Integration vs segregation: network analysis of interdisciplinarity in funded and unfunded research on infectious diseases

TL;DR

This paper tackles how funding shapes interdisciplinarity in infectious disease research by constructing temporal co-occurrence networks across 36 diseases (1995–2022) to compare funded and unfunded outputs. It identifies regime-like phases of topic structure and shows funded IDR tends toward compartmentalisation (stronger within-cluster links) while unfunded IDR leans toward global integration (bridging distant topics). Key findings reveal HIV, TB, and malaria act as global bridges, whereas DTP-related topics reinforce local connections; coronavirus research dominates output post-2019 but has a limited systemic impact on IDR, suggesting delayed or constrained integration. The work offers a generalisable framework for funding policy analysis, with implications for horizon scanning, prioritisation, and incentive design to promote intelligent risk-taking and interdisciplinarity in health research.

Abstract

Interdisciplinary research fuels innovation. In this paper, we examine the interdisciplinarity of research output driven by funding. Considering 36 major infectious diseases, we model interdisciplinarity through temporal correlation networks based on funded and unfunded research from 1995-2022. Using hierarchical clustering, we identify coherent periods of time or regimes characterised by important research topics like vaccinations or the Zika outbreak. We establish that funded research is less interdisciplinary than unfunded research, but the effect has decreased markedly over time. In terms of network growth, we find a tendency of funded research to focus on readily established connections leading to compartmentalisation and conservatism. In contrast, unfunded research tends to be exploratory and bridge distant knowledge leading to knowledge integration. Our results show that interdisciplinary research on prominent infectious diseases like HIV and tuberculosis tends to have strong bridging effects facilitating global knowledge integration in the network. At the periphery of the network, we observe the emergence of vaccination-related and Zika-related knowledge clusters, both with limited systemic impact. We further show that despite the surge in publications related to COVID-19, its systematic impact on the disease network remains relatively low. Overall, this research provides a generalisable framework to examine the impact of funding in interdisciplinary knowledge creation. It can assist in priority setting, for example with horizon scanning for new and emerging threats to health, such as pandemic planning. Policymakers, funding agencies, and research institutions should consider revamping evaluation systems to reward interdisciplinary work and implement mechanisms that promote and support intelligent risk-taking.

Figures (10)

• Figure 1: Bottom-up hierarchical clustering of yearly time slices for the funded, unfunded and aggregated networks. Hierarchical clustering using UPGMA was performed for all (A), funded (F), and unfunded (U) networks respectively. The highlighted identified significant clusters were based on the elbow method as in Figure \ref{['Fig8']} in the appendix.
• Figure 2: Network visualisation of funded and unfunded temporal regimes. The link strength $w_{ij}$ shown in the visualisation of the regime is computed by the average of the link strength across the regime. Node size of disease $i$ represents the proportion of publications on $i$ during the regime. The network layout was produced using R based on the Fruchterman-Reingold method frlayout applied on F1. The community structure was detected based on the Louvain method blondel_fast_2008.
• Figure 3: The evolution of the level of knowledge integration in funded research compared with unfunded research for (A) U1:1995-2003, U2:2004-2015 and U3:2016-2022; and (B) F1:1995-2007, F2:2008-2015 and F3:2016-2022. For each plot, the y-axis represents the average link strength of a pair of infectious diseases in funded research and the x-axis in unfunded research. Each error bar represents the standard error of the link strength of a pair within the regime. The red line represents the 45-degree line: any link lying on the line represents the same level of ID in funded and unfunded research, above (below) indicates more (less) funding is allocated to the pair than should be. A regression line is fitted to the points and the slope of the fitted regression line, the standard error of the slope, and the R-squared value of the fitted regression line were reported in the top left corner. The shaded area around the fitted line represents one standard error of the slope. Yellow labels represent the top seven influential links that appeared in F3 or U3, and their corresponding positions were shown in F2, F1, U2, and U1 plots.
• Figure 4: Temporal change in the ranking of the annual number of publications, strength and betweenness of (a) coronavirus-related publications (b) zika-related publications in the funded and unfunded disease network 1995-2022. The yellow area in (a) indicates the SARS outbreak 2002-2004, and the red area in (a) indicates the COVID-19 outbreak 2019-2022. The red area in (b) indicates the zika outbreak 2015-2016.
• Figure 5: Comparison of the average funded network 1995-2007 including or not including(*) the years 1997 and 2004. There is no major difference in the network structure after including these two years, except that certain links get slightly weakened due to the noise introduced.