High-resolution contact data and infectious disease absenteeism in kindergarten populations

Abstract

Background

Close social interactions are fundamental drivers of infectious disease transmission in early childhood education settings. However, high-resolution empirical contact data from these environments remain scarce, limiting our understanding of transmission dynamics and prevention strategies.

Methods

We deployed name-tag devices using Ultra-Wide Band (UWB) technology in a public kindergarten in Shenzhen, China. We recorded close-proximity contact events (≤ 2 m) during a typical school day. Contact events were classified into four risk categories based on duration and proximity. We then conducted a joint analysis of temporally segmented, risk-stratified networks alongside six months of infectious-disease-related absenteeism records.

Results

The study achieved an 88.68% participation rate (282 of 318 children) and full participation of 30 teachers, resulting in 389,617 contact events. On average, each child interacted with 57.91 distinct individuals daily. Outdoor activities involved fewer and lower-risk contacts. In contrast, nap time—a sedentary activity—accounted for the highest proportion of high-risk contact time (14.38%). While children with and without absenteeism records had similar total contact numbers and durations over the full day, those in the absentee group engaged in more high-risk contacts. Specifically, they had longer and more frequent interactions across the medium-low, medium-high, and high risk layers. This disparity diminished when low risk contacts were included. At the class level, absenteeism rates were correlated with the density of medium-high (r = 0.833, p = 0.005) and high risk (r = 0.733, p = 0.025) contact layers.

Conclusions

Our findings reveal substantial structural and contextual heterogeneity within kindergarten contact networks and provide quantitative evidence associating contact patterns with health outcomes. High-resolution proximity data offer valuable insights into transmission risk and inform the design of targeted prevention and risk management strategies in early childhood education settings.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12879-026-13832-3.

Introduction

Children’s immune systems are not fully developed, and their complex social contact patterns often lead to outbreaks within schools, homes, and the community. Between 2013 and 2018, 58.33% of gastrointestinal cluster outbreaks in Shenzhen occurred in kindergartens [1]; similarly, from 2014 to 2023, Guangdong Province reported that 21.74% of public health events took place in kindergarten settings [2]. Social contacts are widely recognized as crucial for the transmission of various infectious diseases such as monkeypox, COVID-19, and influenza [3–5]. In highly interactive groups such as kindergarten settings, contact patterns are characterized by high frequency, prolonged duration, close proximity, and structured routines with fixed schedules, all of which significantly increase the likelihood and speed of pathogen transmission [6]. The fine-grained structure of such contact networks remains inadequately characterized, particularly in real-world early childhood education settings. Therefore, a precise understanding of contact behaviors is paramount for quantifying disease transmission risk and identifying key nodes for intervention.

Acquiring accurate social contact data from young children is notoriously challenging. Traditional methods, such as self-reported questionnaires or diaries, are often compromised by the limited linguistic and cognitive development of preschoolers, as well as recall bias. Although Bluetooth, RFID sensor-based contact studies have been conducted in primary and secondary schools in Western countries [7, 8], these technologies have not yet been applied in kindergarten setting of mainland China. This has resulted in a critical knowledge gap regarding the fine-grained contact patterns of this high-risk group.

To address this gap, we leveraged Ultra-Wide Band (UWB) technology, which offers centimeter-level positioning accuracy through high-frequency, low-power pulses [9]. Its ability to precisely measure inter-personal proximity by calculating the signal’s time of arrival makes it an ideal tool for this research. To our knowledge, this is the first study to use high-resolution wearable sensors to map the complete, day-long social contact network of children and staff within a kindergarten in mainland China.

The aim of this research is to provide a multi-faceted analysis of social contact risks in this setting, identify potential transmission hotspots, and offer valuable, evidence-based insights for public health authorities to optimize disease prevention and control strategies.

Methods

Data collection

In June 2024, this study was conducted in a public kindergarten affiliated with a residential community in Fuyong Subdistrict, Bao’an District, Shenzhen. The building is used exclusively for the kindergarten and contains no other grade levels. We aimed to collect contact data from students and teachers who were present and voluntarily agreed to participate. The kindergarten occupies a total area of 2,700 square meters, with a building area of 2,420 square meters. It consists of a three-story teaching building where each floor is dedicated to a single grade level, with each grade comprising three classes. The three grades include Junior (ages 3–4), Middle (ages 4–5), and Senior (ages 5–6). A total of 318 students and 30 teachers were included in the study (Supplementary Table 1).

We used the Tb 1000 g UWB tag (TSINGOAL, China) to record close-range interactions. The device offers a ranging accuracy of up to 10 cm, a communication range of > 5 m, and a configurable detection threshold of 1–3 m. About 12 h of continuous operation are supported by the battery at the default sampling rate of 1 Hz. We set the sampling frequency to 0.25 Hz to balance battery life and data granularity.

To maintain anonymity, each participant was assigned a unique identification code that corresponded one-to-one with a specific UWB badge (Supplementary Fig. 1). Upon arrival at 7:30 AM, teachers assisted students in attaching the badges to the back of their collars to ensure continuous use throughout the day. Teachers placed their own badges in clothing pockets. All badges were collected by volunteers at dismissal at 5:00 PM.

A contact event was defined as the continuous time interval over which two individuals remained within a distance of ≤ 2 m. This threshold was selected based on established droplet transmission distances and is aligned with previous wearable-sensor studies [10, 11].

An event began at the first sampling point where the distance was ≤ 2 m and ended at the final consecutive point meeting this criterion. If a sampling point showed a distance > 2 m, the contact was considered terminated. No minimum duration threshold or grace period was applied; thus, any detected encounter, regardless of its brevity, was recorded as a valid event. For each contact event, the device recorded the unique identifiers of the two individuals, the start and end timestamps, and the minimum distance observed during the event. A sample of the raw data is provided in Supplementary Table 2, where each row represents one contact event as defined above.

According to the daily schedule, activities were segmented into seven distinct modules. Based on the predefined functional areas of the school corresponding to each scheduled activity, breakfast, lunch, nap, and classes were classified as indoor activities, whereas morning exercises, outdoor activities, and walks were classified as outdoor activities (Supplementary Table 3, Supplementary Fig. 2). On the day of data collection, 30 students were absent due to leave. An additional 6 students were excluded from the study due to incomplete data resulting from late arrivals and early departures (Supplementary Fig. 3). Written informed consent was obtained from the legal guardians of all children, and participants’ right to withdraw from the study at any time was fully safeguarded.

Student absenteeism data due to illness for the spring semester (February to July) of 2024 was collected from the kindergarten’s medical office. Only absence records caused by fever, cough, bronchitis, hand-foot-and-mouth disease (HFMD), and herpangina were included. These conditions are hereafter collectively referred to as “absence due to infectious diseases”. See Supplementary Text 1 for detailed classification rules.

Contact risk

The risk classification criteria for each contact event are defined in Table 1 below [12]: High risk contact: Duration ≥ 15 min and distance ≤ 0.1 m. Medium-high risk contact: Duration ≥ 10 min and distance ≤ 1.0 m. Medium-low risk contact: Duration ≥ 5 min and distance ≤ 1.5 m.Low risk contact: All contacts not meeting the above conditions, those with duration < 5 min or distance > 1.5 m (or both).

Table 1.

Contact risk classification criteria

Network

In this study, teachers and students were treated as nodes (n), and their contact relationships as edges (m), forming an undirected network. The aggregated network graph was visualized using Gephi 0.10.1, and the following network metrics were calculated: total number of nodes, total number of edges, average degree, and network density.

First, all contact events were stratified into four risk categories (low, medium‑low, medium-high, and high) based on the criteria in Table 1. We then constructed separate undirected networks for each class within each risk stratum, using only the contact events relevant to that specific risk level. The network construction rules were as follows:

• Nodes (n): Only students who engaged in at least one contact event at the given risk level during the observation period were included; those without such interactions were excluded from the specific risk stratum network.

• Edges (m): These represent unique student-student dyads that had at least one contact event at the designated risk level within the same class.

• For each class-specific risk stratum network, the following four key metrics were computed:

• Node Count (n): The total number of participating students in that risk stratum.

• Edge Count (m): The total number of unique student dyads present in the network.

• Average Degree: The average number of connections per node, calculated as , reflecting the average interaction intensity per node in the network.

• Network Density: The ratio of the actual number of edges to the maximum possible number of edges, calculated as. This value ranges from 0 (no connections) to 1 (a complete graph where every pair of participating students is connected), used to assess the tightness of the network.

Sensitivity analysis

To evaluate the robustness of our findings and mitigate the potential influence of brief encounters, we performed two distinct types of sensitivity analyses:

Sensitivity Analysis 1 (Exclusion of brief contacts): We repeated the primary analysis by excluding every single contact event lasting less than 10 min to determine if very short interactions biased the observed associations between contact patterns and health outcomes.

Sensitivity Analysis 2 (Reclassification of risk strata): Instead of excluding data, we validated the rationality of our risk-stratification framework by reclassifying all recorded contacts using reduced duration thresholds. Specifically, the thresholds were lowered to ≥ 10 min for high-risk, ≥ 5 min for medium-high risk, and ≥ 1 min for medium-low risk, Supplementary Table 6.

Statistical analysis

Contact pairs were categorized by population attributes into four types: within-grade contacts (Junior, Middle, and Senior), cross-grade contacts (mix), student-teacher (S-T) contacts, and teacher-teacher (T-T) contacts. We compared the cumulative contact duration across different risk levels among these groups. Furthermore, students were divided into two subgroups based on their history of infectious disease-related absenteeism: the absentee group and the non-absentee group. For each subgroup, we compared contact indicators—including the number of contact events, total contact duration, and the number of unique individuals contacted—across the overall network and within the low, medium-low, medium-high, and high risk contact layers.

For data conforming to a normal distribution, results are presented as mean ± standard deviation (SD). Independent-samples t-tests were used for between-group comparisons. For non-normally distributed data, results are presented as median (interquartile range, IQR) [M (P25, P75)]. Between-group comparisons were performed using the Mann-Whitney U test, with the test statistic (Z) reported. The significance level was set at α = 0.05 (two-tailed).

Use of AI-assisted language editing

During the preparation of this manuscript, the authors utilized Grammarly (2026), DeepSeek(2026), and ChatGPT-4 (OpenAI) to optimize the readability, language quality, and professional tone of the text. These AI-assisted tools were employed solely to ensure that the manuscript was entirely free of errors in grammar, spelling, punctuation, and academic phrasing. They were not used to generate, analyze, or interpret any empirical research data, nor did they influence the study’s scientific conclusions. The authors explicitly confirm that all automated outputs were thoroughly reviewed, verified, and edited by the researchers, who take full public responsibility for the integrity and accuracy of the final published content.

Results

Basic information

A total of 36 students were excluded from the analysis due to incomplete contact data resulting from absence or early departure on the study day (Supplementary Fig. 3, Supplementary Table 4). The final participation rate for contact data collection was 88.68% (282/318). Participating students ranged in age from 3.8 to 6.8 years, while teacher participation reached 100.00% (n = 30).

During the semester, 273 absenteeism records were related to infectious illnesses. Of these, 141 cases involved febrile symptoms, 62 were identified as hand, foot, and mouth disease or herpangina, and 70 were associated with cough or upper respiratory tract infections. These records involved 180 students, accounting for 56.60% of the total student population, as detailed in Table 2.

Table 2.

Basic information of kindergarten students

Category	Overall		In the Study
	Number of Students(N)	Illness-related Absences n(n/N%)1	Participants (N)	Absentee Group (n/N%)2	Non-Absentee Group (n/N%)2
Overall	318	180(56.60%)	282	165(58.51%)	117(41.49%)
Junior Group	89	58(65.17%)	77	48(62.34%)	29(37.66%)
Middle Group	103	58(56.31%)	85	50(58.82%)	35(41.18%)
Senior Group	126	64(50.79%)	120	67(55.83%)	53(44.17%)

Note:¹ Percentage based on total enrollment. ² Percentage based on the number of participants present on the day of data collection

Contact data analysis

This study recorded a total of 389,617 contacts between 282 students and 30 teachers. The duration of a single contact event followed a power-law distribution (Fig. 1). Although contact duration ranged from 1 s to 146 min, the median duration was 43 s, with 59.38% of contacts lasting ≤ 1 min and 94.21% lasting ≤ 10 min. The study observed 18,069 distinct dyads. The median number of contact events per pair was 24 (IQR: 3, 35), while the median total contact duration per pair was 38.70 min (IQR: 1.73, 88.35). On average, each individual contacted 57.91 ± 14.28 unique individuals, engaged in 1,248.78 ± 212.08 contact events, and accumulated 3,227.54 ± 870.68 min of total contact time(mean ± SD).

Fig. 1.

Contact duration distribution

Risk Level

Location

Figure 2 illustrates the temporal distribution of contact risk levels across these settings. Each point represents a single contact event, with different colors indicating different risk categories. The results demonstrate that, although the frequency of high risk contacts was relatively low in three scenarios, their cumulative time was non-negligible. Among these, indoor contacts occurred much more frequently than outdoor contacts. For outdoor contacts, cumulative time decreased as the risk level increased, with low risk interactions accounting for the majority (57.51%). In contrast, indoor contacts exhibited a larger proportion of cumulative time at medium-high and high risk levels (49.21%). Detailed results are provided in Supplementary Table 5.

Fig. 2.

The contact risk distribution across different locations. Note: This percentage refers to the proportion of the cumulative time of all contact events at different risk levels to the total duration of all contact events within the same location

Activity

According to the school schedule, the day was divided into distinct activity periods (Supplementary Table 3). Data analysis revealed significant variation in the distribution of contact risk across activities. Nap time recorded the longest cumulative high risk duration (21,936.77 min), representing the highest proportion (14.38%) of such contact; together, medium-high and high risk levels accounted for nearly 70% of the overall exposure risk during this period. In contrast, morning exercise demonstrated the lowest risk profile, with low risk contacts accounting for 76.13% of the total. While the proportions of high-risk exposure were similar between indoor classes (6.55%) and lunch (6.08%), the cumulative duration at the medium-high risk level was substantially greater for indoor classes (14,686.50 min) than for lunch (9,946.20 min). For both walking and outdoor activities, the proportion of exposure decreased as the risk level increased. Specifically, walking accounted for only 2.25% of high-risk exposure. Meanwhile, outdoor activities yielded a high absolute cumulative high-risk duration (20,142.18 min) due to longer overall exposure time, yet their high-risk proportion (6.69%) remained comparable to that of indoor classes (6.55%)(Fig. 3).

Fig. 3.

The contact risk distribution across different activities. Note: The Y-axis represents the cumulative contact duration. The percentages indicate the proportion of the cumulative contact time for events at each risk level to the total contact time within the same Activity. Color definitions in the figure follow the same scheme as in Fig. 2

Group

Same-grade student interactions exhibit significant clustering at medium-high and high risks levels, which collectively account for 43.30-46.44% of their total contact. The risk structures across grade levels demonstrate high similarity. In contrast, cross-grade (Mix) and teacher-student (S-T) interactions are characterized by a structural low-risk profile; both maintain a low-risk prevalence of > 57%, with proportions decreasing as risk levels escalate. While the percentage of high risk interactions in teacher-teacher (T-T) contacts is somewhat greater (9.19%) than in the Junior group, their overall contact time is significantly lower. Out of all the student groupings, the junior group has the highest proportion (8.87%). Consequently, the actual high risk exposure time for T-T interactions is only 454.17 min, which represents less than 2.3% of the total high-risk duration recorded for the Junior group (Fig. 4).

Fig. 4.

The contact risk distribution across different groups

Among the 282 students who participated in the contact data collection, 165 had absenteeism records related to infectious disease, while 117 had no such records. Table 3 presents a comparative analysis of contact characteristics across different exposure risk levels. At medium-low to high risk levels, students with absentee group demonstrated significantly higher total contact duration, number of contact events, and number of unique individuals contacted compared to those in the non-absentee group (p < 0.05). Conversely, within the low-risk level, students without absenteeism exhibited slightly higher total contact duration and number of contact events, although the difference in the number of unique individuals contacted was not statistically significant. When evaluating full-day interactions across all risk levels, no significant differences were observed between the two groups across any contact indicators.

Table 3.

Comparison of exposure levels between non-absentee and absentee groups across different risk categories

		Non-Absentee Group (n = 117)	Absentee Group (n = 165)	z	p
		P50(P25, P75)	P50(P25, P75)
Summary	Total Contact duration	3120.55(2774.07,3794.73)	3218.18(2849.88,3764.37)	1.295	0.195
	Contact events	1095.00(1018.00,1316.00)	1083.00(980.00,1234.00)	-1.809	0.070
	Unique contacts	53.00(43.00,61.00)	52.00(42.00,64.00)	0.371	0.710
Low risk	Total Contact duration	1146.62(995.95,1319.40)	1008.43(908.63,1170.97)	-5.139	< 0.001
	Contact events	970.00(887.00,1167.00)	929.00(838.00,1065.00)	-2.652	0.008
	Unique contacts	53.00(43.00,61.00)	52.00(42.00,63.00)	0.300	0.764
Medium- low risk	Total Contact duration	679.05(565.78,850.40)	776.22(628.47,932.70)	2.837	0.005
	Contact events	78.00(63.00,93.00)	84.00(73.00,102.00)	2.855	0.004
	Unique contacts	27.00(24.00,34.00)	30.00(25.00,36.00)	2.778	0.005
Medium- high risk	Total Contact duration	1045.95(856.30,1303.35)	1172.12(1003.27,1452.92)	3.413	< 0.001
	Contact events	49.00(38.00,61.00)	53.00(43.00,65.00)	2.525	0.012
	Unique contacts	23.00(20.00,28.00)	26.00(23.00,31.00)	3.941	< 0.001
High risk	Total Contact duration	226.17(138.78,317.12)	264.97(157.97,378.23)	2.006	0.045
	Contact events	6.00(5.00,9.00)	8.00(6.00,11.00)	3.494	< 0.001
	Unique contacts	5.00(4.00,6.00)	6.00(5.00,8.00)	3.025	0.002

Note: Total Contact duration is reported in minutes; Contact events are reported as counts (times); Unique contacts are reported as counts (individuals)

Contact networks

A total of 18,069 distinct contact dyads were observed, of which 9,028 were peer dyads; these formed the basis for the subsequent class-level network analysis. Network topology analysis revealed that the shortest path length between any two individuals was no greater than 3, with an average path length of 2.157 (Fig. 5), indicating strong population connectivity. However, the overall network density was only 0.187. The density of high risk contact networks (0.113–0.270) was significantly lower than that of medium-high risk networks (0.601–0.873, p < 0.05), network density varied considerably across classes. Notably, this variation corresponded to the infectious disease-related absence rates of respective classes (Table 4). Spearman’s rank correlation analysis revealed a positive correlation between infectious disease-related absence rates and medium-high risk network density (r = 0.833, p = 0.005), and a positive correlation with high risk network density (r = 0.733, p = 0.025). Results for medium-low risk and low risk networks are presented in the Supplementary Table 7.

Fig. 5.

The color coding of dots indicates membership in the same kindergarten class. Node size is proportional to the number of contact partners. Edges represent contacts, with thickness corresponding to contact duration

Table 4.

The network characteristics of contact networks for classes at different exposure risk levels

Class	Absenteeism Rate	Medium-high risk				High risk
		Node Count	Edge Count	Avg. Degree	Density	Node Count	Edge Count	Avg. Degree	Density
1 A	83.33%	25	262	20.960	0.873	25	81	6.480	0.270
1B	30.00%	28	289	20.643	0.765	28	71	5.071	0.188
1 C	80.00%	24	240	20.000	0.870	24	62	5.167	0.225
2 A	61.76%	27	275	20.370	0.783	27	74	5.481	0.211
2B	34.28%	30	137	21.133	0.729	30	83	5.533	0.191
2 C	73.53%	28	305	21.786	0.807	28	76	5.429	0.201
3 A	57.14%	38	566	29.789	0.805	38	168	8.842	0.239
3B	47.62%	41	601	29.756	0.744	41	163	7.951	0.119
3 C	48.78%	41	493	23.45	0.601	40	88	4.400	0.113

Sensitivity analysis

Sensitivity analyses confirmed that the primary findings of this study remained robust across different contact definitions and threshold selections.

Scenario 1 (Filtering by individual event duration): After excluding all contact events shorter than 10 min, the total number of contact events decreased from 389,617 to 22,603 (5.8% of the original). However, the comparisons between the absentee and non-absentee groups remained consistent with the primary analysis (Supplementary Table 8). Specifically, in the medium-low, medium-high, and high-risk strata, all contact metrics—including total contact duration, number of contact events, and number of unique individuals contacted—remained significantly higher in the absentee group(all p < 0.01).

In contrast, the low risk stratum, which showed significantly higher contact metrics in the non-absentee group in the primary analysis, was no longer significant (p > 0.3). Notably, the overall contact metrics (total duration, number of events, and number of unique contacts), which were not significantly different between groups in the primary analysis, became highly significant in the sensitivity analysis (p < 0.01), with higher values observed in the absentee group. At the class level, because high risk and medium-high risk networks consist predominantly of longer-duration contacts, their network densities remained unchanged, and their significant correlations with absenteeism rates persisted (r = 0.701 and 0.792, respectively, both p < 0.05). In contrast, the low-risk network became sparser after truncation, and its density became significantly correlated with absenteeism rates (r = 0.767, p = 0.016; Supplementary Table 9).

Scenario 2 (Reclassification without data exclusion): Furthermore, when duration thresholds for risk strata were reduced, the absentee group continued to exhibit significantly higher contact metrics in the medium-high and high-risk categories (p < 0.05), (Supplementary Table 10). At the class level, Spearman’s rank correlation analysis using these reduced thresholds revealed that only high-risk network density maintained a significant positive correlation with absenteeism rates (r = 0.75, p = 0.02; Supplementary Table 11), while correlations for the medium-high and medium-low strata were no longer statistically significant (p > 0.3; Supplementary Table 11). These results collectively indicate that the observed associations between specific contact patterns and absenteeism are robust and not artifacts of brief contacts or specific threshold selections.

Discussion

This study presents the first detailed measurements of close contact patterns among kindergarten children in mainland China (for primary school groups, see previous work by our group [13]). Given the high burden of infectious diseases among kindergarten-aged children, collecting accurate contact data in these settings is particularly important. The resulting contact pattern data can be used to simulate the transmission dynamics of various infectious diseases and to aid in evaluating the effectiveness of specific control measures, such as school closures and immunization strategies. The student participation rate reached 88.68%.

In this relatively enclosed kindergarten environment, each child experienced an average of 1,248.78 contact events per day. The median contact duration for a single event was 43 s, and on average, each child interacted daily with 57.91 distinct individuals, resulting in a total contact time of about 3,227.54 min per person.

Compared to two prior studies employing stricter contact definitions, our study observed substantially higher contact rates per individual. The Gambian kindergarten study [14], which required direct observation of conversation or physical touch lasting ≥ 5 min, reported 15 individuals per child. Similarly, the French study of children aged 5–9 years [15], using contact diaries with a minimum duration of 2 min for face-to-face conversations or skin-to-skin contact, recorded only 10.5 individuals. In contrast, our UWB-based approach yielded 57.91 contacts per individual. This figure was comparable to the approximately 50 individuals identified in a primary school study using high-precision instruments [16], where contact was defined as proximity < 1.5 m for ≥ 20 s. However, our result remained lower than the 70.3 individuals reported in a UK secondary school setting [17], which required face-to-face conversations ≥ 10 min or adjacent seat sharing ≥ 30 min. This difference likely reflects the inherently larger social circles and higher population density typical of secondary school environments compared to kindergartens.

These discrepancies primarily reflect the varying operational definitions of “contact” across studies. Stemming from our choice of definition, this study is inherently more sensitive to brief encounters. This heightened sensitivity results from the UWB badges’ capacity to capture short contact events that self-reports, diaries, or observational methods may overlook, thereby reducing recall bias. Consistent with this interpretation, previous research has demonstrated that high-precision devices record significantly more contact events ≤ 5 min in duration compared to traditional questionnaires [8].

Data on outdoor contacts showed that students had infrequent and brief interactions (Supplementary Fig. $), resulting in a high proportion (57.15%) of low risk contacts. This is consistent with the lower transmission risk in well-ventilated outdoor environments, where pathogen spread is significantly reduced [18]. In contrast, indoor environments carry a substantially higher risk. Enclosed and poorly ventilated spaces allow pathogens to accumulate in the air and on surfaces [19]. This risk is further amplified by shared objects, such as desks and toys, which can act as transmission vectors [20], especially when combined with poor hand hygiene among kindergarten students. These combined factors can lead to outbreaks of respiratory and gastrointestinal illnesses. Therefore, rigorous implementation of ventilation and surface disinfection protocols is essential.

Activity-specific risk levels showed significant heterogeneity (Fig. 3). Morning exercises and post-meal walks, characterized by organized routines and short duration (30 min), had the lowest proportion of high risk contacts (0.47%–2.25%). In contrast, nap time posed the highest risk (14.38%) due to confined spaces and prolonged static contact (Supplementary Fig. 5). The long duration (2 h) of outdoor free activities and the cross-class interactions may be associated with the significantly increased risk observed during those periods, which is consistent with findings from a survey of an American summer camp [21].

Notably, although the contact patterns observed during breakfast and lunch were broadly similar, extending the lunch period by just 10 min increased the cumulative contact duration by 50.38%. Consistent with secondary school studies [22], our data show that indoor lessons, on average, generated fewer contact events than outdoor activities (Supplementary Table 5). This suggests that contact duration plays a role in shaping potential transmission risk. However, unlike the highly structured format typical of primary and secondary classrooms, indoor lessons in the kindergarten setting frequently incorporated unstructured free-play periods. This spatial dispersion reduced prolonged, large group gatherings and resulted in contact patterns that more closely resembled those seen in outdoor settings. This highlights the importance of considering activity duration as a critical parameter in risk management, particularly in densely occupied settings.

Analysis of contact-risk distributions (Fig. 4) and network structures (Fig. 5) indicated that student interactions were mainly confined within the same grade, with relatively few inter-grade contacts. Even within the same grade, cross-class interactions were not predominant. An exception was observed between certain class pairs (e.g., Class 1 A with Class 3B, and Class 1B with Class 3 A), where stable inter-grade links were evident. These patterns may be a product of the school’s spatial and temporal organization : a tightly scheduled timetable, the absence of free recess periods, and strict zoning of outdoor areas appear to limit between-class contacts on the same floor, while promoting frequent contacts between classes that share specific activity areas.

In contrast to previous primary school studies [7], our observations identified substantial teacher-student interactions, potentially attributable to the distinctive pedagogical approaches in kindergartens and children’s dependent behaviors toward teachers [23].

At the individual level, students with infectious disease-related absenteeism records showed significantly higher cumulative contact duration, contact frequency, and number of unique individuals contacted in medium-low, medium-high, and high risk interactions compared with those without such records. This pattern aligns with the hypothesis that prolonged close contact is associated with increases transmission risk [24, 25], however, additional evidence is needed to confirm this relationship. Children with absenteeism records did not differ from peers in overall contact metrics, but they engaged in significantly more frequent and longer contacts at medium-low, medium-high, and high risk levels. This appears to contrast with the conventional view that higher social activity correlates with higher infectious disease risk [26]. This lack of difference may be explained by the inclusion of a large number of short-duration, low risk interactions in the overall metrics. Such interactions, while inflating total contact counts, contribute little to actual transmission risk and thus dilute the underlying differences observed in high-risk interactions. These findings suggest that relying solely on overall contact frequency or duration to assess disease risk may underestimate the role of high-risk interactions in transmission.

The density of high-risk contact networks varies across different classes, with the overall trend showing a positive correlation with class-level infectious disease absenteeism rates. This aligns with the view that “higher network density increases the opportunity for interpersonal contact, which could contribute to larger and faster disease outbreaks“ [27, 28]. However, given the limited sample size in this analysis (n = 9), this association should be interpreted cautiously, as it may reflect confounding factors—such as grade levels or teacher practices—rather than the contact structure itself.

These findings suggest potential strategies for targeted prevention. During peak infectious disease seasons, educators could facilitate more outdoor activities or divide classes into smaller activity groups to “loosen” the network structure, thereby reducing sustained close contacts. Furthermore, enhanced health screening, such as the reinforcement of morning and midday checks, may enable the early identification and isolation of symptomatic students before further transmission occurs.

Limitation

While the UWB badges used in this study provided high-precision measurements of interpersonal distance, several limitations must be acknowledged. First, the technology is unable to capture the relative orientation of individuals, such as whether they were engaged in face-to-face communication or physical contact. This introduces the possibility of overestimating epidemiologically relevant contacts by recording instances where children are simply in close proximity—such as sitting back-to-back—without direct interaction. Furthermore, the badges cannot differentiate between specific types of contact (e.g., brief physical touch versus sustained conversation), which may lead to overlooking exposures that depend on specific transmission routes, such as droplet versus contact transmission.

Second, the study relied on data collected from a single kindergarten on a single day. Although previous research suggests a high degree of consistency in children’s contact patterns across school days [29], the stability of these patterns under varying seasonal conditions or different behavioral contexts remains uncertain. Third, the dataset was limited to interactions within the kindergarten setting, excluding contacts in family and community environments. This omission prevents a comprehensive assessment of the total transmission landscape for these children.

Fourth, the class-level correlation analyses were based on a small sample size of nine classes (n = 9), which limits the statistical power and the generalizability of the findings. The observed associations between network density and absenteeism may be confounded by unmeasured class-level factors, including classroom volume, ventilation efficiency, or specific teacher-led behavioral interventions. Finally, the study lacked data on children’s individual immune backgrounds and vaccination statuses, which are critical biological determinants of infection risk alongside social exposure. Furthermore, the reliance on clinical symptoms recorded by the school medical office rather than laboratory-confirmed pathogen diagnoses means that the findings represent an exploratory, hypothesis-generating analysis.

Conclusion

A total of 389,617 contact episodes between 282 students and 30 teachers were recorded in this first high-resolution contact study in a mainland Chinese kindergarten. On average, each child accumulated over 53 h of contact time and interacted with 57.9 individuals daily, a figure significantly higher than previously documented in comparable contexts. Using a risk-stratification framework based on proximity and duration, we found that outdoor activities were dominated by low-risk contacts, whereas indoor activities—particularly nap time—contributed the largest proportion of high-risk contact.

Despite reporting comparable overall contact totals to their peers, children with infectious disease-related absenteeism records displayed significantly higher contact metrics within the medium-to-high-risk strata. At the class level, the densities of medium-high risk (r = 0.833) and high risk (r = 0.733) contact networks showed a correlation with absenteeism rates. These associations remained robust in sensitivity analyses excluding brief contacts.

Our findings demonstrate that aggregate contact measures alone may obscure meaningful differences in transmission-relevant interactions. By identifying specific activities and network structures associated with elevated risk, this study provides an evidence base for targeted, minimally disruptive interventions—such as modifying nap-time arrangements or promoting outdoor activities—and highlights the value of high-resolution proximity sensing in understanding infectious disease dynamics in early childhood settings.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

UWB

Ultra-Wide Band

COVID-19

Coronavirus Disease 2019

RFID

Radio-Frequency Identification

S-T

Student-teacher contacts

T-T

Teacher-teacher contacts

SD

Standard deviation

M

Median

p25

25th Percentile

p75

75th Percentile

vs.

Versus

p

p-value

Avg. Degree

Average Degree

Author contributions

Fan Jiang: Writing-original draft, Visualization, Methodology, Investigation, Formal analysis, Data curation; Ying Wen: Writing-review & editing, Visualization, Investigation, Methodology, Conceptualization, Funding acquisition; Fangfang Lu: Writing-review & editing, Investigation, Methodology, Conceptualization, Methodology; Deji Ciren, Yingluan Zhang, Huiyang Sun and Rijing Luo: Writing-review & editing, Visualization, Investigation; Zhen Zhang and Qiuying Lv: Writing-review & editing, Methodology, Supervision, Funding acquisition Conceptualization. Fan Jiang, Ying Wen and Fangfang Lu contributed equally to this work and designated as co-first authors.

Funding

This work was supported by the Talent Support Program of Guangdong Provincial Center for Disease Control and Prevention (2025D404); the Key Project of Shenzhen Science and Technology Innovation Commission (JCYJ20210324115411030); the Shenzhen Key Laboratory of Major Infectious Disease Surveillance; the Shenzhen Science and Technology Program (Grant No. SYSPG20241211173921049); and the Guangdong Provincial Medical Science and Technology Research Fund (A2025272).

Data availability

The datasets generated and analysed during the current study are not publicly available due to privacy and ethical restrictions related to participant confidentiality, but are available from the corresponding author on reasonable request and subject to institutional approval.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the Ethics Committee of Shenzhen Center for Disease Control and Prevention (Approval No. Shen-Ji-Lun-Zi (2023) No. 028B). Written informed consent was obtained from the legal guardians of all children, and participants’ right to withdraw from the study at any time was fully safeguarded

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.