Thursday 25.11. at 13:45-15:00
Chair: Diego Flores, MSF – HUMLOG Institute, Finland

Interpretable Feature Selection for Medical Decisions by Evidential Reasoning: A case study in Sepsis

Fatima Almaghrabi, Umm Al-Qura University, Saudi Arabia
Swati Sachan, University of Liverpool, United Kingdom
Dong-Ling Xu, University of Manchester, United Kingdom

Decision support system
Interpretable Feature Selection
Evidential Reasoning
Decision Support Systems
Referential Values

Highlighting the most contributing features and features that could cause noise to a prediction model is becoming of a great
interest to be applied in many fields, including various medical applications. We propose the evidential reasoning (ER) rule for
building an interpretable feature selection and identifying specific thresholds for medical features collected from patients in
diagnosing critical sepsis cases. We provide details for the implementation process for the ER rule, the highlighted features in the
interpretable feature selection process, and the details of the selection of referential values in each feature to identify the optimal thresholds for diagnosis and to highlight the parallel in the selected thresholds between the present work and those presented in literature. The interpretable feature selection using the ER rule has highlighted the following features: age, CCMDS sample days, and length of stay for predicting the severity of sepsis cases.

Mobile Laboratories – an innovative approach to bridging the diagnostics gap in low- and middle-income countries

Thomas Breugem, INSEAD, France
Tim Sergio Wolter, INSEAD, France
Luk Van Wassenhove, INSEAD, France

Mobile laboratories

Praesens Care is a social enterprise start-up that produces mobile laboratories that are used in low- and middle-income countries, predominantly Africa. Following successful deployments in Senegal, Praesens Care recently received a sizeable contribution from the European Commission to scale-up its global fleet to at least eight such mobile laboratories and expand to other countries. First and foremost, the mobile laboratories are used during an outbreak response for epidemic-prone diseases such as Ebola, and they have been used as integral part during the 2017 and 2018 Dengue outbreaks in Senegal. There are, however, many other efficient use cases for such mobile laboratories, especially outside of periods of emergency response. Such applications include surveillance and screening, patient follow-up for chronic diseases (Tuberculosis, HIV/AIDS, chronic heart and liver disease, Hepatitis B, etc.), or providing an integrated testing & care package to rural areas, refugee camps, or following natural disasters. The opportunities are huge, and options of operational set-up are seemingly endless, but options require generate vastly different impact and they also have varying enabling conditions. Praesens Care has a longstanding partnership with INSEADs Humanitarian Research Group, who provide support in the areas of business strategy, service packages, stakeholder management, operational set-up and optimization, as well as impact maximization. In this ongoing research, we discuss the key take-aways in these areas. This is an example of research that provides practical insights to an implementing partner active in humanitarian supply chains, and these insights are directly transferrable to other applications or new technologies in humanitarian supply chains. Amongst many others, research questions include: What services do we offer, and how? Which stakeholders should we engage with, and how? Which locations should the mobile lab service, how often, and for how long, to generate the highest impact? How should impact be measured?

Revealing vaccine hesitancy mechanisms by using systems thinking: insights from two Rwandese communities

Catherine Decouttere, Research Center for Access-to-Medicines, KU Leuven, Belgium
Stany Banzimana, University of Rwanda, EAC Regional Centre of Excellence for Vaccines, Rwanda
Pål Davidsen, System Dynamics Group, University of Bergen, Norway
Carla Van Riet, Research Center for Access-to-Medicines, KU Leuven, Belgium
Corinne Vandermeulen, Leuven University Vaccinology Center, KU Leuven, Belgium
Elizabeth Mason, MGH Institute for Technology Assessment, Harvard Medical School, Cambridge, MA, USA
Mohammad S. Jalali, MGH Institute for Technology Assessment, Harvard Medical School, Cambridge, MA, USA
Nico Vandaele, Research Center for Access-to-Medicines, KU Leuven, Belgium

Objective: To study measles outbreak and vaccine hesitancy dynamics leading to local under-immunization in a low-income country by developing a conceptual, community-level model of behavioural factors.

Methods: We applied systems thinking, human-centred design, and behavioural frameworks to explore local immunization systems in two Rwandese communities, one of which recently faced a measles outbreak. Based on eleven in-depth discussions with health centre staff (providers), 161 interviews with caregivers (beneficiaries) at health centres and nine follow-up validation interviews with health centre staff between 2018 and 2020, vaccine hesitancy factors were categorized using the vaccine hesitancy “3Cs” framework (Confidence, Complacency, Convenience). We developed a conceptual model representing the vaccine hesitancy mechanisms with feedback loops.

Findings: By comparing the perspectives of service providers and caregivers in rural and peri-urban settings, we found similar strengthening factors contributing to vaccine uptake such as high trust in vaccines and service providers based on personal relationships with health centre staff, the connecting role of community health workers, and the strong sense of community. In contrast, challenges contributing to vaccine hesitancy differed amongst providers/beneficiaries and settings such as accessibility of services and vaccination follow-up. The conceptual model was used to explain the drivers of a recent measles outbreak and informs intervention design for vaccine uptake.

Conclusions: Community-level insights analysed with behavioural frameworks and systems thinking revealed vaccine hesitancy mechanisms that show the interrelationship between immunization services and caregivers’ vaccination behaviour. They expose confidence-building social structures and context-dependent challenges that need to be addressed for sustainable and equitable vaccine uptake.

Prioritization-based allocation of bed-nets for malaria control

Fabíola Negreiros de Oliveira, PUC-Rio, Brazil
Douglas Alem, The University of Edinburgh, United Kingdom
Fabrício Oliveira, Aalto University, Finland
Adriana Leiras, PUC-Ri, Brazil


Malaria is a life-threatening infectious disease widely distributed in tropical and sub-tropical regions worldwide. The disease remains one of the most severe public health concerns. Although the burden of malaria has globally decreased, morbidity and mortality remain high, especially in impoverished nations where the population has low socioeconomic conditions. Several studies have associated both environmental and socioeconomic conditions with malaria incidence, suggesting that these aspects should be considered when developing and implementing malaria control interventions, such as bed-nets distribution. For regions where funding and disease surveillance are limited, it is crucial to design prioritization public policies that ensure efficient bed-nets allocation and distribution to those who need them most. This paper proposes a location-allocation model for bed-nets distribution to maximize the expected benefit of prioritizing the most vulnerable locations to malaria transmission. The prioritization model is guided by the Malaria Vulnerability Index (MVI), a composite index that incorporates 12 variables structured in 3 dimensions: epidemiological, socioeconomic, and environmental. For empirical validation, we analyze 310 municipalities of the North region of Brazil, considered the endemic area of the country. Our MVI prioritization model encourages the search for equitable solutions, as it focuses on covering the most vulnerable municipalities to malaria transmission. The model results allow academics and practitioners to have a deeper look into discussions about equity and prioritization measures in public policies for malaria control and prevention.