Census rate as a pragmatic indicator of emergency department crowding: a multicentre prospective observational comparison with CEDOCS against frontline-perceived workload

Abstract

Background

Emergency department (ED) crowding affects patient flow, care timeliness, staff well-being, and frontline workload. Although several crowding metrics are available, their interpretability across EDs of different sizes remains uncertain. We compared Census Rate (CR) and the Community Emergency Department Overcrowding Scale (CEDOCS) against frontline-perceived workload across three French EDs.

Methods

We conducted a multicentre prospective observational study from 15 to 28 May 2023 in one university hospital and two community hospitals. Twice daily, operational variables were recorded and eligible frontline staff completed a 6-point staff workload assessment (SWA). The primary outcome was the association between SWA and each crowding metric. Spearman’s rank correlation was prespecified as the primary analysis, with Pearson correlations and linear regression as sensitivity analyses. Overlapping dependent correlations were compared using Williams’ test. Additional exploratory analyses examined empirical CR tertiles.

Results

Eighty-four site-time observations were expected, of which 81 had complete data for the main analyses. CR showed a strong association with SWA (Spearman’s ρ = 0.744, 95% CI 0.619 to 0.832; p < 0.001), whereas the association for CEDOCS was more moderate (ρ = 0.471, 95% CI 0.289 to 0.616; p < 0.001). The association between CR and SWA was significantly stronger than that between CEDOCS and SWA (difference in correlations 0.273, 95% CI 0.116 to 0.438; Williams’ t = 4.728, df = 78; p < 0.001). In community hospitals, both metrics performed similarly, whereas CR showed a clearer advantage overall. Empirical CR tertiles were significantly associated with SWA and were used for exploratory interpretation within this dataset.

Conclusions

Across EDs of different sizes, CR showed a stronger association with frontline-perceived workload than CEDOCS. These findings suggest the potential value of CR as a simple and interpretable crowding metric to support shared situational awareness and workload-informed discussions, pending validation in broader settings and nurse-specific studies.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12912-026-04902-7.

Background

Emergency department (ED) crowding is a persistent international problem that affects patient flow, timeliness of care, staff well-being, and the overall safety of care delivery. Conceptually, crowding reflects interacting input, throughput, and output constraints rather than a single operational bottleneck [1]. Recent syntheses have continued to show that crowding is associated with delays in care and broad harm across emergency care systems, making robust and interpretable metrics important for both research and service organization [2, 3].

From a nursing perspective, crowding is not only an operational phenomenon but also a work-environment problem. Emergency nurses are central to surveillance, reassessment, prioritization, coordination, and continuity of care in settings marked by uncertainty, rapid turnover, and fluctuating acuity. Better ED work environments have been associated with better reported care quality and safety and with lower burnout, job dissatisfaction, and intent to leave among emergency nurses [4]. In parallel, literature focused specifically on emergency nurses has described crowding as contributing to staffing pressure, stress, violence exposure, and difficulties in providing timely and complete care [5].

This nursing perspective is important because the strain generated by crowding is partly expressed through missed or delayed nursing care. A recent scoping review identified missed nursing care as a significant patient-safety issue in EDs [6], and a cross-sectional study in Swedish EDs reported high levels of missed nursing care together with poorer perceptions of quality of care and patient safety [7]. Accordingly, a crowding metric intended to support frontline situational awareness should not only be easy to calculate, but should also reflect frontline-perceived workload and operational strain.

Several crowding metrics have been proposed, but their interpretability across heterogeneous EDs remains uncertain. Reviews have highlighted the heterogeneity of existing crowding measures and the difficulty of identifying tools that are both practical and comparable across settings [2, 3]. Community Emergency Department Overcrowding Scale (CEDOCS) was developed to adapt crowding measurement to community EDs [8], while Swedish work on workload-oriented models has emphasized the value of anchoring crowding assessment to staff-perceived workload [9, 10]. These issues are particularly relevant for frontline and nursing leadership because indicators used to support situational awareness or escalation discussions must remain understandable across departments of different sizes and organizational models.

In the present study, we examined the Census Rate (CR), an author-defined pragmatic indicator operationalized as the ratio between current ED census and average daily ED volume, as a simple volume-based measure that does not directly depend on bed count. Such an approach may be advantageous when comparing university and community EDs. At the same time, prior work has shown that nursing workload and ED crowding are related but not interchangeable constructs, and that perceived nursing workload can be estimated in real time from routinely available operational data [9, 11]. More broadly, pragmatic nurse-led process redesigns in the ED, such as nurse-initiated imaging pathways, may improve flow and timeliness of care, supporting interest in simple operational tools that are feasible in daily practice [12].

Therefore, this multicentre prospective observational study aimed to compare CR and CEDOCS against frontline-perceived workload across three French EDs of different sizes. We hypothesized that CR, as a simpler and potentially more scalable metric, would show a stronger association with perceived workload than CEDOCS. A secondary objective was to explore whether a pragmatic categorization of CR could support interpretation of frontline-perceived workload across heterogeneous ED settings.

Methods

Study design and setting

This multicentre prospective observational study was conducted from 15 to 28 May 2023 in three EDs within the same French region: one university hospital and two community hospitals. The study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement [13].

The participating sites were selected to reflect heterogeneous ED activity levels and organizational contexts. The university hospital recorded 49,381 annual ED visits, whereas the two community hospitals recorded 16,058 and 12,497 annual ED visits, respectively. Baseline information on annual activity and treatment-bed capacity was collected for site characterization and for derivation of the CR denominator. Annual ED volume was based on the most recent complete annual activity data available before the study period and was divided by 365 to estimate the average daily ED volume at each site.

Participants and data collection

Eligible respondents were nurses and attending physicians working clinically in the ED during the study period. Residents, students, and staff assigned exclusively to the short-stay unit were not eligible. Because respondent profession was not consistently recorded at each site-time observation, rater-specific or nurse-specific analyses were not prespecified and could not be performed.

Data were collected twice daily, between 09:00 and 10:00 and between 17:00 and 18:00. During each measurement period, a designated shift coordinator, typically the charge nurse or local flow coordinator, recorded routinely available operational ED variables and invited on-duty eligible staff to complete a single-item staff workload assessment (SWA).

The SWA used a 6-point Likert scale ranging from 1 (minimal workload) to 6 (extreme workload). SWA was considered a pragmatic proxy for frontline-perceived workload rather than a validated measure of nursing workload or an objective measure of ED crowding. Individual SWA forms were aggregated by site and measurement period using the mean score, yielding one site-time observation per measurement period. The analytic unit was therefore the site-time observation rather than the individual staff response. Measurements were standardized time-point snapshots rather than continuous monitoring.

Measures

Operational variables recorded once per site-time observation were the current ED census, the number of critical care patients, the number of patients not yet seen by a provider, the longest ED length of stay, and the number of available ED treatment beds. The current ED census was defined as the total number of patients under ED responsibility at the time of assessment, including both patients waiting to be seen in the waiting area and those already assigned to treatment spaces. Critical care patients were defined as patients triaged as Emergency Severity Index level 1 or 2.

The CR was defined a priori for this study as the ratio of the current ED census to the average daily ED volume, derived from historical site activity available before study initiation. For exploratory interpretation within this dataset, CR was categorized into empirical tertiles based on the observed distribution of complete site-time observations, with percentile boundaries of 0.204 and 0.290, correspond approximately to low (< 20%), moderate (20%–30%), and high (> 30%) CR categories. The CEDOCS was calculated using the original published variables and interpretation thresholds [8]. The SWA was used as a pragmatic criterion measure of frontline-perceived workload and operational strain. Single-item global workload ratings are feasible for repeated real-time use during routine ED activity and have been used in workload-oriented emergency care research [10, 11].

Outcomes

The primary outcome was the association between SWA and each crowding metric (CR and CEDOCS). Secondary analyses comprised direct comparison of the two overlapping correlations, stratified analyses by hospital type and time of day, and exploratory analyses across CR tertiles to describe variation in frontline-perceived workload across increasing CR levels.

Statistical analysis

Because SWA was ordinal and overall normality assumptions were not met, Spearman’s rank correlation was prespecified as the primary analytical approach. Pearson correlations and simple linear regression models were performed as sensitivity analyses. 95% confidence intervals for correlation coefficients were estimated using bootstrap resampling. The difference between the CR–SWA and CEDOCS–SWA correlations was tested using Williams’ test for overlapping dependent correlations, implemented with the cocor package [14]. Differences in SWA across CR tertiles were assessed using the Kruskal–Wallis test with Dunn–Holm post hoc comparisons; one-way analysis of variance with Tukey post hoc testing was used as a sensitivity analysis. Exploratory analyses included within-tertile correlations and the association between high perceived workload, defined descriptively as SWA ≥ 4, and high CR, defined as the upper empirical CR tertile. The SWA ≥ 4 threshold was not considered a validated cut-off and was used only to describe observations in the upper part of the workload scale. To account for repeated site-time observations within centres, a sensitivity linear mixed model with a random intercept for centre was also examined. All tests were two-sided, and p < 0.05 was considered statistically significant. Data were organized in Excel 2019 (Microsoft, Redmond, WA, USA) and analysed using R version 4.3.5 [15].

Results

Study sample and site characteristics

Across the three participating EDs, 84 site-time observations were expected over the 2-week study period (3 sites × 14 days × 2 daily measurement periods). Of these, 81 had complete data for CR, CEDOCS, and SWA, and were included in the main analyses. Three site-time observations were excluded because the corresponding forms were incomplete or not correctly completed (Fig. 1).

Fig. 1.

Study flowchart. Flowchart describing the distribution of forms across the three participating emergency departments (one university hospital and two community hospitals), the number of missing or incomplete forms, and the final number of complete observations included in the analyses

The university hospital recorded 49,381 annual ED visits and 39 treatment beds, whereas the two community hospitals recorded 16,058 and 12,497 annual visits, with 12 and 8 treatment beds, respectively. Mean CR values were 0.27, 0.28, and 0.19 across the three sites; mean CEDOCS values were 104.5, 44.19, and 28.81; and mean SWA values were 2.58, 3.14, and 2.39, respectively (Table 1).

Table 1.

Characteristics of included study sites

	University hospital	Community hospital 1	Community hospital 2	p-value1
Annual visits, n	49,381	16,058	12,497
Number of treatment beds, n	39	12	8
Census rate, mean (SD)	0.27 (0.09)	0.28 (0.10)	0.19 (0.08)	0.002
CEDOCS, mean (SD)	104.50 (31.86)	44.19 (20.19)	28.81 (30.43)	< 0.001
SWA, mean (SD)	2.58 (0.73)	3.14 (1.34)	2.39 (1.11)	0.039
Missing data (n)	0	2	1

SD= standard deviation; CEDOCS= community emergency department overcrowding score; SWA= staff workload assessment

¹ p-values compare the three participating sites

Association between crowding metrics and perceived workload

CR showed a strong positive association with SWA (Spearman’s ρ = 0.744, 95% confidence interval (CI) 0.619 to 0.832; p < 0.001), whereas CEDOCS showed a more moderate association with SWA (ρ = 0.471, 95% CI 0.289 to 0.616; p < 0.001) (Fig. 2). The association between CR and SWA was significantly stronger than that between CEDOCS and SWA (difference in correlations 0.273, 95% CI 0.116 to 0.438; Williams’ t = 4.728, df = 78; p < 0.001).

Fig. 2.

Correlation between SWA and crowding metrics (CEDOCS and CR). Scatter plots of subjective workload assessment (SWA) against CEDOCS score (left panel) and Census Rate (CR) (right panel). Each panel displays Spearman’s ρ and Pearson’s r with 95% confidence intervals, as well as the linear regression equation and p-value for the slope

Sensitivity analyses using Pearson correlations yielded consistent conclusions. The Pearson correlation between CR and SWA was 0.718 (95% CI 0.602 to 0.813; p < 0.001), compared with 0.374 (95% CI 0.225 to 0.504; p < 0.001) for CEDOCS and SWA.

Stratified analyses

When analyses were stratified by hospital type, positive associations were observed in both settings. In community hospitals, the Spearman correlation between CR and SWA was 0.766, compared with 0.768 for CEDOCS and SWA. In the university hospital, the corresponding correlations were 0.835 for CR and 0.735 for CEDOCS (Fig. 3).

Fig. 3.

Association between SWA and crowding metrics by hospital type. Stratified associations between SWA and the two crowding metrics (CEDOCS score and Census Rate) by hospital type (university hospital vs. community hospitals). Panels report Spearman’s ρ and Pearson’s r with 95% confidence intervals, and the corresponding linear regression equation with p-value for the slope

When stratified by time of day, CR remained more strongly associated with SWA than CEDOCS in both periods. In the evening, the Spearman correlations were 0.805 for CR and 0.576 for CEDOCS; in the morning, they were 0.695 and 0.379, respectively (Fig. 4).

Fig. 4.

Association between SWA and crowding metrics by time of day. Stratified associations between SWA and the two crowding metrics (CEDOCS score and Census Rate) according to the two daily measurement periods (morning vs. evening). Panels report Spearman’s ρ and Pearson’s r with 95% confidence intervals, and the corresponding linear regression equation with p-value for the slope

CR tertiles and exploratory analyses

For exploratory interpretation, empirical CR tertiles corresponded approximately to low (< 20%), moderate (20%–30%), and high (> 30%) CR, with observed cutoffs at 0.204 and 0.290 (Fig. 5). SWA differed significantly across CR tertiles (Kruskal–Wallis test, p < 0.001). Supplementary analyses further explored pairwise differences across tertiles and within-tertile correlations (Supplementary Tables S1–S2).

Fig. 5.

Distribution of census rate by empirical tertiles. This figure presents the distribution of Census Rate (CR) values across empirical tertiles derived from the complete study sample: low CR (< 20%, median = 18%, n = 27), moderate CR (20%-30%, median = 26%, n = 27), and high CR (> 30%, median = 35%, n = 27). These categories were used for exploratory interpretation within this dataset and should not be considered validated crowding thresholds

High perceived workload, defined descriptively as SWA ≥ 4, occurred in 15 of 81 complete site-time observations (18.5%). Most high-SWA observations occurred in the upper CR tertile (13/15, 86.7%). Conversely, high perceived workload was observed in 13 of 27 observations in the upper CR tertile (48.1%), compared with 2 of 54 observations in the low or moderate CR tertiles (3.7%) (Supplementary Table S3).

In an additional exploratory analysis, observations with high perceived workload (SWA ≥ 4) had higher CR values than observations with lower perceived workload (median 0.34 [IQR 0.31–0.38] vs. 0.21 [0.18–0.27]; p < 0.001), whereas CEDOCS values were numerically higher but not significantly different (median 60.00 [52.00–69.50] vs. 44.50 [21.25–92.00]; p = 0.213) (Supplementary Table S4).

Discussion

In this multicentre prospective study across three French EDs, CR showed a stronger association with frontline-perceived workload than CEDOCS overall, while both metrics performed similarly in community hospitals. These findings suggest that CR may provide a simple and interpretable signal of operational strain across EDs of different sizes, although its broader applicability requires external validation. Because the practical value of a crowding metric depends not only on whether it can be calculated but also on whether it reflects the strain perceived by frontline teams, this result may be relevant to shared situational awareness and operational coordination.

Our findings are consistent with previous literature showing that ED crowding is a multidimensional phenomenon and that crowding metrics vary in their feasibility, interpretability, and transferability across settings [1–3, 16]. Prior work has also suggested that predefined crowding thresholds may not be equally applicable across EDs with different structures and patient volumes [17]. In this context, the stronger overall association between CR and perceived workload suggest that a simple census-based indicator may be useful when comparing departments of different sizes. At the same time, our results do not suggest that CEDOCS is without value. In community hospitals, its association with workload was very similar to that of CR, indicating that CEDOCS may still be informative within specific local contexts.

From a nursing perspective, the main contribution of this study is not the promotion of another abstract score, but the evaluation of a practical operational signal against the workload perceived by frontline staff. Emergency nurses work at the intersection of direct care, reassessment, prioritisation, surveillance, coordination, and patient flow. Recent nursing scholarship has emphasized that patient flow and direct care are closely interconnected in emergency nursing practice [18]. Within that framework, a crowding indicator may be useful only if it helps make visible the workload strain under which care is delivered. Our results suggest that CR could help translate operational pressure into a shared language for frontline teams and local leadership, while further nurse-specific validation remains necessary.

This interpretation is consistent with nursing literature linking ED work environments with burnout, job dissatisfaction, and perceived quality and safety of care [4], as well as with studies showing that missed nursing care is an important safety concern in emergency settings [6, 7]. More recent evidence has also shown that better nurse work environments are associated with more favourable flow outcomes, such as lower left-without-being-seen rates and shorter lengths of stay [19], while systematic review data suggest that inadequate nurse staffing is associated with delayed care and other adverse outcomes in EDs [20]. Together, these findings suggest that simple operational indicators could contribute to earlier recognition of rising strain by charge nurses, physicians, and local leadership, although our study did not directly assess staffing decisions, nursing-sensitive outcomes, or escalation processes. This perspective is also consistent with broader interest in pragmatic nurse-led organizational strategies to improve flow, staffing, and quality of care in emergency settings [12, 21].

This study has several strengths, including its prospective multicentre design, inclusion of EDs with different activity levels, and focus on frontline-perceived workload. However, some limitations should be acknowledged. First, SWA was used as a pragmatic proxy for frontline-perceived workload rather than as a validated measure of nursing workload or an objective measure of ED crowding. Second, individual responses were aggregated at the site-time level to match the unit of analysis of the crowding metrics; this pragmatic approach may have reduced individual-level variability and may introduce ecological bias. Third, data were collected at two predefined daily time points and therefore represent standardized snapshots rather than continuous monitoring; nighttime periods were not assessed, and weekday–weekend differences were not specifically examined. Fourth, the study was conducted over a two-week period in May in three hospitals within one French region, which may limit generalisability and may not capture seasonal peaks in crowding. Fifth, respondent profession was not consistently recorded at each site-time observation, preventing nurse-specific analyses, and staffing variables such as nurse-to-patient ratios or skill mix were not collected. Workload perception may also have been influenced by factors not fully captured by the available operational variables, including case mix, patient acuity, team composition, and social desirability bias. Finally, the proposed CR categories were empirically derived from this dataset and should be considered exploratory rather than definitive thresholds. Future studies should validate CR in broader settings and examine its relationship with nurse-specific workload measures, staffing decisions, nursing-sensitive outcomes, and patient-flow outcomes.

Conclusion

In this multicentre prospective study, Census Rate showed a stronger association with frontline-perceived workload than CEDOCS across EDs of different sizes. Although CEDOCS remained informative in some local contexts, CR appeared to provide a simpler and more interpretable signal of operational strain. These findings suggest the potential value of CR as a pragmatic crowding metric to support shared situational awareness and workload-informed discussion, pending validation in broader settings and nurse-specific studies.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

CEDOCS

Community Emergency Department Overcrowding Scale

CI

Confidence interval

CR

Census Rate

ED

Emergency department

STROBE

Strengthening the Reporting of Observational Studies in Epidemiology

SWA

Staff workload assessment

Author contributions

FP conceived the study, designed the protocol, supervised its implementation, and drafted the manuscript. JH collected the data, initiated the study across the participating hospitals, and contributed to the preliminary analysis. VG performed the statistical analyses, interpreted the results, and critically revised the manuscript. All authors approved the final version of the manuscript.

Funding

The study received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data availability

The datasets generated and/or analysed during the current study are not publicly available due to institutional restrictions regarding operational hospital data but are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study received a favorable opinion from the Groupe Éthique d’Aide à la Recherche Clinique pour les protocoles de recherche non soumis au Comité de Protection des Personnes (Ethics Committee in Human Research), Tours University Hospital, France (project No. 2024 105; February 11, 2025). This study was conducted in accordance with the Declaration of Helsinki and with relevant French guidelines and regulations. Participation was voluntary and the questionnaire was anonymous. All participants were informed about the study before participation, and completion of the questionnaire was considered as implied informed consent to participate.

Consent for publication

Not applicable.

Use of generative AI and AI-assisted technologies in the writing process

Generative AI tools were used only for language editing after completion of the scientific content; all outputs were reviewed and revised by the authors.

Competing interests

The authors declare no competing interests.