Fig 1.
Representation of the observed network recorded during the i-Bird study: (a) total network, and (b) patient-patient, (c) staff-staff and (d) patient-staff subgraphs on a single day. The date of 28th of July 2009 was chosen arbitrarily. The layout was calculated using the Kamada-Kawai algorithm, with no weights applied to edges. e) Distribution of individual degrees for the total network per person per day, across the entire study period. The dashed red line indicates the mean degree (13.59). CV: coefficient of variation (standard deviation/mean).
Table 1.
Summary of network characteristics for the observed i-Bird total network, patient-patient subgraph, staff-staff subgraph, and patient-staff subgraph.
Values were estimated for each day of the 28-days period and summarised here with the mean and standard deviation (sd). Transitivity is not calculated for the patient-staff subgraph as triangles of contacts cannot occur in this network.
Fig 2.
Description of contact heterogeneity and recurrence across the facility.
a) Repartition of contacts between grouped staff professions and patient wards. A link between one staff category and one patient ward indicates that, at any point during the investigation period, a staff member from that category had a contact with a patient from that ward. For ease of visualisation, occupational therapists, physiotherapists, and other re-education staff are grouped into “Reeducation”; administrative, animation/hairdresser, logistic, and hospital service agents are grouped into “Other”; and nurses, head nurses, and students/interns are grouped into “Nurses”. Porters, doctors and care assistants are not grouped. b) Distribution of number of wards with which each staff member has had at least one contact with during the study period. c) Distribution of probabilities of recurring contacts. Each observation is calculated over the entire studied period, and corresponds to the average probability for one staff or one patient to form a new contact with a previously-met individual (staff or patient) over the studied period rather than a new individual. Diamonds indicate the mean values.
Fig 3.
Comparison of network characteristics.
The reconstructed networks with observation bias exclude individuals from the network at times when they were known to not wear their sensors. The random networks did not take into account the ward-level structure of the contacts or the probability of recurring contacts. Boxplots for the observed network show the distribution of values calculated for each day. Boxplots for all reconstructed and random networks show the distribution of the median values calculated for each day across 100 networks. The distributions were compared using Wilcoxon tests with Bonferroni correction. ns: not significant; *: p < 0.05; **: p < 0.01; ***: p < 0.001.
Fig 4.
Comparison of network contact number and duration.
a) Distribution of number of unique contacts per hour, separated by type of day (weekday or weekend). Points correspond to the median, and the shaded areas correspond to the interquartile range. b) Distribution of contact durations. For ease of visualisation, outliers are not shown on the graph.
Fig 5.
Comparison of resulting incidence dynamics depending on the networks.
a) Epidemic dynamics for the two random networks. Lines indicate median values, and the shaded areas indicate the interquartile range. b) Epidemic dynamics for the observed and reconstructed networks. Lines indicate median values, and the shaded areas indicate the interquartile range. c) Characteristics of the resulting epidemics for the different evaluated networks. Peak date is shown for the median epidemic curve (solid lines in panels a and b). “Extinction prop.” indicates the proportion of simulations excluded from the analysis, with less than 10 individuals infected in total.