A Time Delay Dynamical Model for Outbreak of 2019-nCoV and the Parameter Identification

1 Introduction

• The introduction describes early interventions as having questionable effectiveness and efficiency amid more than 11,000 confirmed cases in China.It also reports more than 3,000 confirmed cases in Wuhan and cases in multiple other countries.Source: §1 · p. 2
• Latent-period transmission distinguishes this outbreak from SARS and makes classical SIR, SEIR, and SEIJR models inappropriate for describing it, according to the authors.The authors use the available average latent and treatment periods to motivate a time-delay process.Source: §1 · p. 2
• The Chinese health commission reports accumulated diagnosed, cured, and dead cases, which the authors use as model variables because they are more numerically stable than instantaneous increments.The accumulated variables also align with the official data.Source: §1 · p. 2
• The model introduces isolation to represent government measures for suspected cases and contacted individuals.The introduction identifies isolation and travel restrictions among the implemented interventions.Source: §1 · p. 3

2 A Novel Dynamical System with Time Delay

• The model treats I(t) − G(t) − J(t) as the exposed population able to transmit, and β as a fixed spread rate.The authors note that β might depend on time as virus activity and the environment change.Source: §2 · p. 5
• Diagnosed increments depend on the history of infections, with average infection-to-diagnosis delay τ1 and a normalized distribution over prior infection times.The model allows a normal distribution for that delay, or a delta function when individuals share the same latent and treatment period.Source: §2 · p. 5
• Isolation and recovery are also modeled with delay processes, including newly isolated exposed people and cured people arising from infections at an earlier time.The recovery delay reflects the average time from infection to cure.Source: §2 · pp. 5–6
• The model is a time-delay dynamical system whose closed-system formulation is obtained by substituting the infection increment into the delay terms.It uses accumulated infected, diagnosed, isolated, and cured populations as variables.Source: §2 · pp. 3–6
• Parameter identification uses cumulative diagnosed J(t) and cured R(t), which are available in official data, rather than typically unobservable I(t) and G(t).R(t) does not affect the forward evolution of I(t), J(t), and G(t), but its observation may help infer parameters and initial conditions.Source: §2 · p. 6

3 The Reconstruction and Prediction Scheme

• The simulation assumes I(t0)=5 and G(t0)=J(t0)=R(t0)=0, with no isolation measures before T=t0+15.The initial condition represents five infections from unknown sources and no diagnosed, isolated, or recovered people on the initial day.Source: §3 · pp. 6–7
• The scheme treats the average latent and treatment periods as known according to official data.It also sets morbidity γ to 0.99 based on present data.Source: §3 · p. 7
• Reported diagnosed and cured counts from the National Health Commission of China are the observations used in the reconstruction scheme.The paper identifies them as JObs. and RObs.Source: §3 · p. 7
• The scheme fits θ from official diagnosed-count data, then estimates κ from cured-count data using the recovered θ, and numerically predicts J(t), R(t), I(t), and G(t).The paper names Levenberg–Marquardt and MCMC as suitable methods for the optimization problems.Source: §3 · p. 7

Basic simulation

• With isolation rate ℓ=0.6, cumulative infected people tend to about 14,000 after 40 days, while without isolation they tend to infinity.Figure 2 compares simulations with different isolation rates.Source: Figure 2 · p. 8, §4 · p. 8

Global and local outbreak

• The 18th-day predictions are tested after parameter inference from 10 days of national data and 17 days of local city/province data.Figure 3(a) shows the national test; panels 3(b)–3(d) show local cases.Source: §4 · pp. 8–9, Figure 3 · p. 9
• Figure 3 distinguishes observed cumulative diagnosed and cured data used for reconstruction from reconstructed values and observations used to evaluate predictions.The passage names these series JObs., RObs., JReconstr., RReconstr., JObs.2, and RObs.2.Source: §4 · p. 9, Figure 3 · p. 9

Conclusion

• The authors conclude that the time-delay system can predict outbreak tendency and that the situation would improve.This conclusion follows their review of Figure 3 and reconstructed parameters in Table 2.Source: §4 · p. 9
• The estimated parameters indicate that event evolution depends strongly on population size and local-government public policy.Source: §4 · p. 9

• With isolation rate 0.5, the outbreak may end in around 50 days, while medical-resource needs may peak around day 35.Source: §5 · p. 10
• The time-delay term is effective for modeling transmission during the latent period, but the current system assumes time-independent parameters.Future work may let isolation strength and spread rate vary over time to improve agreement between real data and estimated solutions.Source: §5 · p. 10
• J(t) − R(t) represents people in treatment and may also be used in the inverse problem; external source terms may represent process uncertainty.Source: §5 · p. 10