Enhancing emergency blood supply using drones

(1)

Enhancing emergency blood supply using drones

Master Thesis

MSc. Technology & Operations Management

Author: Jannik Krivohlavek (s3908437)

University of Groningen, Faculty for Economics and Business

University supervisor: dr. ir. D.J. van der Zee

Second assessor: dr. N.B. Szirbik

(2)

Abstract

Background

Within blood supply chains, emergency deliveries from blood banks to hospitals occur frequently due to low shelf lives and complex storage conditions of certain blood products. The economical-driven trend to centralize blood supply facilities and increasing traffic levels result in challenges of traditional emergency vehicles to comply with timeliness requirements. As unmanned aerial vehicle (drone) technology has progressed in recent years, they offer a solution to the transportation challenges due to their independence of ground infrastructures. The objective of this paper is to explore the economic and operational value of drones for emergency blood deliveries.

Method

Using a design science approach, the initial problem context is a real-world setting in a rural part of the Netherlands. Next to this scenario, an urban scenario was defined to assess the impact of the level of urbanization on the drone value. Current drone capabilities as well as assumed future drone capabilities were considered. Two sets of experiments were designed comparing drones to existing ground vehicles and analyzing the possibility to increase the operational radius of a blood bank using drones. A simulation study was conducted to perform and evaluate the experiments.

Results

Drones were able to decrease the response times of emergency orders by 35 % depending on the travel distance. This allows to increase the operational radius of blood banks by 72 % based on assumed future drone capabilities. Economically, drones could only compete with existing land transportation in urban regions when the number of missions is sufficiently high. Including additional less critical blood deliveries in the operations resulted in 15 % cost savings per mission of drones.

Conclusions

(3)

(4)

List of figures

Figure 2.1: Different phases in the blood supply chain (green parts lie within scope of this research) ... 14

Figure 2.2: Schematic picture of the Point-to-Point and Hub & Spoke model (the hub is representing a blood bank as an intermediate distribution point) ... 16

Figure 2.3: Drone network in Rwanda (Ackerman and Koziol, 2019) ... 20

Figure 3.1: Work packages integrated in research design ... 23

Figure 3.2: Using a framework to systematically describe a case example ... 24

Figure 3.3: Conceptual visualization of quantifying cost values for the economic analysis ... 26

Figure 4.1: Overview over the activities in the blood supply chain in the Netherlands ... 28

Figure 4.2: Blood bank and served locations in the case example ... 30

Figure 6.1: Model set-up representing emergency blood distribution (Hospital stage is not included in model) ... 44

Figure 6.2: Scenarios considered in the simulation ... 46

Figure 6.3: Demand distribution for each scenario depending on distance to blood bank ... 47

Figure 6.4: Visualization of the demand points in scenario A and different radii for the dummy points in scenario B 48 Figure 6.5: Conceptual depiction of the sets of experiments allocated to the scenarios ... 50

Figure 6.6: Experiment tree for vehicle comparison experiments (C_Drone = Current drone capabilities, F_Drones = Future drone capabilities) ... 51

Figure 6.7: Coverage for a radius of 65 and 95 km ... 52

Figure 6.8: Experiment tree for the coverage area experiments (F_Drones = Future drone capabilities) ... 53

Figure 6.9: Default demand distribution B.1 for scenario B and an alternative distribution B.2 ... 54

Figure 7.1: A1 response time plateau values for scenario A, only A1... 56

Figure 7.2: A1 response time plateau values for scenario A, A1 + A2 ... 56

Figure 7.3: A1 response time plateau values for scenario B, only A1 ... 57

Figure 7.4: A1 response time convergence behavior for vans ... 58

Figure 7.5: A1 response time convergence behavior for current drones ... 58

Figure 7.6: A1 response time convergence behavior for future drones ... 59

Figure 7.7: A1 response time plateau values for scenario B, A1 + A2 ... 59

Figure 7.8: A1 response times with required number of vehicles, scenario A, A1 + A2 ... 61

Figure 7.9: A1 response times with required number of vehicles, scenario B, only A1 ... 61

Figure 7.10: A1 response times with required number of vehicles, scenario B, A1 + A2 ... 62

Figure 7.11: A1 response time plateau values for the coverage area experiments ... 64

(9)

Figure 7.14: Using future drones for the payload size (A2) sensitivity experiments ... 67

Figure 7.15: Impact of demand distribution when vans are used ... 68

Figure 7.16: Impact of demand distribution when future drones are used ... 69

Figure 11.1: Framework to describe and analyze a transportation system ... 81

(10)

List of tables

Table 2.1: Characteristics of routine and emergency missions ... 17

Table 3.1: Type of costs included in the cost analysis ... 27

Table 4.1: Locations of hospitals that the blood bank Groningen serves and the number of A1 and A2 missions in 2019 (Sanquin, 2020c) ... 29

Table 4.2: Shipment characteristics of the case example based on internal documents, personal interviews and estimations ... 31

Table 4.3: Geographical characteristics of the case example ... 31

Table 4.4: Characteristics of the vehicle currently used in the case example ... 32

Table 4.5: Operational characteristics as a sub-group of the vehicle characteristics ... 33

Table 4.6: Analysis of the suitability of the case example for drone usage ... 35

Table 5.1: Classification of different drone concepts, based on USAID Global Health Supply Chain Program-Procurement and Supply Management (2017) ... 36

Table 5.2: Overview over selected available drone solution (* linearly interpolated values) ... 38

Table 5.3: Comparison of current drone capabilities with assumed future capabilities ... 39

Table 5.4: Cost estimations for the “Manta Ray” drone used in this study (*: one operator is assumed to be able to operate 5 drones at the same time based on (HEMS, 2020b))... 40

Table 6.1: Overview of scenario parameters ... 46

Table 6.2: Demand points and their distance to blood bank for both scenarios ... 48

Table 6.3: Congestion factors for each scenario ... 49

Table 6.4: Experimental factors for the vehicle comparison experiments ... 50

Table 6.5: Experimental factors of the coverage area experiments ... 52

Table 7.1: Minimum required number of vehicles to comply with the A1 time constraints ... 60

Table 7.2: Total annual travel distances and number of flights/rides for experiments ... 63

Table 7.3: Results of the cost analysis ... 63

Table 7.4: Minimum required number of vehicles to comply with A1 time-constraints in coverage area experiments ... 65

Table 7.5: Annual travel distances for coverage area experiments ... 65

(11)

1 Introduction

Emergency Medical Services (EMS) in the healthcare sector, such as rescue operations, patient or doctor transportation and delivering biological/medical goods, are increasingly facing operational and economic challenges. The trend of centralizing resources (e.g. hospitals, departments, special treatments) leads to longer travel distances. As a result of this, the travel times for EMS are increasing such as the arrival time of emergency doctors in Germany (+ 31 % over the past 25 years) (ADAC Luftrettung gGmbH, 2020). Furthermore, increasing traffic volumes put additional strain on transports related to EMS. The amount of medical transports is increasing as the amount of medical treatments increase which cannot be expected to decline because of aging populations in many parts of the world, especially developed countries.

In order to cope with the above mentioned transportation challenges, a disruptive transportation technology using the airspace is recently gaining more and more attention due to technological improvements. Electrical vertical take-off and landing aircrafts (eVTOLs) are not only independent of the ground traffic but also allow a high level of autonomy and convince in terms of noise and sustainability. Travel times of EMS can be decreased using EVTOLs. When eVTOLs are unmanned they are classified as drones. Several international working groups bringing together various stakeholders aiming to unleash the potential of drones for medical purposes are increasingly emerging as, for example, the Unmanned Aerial Vehicles for Payload Delivery Working Group (UPDWG) or the ISG Unmanned Aircraft Systems (UAS) Coordinating Body.

(12)

(13)

of this study is to explore the economic and operational value of drones for emergency blood supply system by analyzing a real-world system. The research objective is:

Explore the potential of drones to decrease the response time and costs of the emergency blood supply in developed regions

The study uses a design science approach which aims to transfer the traditional emergency blood distribution to drone-based emergency blood distribution. Different problem contexts and different type of drones are input to a simulation study which explores the value of drones over traditional land vehicles. Primary contribution is the quantification of response time and cost benefits of drones based on a real-world setting. Furthermore, implications on a strategical level are reported. Secondary contributions are a framework to systematically describe the key features of a given transportation system and a simulation model which can be used a tool for stakeholders willing to analyze their specific system. The outcomes of the study lead to recommendations concerning drone integration into emergency blood supply chains and ultimately enhances the implementation.

(14)

2 Theoretical background

2.1 Characterizing blood supply chains

The objective of blood supply chains is to ensure the availability of usable blood products for medical services while keeping a balance between supply and demand. The reduced shelf-life of most blood products paired with special cold chain storage requirements and fluctuating demand and supply impose strict constraints on the supply chain (Osorio et al., 2015). The shelf life varies from 5 days for platelets (generally needed for cancer patients) to 42 days for red blood cells (RBCs) and 1 year for frozen plasma. If shortage occurs, surgeries might be postponed or acute patients cannot be treated. Wastage, on the other hand, has to be kept at a minimum to reduce costs. A structured review of the key constructs and phases of a blood supply chain, broken down by hierarchy level (strategic – tactical – operational), was done by Osorio et al. (2015). They divide the blood supply chain into the categories collection, production, inventory and distribution, see Figure 2.1. The focus of this research lies in the means of transportation. This has, in the first place, an impact on the distribution phase but also on the inventory phase as elaborated in the following.

Figure 2.1: Different phases in the blood supply chain (green parts lie within scope of this research)

2.1.1 Inventory phase

After the donor blood has been collected and further processed into usable blood products, it has to be stored somewhere until it will be needed. The products can be stored in the points of use (i.e. hospitals) or in storage facilities (i.e. blood banks). The location of storage depends mainly on two factors, namely the demand forecast and the storage capabilities of the hospitals. The storage requirements of products impact the inventory suitability of these products.

(15)

Hospital size generally correlate with higher demand level, hence, bigger hospitals have a higher demand. Since they can assume a certain minimum of demand, certain inventory levels are likely to be hold. (Thiels et al., 2015). For hospitals with less or highly fluctuating demand it is often undesired to maintain inventories due to the risk of wastage or shortage. In the UK, for example, hospitals with an annual demand of less than 500 platelets pools usually do not store them at the hospital but request them from the national blood service (Elmi et al., 2017).

Storage requirements and capabilities

Different blood products require different storage conditions. The required temperature range is from 20-24 °C for platelets to lower than -65 °C for frozen RBCs. Platelets require in addition a continuous gentle agitation in order to keep the product usable. The presence of storage equipment such as ultra low freezers or agitation machines limit the inventory possibilities of hospitals. The expiry date of blood products is also an important factor which limits or enables inventories. While frozen products such as RBCs or plasma generally have a lifetime of 1-10 years, platelets expiry after only a few days after collection (Basu and Kulkarni, 2014). The shorter the expiry date the more limited the inventory potential of blood products.

2.1.2 Distribution phase

(16)

Figure 2.2: Schematic picture of the Point-to-Point and Hub & Spoke model (the hub is representing a blood bank as an intermediate distribution point)

Transport modality

Several means of transportation exist to distribute blood. The most common means are cars and motorbikes. The choice of vehicle depends on the national context. Generally, in regions with a low-quality infrastructure and barely accessible roads (e.g. parts of Africa), motorcycles are preferred (Wright et al., 2018). In regions with well-developed infrastructure, typically cars are used as in Austria or the Netherlands (Hemmelmayr et al., 2009) (and Sanquin talk). Helicopter used for trauma missions carry blood in this kind of missions, however, a helicopter flight only for the purpose of blood delivery is prohibited by cost inefficiencies (Thiels et al., 2015).

2.1.3 Blood supply missions

(17)

Table 2.1: Characteristics of routine and emergency missions

2.2 Challenges faced in blood distribution and transportation

2.2.1 Realizing the benefits of centralized storage – coping with longer travel distances

Traditionally, healthcare organizations have a large amount of medical goods in their inventory. Mostly, medical supplies are ordered in bulks in order to keep the total number of deliveries low, reducing transport and order costs. In this way, medical facilities always have their medicine, vaccines and blood products readily stored for emergency cases. However, this comes at high costs regarding wastage and inventory effort. Inventories need to always be kept at a sufficient level because the need of specific medical goods heavily fluctuates, and yet still needs to be available at all times. As a consequence, every year approximately $10 billion worth of unused medication is thrown away due to fast approaching expiry dates (Rijnierse, 2020). Basically, high transport costs of conventional ground vehicles result in less frequent bulk deliveries which results in higher inventory levels to safeguard availability which ultimately results in costly wastage.

(18)

A trade-off between an economic and efficient inventory system and short delivery distances can be observed.

2.2.2 Realizing the benefits of higher resupply frequency – coping with traffic

Increasing the delivery frequency is stated by VillageReach (2020) to be one of 5 main concepts in supply chain designs to address common supply chain challenges. More frequent deliveries increase the flexibility and allow to lower the inventory levels. In an extreme case, the resupply can be organized in an on-demand manner as it is done by Zipline in Ghana (see section 2.3.2.1). However, the number of deliveries would be increased.

Deliveries of medical goods have to meet certain timeliness or response time requirements.

Traffic and infrastructure conditions are main factors that cause delays of deliveries which can

be particularly serious for emergency requests because the patients’ clinical outcome can depend on the delivery time of a medical good. Traffic data prove the trend of constantly higher traffic (jam) volumes, globally and in the Netherlands (ANWB, 2019; TOMTOM, 2020). Traffic does not only delay the deliveries but also causes additional risk and stress for the operation as well as for the other road users. The emergence of drone pilot projects, as for example in Hamburg and the Netherlands, is partly justified by increasing traffic stress.

2.3 Considering alternative vehicles

2.3.1 Drones as a solution to blood distribution and transportation challenges

(19)

Blood supply chains are an example of supply chains in the healthcare system that are increasingly strained by cost and performance pressure due to the “lean” healthcare concept. Drones have the potential to play a crucial role in the future design of blood supply chains considering the above mentioned implications. Leading working groups prioritized drones for the use in emergency blood deliveries based on key stakeholders which underpins the potential of drones for medical emergency deliveries including blood deliveries (ISG UAS Coordinating Body, 2018). 2.3.2 Efforts enhancing medical drone implementations for blood supply

2.3.2.1 Real-world (pilot) projects

A great number of projects have been carried out in Africa. In 2016, Phillips et al. (2016) investigated a case in Malawi where efficient and timely transportation of laboratory samples and results between health facilities and laboratories. They analyzed four scenarios where drones and motorcycles were used to transport laboratory samples. They concluded a cost increase for drones in three of the scenarios but also mentioned the immaturity of the drone technology which might make drones more cost competitive in the future. In 2019, VillageReach collaborated with the Malawi Ministry of Health and Population and the Malawi Blood Transfusion Service and investigated another use case in Malawi, namely the transport of blood and oxytocin to treat maternal bleeding. This time they concluded that drones can have lower monthly costs compared to motorcycles due to lower vehicle and personnel costs. This shows that the development of drones potentially can increase the cost competitiveness over time.

(20)

motorcycles. According to Zipline, however, the transport costs for emergency deliveries are already cost-competitive. It is not clear though, to what extend the drones decreased other system costs such as wastage or storage costs at health care facility level. In the end, Zipline is relying on subsidies to operate their service (Ackerman and Koziol, 2019). Though, the positive impact on the health care is enormous which lead to a contract extension with the government of Rwanda and in the expansion into the blood delivery market in Ghana and Tanzania (Ackerman and Koziol, 2019).

Figure 2.3: Drone network in Rwanda (Ackerman and Koziol, 2019)

The case of Zipline shows that a blood delivery network can be established on a large-scale leading to significant health benefits. Though, economics of scale did not yet lead to savings in transportation costs compared to existing ground transportation. It also shows that settings with difficult geographical conditions are likely to be exploited first by drones.

(21)

phase investigating the feasibility or conducting first test flights without cargo in unpopulated areas. Both projects report an increasing demand for customized healthcare and decreasing accessibility of hospitals and healthcare locations which led to the creating of the projects (Medical Drone Service, 2019; Medifly Hamburg, 2020). Additionally, several joint projects within the EU, funded by the EU, have been initiated.

2.3.2.2 Theoretical analyses of medical drone value

The literature relevant for drone blood delivery can be divided into three sub-categories. The first category includes general research on drone potential for medical deliveries. These papers are mostly of a descriptive format explaining why drones can be beneficial for certain applications. They often conduct SWOT-analyses and present recent pilot projects. To mention is here the work of (Thiels et al., 2015; Scott and Scott, 2017; Haula and Agbozo, 2020). The second sub-category includes paper that investigate important distribution-related design parameters such as the network design of possible drone bases or routing decisions. Notable is here, for example, the work of Dhote & Limbourg (2020) who determine the optimal drone network design for blood transportation in the city of Brussels.

The third and most relevant sub-category for this project includes paper that analyze the operational and economic value of drones for the distribution of blood products or similar medical goods. Haidari et al. (2016) used a simulation model to assess the impact of using drone systems for routine vaccine distribution in low and middle income countries. They found that implementing drones can increase vaccine availability and decrease costs in a wide range of settings and circumstances if the drones are used frequently enough. Similar work has been done by Wright et. al (2018) who suggest a flight number of 5000 per year per drone in order to become cost-competitive over motorcycles. Using expert knowledge and experiences of drone projects in Africa they provide a set of factors that are indicative for a value-adding use case for drones in terms of transportation and system costs.

(22)

(23)

3 Methodology

3.1 Research design

This research is organized according to a design science (DS) approach based on the work of Wieringa (2014). Design science is the design and investigation of artifacts in context. The design cycle consists of a Problem Investigation, Treatment Design and Treatment Validation phase. The overall design science goal is to transform the land transportation-based emergency blood supply system to a drone-base supply system by iterating over the design cycle. The focus in this study lies on the consideration of different problem contexts related to the same type of system rather than on a highly detailed artifact (= treatment) design.

For each defined problem context the value of the treatment, which is a drone-based distribution system, is evaluated using discrete-event simulation (DES) experiments. The DES experiments are organized in a simulation study which generates the required outcomes to answer the research question. A discrete-event simulation (DES) method is chosen for the evaluation phase due to multiple stochastic processes in the respected system and due to the simplicity to analyze various problem contexts. Pierskalla (2004) underlines this justification by naming the DESs in the healthcare context a “textbook example”.

(24)

3.2 Problem investigation

3.2.1 Case (example) description

This step lays the knowledge foundation for the subsequent steps. Together with step 2, this step corresponds to the problem investigation phase. First the general use case “blood supply in the Netherlands” is introduced. Subsequently, the blood bank in Groningen, the Netherlands acts as a case example and the main design parameters of it are described using a framework, see Figure 3.2. This framework is is based on literature information on transportation systems.

Figure 3.2: Using a framework to systematically describe a case example

3.2.2 Analysis of case example

Input for this step is the list of main design parameters of the case example. The parameters are analyzed in terms of their suitability for drones. Based on this, a general assessment of the overall suitability of the case example for drones is being given. Furthermore, important influence factors on the suitability are qualitatively identified using literature and domain experts.

Based on these factors, other problem contexts are described where the value of drones is expected to be different compared to the initial case example problem context.

3.3 Treatment design

(25)

scenarios constituting a rural and an urban region. Different types of drones representing the state-of-art capabilities and likely future capabilities are considered.

3.4 Treatment validation

This step investigates the effect of the treatment design on the problem. Since multiple problem contexts are being considered, the effects are being compared among the different problem contexts. The method to evaluate the defined experiments according to the experimental design is a discrete-event simulation software. A simulation model is set up that represents the involved processes in the emergency blood distribution based on the case example. Simplifications and assumptions have to be made due to data uncertainties. The simulations model generates the same type of outcomes for all experiments which allows for comparisons among the experiments. In that way, the influence of the problem contexts on the outcomes can be quantified and validated.

3.5 Outcome parameters

3.5.1 Operational performance - response time

The response time of emergency missions can be broken into three segments (van Werven, 2012). o activation duration

o mobilization duration o travel duration.

(26)

chain). The travel duration means the time after departure from the origin to the arrival at the destination. In this research the response time is made up of the mobilization and travel time. Since the response time is effected by stochastic or time-dependant processes (e.g. traffic, preparation time), the variance of the response time has to be considered. Hence, for every hospital the average response time as well as the confidence interval (95 %) is regarded.

3.5.2 Operational performance – required number of vehicles

In every experiment a certain number of vehicles is required to fulfill the A1 time-constraints which allow a maximum of 1 h response time for every hospital in the Netherlands. In this study, the required number of vehicles is the number of vehicles so that 95% of the A1 missions to the furthest hospital can be served within 1 h. The furthest hospital is assumed to be the bottleneck in the system, hence only this one will be regarded.

3.5.3 Economic performance - costs

Costs can be divided into fixed and operational costs, see Table 3.1. Both costs is account for in this study. The economic analysis is carried out based on the number of vehicles used in the experiments and on the total travel distance. Both variables are (intermediate) outcomes of the simulation study and are used to calculate the corresponding total costs, see Figure 3.3.

Figure 3.3: Conceptual visualization of quantifying cost values for the economic analysis

(27)

medical use cases. However, while the spreadsheet of Wright et. al (2018) takes the number of flights as a metric to calculate the variable costs of drones, this spreadsheet takes the actual traveled distance as a result of all operated (short and long) flights as a metric. This can be seen as an improvement of the calculations, as no universal assumption on the distance of a flight has to be made.

Table 3.1: Type of costs included in the cost analysis

Fixed costs Operational costs

Purchase costs of vehicle Fuel

Operator salary Maintenance/parts replacement Other (Insurance, licenses etc.)

3.6 Data collection and analysis

Data concerning the case example as for example, the locations of served hospitals and the quantity of A1 and A2 missions is collected from the responsible organization (Sanquin). Internal information could successfully be requested. Furthermore, domain experts in the field of drone operation and blood distribution have been interviewed. These experts work in the Netherlands and are highly involved in the current blood distribution or drone implementation projects, such as the joint project “Medical Drone Service”. In total 7 unstructured interviews have been conducted throughout the project timeline. E-mail correspondence with the interviewed domain experts helped to clarify or request additional information. Next to domain expert opinion and internal information from Sanquin, the existing literature has been used to collect data to for example set up the framework mentioned in section 3.2.1 or to estimate cost parameters for cars and drones.

(28)

4 Case description

4.1 Blood supply chain in the Netherlands

In the Netherlands the blood supply is entirely safeguarded by the non-profit organization Sanquin. A sketch of the different activities throughout the supply chain is shown in Figure 4.1. Approximately 2% of the inhabitants are registered as blood donors (Sanquin, 2020a). There are around 100 sites throughout the country where blood is being collected. The collection strategy is planned by Sanquin according to the past demand in order to maintain the desired inventory levels of blood.

Figure 4.1: Overview over the activities in the blood supply chain in the Netherlands

(29)

Blood is distributed from the distribution centers to the hospitals with 3 different missions, namely routine, A2 and A1 missions. Routine missions are planned deliveries to maintain the blood inventories in the hospitals. 30 out of the 90 hospitals are being delivered once per day during the week and not at the weekend. The other 60 hospitals are being served twice a day during the week and not at the weekend. The 8 academic hospitals in the Netherlands are additionally being served at the weekend. Next to routine missions, there are A2 and A1 emergency deliveries which occur due to unexpected circumstances. Per definition, an A2 request cannot wait until the next routine delivery and has to meet an agreed deadline (typically < 3 hours) for every hospital in the Netherlands. A1 emergency requests have to be delivered within 1 h from the time of request to every hospital. In the whole Netherlands around 100 A1 missions and around take place per month. For the A1 emergency missions vehicles are allowed to make use of “blue-light” and sirens to reduce the travel time.

4.2 Case Example – North of the Netherlands

The blood bank in Groningen is one of the 7 distribution centers run by Sanquin. It is assigned to 11 hospitals in the North of the Netherlands covering all of the hospitals in the provinces of Groningen, Friesland and Drenthe, see Figure 4.2.In 2019, 175 A1 emergency missions and 1525 A2 emergency missions have been reported, see Table 4.1. The blood bank is located in the same building as the UMCG, hence the air distance to the UMCG is 0 km.

(30)

Figure 4.2: Blood bank and served locations in the case example

To describe key features of the Sanquin transportation system, a framework (see Appendix) is being applied. According to the framework the transportation system will be characterized according to vehicle characteristics, shipment characteristics, geographical characteristics and operational characteristics, also see Table 4.2 - Table 4.5. The description of the case example according to these characteristics is based on literature and domain experts.

4.2.1 Shipment characteristics

(31)

The capacities of the missions differ, where A2 requests show a bigger range of possible payloads (Sanquin, 2020b). The shipment distances are independent of the type of mission.

Table 4.2: Shipment characteristics of the case example based on internal documents, personal interviews and estimations

4.2.2 Geographical characteristics

The covered are of the blood bank Groningen are the Dutch provinces Groningen, Drenthe and Friesland. The covered population is low compared to the rest of the Netherlands. Two important factors to describe the geographical characteristics in the respect of road density and traffic conditions are the road circuit factor and congestion factor. The derivation of the values for the case example, shown in Table 4.3, are explained in the appendix. Weather conditions are characterized by temperate climate with occasional strong wind events.

(32)

4.2.3 Vehicle Characteristics

The distribution center in Groningen uses exclusively vans for both routine and emergency missions such as the VW caddy type. The capacity of these type of cars is multiple times bigger than the required capacity for A1 or A2 missions. The range of cars is practically unlimited because of the high petrol station density in the Netherlands which allows the vehicles to extend their range at any time.

The average velocities of the cars are assumed to be different for short urban routes and long intercity routes, see appendix. Furthermore, only A1 missions are allowed to make use of emergency tools (blue lights and sirens) which results in higher average velocities compared to civilian vehicle velocity (Poulton et al., 2019). Reasonable estimates of average velocities are based on Petzäll et al. (2011) and shown in Table 4.4.

Cost estimations are mostly obtained from the car costs calculation tool provided by the ANWB Nederland. The reference car is a VW Caddy from 2019 because Sanquin uses this type of car amongst others.

(33)

4.2.4 Operational characteristics

The operational characteristics can be seen as a sub-group of vehicle characteristics. The impact on other road users and pedestrians can be described as significant due to the noise and the urgency of the missions. Besides the normal engine noise, sirens that are being used for A1 missions, generate relevant noise. The urgency of the missions and the sometimes unpredictable behavior of other road users entail a substantial risk for accidents as elaborated in the appendix. The implied traffic delay and stress is expected to scale with the amount of other road users which in the case example is rather low due to rurality.

Table 4.5: Operational characteristics as a sub-group of the vehicle characteristics

4.3 Performance analysis case example

4.3.1 Benefits of drones

(34)

The use of drones provide the advantage of greater flexibility, particularly due to the absence of a pilot, and therefore enables easier human resource management. The autonomy of drones enables vehicle to operator ratio that can, based on domain experts, be 5 to 20 times higher compared to cars which additionally reduces labor costs which are especially dominant in developed countries. Further, cost benefits of drones are present, for instance, in the fuel costs. According to as use cases in Africa, the fuel costs of drones made up to 0,05% of total operating costs compared to 20% for motorcycle (Phillips et al., 2016).

4.3.2 Assessment case example

This analysis highlights factors of the case example which favor the use of drones and factors which favor the use of ground vehicles (as being in operation). Wright et al. (2018) gives a set of factors that indicate a value-adding use case for drones. These factors are based on a rough-cut analysis with expertise and experiences mainly from use cases in Africa. These factors are:

 High demand within range of drone  Difficult to access by road facilities

 High financial value, scarce, high health value (e.g., life-saving) product  Unpredictable demand (at level of individual facility) product

 Product difficult/expensive to store at last-mile or has a short shelf-life

(35)

Table 4.6: Analysis of the suitability of the case example for drone usage

The usage of drones in the case example can clearly be recommended from a product and operational perspective. The strict timeliness requirements and relatively long distances (< 65 km) are other pro factors because according to domain experts the value of drones grows with the travel distance. However, the geographical conditions as well as demand quantity in the case example are value-decreasing factors. This leads to the conclusion that the overall suitability of the case example for drones is only limited.

(36)

5 Selecting and detailing drones that fit case example needs

5.1 Types of drones

There are three different types of drones which have distinct advantages and disadvantages, see Table 5.1. While multicopters are preferred for short distance use within cities, hybrid and fixed-wing drones are more suitable for medium to long-distance missions due to their aerodynamically efficient flying manner. Fixed-wing drones have proven to be a suitable vehicle to transport blood and other medical goods in Rwanda and Ghana by the American company Zipline. Hybrid drones, on the other hand, are increasingly involved in pilot projects and gaining relevance especially in developed countries. They offer the advantages of Vertical Take-off and Landing (VTOL) capabilities while being able to compete with fixed-wing drones in terms of range and payload.

(37)

5.2 Drone specifications

5.2.1 Current market situation of hybrid drones

In summer 2020 several drone manufacturers took part in the “Lake Kivu challenge” which evaluated the performance of hybrid drones in specific use cases in Rwanda (McNabb, 2020). The winning drones are taken here as representative for the state-of-art hybrid drones while the drone of Zipline is taken as the state-of-art fixed-wing drone. The drone “Wingcopter 178 Heavy Lift” from the drone manufacturer WingCopter and the “Manta Ray SR” from Phoenix Wings won the “Emergency Delivery” and “Sample Collection” challenge in Rwanda, respectively. The third relevant drone representing the current market situation is the “Avy Aera” from the manufacturer Avy that won a safety award in the Lake Kivu Challenge and also is the drone used within the Medical Drone Service project in the Netherlands. Data about the technological capabilities have been gathered from their publicly available websites or magazine articles and is shown in Table 5.2.

(38)

Table 5.2: Overview over selected available drone solution (* linearly interpolated values)

5.2.2 Drone requirements in case example

The case example sets some minimum requirements to the drone capabilities. The range of the drone has to be at least as big as twice the air distance of the furthest hospital. The furthest hospital is Sneek with a distance of 65 km from the blood bank which requires a drone range of 130 km to allow for a return flight without battery change. This is seen as a crucial requirement as a battery exchange somewhere else than the drone base would increase the cost and complexity of the system. Another requirement is present concerning the type of payload drop-off. According to Sanquin air drops of blood bags are not desired which eliminates the use of fixed-wing drones. Hence, hybrid drones are favored. Another reasons favoring hybrid drones is the opportunity to add future use cases as, for example, sample pick up missions to the portfolio of drone missions. Thus, hybrid drones allow for an easier reaction to future developments/demands. The mentioned requirements are only fulfilled by the Manta Ray LR drone of the German Drone manufacturer Phoenix Wings who are focusing on the medical delivery market. The Manta Ray LR drone represents the state-of-art hybrid drone technology in this study.

5.2.3 Future improvement of drone capabilities

(39)

Considering this inevitable delay before drone implementation become realistic and the fact that technological development has key impact on the drone market (Dhote and Limbourg, 2020), a “future” version of the Manta Ray drone will be considered next to the “current” version in the subsequent analyses.

The industry of drones is still in an immature phase which leads to steady improvement in terms of performance, costs and operability. Wright et al. (2018) investigated the annual improvement of drones based on literature and predicts an improvement rate of annually 3% for the battery density and of 20% for the cost performance. Next to battery improvements it can be expected that design optimization and weight savings will lead to further improvement. The example of Zipline demonstrates how fast drones can develop. After they started operating in Rwanda 2016 with their drone “Zip 1” they released a new version “Zip 2” in 2018. The new drone can carry a payload of 1,75 kg compared to 1,2 kg (+46%) with “Zip 1”. Furthermore, the maximum velocity could be increased from 100 to 128 km/h (+28%) and the operational procedure to launch the drone could be reduced from 10 to 1 min (Giles, 2018; Ackerman and Koziol, 2019). Based on the finding we assume an improvement of 25 % of speed and range for the Manta Ray drone within the next 3-5 years, see Table 5.3.

Table 5.3: Comparison of current drone capabilities with assumed future capabilities

5.2.4 Cost estimations

(40)

accurately estimated because mass production has not yet been widely implemented. It is assumed that this will decrease costs considerably.

Cost estimations, as shown in Table 5.4, are based on HEMS (2020a) and Wright et al. (2018) . The former is a domain expert involved in the Medical drone service project in the Netherlands and was able to give rough estimations for the AVY drone which is being used in this project. Knowing that the “Manta Ray” drone used in this project is designed for longer and heavier payloads, the cost assumptions for the AVY drone have to be adjusted. The latter provided cost estimations for various type of drones based on expert experience in drone projects in Africa. To account for future cost improvements, the purchase cost and the lifetime of batteries will be considered twice, once for current scenarios and once for future scenarios.

(41)

6 Design of the Simulation Study

6.1 Research questions and hypotheses

6.1.1 Meeting service requirements: Response time

The response times of cars and drones are a central outcome in this study. While drones are independent of the ground infrastructure, cars are subject to traffic which delays the response time. These delays are expected to be more present in urban regions due to a higher number of road users. Urban areas also pose more traffic obstacles such as intersections or sharp turns. On the other hand, fixed starting and landing delays of drones are expected to affect short routes more severe than long routes. The determination of the value of drones for short and long routes is therefore, one sub-research question:

SRQ1:

o What is the impact of drones on the response time of emergency missions and how does it depend on the type of route (i.e. short urban or long intercity) and level of urbanization? The average velocity of cars in cities is lower than on highways. The increasing traffic is expected to decrease the speed of cars in cities by 30 – 50% in the next 3 years AVY BY (2020). Drones can then save between 30 – 60% of travel time depending on the time of the day compared to ground vehicles (AVY BY, 2020). For longer distances with a significant ratio of highways the average velocity of cars and drones may be less different. However, a longer distance with a small speed advantage of drones may add up to a significant time saving ultimately. The technical development of drones is also considered as being important. The hypotheses posed concerning SRQ1 are:

(42)

6.1.2 Costs

According to the analysis in section 4.3 the rurality of the case example is a non-advantageous factor for the suitability of drones because it results in a low number of flights. The importance of flight numbers in terms of costs has been stated by Wright et. al (2018) who state that “fixed costs need to be defrayed over a large number of flights from a single hub for drones to be cost-competitive…” Hence, a sub-research question is:

SRQ2:

o How can the number of flights in the case example be increased and to what extent does it improve the cost-effectiveness of drones?

According to Wright et. al (2018) layering use cases is a powerful tool to increase the number of flights. In the case of the emergency blood supplies in the Netherlands, the A2 missions can be regarded as a use case which can be added to the drone portfolio. While the A1 missions have the highest priority due to their urgent nature, including A2 missions might lead to a higher utilization and better cost-effectiveness of drones. Furthermore, the demand in the case example is rather low. Assuming a more urbanized context would induce a higher number of flights. Based on the statement of Wright et. al (2018) and the mentioned opportunities to increase the number of flights, the hypotheses are:

 A higher level of urbanization of a region and/or including A2 missions to the operation spectrum will at some point make drones more cost-effective than ground vehicles. The hypothesis is underpinned by the fact that multiple drones can be operated by a single operator which does, in case of multiple required drones, have a strong impact on labor costs. Furthermore, drones have lower operational costs which results in a bigger absolute cost saving when increasing the number of missions.

6.1.3 Coverage area

(43)

The timeliness requirements limit the coverage area of blood banks using traditional land transport. Considering future drones which do have higher average speed than ground vehicles, another sub-research question is:

SQR3:

o How much can the coverage area of a blood bank be increased with drones while satisfying

the A1 time-constraints?

The theoretical range of the assumed future drone is 196 km for a payload of 2 kg with a speed of 125 km/h. This could allow a covered radius of 98 km. However, for these long trips a drone would be occupied a significant amount of time which would result in a higher number of drones to meet the service requirements (95 % under 1 h). This phenomena is even more critical if A2 missions are included in the drone operations because multiple long distance flights might be required to fulfill an A2 mission. The following hypothesis is made:

 Coverage radii up to 95 km are achievable but required fleet size increases disproportionally with radius increase.

6.2 Simulation Model

6.2.1 Model Set-up

(44)

the order enter the transport cycle which include for both ground and air vehicles the journey to the destination and the relocation journey to the blood bank. Ground vehicles can always transport A1 and A2 orders within one transport cycle. Drones can transport A1 orders in one cycle but for A2 missions they may go multiple times through this cycle depending on the order size (i.e. number of blood bags) the travel distance and drone capabilities.

For ground vehicles, the travel time is calculated using the assumed average speeds, the traffic factor and road circuit factor, see section 4.2.2. For drones the travel phase is divided into starting, cruising and landing phase. Starting and landing delays are fixed times regardless of the length of trip and based on expert opinion (HEMS, 2020b). The cruising period is always flown with maximum drone speed both ways.

Figure 6.1: Model set-up representing emergency blood distribution (Hospital stage is not included in model)

6.2.2 Outcome measures

From the simulations the following information are obtained for every experiment: o For every demand point

o Average response time of A1 deliveries (primary) + confidence intervals (95%) o Average response time of A2 deliveries (secondary) + confidence intervals (95%) o Total traveled distance

(45)

6.2.3 Software, Warm-up, run length and observation number

The simulation will be run with the simulation software PlantSimulation 15. The run length of an observation depends on the scenario because of the different demand levels. For scenario B a run length of 2 years was found to be sufficient. For scenario A a run length of 10 years was chosen due to the low number of orders, especially for long distances (2 annual orders for Sneek). Increasing the run length to more than 2 or 10 years respectively had no significant effect on the outcome variation. The warm-up period is set to 10 days. An extensive warm-up is not regarded as necessary due to the low system occupancy during the night which “resets” the system every day. Every experiment consists of 10 observations, each with a different random seed value. The coded simulation model can be seen in the appendix.

6.3 Experimental design

6.3.1 Scenario description 6.3.1.1 Rural and urban scenario

(46)

Figure 6.2: Scenarios considered in the simulation

(47)

6.3.1.2 Demand distribution over the radius

For the case example region, demand data for known demand points is available. It has to be noticed, that the UMCG plays a dominating role as it accounts for 46% of all emergency orders in the case example, see Figure 6.3. Since the blood bank in Groningen is located within the same location as the UMCG and inter-facility transport is not required, these missions have to be excluded from the simulation. For the scenario B, such a dominating behavior cannot be expected because there are multiple cities spread over the covered area. In this scenario, an equal demand distribution as in Figure 6.3 (orange line) will be assumed. This means, that after the demand has been extrapolated based on the covered population, 12.5 % instead of 46% of the total demand will be neglected in the simulation.

Figure 6.3: Demand distribution for each scenario depending on distance to blood bank

(48)

that the results can be applied afterwards to specific hospitals with known locations. One dummy point is assumed every 10 km starting from 5 km distance to blood bank.

(49)

6.3.1.3 Traffic modeling

The average velocities for cars on short urban and long intercity routes have been mentioned in section 4.2.3. These values given by Petzäll et al. (2011) do not explicitly mention an incorporation of traffic. In order to account for possible traffic related delays, a congestion factor is applied to the average velocity. This factor is based on the type of route and the time of the day, see Table 6.3. For city areas these values can be obtained by literature as for instance by Poulton et al. (2019) who reported a decrease of 20% of the average speed of ambulances during the day in the London area or by publicly available traffic information as for example by the TomTom Traffic Index (TOMTOM, 2020) that reports a trip delay in Dutch cities of around 40 % for the civilian traffic during rush hour. Congestion effects are assumed to be relatively less present in long distance routes because long distance trips constitute more rural highway roads and less urban non-highway roads. The following congestion factors are used in the simulation:

Table 6.3: Congestion factors for each scenario

6.3.2 Experimental Factors

(50)

Figure 6.5: Conceptual depiction of the sets of experiments allocated to the scenarios

6.3.2.1 Vehicle comparison experiments

These experiments compare the performance of different vehicle types in different contexts. 4 contexts determined by the scenario and the inclusion of A2 missions are examined. The experimental factors for the vehicle comparison experiments are listed in Table 6.4. A total of 12 different experiments will be simulated, see Figure 6.6. For every experiment the number of applied vehicles will be varied in order to obtain the best performance but also in order to obtain the minimum required number of vehicles, compare section 3.5.2.

(51)

Figure 6.6: Experiment tree for vehicle comparison experiments (C_Drone = Current drone capabilities, F_Drones = Future drone capabilities)

6.3.2.2 Coverage area experiments

(52)

The experiments incorporate the operational radius as an experimental factor, see Table 6.5. The range of 65 km is the assumed maximum range that Sanquin can serve with current ground vehicles. The range of 95 km is the maximum that can technically be done considering the assumed future drone range. Only scenario B will be considered because the approach of using dummy points can easily adapted to these experiments. The covered surface of such a radius can be seen in Figure 6.7. The overall demand is not changed in these experiments and the demand distribution remains equal over the radius according to Figure 6.3.

Table 6.5: Experimental factors of the coverage area experiments

Figure 6.7: Coverage for a radius of 65 and 95 km

(53)

Figure 6.8: Experiment tree for the coverage area experiments (F_Drones = Future drone capabilities)

6.3.2.3 Sensitivity analysis

Assumptions had to be made in the experimental design for certain parameters due to the lack of data. To account for lack of data or data uncertainty, a sensitivity analysis is conducted. The parameter that will be analyzed are the payload size of A2 requests and the demand distribution over the radius.

According to Sanquin the payload size for A2 missions can vary between 1 and 25 blood bags but is not further recorded. A triangle distribution with 12 bags as mean value have been assumed in the vehicle comparisons and coverage potential experiments. For the experiments 1.4, 1.5 and 1.6, a mean of 7 and a mean of 17 blood bags are simulated and compared with the default mean of 12. The aim is to find out if the response times and confidence intervals change.

(54)

Regions such as for example the Paris, London or Istanbul region would result in a different demand distribution with a significantly higher share in shorter distances (< 20 km). An alternative distribution B.2 is assumed according to Figure 6.9 for the experiments 1.10, 1.12 and 2.6.

(55)

7 Results

7.1 Vehicle comparison experiments

When the number of vehicles are increased, the response times reach a plateau at some point. The level of the plateau and how fast (i.e. with how many vehicles) it can be reached depends on the problem context (i.e. experiment) and vehicle type. First, the level of the plateaus are being compared for the vehicle types. This enables to assess the maximum potential reduction in response time when using drones compared to cars. Second, the numbers of required vehicles for each defined experiment, in order to comply with the A1 time-constraints, are compared. Subsequently, the number of required vehicles is used for a cost analysis.

7.1.1 Response times

Scenario A

The lowest possible response times that can be achieved (i.e. the plateaus) in scenario A are shown in Figure 7.1 and Figure 7.3. Additionally, ranges (confidence intervals) are shown that include 95% of the corresponding missions. All vehicles are able to achieve response time and ranges below 1h. For short routes, drones offer no or only a small advantage whereas for longer distances the advantage becomes significant. The fixed starting and landing delay of drones impact the short distances relatively more than the long distances. For the demand points Emmen and Sneek the response times could be reduced by around 15 and 30 % using current and future drones respectively. The size of the confidence intervals is slightly higher for cars than for drones which can be explained by the influence of different time-dependent traffic levels.

(56)

Figure 7.1: A1 response time plateau values for scenario A, only A1

(57)

For scenario B when only A1 missions are included, the plateaus of the response times are shown in Figure 7.3. Again, the advantage of drones becomes bigger the longer the travel distance is. For the demand point 65 km the time saving are 22 and 35 % when using current and future drones and, therefore, higher than the time savings in scenario A. Furthermore, it is noticeable that the confidence interval of the response times for drones (3.4 min) is independent of the demand point and only influenced by the stochastics of the order preparation process. The absolute confidence interval size for vans is clearly increasing with distance from 3.4 min to 11.6 min as it is additionally influenced by different time-dependent traffic levels that have a bigger impact for longer distances.

Figure 7.3: A1 response time plateau values for scenario B, only A1

(58)

Figure 7.4: A1 response time convergence behavior for vans

(59)

Figure 7.6: A1 response time convergence behavior for future drones

(60)

drones offer no substantial benefit. However, long distances are the bottleneck in such system as the 1 h time-constraints are being hurt their first. Based on the reduced response times, it is expected that the number of required vehicles differ from cars to drones for a given setting. 7.1.2 Required number of vehicles

The required number of vehicles for each experiment is defined as the number of vehicles with which all average response times are below 1 h. Additionally, in order to ensure a high service level, 95% of missions to the furthest hospital have to be accomplished as well under 1 h. For scenario A these requirements could be reached by all type of vehicles. For scenario B, due to higher traffic delays, cars were not able to reach the 95 % requirements. Only around 56 % of A1 missions to the furthest hospital could maximum be reached, see appendix. Therefore, the number of required vans for scenario B is set to the number when no further significant decrease in response times can be achieved. The required number of vehicles for these experiments are shown in Table 7.1.

Table 7.1: Minimum required number of vehicles to comply with the A1 time constraints

(61)

Figure 7.8: A1 response times with required number of vehicles, scenario A, A1 + A2

(62)

Figure 7.10: A1 response times with required number of vehicles, scenario B, A1 + A2

It can be seen that although the number of vehicles is lower for drones compared to cars, the response times in all settings are also lower. The reduced number of vehicles is noticeable in the higher ranges of response times. This is a result of higher utilizations which entail waiting times of orders that increase the overall variance. Furthermore, the response times of vans for long distances (65 km) in scenario B is around 1 h in average with a significant part of the range being above that threshold. This underlines the remark made in the beginning of the section and clearly shows that cars are not able to serve hospitals at 65 km within 1 h in scenario B.

7.1.3 Economic analysis

(63)

Table 7.2: Total annual travel distances and number of flights/rides for experiments

Based on the total travel distance, the number of required vehicles and the cost assumptions (Table 4.4, Table 5.4) the total annual costs can be calculated, see Table 7.3. The costs are displayed in absolute values and in cost per mission. For every vehicle type the costs per mission decrease when increasing the demand. Vans were found to be the most cost-effective option in scenario A. However, in scenario B, drones with future specifications could save 6 % of missions cost compared to vans when only A1 missions were incorporated. When including A2 missions the cost savings could even be increased up to 15 % per mission. The calculations including the current cost assumptions are implemented in an Excel spreadsheet which allows to easily change the influencing parameters.

(64)

7.1.4 Statistical reliability

In every experiment 10 observations have been simulated each with a different random seed value. Hence, all observations are independent of each other. Every observation calculates average response times including 95 % confidence intervals. The average response time and the size of the confidence intervals have then been averaged throughout the 10 observations. This approach is seen as statistically reliable which is underlined by the relatively small confidence intervals, compare Figure 7.1, Figure 7.2, Figure 7.3 and Figure 7.7. The confidence intervals have a certain size in any case due to stochastic processes in the model.

7.2 Coverage potential experiments

7.2.1 Required number of vehicles and response times

The achievable response times (i.e. plateaus) for all distances lie under 1 h. This is true for the average values as well as for the 95 % confidence intervals, see Figure 7.11. Incorporating A2 missions does not change the results significantly.

(65)

The required number of drones for different coverage radii in order to comply with the A1 time-constraints is shown in Table 7.4. While the number of vehicles increases to maximum 2 when including only A1 missions, the number of required vehicles increases to maximum 7 when A2 missions are also included. It can be noticed that 2 (future) drones are enough to cope with the A1 demand even for a coverage radius of 95 km. This is interesting when considering that for a 65 km radius the same amount of vans (2) was needed for the same setting, see Table 7.1.

Table 7.4: Minimum required number of vehicles to comply with A1 time-constraints in coverage area experiments

When increasing the coverage radius, A1 and A2 missions have a different impact on the total travel distance, see Table 7.5. This is due to the bigger payload size of A2 missions and the payload restrictions of drones for bigger ranges, see Table 5.3. For an A2 delivery to a 95 km far away demand point with a payload of 25 blood bags, 8 flights become necessary whereas an A1 delivery requires 1 flight to the same demand point. This results in an overall higher utilization of the vehicles, a higher total travel distance and in A2 response times of more than 2 hours for long distances, see Figure 7.12. This effect is also visible in the response times of A2 missions which become greater than 2 h for long distances, see Figure 7.12.

(66)

Figure 7.12: Response times of A2 missions when operational radius is 95 km

7.3 Sensitivity Analysis

7.3.1 Payload size of A2 missions

(67)

Figure 7.13: Using current drones for the payload size (A2) sensitivity experiments

(68)

7.3.2 Demand distribution scenario B

Assuming an indefinite number of vehicles an alternative demand distribution does not change the plateau response times. However, if limited vehicles are available, the influence of another demand distribution can be important. This influence is shown in Figure 7.15 and Figure 7.16 where the response times and confidence intervals are depicted for the scenario B with A2 missions for the exact same amount of vehicles. The average response times and confidence intervals are significantly smaller for the alternative demand distribution B.2. This is reflected in the percentage of missions that cannot be completed within 1 h, see Table 7.6. The reason is a higher share of short routes which decreases the utilization of vehicles. This leads to lower average waiting times of orders which, ultimately, decreases the average response times. The demand distribution has, therefore, a significant impact on the results and should be modeled carefully.

(69)

Enhancing emergency blood supply using drones