOVA/Tukey inference given the sampling structure Critical 3 Directly address uncertainty arising from estimating iAs from total As Critical 4 Directly address uncertainty arising from assuming 100% of Hg is MeHg Major 5 Reassess body-weight assumptions; higher body weight does not yield conservative dose/kg estimates Major 6 Distinguish modeled RNI contributions from actual dietary intake Major 7 Provide stronger evidence for the claim that SDF are "affordable" Major 8 Temper national/general population claims because sampling is concentrated along RN7 Major 9 Clarify the statistical/observational unit for commercial types, batches, and samples Major 10 Provide uncertainty estimates/raw replicate data and improve reproducibility Major

   Overall assessment

   Domain Assessment Novelty Good — integrated nutritional and contamination assessment of Malagasy SDF is valuable Importance High — directly relevant to food security and micronutrient deficiencies Study design Moderate — useful field design but limited geographic and analytical sampling Taxonomic methods Good, with limitations Nutritional analysis Potentially strong, but interpretation needs to distinguish composition from dietary adequacy Contaminant analysis Important, but risk estimates rely heavily on assumptions Statistics Needs substantial revision Reproducibility Moderate, with important missing details/data Figures/tables Generally strong, but uncertainty and methodological distinctions should be clearer Interpretation Promising but currently somewhat overstated Public-health relevance High, provided the risk estimates are presented appropriately Overall Major revision

   Recommendation: Major Revision

   I would not recommend rejection. The manuscript addresses a worthwhile question, contains substantial original field and laboratory work, and has the potential to make a useful contribution. However, the current version should not be accepted without substantial methodological clarification and revision.

   The five most important revisions are:

   1. Resolve the pseudoreplication/experimental-unit problem.

   2. Reanalyze or appropriately qualify the ANOVA/Tukey results.

   3. Perform sensitivity analyses for iAs and MeHg fractions.

   4. Correct/reconsider the body-weight rationale for "conservative" exposure estimates.

   5. Reframe the conclusions so that modeled nutrient contributions and contaminant reference-point exceedances are not presented as demonstrated population-level health effects.

   A particularly important positive feature is that the authors already recognize some of the uncertainty around the arsenic reference value; strengthening that transparency throughout the manuscript would substantially improve its scientific credibility.

   Competing interests

   The author declares that they have no competing interests.

   Use of Artificial Intelligence (AI)

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

   Read the original source
2. 

   Version published to 10.64898/2026.08.28.26361595 on medRxiv

   Sep 1, 2026