Using contact tracing records provided by the Cyprus Ministry of Health, the authors reconstructed directed infection trees for the first four waves of the COVID‑19 epidemic in Cyprus. In these trees each node represents an infected individual and directed links indicate the inferred direction of transmission between nodes. For every infection tree of size N they computed the hopcount distribution, defined as the distribution of distances (number of hops) from the root case to all other nodes in that tree.
The reconstruction and summary of these empirical trees provided the basis for a topology‑driven investigation into transmission-time dynamics and for deriving measures related to transmissibility across epidemic waves.
To interpret the empirical hopcount distributions, the study compared them with hopcount distributions obtained from infection trees generated by a non‑Markovian SI (susceptible→infected) process run on a complete graph. In the simulated process individual infection times were drawn from a Weibull distribution, which is characterized by a shape parameter, denoted α. The special case α = 1 corresponds to an exponential infection-time distribution and therefore to a Markovian (memoryless) infection process.
The procedure consisted of varying α in the SI simulations, generating infection trees, calculating hopcount distributions from those simulated trees, and then determining which α values produced hopcount distributions that best matched the empirically observed hopcounts from the Cyprus contact tracing trees.
Fitting simulated hopcount distributions to empirical hopcounts produced best‑fit values of the Weibull shape parameter α. Across the reconstructed infection trees for the four waves, the authors report that the fitted values were greater than 1 only. A Weibull infection‑time distribution with α > 1 yields a unimodal density with a peak at finite time, in contrast to α = 1 where the density is monotone decreasing (exponential).
Because α > 1 corresponds to infection times that are concentrated around a characteristic delay rather than memoryless waiting times, this result points toward transmission dynamics with temporal structure that departs from simple Markovian assumptions. The abstract notes that this finding is consistent with previous literature reporting peaked generation or infectiousness profiles.
A key message of the analysis is that non‑Markovianity of the spreading process can be detected using only the topology of infection trees — specifically through the hopcount distributions derived from contact tracing trees — without requiring direct observation of individual infection timing. The congruence between empirical hopcounts and simulations with α > 1 led the authors to conclude that the Cyprus spread process is most likely governed by non‑Markovian dynamics.
This approach leverages network topology and contact‑tracing linkage information to infer temporal properties of transmission that are otherwise challenging to estimate from case surveillance alone.
Beyond hopcounts, the authors analyzed the empirical distribution of the number of secondary infections produced by each node in the infection trees. They did this across different time windows to estimate an effective reproduction number over time.
According to the abstract, the simple average number of secondary infections per node often provided a lower bound for the reproduction number computed by the Cyprus Ministry of Health. The authors observed that excluding the last level of the infection trees — which is dominated by terminal nodes that do not generate further infections — produced estimates that more accurately reflected the epidemic dynamics. This suggests that naive averaging over entire trees can underestimate transmissibility because terminal cases, including those identified late or isolated, depress the mean number of onward transmissions.
No numerical reproduction‑number time series or confidence intervals are given in the abstract; the summary reports qualitative relations between the tree‑based averages and the Ministry’s estimates.
The study demonstrates a method to use contact tracing linkage data and infection‑tree topology to infer whether spread dynamics follow non‑Markovian laws and to derive tree‑based estimates of the effective reproduction number. The finding that the best‑fit Weibull shape parameter was α > 1 across examined waves supports the presence of a peaked transmission timing distribution in Cyprus COVID‑19 data.
Limitations apparent from the abstract: the detailed numerical results, uncertainty estimates, and full methodological parameters (for example, sample sizes per wave, statistical fit criteria, or sensitivity analyses) are not reported in the abstract. Therefore, readers should consult the full text for the precise simulation settings, fit metrics, and any additional caveats about data completeness, contact‑tracing coverage, or temporal windows used for reproduction‑number estimation.
Overall, the work highlights that topology of traced infection trees contains information about temporal transmission structure and that simple tree‑based diagnostics can inform whether Markovian assumptions are appropriate when modeling epidemic spread.