A recent cluster randomised controlled trial in Mexico and Nicaragua showed a convincing impact of evidence-based community mobilisation on risk of dengue [1]. The Camino Verde intervention had a positive effect on all outcome parameters (serological evidence of dengue virus infection in children, self-reported dengue cases and all conventional vector indices). Earlier trials of community-based interventions have shown impact on entomological indices [2, 3, 4, 5, 6, 7, 8, 9], but the Camino Verde trial was the first to show impact on dengue virus infection. It is likely that there will be increased emphasis on community-mobilisation for prevention of dengue and other diseases borne by the Aedes aegypti vector, especially with evidence from the Camino Verde trial and from other studies [10, 11, 12, 13] of the lack of effectiveness of the use of the organophosphate pesticide temephos in household water stores in reducing vector indices and dengue infection [14].

While the Camino Verde intervention had impact, it did not eradicate the vector or dengue virus infection in the intervention clusters. It is important to know how interventions of informed community mobilisation for dengue prevention achieve an impact and how they might be fine-tuned to increase their impact. Complex interventions can change many aspects of human behaviour. In the impact study of the Camino Verde trial, we measured a number of intermediate behavioural outcomes. These are described in the CASCADA model, based on the theory of planned behaviour [15, 16]. This extends the knowledge, attitudes and practices (KAP) model of behaviour change popularised in the 1960s [17].

Recognising the well-documented limitations of the KAP model [18, 19], our acronym CASCADA [20, 21] describes a partial order of intermediate outcomes through Conscious knowledge, Attitudes, Subjective norms and positive deviation from these, intention to Change, Agency (individual and collective) to make change, Discussion of possible action and, finally, Action or change of practice. We used the CASCADA model as the basis of a randomised trial of an intervention to increase vaccination uptake in Pakistan [22, 23] and in analysis of a cross-sectional study in southern Africa [24]. A qualitative analysis of narratives of young women in three southern African countries validated the CASCADA model in the case of action to prevent gender violence [25].

Our CASCADA model uses the constructs of the conventional theory of planned behaviour. Figure 1 shows the differences between CASCADA and the conventional theory of planned behaviour [15]. First, we propose CASCADA as a partial order, not an invariable or linear sequence, but different to the parallel constructs seen in the conventional theory: knowledge generally precedes attitude; attitude shifts generally precede a positive deviation from a negative subjective norm; and so on. Second, the CASCADA model adds “Discussion/Socialisation” as a measurable construct between Agency (perceived behaviour control) and Action (behaviour). We base this on the observation that where behaviour change is an essentially SOCIAL outcome, it is reasonable to expect discussion or socialisation one step before implementation.

The Camino Verde trial reported unremarkable differences in CASCADA outcomes between intervention and control clusters: there was a significant difference in only one intermediate outcome, Agency or collective self-efficacy (RRR 9.6% 95%CI 3.4–15.8) [1].

Here we report a secondary analysis of the impact of the Camino Verde intervention on the outcomes in the CASCADA results chain in all respondents and in men and women separately, in the Mexican arm of the trial, as part of our efforts to understand the gender-specific individual responses to the community-led intervention.

The methods and results of the Camino Verde trial are described in detail elsewhere [1]. This secondary analysis used data from the impact survey of the trial in Mexico in 2013. Fieldworkers interviewed household respondents in 45 intervention and 45 control clusters in the coastal region of Guerrero state. The Camino Verde CASCADA model included the following variables: Conscious knowledge (C) was the ability to identify a physical sample of a mosquito larva (the interviewer showed a physical larva and took mention of “mosquito” in the response to indicate conscious knowledge of the mosquito life cycle). Attitude (At) to vector control was the response to the question: Do you consider it worthwhile (vale la pena) to spend time and money every week to eliminate mosquito breeding sites (criaderos) in your house. Subjective norm (S) was derived from the question: Do you think your neighbours consider it worthwhile to spend time …. Intention to Change (iC) used the question: Do you plan to spend time and money each week to eliminate mosquito breeding sites in your house. Agency (Ag) derived from answers like “myself” or “ourselves” to the question: Who is responsible for control of mosquito breeding sites. Discussion (D) about prevention derived from the response to the question: How often do you chat with your neighbours about how to control mosquitos. These two constructs, Agency and Discussion, replace perceived behaviour control in a more usual theory of planned behaviour [26]: Agency refers to both self-efficacy and collective efficacy; we included Discussion as a linkage to factors that facilitate the performance of the behaviour. Action (A) focused on participation in collective activities in the neighbourhood to control mosquito breeding sites.

We analysed responses of 6185 people (5042 women and 1143 men) in the intervention group and 6204 (5025 women and 1179 men) in the control group. The left side of Fig. 2 shows the net transitive closure of the CASCADA map derived from responses in the 45 control communities. Green arcs represent the positive net influences and red arcs the negative net influences, all in the direction of Action. Thicker lines indicate stronger influences. Conscious knowledge has an influence on Attitude, Subjective norm, and each subsequent step in the CASCADA. Attitude influences Subjective norm, Change intention, and each subsequent step in the CASCADA. In the control clusters, there is a negative influence of intention to Change on Discussion and on Action; also of Agency on Discussion; and of Agency on Action. The cumulative net influence across the CASCADA sequence is 0.88 out of a possible maximum value of 6 (which would be attained if every outcome in the CASCADA results chain had the strongest possible positive influence on the next factor in the chain).

The right side of Fig. 2 shows the analysis of the net transitive closure of the CASCADA map of outcomes in the 45 intervention clusters. The negative influences in the control clusters – red arrows between intention to Change, Agency and Action – do not appear in the intervention responses; instead there are strongly positive influences in the intervention group. The net influence of these variables on Action is consequently much higher in the intervention than in the control clusters (1.92 versus 0.88). The only negative arc is between Conscious knowledge and Subjective norms, the implication being that the more people in intervention sites knew about dengue prevention, the less they recognised incompatible norms of their neighbours.

Among female respondents (Fig. 3), the control group showed much the same picture as did the overall control group. The intervention group, again like the overall picture, showed positive arcs replacing the negative arcs from Agency to Discussion and from Agency to Action. Two negative arcs emerged: between Attitudes to prevention and Agency, and Subjective norms to Agency. The cumulative net influence increased from 0.78 to 1.73 with the intervention.

Figure 4 shows the findings for male respondents. Male control responses showed similar patterns to female control responses but male intervention responses were very different to female intervention responses. Four strongly negative arcs emerged among men in intervention clusters, with Conscious knowledge negatively related to Subjective norms, intention to Change, Discussion and Action. Among male respondents, the cumulative net influence decreased from 1.38 in the control clusters to 1.11 in the intervention clusters.

Fuzzy transitive closure allowed scrutiny of the proposed results chain underlying dengue prevention action in the Camino Verde trial. The cumulative net influence across the CASCADA partial order of intermediate outcomes is compatible with the main finding of the trial, illustrating a likely mechanism of the positive influence on participant behaviour. Positive arcs between Change intention or Agency and Discussion or Action replaced negative arcs in the control clusters, probably related to the multiple opportunities to discuss prevention offered by the intervention, and reflected in the increase in collective efficacy (Agency) reported in the original study [1].

The transitive closure of the CASCADA model emerging from male and female responses, however, is quite different. Among women, the intervention appears to have overcome blocks in the distal partial order and increased the net influence of Conscious knowledge on Action. It seems plausible that this accounts for most of the measured impact in the trial.

Responses from men suggest the Camino Verde intervention did not work well for them, in the important sense that it did not generate or reinforce positive prevention behaviour in this group. If anything, the intervention seems to have weakened the existing influence of knowledge on action shown in the control group, quite possibly demotivating men.

It is possible that the male responses included in our analysis represent a selected and minority view – there were many fewer male than female household respondents, and the male responses came from men who were at home at the time of the survey, when most men were out at work. The findings may also be related to the intentional emphasis of the intervention on women’s participation and empowerment. The Camino Verde intervention was gendered in the sense that it engaged, mobilised and empowered women. The negative reaction of men might have represented some sort of push-back in the context of women engaging in community issues that might ordinarily have been a male domain. This phenomenon is well recognised in other spheres of development involving women’s empowerment [32, 33, 34]. In a broader and probably more correct sense of gendering interventions, we should also have considered and addressed the potential effect of women’s engagement on their spouses.

Fuzzy transitive closure operationalises the CASCADA partial order by quantifying the sequential relationship (for example, Knowledge leading to Attitudes) in the context of all other relationships between Knowledge and Attitudes, in both directions. The resulting quantification is context and subgroup specific, as the relative influence of the network of different constructs will change with context. Fuzzy transitive closure incorporates the influences of all other arcs, to generate the summary that is very different for men and women even in the same communities. The difference is informative, fuelling discussions about intervention gender dynamics.

As far as we can ascertain, this is the first use of transitive closure to analyse gender specific results chains in behaviour change interventions. By treating the intermediate outcomes as a network with each node potentially related to every other node, transitive closure provides a coherent summary of the overall relation between knowledge and preventive action. It allows us to evaluate the results chain in a way that appropriately emphasises the weakest link in the chain. It also illustrates in a compelling way where “blocks” in the CASCADA need to be confronted and, in the counterfactual comparison allowed by the Camino Verde trial, whether the intervention overcomes these blocks.

Our conclusion is that future dengue prevention interventions have to work effectively with both women and men, perhaps beginning with gender-stratified action planning and then bringing together the two groups for strategic discussions about who tackles what part of the problem.