Our review of evidence unveiled only six articles published that assessed the effectiveness of agricultural interventions on food security and nutrition in CHES, and none published before 2017. Clearly, this is a relatively low number of articles identified as compared to the increasing number of countries in need of humanitarian assistance and/or experiencing high political instability [32, 40]. Additionally, the geographic coverage of these studies was limited to Africa (one study in South Sudan and North-East Nigeria, and two studies in DRC), and none were from the Middle East, Asia, or South America.

Of these six articles, four were peer-reviewed and two were published reports, as compared with a relatively high number of peer-reviewed articles published from stable settings [23,24,25]. Our review applied stringent criteria for inclusion of studies and did not include observational designs that previous reviews considered which could explain the low numbers of studies identified. This indicates that although experimental and quasi-experimental designs are possible to implement in CHES, very little rigorous research linking agricultural interventions to food security and nutrition has been conducted in such settings, and the majority of studies were conducted in prolonged relief or recovery (protracted crises and post-conflict) rather than acute phases, highlighting a major research gap.

Homestead food production, agricultural extension, and livestock support alone or in combination were the only nutrition-sensitive agricultural interventions identified in CHES. These interventions were also common in stable settings, but the latter also often included development-oriented interventions such as biofortification, irrigation, and value chain support, alone or in combination to food production interventions. It is likely that agricultural input provision is the main agricultural intervention type implemented in CHES as it provides tangible assets to households, it is easy to distribute, and it generates immediate socio-economic and nutritional benefits. These interventions enable vulnerable households to establish and profit from small-scale local agricultural production during a crisis to improve their food security. Local production in CHES is essential to ensure adequate food supply particularly that the agricultural sector deteriorates significantly during complex humanitarian emergency periods [8].

The reviewed studies included the primary outcome indicators: food security, nutrition, and health. In addition we considered outcomes on the impact pathway: agricultural production, asset ownership, and income. Overall, the interventions showed a positive impact on the use of agricultural input and techniques, but no impact on agriculture production and yields [36,37,38]. Agricultural interventions increased income and savings and decreased the need to rely on credit but resulted in mixed evidence in regard to its impact in the sale of productive assets to deal with income shocks [36, 39]. The majority of the interventions demonstrated a positive effect on household dietary diversity and food security, yet one study did not demonstrate any significant impact [38].

The studies also report a modest increase in children’s dietary diversity, yet only two articles investigated the impact on prevalence of stunting and underweight among children, where none find any detectable significant impact [34, 36]. In fact, Doocy et al. [34] finds that incorporating a behavior change communication (BCC) component led to an increase in children’s minimum diet diversity and minimum meal frequency, consistent with a recent meta-analysis that finds a positive impact of nutrition-sensitive agriculture on diet diversity in children in stable settings, that is augmented when interventions include BCC [41].

Our results are largely consistent with findings from previous reviews conducted on studies in stable developing settings, which demonstrate a positive effect on the use of agricultural inputs and practices, and some mixed evidence on food production, consumption, and dietary diversity [16, 17, 21,22,23,24,25, 41].

However, it is imperative to differentiate the underlying mechanisms through which agricultural interventions impact these outcomes across the two settings. Complex humanitarian emergencies are the main driver of food and nutritional insecurity [6]. Thus, the latent factors which affect the impact pathways and outcomes of agricultural interventions in CHES are also likely to be impacted by violent conflict itself. CHES-driven factors such as restriction to access land and water resources, loss of productive and livestock assets, agricultural, crop damages, and agricultural labor shortages driven by displacement of people from rural areas are essential determinants of these nutrition and welfare outcomes [4]. In addition, CHES limit access to output and value chain markets for selling agricultural produce, which limits income-generation, availability, and supply of fresh produce in markets [4]. Agricultural intervention in CHES, hence, are theorized to improve auto-consumption of livestock and crop produce but not local production and consumption. Conflict could also lead to poor child nutrition through the lack of accessibility, availability, and affordability of healthcare facilities, and access to healthcare was not accounted for in any of the studies that assessed child nutrition outcomes. Apart from Vallet et al., [39] who investigated the role of rural markets, the role of contextual factors and the potential mechanisms of action in CHES were insufficiently explored. In addition, exposure to conflict directly shapes decision-making and risk-taking [42–43]. Displacement and population movement caused by conflict decreases farmer’s ability and willingness to invest in agriculture and can influence household participation and uptake of these interventions, as well as how they benefit from it. These factors are not prevalent in non-CHES.

As a result, this review was not able to determine specificities of the impact pathways linking agricultural intervention to nutrition, food security, and health in CHES. Therefore, these mechanisms and their implications on outcomes along the causal pathway in CHES need to be better investigated in future studies.

Finally, the low number of rigorous studies in CHES could be explained by two factors: (1) the lack of funding towards agricultural interventions in CHES and (2) the scarcity of good quality data in these settings.

First, development funding timelines and objectives differ substantially from humanitarian funding which tends to focus on responding to immediate and acute relief rather than building long-term resilience. Therefore, funding allocations to agriculture in CHES make up a fraction of that allocated to development programs and their evaluation [44].

The lack of studies and data emanating from CHES may also result from the reluctance of participants to accurately report production, consumption, and income in challenging settings. For example, respondents may under-report due to fear of losing assistance or no longer qualifying to receive it [45]. Another explanation that could apply to both types of settings, but is more accentuated in CHES, is the decrease in sample size mainly caused by attrition and access difficulties in the field, which could have prevented the identification of any effect. And although we find a similar lack of impact as previous reviews with respect to child stunting and underweight, reasons identified by authors are different, and include measurement bias and the inability to correctly estimate children’s dates of birth which are essential for the accuracy of anthropometric status indicators. Also, the constant movement of households, particularly of older children, who are often relocated to live with relatives can further decrease the sample of children available for follow up during surveys. It is also likely that in both types of settings, follow-up durations are not sufficient to identify an impact on anthropometric indices [34, 36]. Impact evaluation studies are also challenged by a myriad of methodological, ethical, and practical challenges, especially in CHES [46, 47]. Our review identified selection bias, spillover effect, attrition bias, information, recall, measurement biases, and non-random response as threats to internal validity. Moreover, many studies reported that data collected in such settings face logistical and practical challenges, which not only reduced sample sizes and underpowered the studies to identify any effect but also limited the study’s ability to measure, through process evaluation and intervention mapping, the implementation fidelity and the extent to which the impact could be attributed to the intervention itself [35,36,37,38, 48].

It is possible and feasible to use and adapt existing tested methods implemented in research studies from stable developing settings, including the use of RCTs, yet there is a need for exploring novel approaches to conducting impact evaluation in complex humanitarian emergency settings, which address some of the contextual ethical and practical challenges [46–47, 49]. In contexts where traditional face-to-face household surveys are difficult to conduct, alternative remote-based tools such online or mobile surveys [50], crowdsourcing [51–52], geospatial data [53], satellite data, and remote sensing [54] can be used to measure and assess outcomes (e.g. plot or land area, land and water use, crop production and productivity and market access). In addition, various studies included in this review have underscored attrition rates, potentially leading to smaller sample size at follow-up. To overcome this common challenge, impact evaluations in these types of contexts could consider oversampling techniques to prevent loss of statistical power and maintain the robustness of research findings. The included studies could also be strengthened by adopting mixed-method approaches, particularly involving stakeholders, which can provide a more complete understanding of the complex nature of resulting behaviors, experiences, differential impacts, and potential unintended consequences.

To our knowledge, this systematic review is the first to explore the potential impact of agricultural interventions on food security and nutrition outcomes in CHES. Our analysis focused exclusively on studies that compared outcomes between different groups, or before and after the intervention within the same group, which strengthened our findings, compared to studies with no control or comparator group. However, this study is subject to several limitations. The review was limited to studies written in English as the inclusion criteria, which could have excluded relevant studies in other languages. Furthermore, the interventions of the included studies were carried out in three countries (DRC, Northeast Nigeria, and Sudan), which limited the generalizability of the review to other countries and regions. From the articles identified, four were subjected to moderate risk of bias while two were deemed to be at a serious risk of bias, lowering the certainty of evidence of the impact of agricultural interventions on food security and nutrition outcomes in CHES. Finally, considering the aforementioned limitations and the nature of the studies included, we were not able to conduct a meta-analysis.