Leveraging System Dynamics Modelling for Humanitarian Response

Aug 27

At the MSF Sweden Innovation Unit, we endeavour to stay ahead of the curve and look at how new technologies could be implemented in MSF contexts. As part of this, the SIU is publishing a series of guest blogs in which experts in a particular technology can write about their areas of expertise in relation to MSF.

System Dynamics modelling — which aims to create complex simulations of operational decisions to determine the best course of action — is one such area. Here, two experts in the field, Dr. Muhammad H. Zaman and Anuraag Gopaluni, explain the concept in more detail...

The delivery of healthcare in conflict and humanitarian crises is a complex challenge. Humanitarian agencies face an uphill task in delivering care to the vulnerable populations and are often themselves in the cross-hairs. As a whole, they face challenges due to the fluid and unpredictable environment they operate in, especially in the realm of logistics and operational efficiency. The inherent complexity that arises from the nature of the work is such that the exact ramifications of every decision made are virtually unknowable. The number of moving parts in the system, including those that are highly unpredictable, influence not only logistics but critical decisions at the bedside. How do we decide on the most efficient and strategic way to channel efforts and resources for both robust outcomes and resource efficiency at the time of crisis?

To answer these questions, we must first recognise the underlying dynamic complexity in emergency aid and relief operations; that is, there are so many interconnected and interrelated elements, and it is the nuanced, ambiguous interactions among those elements that create an overall system response. Additionally, this system response is often in the form of counterintuitive, unexpected, and delayed behaviours and consequences. Thus, it is absolutely imperative that decision-makers have the tools to be able to address these challenges and complexities by adopting a systems thinking mindset and craft appropriate strategies. While the dynamic nature of the problem means that all decisions will have an element of risk, modelling through robust mathematical approaches, including System Dynamics (SD) enables the stakeholders to capture the complexity, minimise risk and improve efficiency in delivery. While SD has been used effectively in improving hospital function and service delivery in high resource environments, the application of SD has been generally lacking in humanitarian emergencies. Here, we propose how to bridge that gap in knowledge.

What is System Dynamics?

First, let us understand what we mean by System Dynamics as a modelling strategy. System Dynamics is a well-established analytical simulation modelling technique that is suited perfectly to capture the dynamic complexity of health systems and model their many interconnected and interdependent elements. Crucially, System Dynamics allows us to look at the entire system in a holistic manner, where a change in the system is not always the result of a simple, linear change in one other part but instead where it is the result of a complex, non-linear combination of actions. So how does it work exactly? SD models consist of both a qualitative and quantitative aspect. The qualitative aspect of system dynamics is represented as a causal loop diagram (CLD) (shown in Figure 1 below), which captures the interactions between the components of the system; the components are connected by arrows with signs depending on increasing or decreasing relationships. The quantitative aspect of system dynamics is represented as a stock-flow diagram—derived from the CLD and built and simulated by computer software—that expresses the flow of the continuous entities of the system. Also embedded in SD models—in both the causal loop diagrams and stock and flows—are feedback loops, which capture the self-reinforcing and self-correcting behaviours of systems, and time delays.

Let us consider a brief, simple illustration. Imagine, for example, that we are concerned with modelling medical care delivery. In this case, the stocks would be medical personnel, volunteers, and supplies or resources such as disaster kits and vehicles, all of which either increase with further recruitment or decrease with deployment. An example of a behaviour that the causal loop diagram in Figure 1 captures is the intuitive process of there being an influx of medical personnel and volunteers, which increases the distribution rate and allocation of relief, which in turn reduces number of personnel and amount of resources to begin with. Another intuitive mechanism would be the process of a certain site meeting its needs, which reduces the distribution of personnel and supplies to it, thereby increasing their potential to be distributed elsewhere. This is not a comprehensive diagram by any means, and of course, there are many more stocks and flows to consider and other details and nuances to take into account—delays, for example—to develop a causal loop diagram and then convert it to a stock and flow diagram by bringing in the quantitative elements, but compared to other simulation methods like discrete event simulation (DES), SD models are far more approachable. One may also wonder about or question, very reasonably, the robustness and usefulness of an SD model when, as is often the case in humanitarian settings, there is a lack of data or high-quality data to support building the model. Fortunately, to develop causal loop diagrams and capture the qualitative dynamics of the system, expert opinion is all that is needed. To validate the quantitative aspects of the SD model, ideally historical data on resource allocation and relief are available, but SD models do not require large amounts by any means. Moreover, understandably the data may be of low quality, but SD models still work in such scenarios; unlike other models, SD models are very robust to limitations in the size and quality of datasets. Ultimately, as long some data on outcomes exist, SD models can be very easily built, calibrated, especially with the help of experts to weigh or eliminate different scenarios, and ultimately prove to be very useful.

The utility of SD in humanitarian response

Now, let us analyse the real, tangible utility of this approach, particularly for field hospitals in conflict and humanitarian emergencies. The real value of system modelling is in simulation: being able to assess the consequences of different courses of action so that an intelligent, truly well-informed decision can be made. SD models can allow managers to understand the interactions between the different components, experience the long-term effects of decisions, and explore new strategies. Thus, the utility of SD models extends beyond capturing the behaviour of the health system as was briefly described in the above paragraph; SD models can really be leveraged for intervention. Efforts to apply System Dynamics to humanitarian responses are well-documented (as mentioned in the references below), especially as it relates to supply chain management and operational efficiency. For example, SD was previously used to model and optimise field vehicle fleet management as well as transportation of relief supplies. Additionally, System Dynamics has been shown to be valuable in healthcare, whether it is in the realm of patient flow modelling or hospital resource management. Ultimately, SD offers a system-wide perspective that can aid strategic decision-making.

Accordingly, we believe it can be applied to an MSF hospital and that it can provide managers and decision-makers with a set of tools to really tackle dynamic complexity and make the best decisions possible with the best possible consequences. The goal here would be to analyse the hospital function in terms of its key constituent elements (e.g. number of patients and the variation in that number, number of clinical staff, infrastructure and clinical resources and the factors that influence these elements). By modelling the flow (or lack thereof) of these factors, we should be able to simulate various realistic scenarios seen at the field hospital. These scenarios would then provide us with the necessary knowledge about not only the bottlenecks to efficiency but also the best application of limited resources to overcome the said bottlenecks. For example, say there are multiple sites that are in need of immediate assistance but with different circumstances; perhaps one site is a greater distance away but is in a direr need for doctors. Using simulation modelling, doctors and decision makers can explore sending different combination of teams to the different sites, assess how that affects desired outcomes, and develop strategies to optimise and maximise those desired outcomes. Not only will the simulation environment provide an analysis of current scenarios most often seen, and an analysis of whether the current approach is most effective, but it will also provide information of future realistic (or even rare) instances which may strain the system. This would create an opportunity to examine current practices, improve upon them and provide guidance for preparing to deliver the best care for the most vulnerable populations.

Authors

Dr. Muhammad H. Zaman
Professor, Departments of Biomedical Engineering and International Health, Boston University; Professor Howard Hughes Medical Institute; zaman@bu.edu

Anuraag Gopaluni
Department of Mathematics and Statistics, Boston University; raag@bu.edu