Multi‐Criteria Decision Analysis for Designing Optimal Aflatoxin Monitoring Schemes in the Dairy Supply Chain

ABSTRACT

Designing effective food safety monitoring schemes is a complex task involving multiple, often conflicting, criteria. This study applied Multi‐Criteria Decision Analysis (MCDA) to evaluate and identify optimal aflatoxin monitoring schemes along a Dutch dairy supply chain—a critical context where aflatoxin B1 (AFB1) contamination in feed can lead to aflatoxin M1 (AFM1) in milk, potentially posing public health concerns and economic losses. Monitoring schemes differed in detection intensity at feed mills and dairy farms, defined as the probability of identifying contaminated batches (low: 50%, medium: 80%, high: 90%) and determined by the number of monitoring batches and corresponding sample sizes used for AFB1/AFM1 sampling and analysis. Performance scores for each monitoring scheme were derived from quantitative models, scientific evidence, and expert consultation, while preference weights for criteria were elicited separately from representatives of the feed industry, dairy industry, and from a combined supply chain perspective. Results revealed that all stakeholder groups prioritized public health, but differed in their weighting of monitoring costs, production losses, customer trust, and implementation complexity. The feed industry preferred high‐intensity detection at both control points, while the dairy industry preferred medium‐intensity at feed mills and high‐intensity at farms. Overall, the MCDA framework facilitated a transparent and evidence‐based approach to identify an optimal monitoring scheme, highlighting the importance of stakeholder engagement in designing programs that are not only scientifically robust but also socially responsive and aligned with the WHO Global Strategy for Food Safety and the Sustainable Development Goals.

1. Introduction

Food safety monitoring is critical not only for detecting and controlling hazards but also for safeguarding public health and supporting resilient food systems. It facilitates establishing baseline contamination levels, tracking spatial and temporal changes, and ensuring compliance with quality systems and regulatory standards (Fazil et al. 2008; Focker et al. 2018; Wang et al. 2021). However, the increasing complexity of modern food supply chains poses challenges to food safety monitoring, necessitating coordinated and collaborative approaches across multiple stakeholders to manage risks effectively (Mu et al. 2021; Wang et al. 2020). Beyond technical food safety control, robust food safety monitoring directly supports several United Nations Sustainable Development Goals (SDGs), including access to safe and nutritious food (SDG 2), reduction of hunger, malnutrition, and foodborne diseases (SDG 3), alleviation of poverty‐linked economic and health burdens (SDG 1), minimization of food loss through sustainable production practices (SDG 12), promotion of decent work and economic growth (SDG 8), and strengthening multi‐stakeholder partnerships (SDG 17) (Fanzo et al. 2021; FAO 2023; Grace 2017; Manzoor et al. 2024) .

Given limited resources in food safety governance and the growing intricacy and interconnectedness of global food production networks, the efficient design of monitoring schemes has become an important research area. Previous research has evaluated monitoring strategies primarily using two equally weighted criteria: effectiveness (or benefits) and associated costs (Focker et al. 2018; Focker and van der Fels‐Klerx 2020; Wang et al. 2020; Wang et al. 2023). However, for different stakeholders in the food supply chain, such as regulators, industry, and consumers, other criteria come into play in the decision‐making process of food safety management (Ali et al. 2022; Ze et al. 2024). These criteria may include organizational aspects, legal aspects, or customer demands (Aenishaenslin et al. 2013; Focker et al. 2019b; FAO and WHO 2014). In addition, stakeholders¹ may perceive the relative importance of these criteria differently (Banach et al. 2021). The WHO Global Strategy for Food Safety further emphasizes that successful food safety systems depend on integrating these diverse factors into collaborative, evidence‐based frameworks, reflecting interconnected and sometimes conflicting criteria (WHO 2022). Thus, capturing diverse criteria and stakeholder preferences within food safety monitoring is essential for designing strategies that are technically robust, economically feasible, and socially acceptable across the food supply chain.

Multi‐Criteria Decision Analysis (MCDA) provides a transparent framework to support the making of complex decisions, while accounting for multiple and conflicting criteria, as well as the preferences of multiple stakeholders, affected by the decision (Agrawal 2015; Mergias et al. 2007; Saarikoski et al. 2006). Recent applications of MCDA in food safety management have demonstrated its efficacy in ranking hazards and supporting decision‐making for various food safety issues (Ali et al. 2022; Banach et al. 2021; Ferla et al. 2024; Garre et al. 2020; Phan et al. 2023). Notably, FAO and WHO (2014) used MCDA to evaluate 24 foodborne parasites worldwide, scoring them across seven criteria, including disease burden, geographic spread, trade relevance, and socio‑economic impact. Despite these advances, little attention has been paid to the design of monitoring schemes that explicitly balance multiple criteria and stakeholder concerns, highlighting a critical gap in current research.

Against this background, aflatoxin control in the dairy supply chain provides a particularly relevant case for applying MCDA to the design of monitoring schemes. Monitoring decisions for aflatoxins must simultaneously balance public health protection, compliance with regulatory limits, economic impacts, and practical feasibility across multiple supply‐chain actors, making them inherently multi‐criteria and stakeholder‐dependent. Aflatoxins B1, B2, G1, and G2 are carcinogenic toxins produced by fungi on food and feed crops, with Aflatoxin B1 (AFB1) being the most prevalent in feed for dairy cows (IARC 2002; Ismail et al. 2019; Trevisani et al. 2014). When ingested by dairy cows, AFB1 is metabolized into Aflatoxin M1 (AFM1), which is excreted into the milk and can spread through the dairy supply chain, posing significant health risks to consumers (Van der Fels‐Klerx and Camenzuli 2016). To mitigate these risks, the European Commission has established legal limits for AFB1 and AFM1 in feed and food products (European Commission 2019, 2021), with national governments and industry actors implementing monitoring programs to ensure compliance. These programs involve sampling, sample preparation, and analysis, following procedures outlined in Commission Regulation (EC) No 691/2013 (European Commission 2013). Depending on the monitoring objective and the criteria considered, optimal aflatoxin monitoring schemes may involve varying detection intensity levels at different control points in the dairy supply chain (Focker et al. 2019b; Trevisani et al. 2014; Wang et al. 2020).

This study aims to evaluate and identify optimal food safety monitoring schemes using MCDA to balance multiple criteria and stakeholder concerns. As a case study, we examine aflatoxin monitoring along the Dutch dairy supply chain, focusing on AFB1 in feed and its metabolite AFM1 in milk, with particular attention to the highly heterogeneous contamination patterns of these potent carcinogens at feed mills and dairy farms. This uneven distribution of contamination leads to substantial variability in health impacts, economic consequences, and customer trust, alongside differing monitoring complexity and costs among stakeholders. These complexities highlight the need for an evidence‐based, stakeholder‐informed multi‐criteria framework to design aflatoxins monitoring schemes that are transparent, equitable, and resource‐efficient.

2. Methodology

The process of the MCDA consists of three consecutive stages: a pre‐decision, a decision and a post‐decision stage (Greco et al. 2016; Mourits et al. 2010). The application of MCDA to optimizing aflatoxin monitoring in the Dutch dairy supply chain is illustrated in Figure 1. It includes the following key steps: establishing the decision context, identifying alternatives and criteria, assigning scores and weights, calculating overall values, examining results, and sensitivity analysis.

FIGURE 1.

Research design to rank aflatoxins monitoring schemes along the Dutch dairy chain using Multi‐Criteria Decision Making.

2.1. Establishment of Decision Context

The decision context was defined by optimizing AFB1/M1 monitoring strategies within the Dutch dairy supply chain while accounting for multiple, sometimes conflicting criteria, with stakeholder preferences elicited to ensure the relevance of these criteria to monitoring scheme design. The stakeholders considered in this study are feed companies (represented by the umbrella organization SecureFeed²) and dairy companies (represented by the umbrella organization by NZO³), These two umbrella organizations have an overarching aflatoxin monitoring program along the Dutch dairy supply chain in place. A large majority of Dutch feed companies and dairy companies are members of SecureFeed and NZO, respectively. In order to supply compound feed to the Dutch dairy companies (members of NZO), feed producers need to be members of SecureFeed and follow their monitoring protocol. To simplify the case study, we assumed that aflatoxin monitoring in a hypothetical Dutch dairy supply chain (Table B1 in the Supporting Information) takes place at the feed production and dairy farm levels, focusing solely on Dutch consumers and excluding international imports and exports (Figure 2). The control point at feed is the production units of the finished compound feed before transport to the dairy farms. The control point at the dairy farm is the raw cows’ milk in the collecting truck before transport to the processing plant, thus representing bulk milk produced at one dairy farm.

FIGURE 2.

The hypothetical dairy supply chain including the transmission of AFB1 to AFM1 and the two considered control points (CPs).

2.2. Identification of Alternatives and Criteria

The evaluated monitoring schemes along the dairy chain differed in detection intensity at two control points. The detection intensity was defined as the probability of identifying contaminated production units (with AFB1/AFM1 concentrations above the threshold), determined by the number of monitoring batches and the corresponding number of samples for collecting and analyzing AFB1/AFM1 under a given contamination case. According to experts’ opinion, we assumed that 1% of the total number of compound feed batches during the monitoring period were contaminated with AFB1 from an individual feed mill, and the contamination would spread along the following supply chain stages (Tables B2 and B3 in the Supporting Information). Three detection intensity levels were considered at each control point: low intensity (50%), medium intensity (80%), and high intensity (90%). An integrated chain approach was used to estimate the probability of detecting a contaminated production unit from the perspective of the integrated supply chain, that is, both stakeholders jointly monitor the compound feed and the raw milk and share the responsibility of identifying a contaminated production unit (PI chain). The combination of the intensity levels at the feed and milk levels results in nine alternatives. The addition of a “no monitoring” alternative at both control points resulted in a total of ten alternatives appraised in our study. The monitoring alternatives with corresponding numbers of monitoring batches at feed and dairy control points are shown in Table 1, while the respective sample sizes per batch and in total used for AFB1/AFM1 sampling and analysis are provided in Tables B4 and B5 in the Supporting Information.

TABLE 1.

Aflatoxins monitoring alternatives with detection intensity level, corresponding monitoring batches and PI ^a at each control point along the dairy supply chain during one period ^a .

Monitoring alternative			PI b			NMB b
	Feed	Milk	Feed c	Milk c	Chain c	Feed	Milk d
1	Low	Low	50%	50%	75%	16	70
2	Low	Medium	50%	80%	90%	16	160
3	Medium	Low	80%	50%	90%	25	70
4	Medium	Medium	80%	80%	96%	25	160
5	Medium	High	80%	90%	98%	25	228
6 d	High	Medium	90%	80%	98%	28	160
7	High	High	90%	90%	99%	28	228
8	High	Low	90%	50%	95%	28	70
9	Low	High	50%	90%	95%	16	228
10	No	No	0	0	0	0	0

^a One period is 10 working days in 2 weeks.

^b PI means the probability to identify a contaminated production unit; NMB is the number of monitoring batches selected for sampling and analyzing, and the detailed costs and number of collected samples and analyzed pool samples presented in Table B4 and B5.

^c PI Feed and PI Milk are the probability to identify a contaminated feed mill and dairy farm; PI chain is the overall probability of the industry to identify the contaminated production unit since both are responsible for monitoring chemical contaminants along the dairy supply chain. PI chain = 1 − (PI feed × PI milk).

^d No. 6 were regarded as current used monitoring scheme which could achieve 98% of identifying contaminated production unit (according to experts’ opinion).

To ensure relevance to decision makers involved in aflatoxin monitoring along the Dutch dairy chain, criteria were identified and selected through an iterative process, during which their quality was assessed based on measurability, redundancy, exhaustiveness, judgmental independence, duplication, and size. Criteria and related indicators were identified using scientific literature, followed by in‐depth interviews with experts from the feed and dairy food umbrella organizations, as well as food safety scientists. Three general decision criteria (effectiveness, acceptability, and implementation) were used to structure the identification of criteria, relevant in the food safety decision‐making process and provided by the FAO (2017). The final set of criteria was selected by the authors, based on the quality of criteria (Table A1 in the Supporting Information). The selected evaluation (decision) criteria include: Monitoring Costs; Public Health; Production Losses; Customer Trust, and Complexity of Implementation. The detailed explanation for each criterion was presented in Appendix D in the Supporting Information.

2.3. Scoring and Weighting

Performance scores for each of the criteria under the given detection intensity levels were established using three sources: available quantitative models, expert judgment, and expert interviews. The input data related to the dairy chain structure, the contamination case and monitoring scheme alternatives is found in Appendix B in the Supporting Information. The indicators of the decision criteria were as follows. Monitoring costs and production losses were expressed in Euros. Public health was quantified in DALYs (Disability Adjusted Lost Years) as a measure of the disease burden (Appendix C in the Supporting Information). The calculation of DALYs was based on an earlier modelling procedure applied to obtain an optimal sampling for monitoring AFB1/M1 and dioxins along the dairy supply chain (Wang et al. 2020). The scoring of qualitative indicators (customer trust and complexity of implementation) used expert knowledge elicitation and scientific literature on its expected performance with a five‐point scale, see scoring scale in Table 2. The five‐point Likert scale has the advantage over a two‐ and three‐point scale as it allows to express nuances among alternatives, whereas it keeps flexibility to add alternatives in the future. New alternatives may perform better or worse on certain criteria than existing alternatives and by using a 5‐point scale re‐scaling can be avoided (Revilla et al. 2014).

TABLE 2.

Scoring scale for qualitative criteria.

Scale	Score
Very low	1
Low	2
Moderate	3
High	4
Very high	5

Inputs for the priority weights in the MCDA were derived by asking representatives of 20 Dutch feed companies and 18 dairy processing companies about the relative importance of the criteria in the design and implementation of an aflatoxin monitoring scheme along the Dutch dairy supply chain. The weighting of criteria in our study was based on the results of an online questionnaire. This questionnaire asked the industry representatives to divide 100 points among the five criteria, with the number of points reflecting the relative weight. Respondents were guaranteed anonymity and were informed that the data were intended for scientific publication.

2.4. Calculation of Overall Value

After the results of the questionnaire were obtained, the preference weights were transformed into a Microsoft Excel worksheet. The following descriptive statistics were calculated: mean, median, mode, and standard deviation. The raw data can be found in Table C2 in the Supporting Information. Within the MCDA framework, the Preference Ranking Organization Method for Enrichment Evaluation (PROMETHEE) has been frequently used in food safety risk management to produce the operational ranking (or partial ordering) of decision alternatives, as it supports structured and transparent decision‐making through pairwise comparisons of alternatives based on the concept of dominance (Ali et al. 2022; Brans and Mareschal, 2005; Banach et al. 2021). PROMETHEE closely mirrors real decision‐making through straightforward evaluation and ranking procedures, making it particularly well‐suited for food safety applications (Behzadian et al. 2010; Fazil et al. 2008). Unlike fully compensatory methods such as the Analytic Hierarchy Process (AHP) and Weighted Sum Method (WSM), which require converting data to a common scale and may diminish meaningful differences, PROMETHEE retains natural units, applies preference functions for pairwise comparisons, and uses a partially‐compensatory structure that more realistically reflects human judgment (Ali et al. 2022; Ruzante et al. 2017). While Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) may obscure decision rationale through normalization and distance‐to‐ideal metrics, and Elimination and Choice Expressing Reality (ELECTRE) often produces partial rankings that are less intuitive, PROMETHEE overcomes these limitations by providing clear and complete rankings based on outranking flows and defined preference thresholds (López‐García et al. 2025; Martin and Smith 2025; Macharis et al. 2004; Behzadian et al. 2010). In PROMETHEE, preference functions were used for conducting a pairwise comparison. The difference between two alternatives a and b on criterion k can be represented as d_k (a,b) = x_a,b,k . The degree preference of option a over option b on criteria k is calculated according to the following function:

$$Pr\mathrm{k}\left(a,b\right)=\left\{\begin{array}{c}0,\\ \frac{x\mathrm{ab},\mathrm{k}-q\mathrm{k}}{p\mathrm{k}-q\mathrm{k}},\\ 1,\end{array}\right.\begin{array}{c}ifx\mathrm{ab},\mathrm{k}\le q\mathrm{k}\\ ifq\mathrm{k}<x\mathrm{ab},\mathrm{k}\le p\mathrm{k}\\ ifx\mathrm{ab},\mathrm{k}>p\mathrm{k}\end{array}$$ (1)

Where: P_k and Q_k are the indifference and preference thresholds, respectively.

Next, two outranking flows can be calculated: positive and negative outranking flows. The positive flow is the sum of the preferences in favor of alternative a, while the negative flow is the sum of all preferences degrees against. In PROMETHEE I (partial ranking) an alternative outranks another alternative if its positive flow is higher and its negative flow is lower than another.

$${\varnothing }^{+}\left(a\right)=\sum _{b\in A}\sum _{k=1}^{c}Pr(a,b)\times \mathrm{w}$$ (2)

$${\varnothing }^{-}\left(a\right)=\sum _{b\in A}\sum _{k=1}^{c}Pr(\mathrm{a},\mathrm{b})\times \mathrm{w}$$ (3)

where A = {a ₁ … a_n }, the set of n alternatives to be considered; K = {k ₁ … k _c}, the set of c criteria used to evaluate the alternatives; Pr_k (a,b) is the degree of preference of alternative a over b on criteria k; W_k is the weight associated with criteria k.

In PROMETHEE II (complete ranking) these flows are combined into a net flow:

$$\varnothing \left(a\right)={\varnothing }^{+}\left(a\right)-{\varnothing }^{-}\left(a\right)$$ (4)

The performance matrix to obtain the final ranking was analyzed with the software Visual PROMETHEE‐GAIA (VP) (Non‐profit academic Edition, Version 1.4). Performance of each monitoring scheme was computed using preference functions that allow for combining the quantitative and qualitative parameters. The selection of the preference functions was based on the Visual PROMETHEE manual by Mareschal (2015) and their corresponding thresholds were assessed by the Preference Function Assistant, available from the software. Preference functions and their thresholds per criteria are presented in Table 3.

TABLE 3.

Preference functions and their thresholds of the criteria.

			Thresholds a
Criteria	Measurement scale	Preference function	Q	P
Monitoring costs	Continuous	Type V linear	€71,565.86	€174,161.86
Public health	Continuous	Type III V‐shape	N/A	13.1581 DALYs
Production losses	Continuous	Type V linear	€89,103.62	€191,434.09
Customer trust	Qualitative	Type I usual	N/A	N/A
Complexity implementation	Qualitative	Type I usual	N/A	N/A

^a Indifference (Q) threshold: largest deviation considered as negligible; preference (P) threshold: smallest deviation considered as sufficient to generate full preference; for Type I thresholds are not applicable (N/A).

2.5. Examination of Results

The overall dominance scores of the evaluated monitoring schemes alternatives were calculated using the averaged preference weights from the feed industry, dairy industry, and the supply chain level (combining feed and dairy industry). The ranking of alternative monitoring schemes was derived from these overall dominance scores.

2.6. Sensitivity Analysis

To test the robustness of the results, a sensitivity analysis was carried out in VP. First, visual stability intervals (VSI) of the criteria weights were used to inspect intervals. VSI shows the range in which the weight of one criterion can change without altering the obtained ranking of the monitoring schemes. Second, to assess how rankings might vary under different weighting settings for the five criteria, two hypothetical scenarios were compared with the baseline case (the supply chain perspective using average respondent weights). Scenario 1 (S1) assigned higher weights (50 and 30, respectively) to public health and customer trust and lower weights (both 5) to production losses and monitoring costs. Scenario 2 (S2) assigned an equal weight (20) to each criterion. Lastly, the differences in ranking by changing one specific criterion were assessed.

3. Results

3.1. Weighting Factors Reflecting Industries' Preferences

A questionnaire was distributed among representatives from two stakeholder groups along the Dutch dairy supply chain. A total of 18 feed industry and 14 dairy industry representatives completed the survey, representing an overall response rate of 84% across all invited companies (Table C1 in the Supporting Information). Table 4 summarizes the average preference weights assigned by each industry group to the five evaluation criteria. Across both groups, “public health” (ranked 1st) and “customer trust” (ranked 2nd) emerged as the most critical criteria, receiving 45 and 27⁴ points respectively out of a total of 100. The dairy industry placed a slightly greater emphasis on public health (49 points) compared to the feed industry (41 points), whereas the feed industry valued customer trust more (32 points versus 21 points by dairy industry). Interestingly, from the dairy industry perspective, production losses, and monitoring costs were considered equally important. Although both industries largely agreed on the order of criteria importance, differences arose regarding the relative priority of monitoring costs versus complexity of implementation. For the feed industry, implementation complexity slightly outweighed monitoring costs (8 vs. 7 points), while the dairy industry prioritized monitoring costs (11 points) over complexity (9 points). Notably, a few feed industry respondents assigned zero weights to production losses and monitoring costs, with three respondents attributing the entire 100 points exclusively to public health, likely reflecting a strongly risk‐averse attitude or a professional prioritization of health protection over economic and operational considerations. This variation in stakeholder preferences is further illustrated by the relatively high coefficients of variation.

TABLE 4.

Average, rank position, median, mode, standard deviation (SD), and coefficient of variation (CV) of preference weights as indicated by the industry representatives. Number of respondents per chain actor given between brackets (n).

Criteria for different stakeholders	Weights matrix
1. Supply chain (n = 32)	Average a	Rank	Median	Mode	SD	CV b
Monitoring cost	8.79	4	10.00	0.00	8.31	96.74
Public health	45.34	1	42.50	50.00	25.87	57.70
Production losses	10.83	3	10.00	0.00	11.35	103.77
Customer trust	26.37	2	25.00	30.00	22.00	81.40
Complexity of implementation	8.67	5	10.00	10.00	6.64	77.29
2. Feed section (n = 18)
Monitoring cost	7.22	5	5.00	0.00	8.03	111.21
Public health	41.39	1	40.00	50.00	21.85	52.78
Production losses	11.67	3	0.00	0.00	12.39	106.19
Customer trust	31.67	2	30.00	30.00	24.29	76.72
Complexity of implementation	8.06	4	10.00	10.00	6.57	81.60
3. Dairy section (n = 14)
Monitoring cost	10.36	3	10.00	10.00	8.34	80.50
Public health	49.29	1	50.00	25.00	25.34	51.42
Production losses	10.00	4	10.00	10.00	8.86	88.64
Customer trust	21.07	2	20.00	20.00	11.83	56.14
Complexity of implementation	9.29	5	10.00	10.00	6.78	72.98

^a Sum of weights five criteria = 100. Average supply chain (overall average) = sum of average preference weight feed and dairy divided by two.

^b Coefficient of variance, calculated by standard deviation (SD) / average × 100.

3.2. Performance Matrix Monitoring Scheme Alternatives

Table 5 presents the performance matrix comparing the proposed aflatoxin monitoring schemes. The matrix highlights the inherent trade‐offs among alternatives, as each performs differently across the five evaluation criteria. Schemes characterized by higher detection intensity at both feed mills and dairy farms (high–high alternative 7, medium–high alternative 5, and high–medium alternative 6) incurred greater monitoring costs but scored strongly on customer trust and public health criteria. Conversely, these higher‐intensity schemes were the most challenging to implement. Schemes with lower intensity levels (Alternatives 1 and 10) were easier to implement but ranked poorly regarding customer trust and public health. No single monitoring scheme dominated across all criteria.

TABLE 5.

Performance matrix of aflatoxin monitoring alternatives for each criterion.

Monitoring alternatives with detection intensity at each control point			Monitoring costs a			Public Health b	Production losses c			Customer trust d	Complexity of implementation d
No.	Feed	Milk	Feed	Milk	Total	Total	IFeed c	IMilk c	Total	Total	Total
1.	Low	Low	45,760	65,000	110,760	5.525	2625	165,131	167,756	1	1
2.	Low	Medium	45,760	145,600	191,360	2.21	2625	264,210	266,835	2	2
3.	Medium	Low	71,500	65,000	136,500	2.21	4200	165,131	169,331	2	2
4.	Medium	Medium	71,500	145,600	217,100	0.884	4200	264,210	268,410	3	2
5.	Medium	High	71,500	207,480	278,980	0.442	4200	297,236	301,436	4	2
6.	High	Medium	80,080	145,600	225,680	0.442	4725	264,210	268,935	4	3
7.	High	High	80,080	207,480	287,560	0.221	4725	297,236	301,961	5	3
8.	High	Low	80,080	65,000	145,080	1.105	4725	165,131	169,856	2	2
9.	Low	High	45,760	207,480	253,240	1.105	2625	297,236	299,861	3	2
10.	No	No	0	0	0	22.1	0	0	0	1	1

^a Annual monitoring costs along the hypothetical dairy supply chain, measured in Euros.

^b Calculated for the population of 17,000,000 people, assessed in DALYs.

^c Production losses measured in Euros. Ifeed means if the contaminated feed mill is identified, the production losses for the supply chain. IMilk means if the contaminated dairy farm is identified, the production losses for the supply chain.

^d Assessed on a 5‐point scale. 1 = very low, 2 = low, 3 = moderate, 4 = high, 5 = very high. For CT a high value is preferred, for CI a low value is preferred.

3.3. Ranking of Monitoring Schemes

Table 6 presents the overall rankings of the monitoring schemes from three perspective: a combined supply‐chain perspective (averaging feed and dairy industry preferences), a feed industry perspective, and a dairy industry perspective. The net flows, representing aggregated preference scores based on criteria performance and stakeholder weighting, were computed for each perspective. From the perspective of the supply chain, the Alternative 7 ranked best, while no monitoring (alternative 10) ranked worst. Note that the two best performing alternatives have an equal order from the feed industry perspective but different from the dairy industry perspective. From the dairy industry perspective, the ranking of Position 1 and 2 is swapped: Alternative 5 moves to the first place, and Alternative 7 is the second‐best from the dairy industry perspective. For the remaining alternatives (Position 3–10), the ranking is exactly the same from three perspectives. However, as can be seen from Table 6, the scores (net flows) are higher for all alternatives from the feed industry perspective than from the dairy industry perspective, except for low–high monitoring (Alternative 9) and Alternative 10. Three alternatives are least preferred by the industry: Alternatives 1, 2, and 10. The poor performance of Alternative 10, on public health and customer trust, results that all other alternatives have a higher net flow.

TABLE 6.

Net flows and ranks of the evaluated aflatoxins monitoring alternatives, based on the preference weights of, respectively, the supply chain, feed, and dairy industry.

	Net flow			Rank
Monitoring alternatives with detection intensity at each control point	Chain	Feed	Dairy	Chain	Feed	Dairy
1. Low, low	−0.2026	−0.2522	−0.1528	9	9	9
2. Low, medium	−0.0814	−0.1015	−0.0613	8	8	8
3. Medium, low	−0.0355	−0.0576	−0.0134	7	7	7
4. Medium, medium	0.1030	0.1096	0.0964	4	4	4
5. Medium, high	0.2046	0.2359	0.1732 a	2	2	1 a
6. High, medium	0.1557	0.1904	0.1209	3	3	3
7. High, high	0.2198 a	0.2741 a	0.1654	1 a	1 a	2
8. High, low	−0.0028	−0.0274	0.0218	6	6	6
9. Low, high	0.0740	0.0822	0.0657	5	5	5
10. No monitoring	−0.4348	−0.4535	−0.4159	10	10	10

^a Bold figures reflect alternative with highest score and rank.

Figure 3 further explores the relative strengths and weaknesses of the top‐performing monitoring schemes by illustrating net flow scores for each criterion independently, with values ranging from −1 (worst) to +1 (best). Economic criteria, namely production losses and monitoring costs, negatively impacted all evaluated schemes. Alternative 7 showed the largest negative impact in terms of monitoring costs, yet simultaneously achieved the highest positive scores in public health and customer trust, underscoring critical trade‐offs. Although Alternatives 5 and 6 schemes demonstrated comparable public health benefits, Alternative 6 notably outperformed Alternative 5 in mitigating production losses. Customer trust reached its highest positive impact with Alternative 7. Despite its relatively low score regarding implementation complexity, Alternative 7 emerged as the strongest overall choice based on aggregated preference scores. Alternative 5, the second‐best choice, slightly under‐performed in public health and customer trust compared to Alternative 7, yet exhibited stronger outcomes regarding monitoring costs, production losses, and ease of implementation.

FIGURE 3.

Comparison of the four most preferred monitoring schemes, in descending order from left to right, preference in the proportion of evaluation criteria (MC‐ Monitoring Costs, PH‐ Public Health, PL‐ Production Losses, CI‐ Complexity of Implementation, CT‐ Customer Trust).

3.4. Sensitivity Analysis

Table 7 illustrates the sensitivity of the overall ranking to changes in the assigned weights. The stability intervals indicate the range within which individual criteria weights can vary without altering the final ranking presented in Table 6. From the supply chain perspective, the ranking of alternatives (as presented in Table 6) will not change with a preference weight between 24 and 100 for public health and between 23 and 100 for customer trust, respectively. Adjustments to the weight of any single criterion must be balanced by corresponding adjustments in other criteria, as all weights must cumulatively sum to 100. Notably, the overall ranking showed the highest sensitivity to changes in the weighting of implementation complexity, while rankings were least sensitive to variations in weights assigned to public health and customer trust. The VSI under the average industry preference weights can be found in Appendix D in the Supporting Information.

TABLE 7.

Stability intervals of the average whole industry preference weights (from the supply chain perspective).

Criteria	Weight %	Min	Max	Interval range
Monitoring costs	8.79	0.00	23.44	23.44
Public health	45.34	23.72	100.00	76.28
Production losses	10.83	0.00	27.40	27.40
Customer trust	26.37	22.86	100.00	77.14
Complexity of implementation	8.67	3.44	10.11	6.67

Two hypothetical scenarios (S1 and S2) were compared with the baseline, which applied average respondent weights from the supply chain perspective (Table 8). S1 reflects the most frequently observed stakeholder weighting pattern (the modes) from the supply chain perspective, assigning higher weights to public health and customer trust, and complexity of implementation based on mode values from elicited responses. In contrast, production losses and monitoring costs receive comparatively low importance and are jointly assigned a total weight of 10, reflecting their limited influence on the resulting rankings and their relatively minor contribution among the economic criteria in this scenario. S2 assigned equal importance to all criteria. The monitoring schemes of high–high and medium–high still ranked the best and second best in both S1 and S2. Lastly, an extensive single‐criterion sensitivity analysis was conducted using the Walking Weights method in PROMETHEE‐GAIA. This method systematically varied individual criterion weights across their minimum, maximum, mode, and median values to assess the robustness of rankings. The detailed outcomes of this analysis, illustrating how alternative rankings changed in response to varying criterion weights, are summarized in Table D1 in the Supporting Information.

TABLE 8.

Complete ranking for two hypothetical scenarios.

Ranking	Baseline	Scenario 1	Scenario 2
1	High, high	High, high	High, high
2	Med, high	Med, high	Med, high
3	High, med	High, med	High, med
4	Med, med	Med, med	Med, med
5	Low, high	Low, high	High, low
6	High, low	High, low	Low, high
7	Med, low	Med, low	Low, med
8	Low, med	Med, low	Med, low
9	Low, low	Low, med	Low, low
10	No	No	No

Note: Baseline uses average respondent weights from both feed and dairy industries for five criteria. Scenario 1 takes the weights for each criterion: costs = 5, public health = 50, production losses = 5, customer trust = 30, complexity of implementation = 10. Scenario 2: all five criteria are equally weighted. Abbreviations: med = medium, int = high, no = no monitoring.

4. Discussion and Conclusion

Ensuring food safety within increasingly complex food systems presents significant challenges, particularly as contaminants such as aflatoxins are exacerbated by climate change, trade interdependencies, and supply chain vulnerabilities (Garcia‐Cela and Gasperini 2024; Mu et al. 2021). While international frameworks like the WHO Global Strategy for Food Safety emphasize the need for pro‐active, risk‐based approaches grounded in multi‐stakeholder cooperation, operationalizing these principles into effective monitoring strategies remains a gap in the literature (WHO 2022). This study demonstrates the application of MCDA as a structured and transparent approach to support decision‐making in food safety monitoring. International food safety authorities, including FAO, EFSA, WHO, and the Codex Alimentarius Commission, have acknowledged MCDA as a valuable tool for improving food safety risk management (Guarini et al. 2018; JECFA 2017; Stephen et al. 2017). While MCDA has been previously applied to food safety cases, such as hazard prioritization and risk management interventions (Ali et al. 2022; Dunn 2014; Fazil et al. 2008; Ruzante et al. 2010), our study is the first to demonstrate its utility in the operational domain of monitoring system design.

In line with recommendations from Dunn (2014) and FAO (2017), this study confirms that food safety decision‐making is inherently context‐specific and multi‐dimensional, shaped by economic, legal, organizational, and public health considerations. Decision of risk managers on food safety management are affected by the country context with its socio‐economic and cultural aspects, as well as profits of food industry (Dreyer et al. 2010). Therefore, there is not a general list of criteria based on which choices in food safety management can be made. In this study, the selection of criteria was informed by both literature review and interviews with industry experts, resulting in five relevant criteria: public health, customer trust, production losses, monitoring costs, and implementation complexity. This selection aligns with criteria previously reported in other food safety MCDA applications (Banach et al. 2021; Fazil et al. 2008).

A key finding of this study is the strong and consistent prioritization of public health across both the feed and dairy industry stakeholders. The primary objective of the monitoring scheme at the industry is to assure feed and food safety, and this importance is reflected by the relative weights for the criteria. The costs for monitoring received low weights (on average 9%) by both the dairy and feed industry. This shows that the choice for the detection intensity depends on other criteria than costs. Including the criteria “production losses” aimed to account for the increasing economic losses (recall costs, transportation costs, lost revenue) when the contamination is only found downstream the supply chain. In previous modelling studies, monitoring at earlier control points in the chain is found to be more cost‐effective than at stages downstream the chain. This was also highlighted during the expert interviews: putting emphasis on feed monitoring is important to assure milk safety (Focker et al. 2019a; Lascano‐Alcoser et al. 2014). However, the participants considered production losses to be not sufficiently important (11%), and alternatives with higher detection intensity at feed level did not (always) outrank alternatives with a higher intensity at milk level. A case description (Appendix C in the Supporting Information) was provided as annex to the questionnaires used to elicit the relative weights for the five criteria. This case description could have influenced responses towards socially desirable answers and this may be an explanation for the low importance of monitoring costs. Costs may become an important criterion at a point where the effectiveness of the monitoring systems is considered already sufficient and no more incidents are reported.

The results also reveal nuanced differences between stakeholder groups. The similarity in the preferences of the two stakeholder groups implies a common ground between dairy and feed industry representatives regarding aflatoxin monitoring along the supply chain. Despite the high degree of similarities, public health (+7.9%) and customer trust (+10.6%) received higher preference weights from representatives of the feed industry as compared to dairy industry. A possible explanation for the relatively higher weights given by dairy industry to these two criteria might be that total production losses of identifying one contaminated dairy farm are higher than those of identifying one feed mill, and also monitoring costs are much higher for the dairy industry. An explanation for the higher weight on customer trust from the feed industry is that all feed producers supplying dairy feed to the members of NZO need to be members of SecureFeed. The dairy industry trusts the feed industry to check the safety of the supplied feed. As a result, customer trust is very important for feed producers.

It is important to acknowledge several limitations of this study. First, the use of a fixed set of predefined criteria and a specific stakeholder group could be extended. Involving additional groups, such as regulatory authorities, consumer organizations, or academic experts, could result in different weighting patterns and potentially alter the outcomes. Future research could adopt more iterative and participatory approaches, such as Delphi techniques or stakeholder workshops, to refine both the criteria and the associated weights. Second, the scope of the analysis was limited to private monitoring activities at two control points: feed mills and dairy farms. Regulatory sampling and downstream monitoring at processing or retail stages were not included and should be considered when interpreting the findings. Third, although the sensitivity analysis demonstrated that the overall rankings were robust under most weighting scenarios, changes in the weight assigned to certain criteria, particularly implementation complexity, did affect the results. This highlights the need for careful elicitation and validation of stakeholder preferences in future MCDA applications. Finally, the dynamic nature of food safety risk perception adds complexity to decision‐making processes. In the context of aflatoxin management, the perceived importance of criteria such as public health, customer trust, and economic impacts can shift significantly in response to events like contamination incidents, regulatory changes, or shifts in public attention. This temporal variability sets food safety apart from more stable decision domains and highlights the need to regularly revisit MCDA outcomes to ensure they remain relevant and aligned with current risk priorities.

To advance the use of MCDA as a decision‐support tool in food safety monitoring, several recommendations can be proposed for future research. First, future studies should consider incorporating multiple chemical hazards, such as monitoring AFB1/M1 alongside other contaminants like dioxins, to better reflect the complexity of real‐world food safety monitoring programs. Second, this study focused on industry‐led decision‐making, and the selection of stakeholders was aligned accordingly. However, if the decision‐making authority were to shift to a national food safety agency, the inclusion of additional external stakeholders, such as regulatory authorities, academic experts, and consumer representatives, would be essential to ensure that diverse perspectives are captured in the prioritization process. Broadening stakeholder involvement would provide a more comprehensive and balanced understanding of the decision context. In future applications of MCDA for broader food safety risk management, related criteria such as production losses and monitoring costs could be consolidated into a broader “economic” dimension to improve balance across decision domains and more accurately reflect their combined importance relative to overarching categories like public health.

In conclusion, this study provides empirical evidence that MCDA is not merely a theoretical tool but a practical framework for designing food safety monitoring systems that are adaptive, stakeholder‐responsive, and aligned with global food safety goals. By focusing on aflatoxin monitoring within the Dutch dairy supply chain, the study offers valuable insights into how different criteria, spanning public health, economic impacts, implementation feasibility, and customer trust, influence the performance of monitoring schemes for contaminants in both feed and milk. The results reveal a high level of agreement between the feed and dairy industries regarding the relative importance of these criteria when designing monitoring programs. Compared with traditional cost‐effectiveness approaches, MCDA offers a major advantage in its ability to integrate multiple dimensions, including both quantitative data and qualitative stakeholder preferences, enabling the identification of balanced solutions that accommodate competing interests. Embedding MCDA‐based approaches into food safety monitoring across diverse supply chains represents a critical step toward operationalizing the WHO Global Strategy for Food Safety 2022–2030, and accelerating progress toward the Sustainable Development Goals.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting Information: risa70196‐supp‐0001‐Appendices.docx