PREreview of To cull or not to cull: a model-based evaluation of response strategies against Lumpy Skin Disease outbreaks

Peer review of Jombart and Abbate (2026), pre-print

by Brandon Hayes

Disclaimer: The positions expressed in this review are my own and should not be assumed to reflect those of my employer, hosting institution, colleagues, or other affiliated parties.

Summary

In this article, a stochastic metapopulation compartmental model is used to evaluate alternative intervention strategies against the on-going lumpy skin disease (LSD) outbreaks in France. While the topic is both timely and policy-relevant, the modelling framework is fundamentally flawed and undermines the conclusions and recommendations reached by the authors. Principally:

• The model is not formulated to represent LSD as a vector-borne disease, misrepresenting the effects of removal of infectious cattle and failing to capture the infectious pressure from vectors that remains even after culling;

• The model assumes perfect identification of infectious animals, even though subclinical infections are known to occur, are more likely to escape detection than clinical infections, contribute to transmission and are central to the policy debate at-hand;

• The simplified contact structure removes the spatial components necessary to distinguish the effects of selective from mass culling, making the comparison between strategies uninformative.

The examined scenario is one in which all infectious individuals can be reliably identified and whose removal eliminates transmission entirely. However, these are precisely the assumptions that are contested in the current policy debate, and the reasoning behind the large-scale culling strategy. The conclusion that mass culling is equivalent to targeted culling is likely an artefact of a model structure that fails to represent the conditions in which aggressive and selective culling strategies could actually differ.

The choice of contentious framing is also questionable. In addition to the rhetoric throughout the abstract, saying there exists a need for “clear, well-reasoned, and evidence-driven…decisions” implies current decisions are not, and concluding with the hope that this work “contributes to opening up a rational debate” implies the current debate is irrational. Positioning oneself as the voice of reason while implicitly delegitimising existing discourse is a trope for the lectern, not scientific discussion among colleagues.

The current EU response strategy reflects established veterinary epidemiology, EFSA and WOAH guidance, lessons from recent outbreaks in southeastern Europe, and EU Council Directive 92/119/EEC applicable to all Member States. Disputing the bases of a policy is welcome, but suggesting irrationality is a disservice to the plethora of epidemiologists, veterinarians, public health officials, farmers associations, regulatory advisors, and policy makers who have contributed to the current knowledge base surrounding LSD control in Europe.

While an informative exercise in model dynamics, this preliminary analysis is not evidence supportive of policy change. By framing it as a reasonable alternative and then releasing it into the public sphere, the authors risk exacerbating the very tensions they claim they wish to resolve. The modelling framework needs to be realigned to the relevant complexities of LSD transmission and control, at which point conclusions can begin to be drawn regarding the official intervention strategy in the current context.

Major comments

1. Omission of vector dynamics

Despite recognising that LSD is a mechanically-transmitted vector-borne disease (VBD), the model is formulated as if transmission is driven by direct contact between cattle (using a frequency-dependent transmission term). In such a formulation, removing infectious individuals entirely eliminates the force of infection upon susceptibles, which is fundamentally incorrect for VBDs. Infected vectors do not simply disappear when infectious cattle are culled, meaning the force of infection exerted on a herd is not null even when only susceptible and recovered animals remain. Through assuming away this transmission structure, the duration that a farm remains infectious is underestimated, and an equivalence between mass and targeted culling-based strategies emerges since both strategies quickly remove the sole source of infection.

A secondary consequence of not accounting for vector dynamics is that it forces unrealistic assumptions be made for the impact of insect control, here simplified to that solely of insecticide application. Ignoring the complexity of on-farm insect management (Patra (2018). J. Adv. Agric. Res.), reducing this dynamic to a single parameter that attenuates an abstract transmission term allows no consideration for the conditions under which the intervention would succeed or fail. Given that cattle experience upwards of thousands of biting events per day from Stomoxys and other hematophagous insects, even substantial vector mortality may have only a limited impact without concurrently assuming low population turnover; a generally unrealistic assumption for insect populations. Without explicit vector dynamics, this intervention can be neither reasonably parameterised nor evaluated.

Simplifications of indirect transmission pathways into approximations of direct contact can be justified under specific conditions. When the reservoir timescale is much smaller than the host infectious period, there would exist a timescale separation limit in which the vector compartment equilibrates nearly instantaneously to changes in the infected host compartment (Benson et al. (2021), PLoS Comp Biol.; Gunawardena (2014), FEBS J). However, with LSD, the principal vectors have lifespans measured in weeks, environmental persistence can last a month or longer, and cattle are infectious for similar durations. This is what creates the memory in the vector population, where the force of infection does not disappear along with the removal of infecteds. Correspondingly, this is also why a direct transmission model is not appropriate. This is not just theoretical nuance either: indirect transmission was demonstrated as the only pathway that could explain the dynamics of an LSD outbreak in a dairy herd, with direct transmission being estimated to have made negligible contributions (Magori-Cohen et al. (2012), Vet Res.).

2. Neither surveillance nor detection uncertainty considered

Imperfect case detection defines one of the principal LSD control complexities, yet despite being designed to inform policy, this model assumes perfect surveillance and rapid case detection. In doing so, it fails to capture the scenario that forms the basis for the policy it claims to be examining.

Variations in clinical presentation, diagnostic sensitivity, and differentiating infected from vaccinated animals all contribute to detection uncertainty. Further, sub-clinical infections can constitute a majority of cases within a herd. Though the degree of the contribution to onward transmission provided by sub-clinical cases is debatable (Sanz-Bernardo et al. (2021), J Virol.; Haegeman et al. (2023), Viruses), the role is non-negligible and remains a principal limitation to the effectiveness of targeted control.

By defining targeted culling as the complete removal of all infectious individuals, this limitation is assumed away and the effectiveness of control strategies that depend on rapid detection and response become overstated. Here, the comparison between selective and blanket culling strategies is conducted under the conditions that maximally bias model results towards strategies that depend on reliable detection. Conclusions from a model that does not consider imperfect detection cannot be applied to policies where imperfect detection is a binding constraint. An accurate assessment of the current control measures requires accounting for both surveillance limitations and the consequences of persistent infection.

3. Spatial structure not consistent with the policy under examination

The model uses a static nearest-neighbour lattice, with farms represented as cells on a grid and neighbourhoods defined through rook connectivity. While employing an artificial landscape to examine real-world effects is not inherently limiting (and indeed this comment is secondary to the aforementioned concerns), when the contact structures necessary to reflect relevant transmission patterns are excluded, the suitability of the artificial structure fails.

Part of the value of blanket strategies like mass culling is in the abilty to rapidly eliminate transmission in dense networks. Here, all farms share the same sparse homogenous neighbourhood. Through removing the structure that would differentiate outcomes between targeted and mass culling, the equivalence between culling strategies becomes artefactual to the chosen landscape.

The limitation of the spatial construct is presented in the discussion (L258–261), however it is framed around simply restricting the ability to forecast epidemic trajectories. Rather, the primary limitation of considering local-only spread with uniform connectivity is that the two culling strategies are compared in a context that does not permit a difference to emerge between them.

Simplifications in spatial structure are a necessary compromise in most epidemiological models. However, simplification is permissible only to the degree that it doesn’t abstract away the very feature that provides differentiation. Preserving farm network heterogeneity, even abstractly, is likely necessary for an effective comparison between culling strategies.

Secondary comments

1. Lack of seasonality

It is noted in the introduction that LSD transmission, like that of other VBDs, is seasonal (L59–61). Despite this recognition, a full-year simulation is performed with non-seasonal transmission parameters. Ignoring the effects of seasonality on a seasonal disease will likely bias model outcomes, especially regarding estimations of intervention effectiveness.

2. Unanalyzed ad-hoc intervention parameters

Given that the primary objective of the manuscript is to critique the current response strategy, the parameterisation of interventions should be better supported. While personal communications with stakeholders are valuable, they are not a substitute for empirical data when that data is necessary to inform effectiveness. This limitation can be partially alleviated through sensitivity or uncertainty analyses, however neither of these have been performed.

Brief points

The methods state all interventions occurred 4 days after the first case (L162–163), yet the Table 4 footnote states that all responses take place 7 days after first symptom onset.

L171 references Table 6, though I suspect it means Table 4.

Rhetorical criticism

The language in the abstract (“government-imposed”, “critically evaluate the evidence”, “drastic measures,” “question the response strategy”), introduction (“need for clear, well-reasoned, and evidence-driven…decisions”) and conclusion (“rational debate”) closely mirrors the rhetoric that has become characteristic of politicised, populist, and contrarian public health debates, aimed at pre-empting support through evocative imagery in lieu of objective analysis. Such framing is at-odds with the standard of a scientific manuscript, whose merit is expected to arise from the weight of the evidence generated rather than the emotional response it attempts to elicit. Removing these value judgements would provide for a more-measured composition.

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.