Evaluating the impact of incomplete contact tracing data on urban epidemic dynamics

Min-Kyung Chae; Sang Hoon Lee; Woo-Sik Son

Evaluating the impact of incomplete contact tracing data on urban epidemic dynamics

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Contact tracing collapses once more than about 4% of confirmed cases' movements are never traced in a large city, while smaller Busan holds until about 10%.

desk verdict Useful IO/CO decomposition, but the headline 4%/10% thresholds are not yet supported — needs a defined threshold criterion and sensitivity analysis. read the letter →

arxiv 2601.14632 v3 pith:MAMT7CTE submitted 2026-01-21 physics.soc-ph cond-mat.stat-mechq-bio.PE

Min-Kyung Chae , Woo-Sik Son , Sang Hoon Lee This is my paper · ORCID

classification physics.soc-phcond-mat.stat-mechq-bio.PE

keywords contacttracinginfectoromissionagent-basedmodeltransmissionnetworkdiameterepidemiccontainmentSeoulBusan

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper asks what happens when contact tracing is imperfect in different ways, and claims the answer depends sharply on which information is missing. Its simulations of a synthetic Seoul and Busan show that untraced confirmed cases — infector omission — behave like a switch: below a city-specific threshold containment holds, and above it the outbreak grows abruptly. Missed contacts, by contrast, only slow or delay control gradually. This asymmetry leads the authors to conclude that reconstructing a confirmed case's full movement history matters more than notifying every identified contact. The finding matters because it gives public-health planners a concrete target: keep infector omission low, and set that target by local population and mobility structure, not by a national average.

What carries the argument

The argument is carried by a stochastic agent-based model of epidemic spread on a multilayer contact network—households, school classrooms, workplaces, friendships, and local communities—built from census-based synthetic populations. Its central object is the directed transmission network: a node for each infected agent and an edge from infector to infectee. Two observables do the work: cumulative infections and peak timing for severity, and the network diameter for the depth of transmission chains. Across omission scenarios, the paper tracks how these observables respond to the infector-omission rate versus the contact-omission rate, and reads a small diameter as a signature that tracing is

What would settle it

Use contact-tracing records from a real jurisdiction that include whether each confirmed case's full movement history was obtained. If cumulative infections and transmission-chain depth do not jump once untraced cases cross the 4–10% range, the claimed threshold phenomenon is refuted; if the jump appears, it supports the model.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that infector omission and contact omission are not symmetric failures. Infector omission, where a confirmed case is isolated but their trajectory and all downstream contacts are lost, produces a sharp transition: in the Seoul model cumulative infections jump once the omission rate exceeds about 4%, and the directed transmission network grows a longer diameter, meaning chains of infection run deeper. In the Busan model, with roughly a third of Seoul's population and an older age structure, the same transition appears at about 10%. Contact omission—losing individual notifications rather than whole trajectories—causes only gradual increases in pea

Load-bearing premise

The quantitative thresholds rest on model parameters—asymptomatic share 0.2, self-testing probability 0.5, contact frequencies, homophily, and COVID-19-borrowed infectiousness—that the paper acknowledges it could not validate, so a different set of values would move the 4% and 10% breakpoints.

Editorial extensions

If this is right

If the thresholds are right, a tracing system in a dense metropolis must keep untraced confirmed cases below roughly 4% to avoid losing containment, even if contact notification is near complete.
When resources are tight, effort should go to reconstructing the movement trajectories of confirmed cases rather than to chasing every last contact, because contact omission degrades control only gradually.
The city-specific thresholds imply that a one-size-fits-all tracing standard is unsafe; smaller or older cities can tolerate a higher omission rate.
A growing transmission-network diameter could serve as an early warning of hidden chains, visible before cumulative case counts surge.
The same model suggests a quantifiable design target for manual and digital tracing systems: they should be engineered to keep infector omission below the local threshold.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The exact 4% and 10% numbers are model outputs, not measured facts; because the paper does not validate its parameters against real tracing logs, they are best read as qualitative evidence of a threshold, not as precise operational limits.
The abrupt IO threshold hints at a more general principle: any intervention that removes the root of a transmission chain (such as early isolation of index cases) will show a critical coverage level, whereas interventions that only prune individual edges will degrade smoothly.
The diameter effect suggests a testable empirical signature: real outbreak datasets with good tracing coverage should show shorter generation-interval chains, and chain length should jump when tracing coverage falls below the local critical level.
For digital contact-tracing apps, partial adoption creates an infector-like omission at the system level, so the model implies an adoption threshold below which app-based tracing silently fails.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

Quick take: this is a competent ABM study with a genuinely useful conceptual split — infector omission (IO) vs contact omission (CO) — and a nice diagnostic in the transmission-network diameter. The qualitative finding that IO is much more damaging than CO, and that a smaller city (Busan) tolerates more loss, is believable and coherent. They also ship code and data, which is more than many papers do.

The soft spot is exactly where the reader's stress-test lands. The 4% and 10% thresholds are the headline outputs, but they are extracted from a coarse grid — Figure 4 shows IO rates 0%, 2%, 4%, 10% — and no formal threshold-extraction rule is given. The text says 'increases sharply once the IO rate exceeds 2%' and later identifies a 'threshold of 4%'; those are not the same thing. Without a breakpoint regression or a pre-specified operational criterion, the values may be visual impressions. That matters because the parameters driving them are admittedly arbitrary: PA=0.2, Pt=0.5, friend contact 1/7, homophily h=0.9, fixed 50% local-community omission, COVID-borrowed shedding. The authors themselves state in the Discussion that they were 'unable to quantitatively validate the realism or accuracy of the CT processes reproduced by the model.' So the precise 4%/10% numbers should not be read as public-health constants.

I don't think this is fatal. The qualitative ordering is robust enough across their runs (100 repetitions, confidence intervals), and the distinction between missing an infector (an entire subtree) and missing a contact (a single edge) is structurally sound. The paper would be strengthened by a defined threshold criterion, a one-parameter-at-a-time sensitivity sweep, and a clear statement of which results depend on the unvalidated parameters. The self-citations to their prior ABM are fine given the shared framework.

Who's this for? Modelers working on contact-tracing policy and public-health planners who want an intuition for why trajectory tracking matters more than perfect contact notification. A serious referee should see it, with major revision. I would not accept the thresholds as stated.

Referee Report

3 major / 5 minor

Summary. The paper presents a stochastic agent-based model of contact tracing (CT) in two South Korean metropolitan areas, Seoul and Busan, built from synthetic populations and multilayer social networks. It simulates two types of information loss—infector-omission (IO), where confirmed cases' movement trajectories are not traced, and contact-omission (CO), further split into selective (SCO) and uniform (UCO) variants—and reports that IO causes a sharp, threshold-like degradation of CT effectiveness (approximately 4% IO rate in Seoul and 10% in Busan), while CO produces gradual degradation. The paper also analyzes the directed transmission network's diameter and out-degree distribution, finding that information loss increases the diameter and that the IO effect is structurally more disruptive than CO. The authors conclude that accurate trajectory reconstruction is more decisive than complete contact notification and that CT strategies should be tailored to regional population structure.

Significance. If the quantitative thresholds were robust, the paper would provide concrete, actionable targets for CT completeness and a clear argument for prioritizing infector-trajectory reconstruction over exhaustive contact notification. The strengths include a detailed synthetic-population construction from census and survey data, stochastic simulation with 100 runs and reported confidence intervals, and public code/data availability. However, the headline thresholds (4% and 10%) rest on an unspecified threshold-extraction criterion and on a parameter set that the authors themselves acknowledge cannot be quantitatively validated. Because these thresholds are emergent simulation outputs, the paper's central quantitative claims are not yet established at the level of confidence the presentation implies.

major comments (3)

[Results, 'Impact of infector-omission' and Fig. 4] The threshold values are not defined by any formal procedure. The text states that the cumulative number of infections 'increases sharply once the IO rate exceeds 2%' and later refers to 'the threshold of 4%,' while Fig. 4 shows only four IO rates (0%, 2%, 4%, 10%). No pre-specified criterion (e.g., breakpoint regression, crossing of an outbreak-size threshold, or sustained-transmission probability) is given. Without a reproducible extraction rule, the 4% Seoul and 10% Busan thresholds are visual interpretations of a coarse grid and cannot be independently verified or compared across scenarios.
[Table 1 and Discussion (Limitations)] The thresholds depend on parameters that are largely arbitrary or borrowed from COVID-19 without sensitivity analysis: PA=0.2, Pt=0.5, friend/local contact probability 1/7, homophily h=0.9, the fixed 50% local-community omission rate, and the infectiousness distributions. The Discussion explicitly states that the authors were 'unable to quantitatively validate the realism or accuracy of the CT processes reproduced by the model.' Since the IO threshold is an emergent property of this parameter set, plausible changes in any of these values could shift the 4%/10% numbers. A sensitivity analysis, even over a subset of the most influential parameters, is essential to support the paper's quantitative claims.
[Fig. 6 and city comparison] The Busan threshold of 10% is inferred from panels showing only peak time and peak height, with the same unspecified threshold criterion. Moreover, the paper attributes the threshold difference to the 'three-fold difference in population size,' but Seoul and Busan also differ in age structure, commuting patterns, and contact density. No controlled experiment varying population size while holding other factors fixed is performed, so the causal attribution to population size is not supported. The city comparison should either be framed as a case study with demographic covariates or include such a controlled analysis.

minor comments (5)

[Conclusion] Typo: 'missingnsuch' should be 'missing such.'
[Table 1] The relative infectiousness distribution is cited to reference [3] (Ferguson et al.), but that reference does not appear to provide such a distribution. Please verify the citation or supply the correct source.
[Figures 4 and 7] The figures use a 50% confidence interval. Please justify this choice or show standard 95% intervals for the central quantities, as 50% CIs convey less uncertainty information.
[Methods, 'Contact Tracing'] The sentence 'The results in potentially infected agents moving freely within society' is grammatically incomplete; likely 'result in.'
[Discussion] The claim that the model 'can be easily adapted by updating these epidemiological parameters' would be strengthened by a brief note on identifiability or a calibration strategy, especially given the acknowledged lack of validation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the thresholds are emergent simulation outputs, not fitted or derived by construction.

full rationale

The paper's central claims—the city-specific IO thresholds (about 4% in Seoul, 10% in Busan), the gradual effect of CO, and the increase in transmission-network diameter—are emergent outputs of a stochastic agent-based simulation. They are not parameters fitted to the target results, and no fitted input is renamed as a prediction. The IO and CO scenarios are defined differently (node vs. edge omission), but the qualitative comparison is evaluated by simulation rather than reduced to an identity; the paper itself explains the disparity in terms of transmission-chain visibility, which is a mechanism, not a tautology. Self-citations (Chae et al. 19 for synthetic-population generation, Son et al. 32 for contact-duration survey data) supply data and methods from prior published work and are paired with independent sources (census data, Ye et al. 20); they do not by themselves entail the threshold numbers. The Discussion explicitly acknowledges the inability to validate the realism of the CT processes ('we were unable to quantitatively validate the realism or accuracy of the CT processes reproduced by the model'), which is a validity/calibration limitation, not a circular step. The threshold extraction lacks a formal breakpoint rule and rests on a coarse grid (0%, 2%, 4%, 10%), a robustness concern rather than a by-construction circularity. Therefore no circular steps are identified.

Assumptions & free parameters 11 free parameters · 6 assumptions · 0 invented entities

The quantitative thresholds are not calibrated to real outbreak data; they emerge from a model whose dynamics are governed by a dozen hand-chosen or COVID-borrowed parameters, none fitted to this paper's target outcome. This does not make the paper circular, but it means the exact numerical thresholds are conditional on this ledger of assumptions.

free parameters (11)

asymptomatic probability PA = 0.2
Methods/Table 1: 'asymptomatic ratio were assumed to be arbitrarily determined'.
self-testing probability Pt = 0.5
Methods/Contact Tracing: '50% of symptomatic agents are assumed to voluntarily seek testing'.
friend contact probability = 1/7 per day
Table 1 and Methods: 'Each day, every agent independently decides whether to meet friends with a probability of 1/7'.
local contact probability = 1/7 per day
Table 1 and Methods: local-community layer follows the same 1/7 mechanism.
homophily parameter h = 0.9
Social Networks: homophilic BA network with homophily parameter h=0.9.
local community omission rate = 50%
All scenarios: 'the omission rate for the local community network is fixed at 50%'.
initial exposed agents E0 = 20 (40 for robustness)
Results: 20 initial E agents, with E0=40 checked in SI.
relative infectiousness distribution ξ_j = Gamma(1, 0.5)
Table 1 assigns this distribution; the cited source [3] is questionable.
viral shedding distribution = Gamma(3.067, 2.109)
Adopted from COVID-19 ref [29] but treated as applicable to a generic emerging pathogen.
exposed period distribution = Gamma(1.926, 1.775)
Adopted from COVID-19 ref [29]; arbitrary for a novel disease.
infectious period = 8 days
From ref [31], but arbitrary for a novel disease.

assumptions (6)

domain assumption The synthetic population projected from a 2% census via iterative proportional updating represents the full populations of Seoul and Busan.
Methods/Agents: the 2% census records are projected to 9,529,266 (Seoul) and 3,313,542 (Busan) agents.
domain assumption The multilayer contact network (household, workplace, classroom, friendship, local community) approximates real social contact structure relevant to disease transmission.
Methods/Social Networks: the model builds these five layers as the contact substrate for transmission.
domain assumption Per-contact transmission probability is P=1-exp(-λ) with λ = contact duration × infectiousness, with no additional network-specific scaling.
Methods/Epidemiological Parameters, Eq. (1): 'no additional network-specific scaling factor is introduced in λij'.
domain assumption School classrooms are isolated units; students from different classrooms within the same school do not interact.
Methods/Social Networks: 'The model represents classrooms as isolated units, such that students from different classrooms within the same school do not interact.'
domain assumption The BA friendship network with homophily h=0.9 within decade age cohorts captures Korean friendship contact structure.
Methods/Social Networks: friendship network is generated using a homophilic BA model within age cohorts.
domain assumption The implemented CT protocol (voluntary testing, quarantine, isolation, recursive tracing) approximates real manual contact tracing.
Methods/Contact Tracing: the described decision process is assumed to capture the recursive nature of CT; the Discussion admits it was not validated against real CT logs.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Evaluating the impact of incomplete contact tracing data on urban epidemic dynamics." pith.science (2026). https://pith.science/paper/MAMT7CTE

@misc{pith2026260114632,
  author       = {Pith},
  title        = {Pith review of: Evaluating the impact of incomplete contact tracing data on urban epidemic dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MAMT7CTE}},
  note         = {Machine review of arXiv:2601.14632}
}

read the original abstract

Contact tracing (CT) is a frontline measure against emerging epidemics, yet in practice it is never complete. The quantitative impact of missing information -- such as untraced cases or unnotified contacts -- on the effectiveness of CT remains insufficiently understood. Using a stochastic agent-based model with sociodemographics from metropolitan areas in South Korea, we simulate how different forms of information loss affect epidemic spreading dynamics. We construct information-loss scenarios based on two types: infector-omission (IO), the omission of infected individuals from the tracing process, and contact-omission (CO), the omission of specific contact events even when the infected individuals themselves are identified. The sensitivity of epidemic dynamics to increasing omission rates differs markedly between the two types: IO produces substantially stronger and more abrupt changes in transmission structure and epidemic outcomes, whereas CO produces more gradual effects. Notably, CT effectiveness breaks down beyond a city-specific threshold -- an IO rate of approximately 4% in Seoul but about 10% in less populous Busan -- underscoring that CT strategies must be tailored to regional population and mobility structure. Both IO and CO scenarios also lead to an increase in the transmission network diameter as information loss grows, indicating that a small network diameter reflects effective contact tracing that limits the depth of transmission chains. Collectively, our results offer threshold estimates and practical guidance for designing robust CT systems in the real world.