Efficient detection of contagious outbreaks in massive metropolitan encounter networks

Physical contact remains difficult to trace in large metropolitan networks, though it is a key vehicle for the transmission of contagious outbreaks. Co-presence encounters during daily transit use provide us with a city-scale time-resolved physical contact network, consisting of 1 billion contacts among 3 million transit users. Here, we study the advantage that knowledge of such co-presence structures may provide for early detection of contagious outbreaks. We first examine the “friend sensor” scheme — a simple, but universal strategy requiring only local information — and demonstrate that it provides significant early detection of simulated outbreaks. Taking advantage of the full network structure, we then identify advanced “global sensor sets”, obtaining substantial early warning times savings over the friends sensor scheme. Individuals with highest number of encounters are the most efficient sensors, with performance comparable to individuals with the highest travel frequency, exploratory behavior and structural centrality. An efficiency balance emerges when testing the dependency on sensor size and evaluating sensor reliability; we find that substantial and reliable lead-time could be attained by monitoring only 0.01% of the population with the highest degree.

💡 Research Summary

This paper investigates how city‑wide public‑transport co‑presence data can be exploited for early detection of contagious outbreaks. Using a week‑long smart‑card dataset from Singapore’s bus system, the authors reconstruct a high‑resolution temporal contact network comprising roughly one billion physical encounters among three million commuters. Each encounter is weighted by its duration (recorded in 20‑second intervals), enabling a realistic Susceptible‑Exposed‑Infected (SEI) simulation where transmission probability on an edge i‑j is p = β · d_ij. The contagion rate β is varied (0.001–0.005) and ten random index cases are seeded on a Saturday.

Two sensor strategies are compared. The first, the “friend sensor” scheme, selects 1 % of the population at random as a control set C and then picks three random neighbors of each member (allowing repeats) to form the sensor set S. This approach relies only on the friendship paradox—neighbors tend to have higher degree in heterogeneous networks—and requires no global network knowledge. Across 20 simulation runs, the friend sensors consistently reach a 5 % infection prevalence earlier than the whole population, with an average lead‑time of 10–20 hours depending on β. The authors note that despite the temporal nature of the process, the paradox still holds because the underlying degree distribution is highly skewed.

The second approach leverages full network information. Four centrality metrics are computed for every individual: (1) degree k (total number of contacts), (2) travel frequency f (number of transit trips), (3) k‑shell index k_s (from k‑core decomposition), and (4) encounter entropy S (temporal diversity of contacts). For each metric the population is sorted and divided into 100 percentiles; the top 1 % of each ranking is taken as a sensor group. Simulations reveal that all four “global” sensor groups outperform the friend sensors, delivering lead‑times of roughly 19–22 hours, with degree‑based sensors achieving the highest (≈22 hours). When the sensor size is reduced to the top 0.01 % (≈300 individuals), lead‑times increase further, reaching up to 24 hours, while the variance across runs drops markedly. This demonstrates that a very small, well‑chosen subset can provide both early warning and reliability.

The authors also examine the trade‑off between sensor size and detection reliability. Very small sensor fractions (<0.001 %) yield unstable lead‑times, whereas fractions around 0.01 % strike a balance: they capture a substantial portion of the network’s centrality while keeping the monitoring cost low. Importantly, the study shows that degree correlates strongly with the other metrics; high‑degree individuals also travel more frequently and exhibit higher temporal encounter entropy, suggesting that degree alone is a practical proxy for sensor selection.

From a policy perspective, the paper argues that in settings where constructing a full contact network is infeasible (due to cost, privacy, or computational constraints), the friend‑sensor method offers a low‑cost, universally applicable alternative. However, when richer data are available, selecting the top 0.01 % of commuters by degree (or a similar centrality measure) yields the best early‑detection performance with modest resource requirements.

Overall, the study provides the first empirical demonstration—at a metropolitan scale—of how high‑resolution transit co‑presence data can be turned into actionable epidemiological surveillance. It quantifies the benefits of local versus global sensor strategies, clarifies the impact of sensor size, and delivers concrete recommendations for public‑health agencies seeking cost‑effective early warning systems in dense urban environments.