The authors reconstructed time-varying, age-specific contact matrices covering 28 EU/EEA countries for the period 2020–2022. They produced a harmonised set of matrices intended to fill incomplete empirical observations from the pandemic. The output spans the full study window and provides a consistent basis for comparative epidemic modelling across diverse national contexts. The study team stressed that these reconstructions are estimates derived from a modelled relationship between interventions and observed mixing, not raw survey counts.
Investigators linked changes in contact patterns to non-pharmaceutical interventions recorded in the ECDC-JRC Response Measures Database. They used the recorded intervention timeline to explain observed reductions and recoveries in social contacts where CoMix data existed. The analysis estimated how different NPI implementations correlated with temporal shifts in contacts. The study team reported that this relationship allowed prediction of contact levels during intervals without direct CoMix observations, and it underpinned reconstruction in countries lacking CoMix data altogether.
The authors developed a model-based framework to estimate time-varying age-specific mixing. Their approach combined three elements: empirical CoMix observations where available, an estimated mapping from NPIs to contact reductions derived from those observations, and a country-matching procedure to transfer patterns to countries without direct CoMix data. The country-matching process selected donor countries based on socio-demographic characteristics and similarity in NPI implementation. Data from donor countries were applied when a recipient country lacked pandemic contact surveys, producing reconstructed contact matrices for those periods.
The study team noted that pre-pandemic contact structure informed baseline matrices; the pre-pandemic matrices used in this analysis were obtained from previously published sources. The modelling steps and transfer rules were applied consistently across all 28 EU/EEA countries to produce a uniform output set. The authors made the analysis code and generated matrices publicly accessible through a project GitHub repository.
Primary empirical observations came from the CoMix social contact study. The team acknowledged that CoMix provided real-time changes in social mixing but left substantial spatiotemporal gaps—some countries and calendar periods had no CoMix observations. The study combined CoMix with the ECDC-JRC Response Measures Database for recorded NPIs and pre-pandemic contact matrices from the literature as baseline structure. All data and code supporting the work are publicly available; links to the data sources and the GitHub repository were provided in the manuscript. The authors declared that all data used are publicly accessible and reported that ethical approvals and participant consent considerations were addressed, given the public nature of the underlying data.
Reconstructed patterns captured substantial temporal variation: contact levels declined and recovered at different rates across the study period, and these dynamics varied by age group. Cross-country heterogeneity emerged as well; the cohort showed differing magnitudes and timing of contact reductions across European settings. The analysis indicated an evolving association between NPIs and social mixing that was not uniform across nations. The study team observed that using the NPI–mixing linkage enabled plausible interpolation of contact patterns during unobserved intervals, producing continuous matrices even through periods lacking direct survey data.
Authors emphasized that the reconstructions provide actionable inputs for epidemic modellers and policy analysts who require age-stratified contact structure across time and space. They argued that a consistent, EU/EEA-wide set of matrices helps address gaps that hamper comparative modelling and retrospective assessment of intervention impact.
The authors reported several important limitations that affect interpretation. CoMix left substantial spatiotemporal gaps; some countries lacked any pandemic contact data, and for many there were intermittent observations. The source did not report precise model parameter values or exhaustive transfer-rule specifics in the abstract; those details are available in the full manuscript and supplementary materials. The reconstruction depends on the assumption that donor countries selected by socio-demographic and NPI similarity provide appropriate proxies for recipients—a choice that can introduce bias if unmeasured factors differ between matched countries.
The study team made all code and generated matrices available to enable external scrutiny and reuse. Remaining unanswered questions documented by the authors include how unobserved behavioral drivers beyond recorded NPIs (for example, informal local measures or cultural practices) influenced mixing, and the extent to which transferred patterns fully capture country-specific responses. These gaps and the dependence on reconstructed estimates should guide cautious use of the matrices in modelling and policy analyses.