Abstract
Contact tracing is critical to controlling COVID-19, but most protocols only “forward-trace” to notify people who were recently exposed. Using a stochastic branching-process model, we show that “bidirectional” tracing to identify infector individuals and their other infectees robustly improves outbreak control, reducing the effective reproduction number ($R_{eff}$) by at least ~0.3 while dramatically increasing resilience to low case ascertainment and test sensitivity. Adding smartphone-based exposure notification can further reduce $R_{eff}$ by 0.25, but only if nearly all smartphones can detect exposure events. Our results suggest that with or without digital approaches, implementing bidirectional tracing will enable health agencies to control COVID-19 more effectively without requiring high-cost interventions.
The Topology of Transmission: Why Forward-Only Tracing Fails
During the height of the COVID-19 pandemic, public health strategies were challenged by high rates of asymptomatic and pre-symptomatic transmission. Standard "forward-tracing" protocols typically look back only 48 hours from symptom onset to identify people a patient might have exposed.
However, this approach has a fundamental flaw in its topology: if the person who infected the patient (the "parent" case) is asymptomatic, they remain unidentified, potentially sparking new clusters. Forward-only tracing effectively "clips" a single branch of the transmission tree but leaves the rest of the branch—and the trunk—intact. This study demonstrates that "bidirectional" tracing—finding both the infector and their other infectees—is a far more powerful tool for stopping outbreaks.
Methodology: Stochastic Branching-Process Modeling
The researchers developed a stochastic branching-process model to simulate the spread of SARS-CoV-2. Each "case" in the model generates new cases based on a negative binomial distribution to account for superspreading events, incorporating realistic distributions for incubation periods and generation times.
The model compared four primary scenarios:
- Manual Forward-Only: The status quo using a 2-day look-back window.
- Manual Bidirectional: Extending the look-back window (up to 7 days) to identify the infector.
- Digital (Smartphone-based): Utilizing Bluetooth "chirps" for exposure notification.
- Hybrid: Combining manual interviews with digital tools.
Parameters were drawn from early 2020 epidemiological data, assuming a 45% asymptomatic rate and that asymptomatic cases were roughly 50% as infectious as symptomatic ones.
The "Parent" Case as a High-Value Target
A key insight of the paper is the "waiting time paradox" in networks: the person who infected a patient is statistically more likely to be a "superspreader" than a randomly selected infected individual. By identifying the "parent" case, tracers can discover all of that individual's other victims simultaneously. This "reverse-tracing" and subsequent "iterative tracing" (finding the parent's other "children") dramatically improves the resilience of the system.
Bidirectional tracing remains effective even under "noisy" conditions, such as when tests are only 60% sensitive or when health agencies only catch 30-50% of symptomatic cases. In contrast, forward-only tracing fails quickly when case detection is imperfect.
Results: Quantitative Impact on Outbreak Control
The quantitative benefits of switching to bidirectional tracing are substantial. The model shows that manual bidirectional tracing reduces the effective reproduction number ($R_{eff}$) by approximately 0.3 compared to forward-only methods.
In median scenarios, bidirectional tracing tripled the probability of successfully controlling an outbreak. While digital-only tracing showed promise, it was found to be highly sensitive to "network fragmentation." If only 50% of the population uses a tracing app, the probability of both parties in a contact event having the app is only 25% ($0.5^2$). This quadratic decline makes digital tools unreliable as a standalone solution unless uptake is near-universal. However, a hybrid system (manual bidirectional + 80% digital uptake) proved most effective, potentially reducing $R_{eff}$ by an additional 0.25 and bringing it below the critical threshold of 1.0.
Implementation and Future Directions
While bidirectional tracing is more labor-intensive—requiring longer interviews and more tracers to manage the expanded 7-day window—it offers a high-return intervention. The authors suggest that future digital tracing algorithms should be explicitly designed to support bidirectional logic, rather than simply acting as a faster version of forward-tracing. By focusing on the trunk of the transmission tree rather than just the most recent branches, health agencies can achieve significantly better epidemic control without relying solely on high-cost interventions like total lockdowns.
Summary updated