Bidirectional contact tracing could dramatically improve COVID-19 control

The study addresses how to improve the effectiveness of contact tracing for COVID-19. Conventional protocols emphasize forward tracing of contacts from a known case, typically within 2 days prior to symptom onset per EU and WHO guidance. However, widespread asymptomatic and presymptomatic transmission, along with superspreading events, allow transmission chains to persist despite forward tracing. The authors hypothesize that bidirectional tracing—using reverse tracing to identify infectors and then tracing their other infectees—can uncover undetected branches of transmission, particularly when case ascertainment is low or asymptomatic infections are common. They also posit that combining manual tracing with digital exposure notification could amplify benefits by increasing speed, scale, and the ability to trace further back in time.

Public health guidance (EU and WHO) recommends identifying potential infectees starting 2 days before symptom onset, reflecting a forward-tracing focus. Prior modeling studies of COVID-19 tracing largely neglected bidirectional tracing, and digital exposure notification implementations have prioritized forward notifications near peak infectiousness. Evidence indicates substantial asymptomatic infection rates (estimates 18–79%, with several surveys around ~45%) and significant contributions from superspreading, suggesting that targeting infectors could be impactful. Digital systems (Bluetooth/ultrasound 'chirps') have been proposed for exposure notification, offering theoretical advantages in speed, scale, and privacy, but their role in bidirectional tracing had not been systematically evaluated.

The authors adapted and extended a stochastic branching-process model of SARS-CoV-2 transmission to compare forward-only versus bidirectional contact tracing strategies across manual, digital, and hybrid implementations. Each case generates secondary cases drawn from a negative binomial distribution; incubation and generation-time distributions are drawn from published literature. Outbreaks are initialized with 10 index cases to limit stochastic extinction, and an outbreak is considered controlled if it goes extinct before reaching 10,000 cumulative cases. Effective reproduction number (R_eff) is computed as the mean number of child cases per case. Symptomatic cases are identified based on symptoms alone or via contact tracing; baseline assumptions include that symptomatic cases require a positive test before initiating contact tracing (aligned with EU/US practice) and that 90% of cases comply with isolation (ceasing further transmission upon isolation). Baseline parameter scenario used for initial analyses assumes: 10% environmental (untraceable) transmission, 48% presymptomatic transmission, 45% asymptomatic cases with 50% relative infectiousness, R0=2.5, 50% symptomatic ascertainment, and 70% test sensitivity. The model evaluates manual tracing with different look-back windows (2 vs 6 days pre-symptom onset), digital exposure notification (smartphone-based) with data-retention typically 14 days, and hybrid manual-digital systems. Delays for manual and digital tracing are short (illustrative values ~0.5–0.6 days). Digital performance depends on smartphone coverage and proportion of diagnosed users who upload their exposure data; scenarios include low uptake (~53% of cases with smartphones, ≈ two-thirds of smartphone users) and high uptake (~80% of cases with smartphones, ≈ nearly all smartphone users), with 90% of diagnosed users sharing data. The model systematically varies key parameters: tracing window width, trace success probability (excluding environmental transmission and network fragmentation), smartphone coverage and data sharing, symptomatic ascertainment rate, test sensitivity (with and without requiring a positive test before tracing), and baseline R0 (1.0–4.0), under median/optimistic/pessimistic disease parameter sets.

• Forward-only manual tracing with a 2-day look-back can reduce R_eff by up to 0.34 versus no tracing baseline; extending forward-tracing beyond 2 days yields negligible additional benefit.
• Switching from forward-only to bidirectional tracing with a 2-day window provides an additional reduction in R_eff of up to 0.2, roughly doubling the benefit over no tracing relative to forward-only.
• Extending the manual bidirectional tracing window to 6 days pre-symptom onset dramatically improves performance: R_eff reductions are up to 0.42 lower than with the 2-day window (~85% improvement over 2-day bidirectional and ~275% over forward-only).
• Digital exposure notification alone can achieve large R_eff reductions (R_diff approaching 1.0) when uptake is near-universal and data sharing is high, but performance is highly fragile to decreases in smartphone coverage and data sharing. With ~80% adult smartphone coverage and 90% sharing, digital-only performance is only slightly below manual bidirectional (2-day) and substantially exceeds manual forward-only with a 6-day window; however, digital alone is unlikely to control COVID-19 due to uptake sensitivity.
• Hybrid manual-digital tracing without bidirectional offers small gains over manual alone (R_eff reductions up to 0.06 with low uptake and 0.12 with high uptake). Incorporating bidirectional tracing increases gains: with a 6-day manual window, additional R_eff drops up to 0.14 (low uptake) and 0.26 (high uptake). With a 2-day manual window, adding digital yields larger gains: up to 0.21 (low uptake) and 0.42 (high uptake) beyond manual alone. Manual bidirectional with 2-day window reduces R_eff by up to 0.18; thus, high-uptake hybrid bidirectional could roughly double the efficacy of current tracing.
• Robustness: Bidirectional tracing remains considerably more robust than forward-only to low symptomatic ascertainment and to reduced test sensitivity, particularly when a positive test is required prior to tracing. Forward-only performance degrades markedly as test sensitivity declines under test-required policies; bidirectional stays relatively resilient.
• Across R0 from 1.0 to 4.0 and across median/optimistic/pessimistic parameter sets, bidirectional (especially hybrid) consistently outperforms forward-only in lowering R_eff and increasing control probability. Hybrid bidirectional shows the greatest advantage for approximately 1.25 < R0 < 3.25. High-uptake hybrid bidirectional can reliably control outbreaks under optimistic assumptions at R0 ≲ 1.75, and under pessimistic assumptions at R0 ≲ 1.25. At higher R0 (≈≥3.25), control becomes unattainable with any strategy.
• Summary across scenarios at R0=2.5: 2-day manual bidirectional achieves ~1.7–2× the R_eff reduction of forward-only; extending to 6 days improves a further ~75–95% relative to 2-day bidirectional; supplementing 2-day bidirectional with digital improves performance by ~40–60% (low uptake) and ~90–105% (high uptake).

The findings support the hypothesis that bidirectional tracing uncovers otherwise-missed transmission branches, substantially improving epidemic control, particularly when asymptomatic transmission and under-ascertainment are prevalent. By identifying infectors and their other infectees, bidirectional approaches preferentially find clusters and superspreaders, yielding greater reductions in R_eff than forward-only protocols that are limited to a 48-hour window. Extending the manual tracing window to 6 days pre-symptom onset or, alternatively, deploying a high-uptake digital exposure notification system capable of long look-back can approximately double the reductions achieved by current 48-hour forward-focused programs. Hybrid systems combining manual and digital tracing perform best when bidirectional logic is used, offering speed and reach while mitigating sensitivity to any single modality's weaknesses. Policy implications include prioritizing backward tracing and testing of early contacts (e.g., 3–6 days before symptom onset) and forward-tracing from identified infectors, as practiced in Japan, as well as tailoring digital notifications to prompt rapid testing for early contacts. The results also highlight the dependence of digital strategies on high participation, and the particular value of bidirectional tracing under conditions of low test sensitivity or reduced ascertainment—exactly when public health systems are most strained.

This study shows that implementing bidirectional contact tracing—by actively reverse-tracing to infectors and then forward-tracing from them—can markedly enhance COVID-19 control beyond forward-only approaches. Extending manual tracing windows from 2 to 6 days pre-symptom onset or deploying high-uptake digital exposure notification can roughly double reductions in R_eff relative to current 48-hour practices, with hybrid manual-digital bidirectional strategies performing best. Digital systems alone are unlikely to control transmission without near-universal uptake. The work suggests practical strategies: prioritize backward tracing and testing of earlier contacts, expand manual tracing windows where feasible, and design digital systems to support bidirectional tracing and rapid testing. Future research should quantify operational costs and resource requirements of expanding manual tracing windows, incorporate realistic contact network structures and recurring contacts, and further evaluate bidirectional strategies across diverse epidemiological models and settings.

• The branching-process model simplifies real-world complexities and does not account for detailed contact network structure, recurring contacts, or resource constraints in tracing operations.
• The cost and feasibility of extending manual tracing windows (e.g., increased number of contacts per case) are not quantified; the model may overestimate achievable coverage under practical constraints.
• Results depend on assumptions about epidemiological parameters (e.g., asymptomatic fraction, presymptomatic transmission, environmental transmission), test sensitivity, and isolation compliance.
• Digital exposure notification performance is highly sensitive to smartphone coverage and data-sharing behavior; network fragmentation and environmental transmission render some contacts untraceable.
• Many jurisdictions require a positive test before tracing; impacts of varying policies and delays were explored but may differ in practice.
• The model initializes outbreaks with 10 index cases and defines control thresholds that may not reflect all real-world scenarios.