Across the world, several countries present serious geographical barriers to distributing medicines promptly, specifically to remote or rural areas. Traditionally, these areas are located in poor regions, making the supply chain very difficult to manage as a result of restrictions on access and public order, among others. This is the case in Colombia, with vast rural areas with poor road infrastructure, underpopulated towns and a lack of supply in healthcare and education. In meeting the healthcare demands and delivering efficient healthcare operations, various challenges are encountered, since it is essential to provide both human and material resources at the right place, time and in the right quantities to improve the lives of rural communities (Peña-Orozco et al. 2023). New healthcare logistics and organisations are therefore required to fulfil the needs of rural areas without affecting the rural populations negatively.

To deal with those challenges and needs, the use of unmanned aerial vehicles (UAVs), commonly known as drones, becoming more promising to assist health systems and medical staff in fulfilling their mission of attending to population needs and saving lives (Nyaaba & Ayamga 2021). The use of UAVs has some benefits, including limiting the restrictions on truck transportation as a result of road problems and reducing the delivery time of medical supplies (Shi et al. 2022).

However, there are some unresolved issues regarding managing the UAV fleet for this type of activities, such as the location of the air bases, the management of capacity, the programming of routes and the potential integration with other modes of transportation. In fact, although technologically, UAVs are operational and the organisational issues that allow such a system to be planned and operated still need to be examined in-depth and defined in terms of a planning perspective.

These issues are directly related to the strategic planning phase of freight transportation systems (Tavasszy, Browne & Piecyk 2024), and mainly to the design of the transportation network (Crainic, Gendron & Kazemzadeh 2022). The success of a transportation network, as in industrial and social systems, is based on the capabilities of the design and planning team to address the main needs and problems of the field, therefore defining the most suitable and adapted networks. In the context of healthcare distribution in remote areas, the use of unmanned vehicles seems a potential alternative. However, in the current works, unmanned vehicles are mainly analysed and modelled for urban and regional last-mile applications, or theoretical and conceptual models without a direct real applicability (Ahmed et al. 2022). However, analysing the potential of such systems should provide direct insight into developments on a practical level and can contribute on research level by identifying the main elements that define a healthcare unmanned distribution system as well as the decision-making problems related to the derived transportation network design.

This paper aims to propose a problem-solving methodology for designing and deploying UAVs last-mile distribution networks into medicinal supply networks, and more precisely a vaccine supply chain for a rural municipality in Colombia. This work is therefore considered to be field research, which answers to a set of practical needs using research methods and contributes to creating knowledge and methodological insight from a real problem-solving case. An abductive operations research method composed of a problem instruction phase, a modelling and problem-solving phase, and a solution validation phase is proposed. In this case, it includes a set of mathematical models that is able to support the implementation of the distribution of vaccines to remote areas by using UAVs and exploring other modes of transportation, such as river transport. Those models are solvable, using commercial tools and are allowed to define various vaccine supply scenarios in the considered area (the Municipality of Bahía Solano, Chocó and Colombia) and produce a set of results to validate the proposed solutions and methodology. The structure of this work is organised as follows: Firstly, a literature review related to the research topic is presented. Secondly the methodology and mathematical formulation of the location and distribution models are described. Thirdly, the analysis of the results of the different models and instances is discussed. Finally, the main conclusions are made, and further research suggestions are given.

In Colombia, there is an important difference between rural and urban areas. Urban areas represent 77% of the population, whereas small towns and rural sparse areas only contain 7% and 16% of the population respectively, according to the 2024 national census.¹ Moreover, rural areas present important deficiencies in infrastructures, with small, poorly maintained roads. In fact, 39% of rural municipalities present important access problems (Así Vamos en Salud 2025). Furthermore, road infrastructures and accessibility are poor (Paz-Orozco et al. 2022), where limited road infrastructure imposes the use of fluvial and air transport that are not always in the best condition (Gonzalez-Feliu et al. 2021). Also, other infrastructures are deficient, as 59% of rural municipalities do not have a sewage system, and 20% of towns that do have, present a sewage coverage lower than 37%, which makes for a very low water sanitation infrastructure (Así Vamos en Salud 2025) and low educational rates (Peña-Orozco et al. 2023).

Another example is food insecurity. In rural areas, food insecurity accounts for more than 40% of the population and in the urban areas for around 25%. According to the Colombian National Department of Statistics (DANE 2025b), food security in Bogotá (the capital city) and its surroundings is lower than 20% and in the rural areas near the Caribbean Sea or the Amazonian region more than 40%. This leads to malnutrition problems, to delays in growth in children and other nutrition issues, which have a direct impact on the general health of populations, making them more prone to illnesses with regard to national means (DANE 2025a).

For those reasons, a set of actions based on three pillars, namely nutrition, health and education, was developed as part of a national programme over more than 40 years to support health, growth and social improvement of those sensible populations. Healthcare systems, strongly linked with nutrition, are the basis of rural development, and need to be efficient to sustain inclusive growth and social improvement of populations (Peña-Orozco et al. 2023). Those systems need to be managed and, to achieve viability thresholds, economies of scale are necessary, which benefit populated areas. However, rural areas need minimum service rates, a global management system; a cost-outcome system therefore needs to be deployed to ensure the effectiveness of the entire healthcare network (Espinosa et al. 2022). However, healthcare networks in rural areas are having difficulties: In 2024, about 37% of rural hospitals were in economic deficit in comparison to 17% in 2022 (Así Vamos en Salud 2025). Moreover, the concentration of general practitioners is very low, since only 0.5% of rural municipalities count for more than 100 general practitioners per 100 0000 inhabitants. This problem increases if we consider healthcare supply chains and the distribution of medicines and healthcare products to rural regions (Piñan & Bardales 2020).

The population who are mainly affected by this situation are rural inhabitants, and in this case, rural patients, who lack both specialised medicines (with longer times of supplies due to poor road infrastructures) and medical workforce (DANE 2025a). The problem is further enhanced by the need for a reduction in costs of the public healthcare system that needs to find a tradeoff between efficiency, via a centralisation of demand and resources, and service levels, with the overall goal of covering the entire population at national level, and also poses an optimisation problem (Espinosa et al. 2022).

Since the demand is not enough to justify it (on a cost basis), a capillary distribution of capabilities is impossible, and new organisations and logistics networks need to be developed to reach similar service levels, while reducing the overall logistics costs (Piñan & Bardales 2020). In fact, nowadays populations in rural areas are not justified, despite specific programmes and means, as a result of the physical barrier of the Colombian rural areas (lack of road infrastructure) and the sparce demand combined with the limited workforce. To that purpose, the definition of efficient healthcare supply systems based on UAVs could represent a potential solution to increase service levels (and global coverage), while not increasing (and even reducing) logistics costs (Gupta et al. 2021). This can take different forms and be of various natures; the main solutions to improve such systems in the world are related to the optimisation of current operations and the management of current resources (Bradley et al. 2017; Peña-Orozco et al. 2023), the re-design of the transportation (and supply) network (Crainic & Hewitt 2021) and the introduction of unmanned technologies to facilitate economies of scales when dealing with changing demand in small quantities but very frequently.

Healthcare distribution systems present a set of particularities that need specific transportation systems (Ananthakrishnan, Kaplan & Ng 2020). The first is the urgency of products as well as the poor anticipation capabilities of some needs (uncertainty, random nature of some illnesses). The second is the characteristics of healthcare products, which have a high heterogeneity in their availability, value, storage conditions and transport coverage needs, among others. For this reason, planning and optimisation methods that rely on problem solving have been studied for decades (Senna et al. 2023). The most widely used methods for the management of supply networks for medicines and supplies for humanitarian emergencies are mathematical optimisation models, such as linear, mixed, integer, and non-linear programming models; Markov chain models or queuing theory and simulation (Ahmadi-Javid, Seyedi & Syam 2017; Cassidy et al. 2019; Syahrir & Vanany 2015; Zgaya, Hammadi & Renard 2016).

Various examples of models and design methods for healthcare supply networks are seen in literature. Franco (2020) presents a system dynamics simulation model to symbolise the pharmaceutical supply chain in a hospital until delivery to the patient. Gómez, Salazar and Rincón (2019) worked on an optimisation model for reverse logistics of surplus medicines in the shortest period of time. Shi and He (2018) propose an optimisation model for the distribution of medical supplies after natural disasters. It is a two-dimensional model: time and cost. Shishebori and Babadi (2018) present an Mixed Integer Problem (MIP) model for designing a logistics network to supply medicine.

Regarding humanitarian logistics, disaster response and medicine, the implementation of drones or UAV has increased the distribution of supplies to areas that are difficult to access due to natural disasters as well as remote and rural areas where geography prevents the use of other vehicles (Scalea et al. 2018). Sanfridsson et al. (2019), Zègre-Hemsey et al. (2018) and Claesson et al. (2017) propose simulation models to compare the delivery time of automated external defibrillators (AED) using UAVs and emergency medical services completed tests for transporting organs by using UAVs between two hospitals. Mesar, Lessig and King (2018) finished a test to deliver life support medical supplies to a remote area using a multirotor UAV. Amukele et al. (2017) made pilots for the evaluation of the difference in the result of chemical and haematological tests transporting laboratory samples in UAVs.

Liu and You (2020) designed a mobile disaster response system to transport medical supplies by UAV to disaster areas in Puerto Rico. Jeong, Song and Lee (2019) did a multi-objective model to evaluate an airship vehicle, a UAV warehouse, to deliver humanitarian and health supplies in conflict and difficult-to-access areas. Chauhan, Unnikrishnan and Figliozzi (2019) present an integer linear model for UAV base location for humanitarian and medical supply delivery. Boutilier et al. (2017) designed a supply network integrating UAVs to transport AEDs to reduce response time. Pulver, Wei and Mann (2016) show an AED-equipped UAV launch network design. Kim, Jung and Rhim (2017) propose models of base location and the routing of UAVs for delivering and collecting medical supplies. Wen, Zhang and Wong (2016) present a multi-objective problem of routing trained vehicles to reduce distances and the number of UAVs used in the blood supply for areas with limited access.

Furthermore, various works deal with the use of unmanned vehicles in healthcare deliveries (Bridgelall & Tolliver 2024; Ghelichi, Gentili & Mirchandani 2021), mainly in the form of case studies to analyse the needs and main implications of using UAVs to deliver urgent medicines and healthcare supplies, and highlight the potential of such solutions, mainly when facing the lack of accessibility. Regarding unmanned transport systems, various works deal with the optimisation of these systems, but they are more related to urban deliveries and not focusing on healthcare products (Khoufi, Laouiti & Adjih 2019; Thibbotuwawa et al. 2020), and developing vehicle routing variants and algorithms, mostly meta-heuristics because of the non polynomial (NP)-hard complexity of the problem.

The introduction of new vehicles, like unmanned ones, implies a rethinking of the entire transport system. Freight transport systems can be defined as structures which function is to transport freight (Sarder 2020), composed by a demand, a set of vehicles, a network with a layout and a set of constraints (Crainic et al. 2023). To define and plan such systems, various methods can be used, in the function of the characteristics of the system and its planning horizon (Gonzalez-Feliu 2013; Russo & Comi 2023; Tavasszy et al. 2024). The strategic planning phase involves location and network design models and methods (Tavasszy et al. 2024), which have paid little attention in healthcare unmanned systems and are a crucial step in the development and planning of such networks. Moreover, the inclusion of unmanned vehicles changes the nature of the transport system by either changing its layout (Gonzalez-Feliu 2013) or some of its constraints, since unmanned vehicles imply a series of considerations that differ from classical distribution in terms of transport mode, capacities, autonomy and driving and security conditions (Gupta et al. 2021). For that reason, a modelling and optimisation method dealing with network design of unmanned healthcare transport systems will be proposed. It deals with a reality: the needs of vaccine distribution in remote areas of Colombia.

To conduct this study, a mixed-methods research approach is designed. For the initial qualitative phase, surveys, semi-structured interviews, and field observation are used to collect the information in order to understand the operation of the drug supply network in the study region. In the second quantitative phase, data analysis techniques and mathematical modelling are used to design the distribution network.

The model description is shown in Table 1. Initially, the transport model that simulates the current distribution scheme in the Municipality of Bahía Solano for extramural vaccination days is shown. This model presents the results with the existing and proposed means of transport in the municipality.

Likewise, a sensitivity analysis is carried out to reduce the uncertainty of the model input information and evaluate the system performance and behaviour under these different conditions.

The sensitivity analysis is based on a second instance of the previous models, where one of the parameters is modified, but the formulation stays the same. This analysis only applies to UAV models since they can be changed and optimised according to the goals of this work. The sensitivity analysis is done by changing the autonomy parameter of UAVs.

Table 2 shows a summary of the results obtained in each of the models presented above, where the flight time (TV) or navigation time (TN) and the total operating time (TO) are compared.

The analysis of results is presented in two steps. The first step compares the models in which UAVs are used, starting with the two locations, and then transport Model 2 is included. The second step considers the model with UAV that has the best performance. The results are compared with the model that considers the use of motorboats and represents the current distribution situation. The section from Quibdó to Bahía Solano cannot be modified because the only connection is by air, and the time is similar for all cases.

When prioritising broad demand coverage and minimized distances by selecting Ciudad Mutis (the nearest township to Bahía Solano) as the UAV launching base, Models 3 and 4 show no significant difference from the current vaccine distribution model in the municipality.

Additionally, Model 4, which aims to reduce the total travel distances, reduces the TO, TV, La and Re necessary to supply the demand, as seen in Figure 1. In contrast, Model 3 does not seek to optimise the routes only to meet the demand. Model 4 reduces the distance by 7.86% and reduces the total operating time (TO) by 11.1% compared to Model 3. About La and Re, Model 4 needs 5 LA and 9 Re less than Model 3 because Model 4 looks for the best combination of routes to reduce the distances in total.

Model 2 is the only one that uses the 3 UAVs and is the model that generates longer trips since it travels 177 km more than Model 3 and 296 km more than Model 4.

Although these three models are based in Ciudad Mutis, which is the central point for all the townships, Model 2 is the model that generates the shortest operating time (TO), as seen in Figure 2. The shorter operating time is the result of the model having an objective function of time reduction and using all the resources (3 UAVs) that are available for the operation. For these reasons, it needs to perform fewer launches, which significantly reduce the recharging time of the vehicle batteries (TR). The configuration of the routes also makes the launch time (TL) shorter. In this way, the total operation time (TO) for the distribution of vaccines is the minimum.

The TL in Model 2 is reduced by 40.4% compared to Model 3, and 35% to Model 4. The TR decreases by 60% and 53% in Model 2, in comparison to Models 3 and 4 respectively. The TO reduces by 40.5% and 33% in Model 2, in comparison to Models 3 and 4 respectively. These results indicate that Model 2 is the best scheme to distribute vaccines in Bahía Solano among the models presented using UAVs.

When comparing the results of Models 1 and 2, which represent the current vaccine distribution scheme, it is evident that it is more efficient to continue using the existing means of transport in the municipality. Since motorboats have a considerably greater transport capacity than UAVs, they generate fewer trips, making them more efficient for operation in terms of time and distance.

In this instance, time and distance are directly related to the number of trips being made. Similarly, there are significant reductions in the operation time. Model 1, in comparison to Model 2, reduces the total time of the distribution day (TO) by 67.7%, the total time in which the vehicles are in motion (TN, TV) by 73.3%, and the distance travelled by 83.9%.

Table 3 compares the results between Instances 1 and 2 of Models 3 and 4. Here, it is evident that, by increasing the autonomy of UAVs, the time of the operation is reduced by 22% (TO) and 36% of the distance travelled.

Regarding Model 3, despite having the same number of UAVs in operation, the total operating time (TO) is reduced by 14% by increasing the autonomy of UAVs. In this regard, the distance being travelled increases by 11%. However, in Instance 2, the TV increases by 23% because longer trips can be made by increasing their autonomy.

Regarding Model 4, by increasing the autonomy of UAVs, the total operating time (TO) is reduced by 33% and the distance travelled by 23%.

In terms of TO and distances travelled, Instance 2 of Model 2 is the model that performs best, which makes it the best model for vaccine distribution in the municipality. However, the difference in the performance between Instance 2 of Model 2 and Instance 2 of Model 4 is not very huge. Also, Instance 2 of Model 4 reduces the distance travelled and days of operation by 37%, in comparison to Instance 2 of Model 3, which seeks to maximise coverage. In Instance 2 of Model 4, the total operating time is reduced by 2 days and 36% of the distance travelled in comparison to Instance 2 of Model 3, as seen in Figure 3, due to the difference in operating times.

When comparing these results with the initial instances of the models, it is notable that performance measures are reduced by increasing the autonomy of the UAVs. Model 2 is the model that performs the best with Instance 2 being the model that reduces the OT for the distribution of vaccines in the municipality the most.

However, despite the significant reductions in time, it is still better for the municipality to continue distributing vaccines by means of transport by boat.

Finally, we have Instance 3 of Model 2, where intermodal transport is accepted. The results of this instance show that the total operating time (TO) is significantly reduced. This finding agrees with Gunaratne et al. (2022), where intermodality shows essential benefits in transportation costs and higher utilisation. In this instance, for the last-mile transportation, one UAV and a boat must do three launches and departures from the base located in Ciudad Mutis to meet the demand.

During these launches, 19 corresponding trips are made to each township. In this model, the UAV is used for 83% of its flight autonomy, while the boat is used for between 80% and 97% of its navigation autonomy. Instance 3 is the model that generates the best total operating time (TO) among the instances of Model 2, as seen in Table 4.

According to these results, the best way to distribute vaccines in the Municipality of Bahía Solano is through intermodal transport, that is, by river, the maritime mode, and UAVs. Combining these two modes makes it possible to significantly reduce the total operating time by configuring distribution times only with boats or UAVs. This study corroborates the benefits of intermodality when designing UAV-based distribution systems as a result of the potential to significantly improve healthcare supply chain performance, as shown in Gunaratne et al. (2022) and Javaid et al. (2022).