Contributors to perceived workload strain in the pediatric intensive care unit Authors: Candace Collins, Dustin Hansen, Amanda Levin, James Fackler, Michael Rosen, Salar Khaleghzadegan, Gayane Yenokyan, David C. Stockwell Published: 2026 ABSTRACT ObjectivesPatient outcomes worsen during periods of high intensive care units (ICU) capacity strain. Common metrics used to determine workload strain are overall patient census, turnover, and various patient acuity surrogates. We aimed to identify factors that increase perceived workload strain in the pediatric intensive care unit (PICU) using the NASA Task Load Index (TLX) and single item stress scores (SISS).MethodsThis is a single-center, prospective observational study with survey and electronic medical record (EMR) data. We used multiple NASA TLX and SISS metrics per PICU shift as markers of perceived workload strain. These data, collected from physician fellows and PICU charge nurses, were compared to various objective unit- and patient-level factors. We used exploratory visual techniques as well as linear mixed-effect models to characterize relationships between strain and individual variables. Data was collected during the COVID-19 pandemic between October 2020 and April 2021.ResultsWorkload strain was perceived to be higher during day shifts than night shifts and higher for physicians than nurses. Patient volume, turnover, and acuity metrics as well as reported strain showed significant variation during data collection. Patient-level factors associated with increased strain included more patients per shift requiring renal replacement therapy or mechanical/positive pressure ventilation. Unit-level factors associated with increased strain included increases in the totals of patients admitted and discharged, number of blood draws, patient hours, and continuous medication infusions.ConclusionsWe find that the PICU is a dynamic environment with variable demands and thresholds beyond which healthcare workers begin to feel increased strain. These findings provide depth and granularity for the study of workload strain in the PICU and could have implications for future improvements in patient care.Supplementary InformationThe online version contains supplementary material available at 10.1007/s44253-026-00126-4. Introduction Strain in the intensive care unit (ICU) is a topic that has recently increased in the critical care literature, though mostly for adult ICUs. ICU capacity strain is a relatively poorly defined concept, but one proposed definition is “a time-varying imbalance between the supply of available beds, staff and/or resources and the demand to provide high quality care for patients who may become or are critically ill [1].” It has also been defined as “strain on the ability of an ICU to provide optimal care to patient’s that day [2].” In the adult ICU literature, data are mixed, but some studies have shown that patients discharged from ICUs during periods of high capacity strain have worse outcomes than those discharged during times of lower strain. Reported outcomes during high levels of strain include longer post-ICU length of stay, higher mortality, and higher total hospital costs [3, 4]. Overall, ICU capacity strain has been associated with perceived compromised ability to provide high-quality, safe healthcare, and increased healthcare worker stress and burnout [1]. There is a paucity of research that specifically evaluates the definition and effects of strain in the pediatric intensive care unit (PICU). One study found that the odds of both patient and unit adverse events increased at times of high census and high patient acuity [5]. Another study found that perceived workload strain of nurses in a neonatal intensive care unit was associated with increased missed nursing care for patients [6]. A more recent study by Fundora et al. showed that “increased bed occupancy and lower staffing were associated with increased mortality, length of stay, imaging orders, and laboratory turn-around time” in the pediatric cardiac ICU [7]. This finding suggests that an imbalance between patient care demands and the ability to meet these demands can have a negative impact on overall quality of care. Strain in the PICU may have seasonal variations where there are higher overall occupancy of PICU beds and turnover [8]. Given the unique patient population and demands of the PICU, we aimed to understand capacity strain in this setting by characterizing relationships between workload strain perceived by healthcare workers and objective patient- and unit-level factors. Understanding the connection between subjective workload strain and objective, measurable EMR data may identify thresholds of strain that can be tracked and assist future resource and staffing allocation to improve care and reduce clinician burnout. To our knowledge, no study in the pediatric critical care literature has identified a relationship between perceived healthcare worker strain and objective data from the electronic health record (EHR), which would include specific patient- and unit-level factors. We hypothesized that as patient demands increased (e.g., patient hours, admissions, labs ordered, medications administered), workload strain perceived by the participating healthcare workers would also increase. Methods Unit structure This is a single-center, prospective observational study in a PICU with survey and electronic medical record (EMR). STROBE guidelines were followed. The unit consists of mixed 40-bed pediatric general medical/surgical and cardiac ICU (during our study, the two units were formally split, but each unit contained charge nurses and fellows in the same roles, so data collection continued). Our ICU has three services: a neurocritical care and trauma service, a general medical and surgical service, as well as a cardiac service. The unit does not have a fixed nurse to patient ratio and staffing depends on patient acuity and nurse availability. There is regularly a combination of both 2:1 and 1:1 nursing-patient assignments. There are two charge nurses during the day and one at night. The unit is staffed by three PICU fellows during the day and two overnight. In addition to providing patient care, fellows are also responsible for facilitating admissions and transfers out of the PICU, working closely with charge nurses on bed management. The fellows are a mix of first-, second-, and third-year fellows with varying levels of experience. Day teams typically have 12–14 patients per fellow. Two of the teams have nurse practitioners as the front-line providers and there is one teaching service comprised of residents as front-line providers and medical students who rotate monthly. The frontline providers serve as the first point-of-contact for bedside nurses for patients, but they are not responsible for administrative or bed management tasks. Participants After approval from the Johns Hopkins Institutional Review Board (IRB00220441), PICU fellows and charge nurses were recruited at separate meetings for study participation. Physician fellows, charge nurses, and nursing team leaders, also called team leads, were chosen for this study because both groups care for multiple critically ill children at a time and are responsible for unit-level administration tasks during a shift. In our unit, charge nurses are more senior-level nurses with multiple years of nursing experience. They do not usually carry a direct patient assignment and are tasked with directing patient flow in and out of the unit, determining assignments/staffing, and assisting with bedside activities, when possible, among other tasks. Nursing team leads are mainly responsible for supervising and assisting the bedside nurses on one of three care teams within the PICU. They may or may not have a patient assignment on a given day. Workload strain assessments To quantify perceived workload strain, we used the NASA Task Load Index (TLX) score and a single-item stress score via digital surveys (supplementary material 2). The TLX scoring system is a validated tool for measuring perceived workload in healthcare settings. To our knowledge, there are no specific numerical cut-offs for level of workload strain using the NASA TLX that represents high or low strain [9]. It includes 12 workload domains that incorporate different parts of the healthcare worker experience (S1). Each domain is scored from 0 to 20, giving a total possible score of 240 [10]. The single-item stress score has also shown utility for quantifying workload strain across different professions [11]. It is simpler than the NASA TLX, but it is designed to be a quick and effective way to measure perceived stress at a single point in time on a scale of 1–10 [11]. Upon agreement to participate, the three daily on-service PICU fellows and one or two charge or team-lead nurses per shift began to submit NASA TLX scoring data over the course of 43 12-hour day and night shifts. For each shift that data were collected, at least one charge nurse participated and a senior nursing team lead also participated, but overall fewer nursing team leads chose to participate in the study. TLX data were collected at the 4-, 8-, and 12-hour marks of each shift. These shifts included weekday, weekend day, and night shifts. The single-item stress score was measured at the 0-, 4-, 8-, and 12-hour time points. All shifts for PICU fellows and charge nurses are 12 h each; therefore, data collection was standardized for both nurses and physicians. Participants received personalized links via text message at specific collection times during their shift and responses were gathered via an electronic survey in real time. Patient- and unit- level factors Patient level factors that were collected included total number of patients cared for that day (at any time) who required mechanical ventilation/noninvasive positive pressure ventilation (MV/NIPPV), vasoactive infusions, extracorporeal membrane oxygenation (ECMO), or renal replacement therapy. These interventions represent ongoing organ dysfunction and require significant resources. This data was reported through electronic surveys completed by fellows at the end of each shift via text message with personalized links. In addition, we collected retrospective data from the electronic heath record (EHR) that included gross lab data ordered and drawn as well as the total number of continuous medication infusions per shift, including vasoactive medications. We chose these variables to start with the most granular and recent data possible regarding patients being cared for in the PICU on any given shift. Unit-level data were also retrospectively collected from the EHR during the shifts when workload strain data were collected from the PICU. EHR data included the total number of admissions, discharges, and patient hours. Patient hours were a surrogate for census, calculated via extraction from the EMR based on amount of time during the shift that patients spent in the PICU. For example, if the shift started at 0700 and the patient was transferred out of the unit at 1200, five patient hours would be added to the total for that shift. To accommodate our study over a period of months, we chose to collect data on a shift-to-shift basis to better characterize our patient population at any given time and attempt to account for as much variation in workload as possible. Data and statistical analysis Data were collected from patients admitted to the PICU at the Johns Hopkins Hospital in Baltimore, Maryland between October 2020 and April 2021 using the Johns Hopkins Precision Medicine Analytics Platform (PMAP). PMAP integrates patient data from electronic healthcare records and physiologic monitors. Data were cleaned and preprocessed using standardized procedures. Survey and EHR summary data were merged using the date and shift designation. Visual displays and other exploratory analysis techniques were used to assess the relationships between the total TLX and the single-item stress measure with the following potential explanatory variables: shift (day vs. night); team role (charge nurse, team-lead nurse, and fellow); month (October-November 2020 and January-April 2021); physician-reported number of patients who required MV/NIPPV, vasoactive infusions, ECMO, or renal replacement therapy; and (from the EHR) number of lab orders, blood draws, admissions, discharges, patient-hours (more accurate surrogate for census), and continuous medication infusions. Continuous medication infusions included vasoactive medications as well as sedatives, analgesics, intravenous fluids, or any other continuous infusion delivered during the shift. To model these relationships, we used linear mixed effect models to account for repeated measures within shift and individual team members. Multiple candidate models were compared by using the Bayesian Information Criterion, including (1) random intercept for shift, (2) random intercept for study member, and (3) a three-level model that accounted for the entire hierarchy in the data: stress scores clustered within shift and shifts clustered per study member. The random effects were assumed to be normally distributed with respective variances. The results of the three-level models are presented, with separate models for single-stress measure and TLX as the dependent variable. Based on the results of the exploratory analyses, four models are shown for each dependent variable, with the following main explanatory variables: number of admitted patients (Model 1), number of discharged patients (Model 2), patient-hours (Model 3), and number of continuous drips (Model 4). All models were adjusted for study member role (fellow vs. charge nurse), month (6 indicator variables), and shift (night vs. day). The models were adjusted for role because of the slightly different responsibilities of physician fellows and nursing leaders. In addition, adjustments for shift (day vs. night) were included because different staffing ratios exist overnight compared to daytime (i.e. 3 physician fellows compared to 2 and 1 charge nurse at night, compared to 2 during the day). Months were included in the adjustment due to the known seasonal variation of PICU admissions. Linear splines were used to reflect nonlinear relationships between explanatory variables and outcomes. Data were analyzed with the STATA statistical software program (StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX: StataCorp LLC.). Results During the study periods (October 2020–April 2021), there were 15 eligible physicians fellows and about 20 eligible charge/nursing leaders with a total of 26 unique healthcare workers (15 physicians and 11 nurses) participating in the study. The physicians were all fellows in the three-year pediatric critical care medicine fellowship with about 4–6 fellows representing each level (first, second, or third year fellows). The participants provided 477 non-missing responses (345 from physicians and 132 from nurses) for the total TLX score and provided 635 non-missing responses to the single-item stress measure (459 from physicians and 176 from nurses). Data from December 2020 is not included because there were not sufficient study participants working shifts that month. Results from the single-item stress score (range 1 to 10) data were analyzed by role and shift timing, respectively. On average, physician fellows reported the highest levels of stress on the single-item stress score, followed by the charge nurses and then the team leads (medians: 5, 4, and 3; means: 4.8, 4.5, and 3.6, respectively). Note that only 7 team-lead nurses were included in this study and results are not included in the figure. On average, team members reported higher levels of stress during day than night shifts (AM responses n = 459, PM responses n = 176; medians: 5 vs. 4; means: 4.9 vs. 4.2, respectively). Results for the NASA TLX score were similar when analyzed by role and shift timing. The total TLX score (range 0 to 240) was highest for the fellows, next highest for charge nurses, and lowest for the team-lead nurses (medians: 117, 110, and 65; means: 115.2, 111.7, and 96, respectively). On average, team members reported higher TLX scores during day than night shifts (AM responses n = 345, PM responses n = 132; medians: 120 vs. 85.5; means: 122.0 vs. 93.2, respectively). Day shifts made up 72.3% of overall SISS and TLX responses. Figure 1 shows the SISS and TLX scores separated into the specific month when data was collected. Winter months (January and February) exhibited the highest perceived level of strain. Fig. 1Stress levels by data collection month as assessed by the single-item stress score (A) and the NASA Task Load Index (B). The horizontal line in the middle of the box represents median score across all shifts. The box represents the 25th − 75th percentiles or quartile 1 (Q1) and quartile (Q3), respectively, and the error bars show the lower and upper “extremes”, defined as Q1–1.5*(Q3-Q1) and Q3 + 1.5*(Q3-Q1), respectively. Individual dots above and below the extremes are the outliers in the data set. (Note that January – April data was from 2021 and October and November data was from 2020) Stress levels by data collection month as assessed by the single-item stress score (A) and the NASA Task Load Index (B). The horizontal line in the middle of the box represents median score across all shifts. The box represents the 25th − 75th percentiles or quartile 1 (Q1) and quartile (Q3), respectively, and the error bars show the lower and upper “extremes”, defined as Q1–1.5*(Q3-Q1) and Q3 + 1.5*(Q3-Q1), respectively. Individual dots above and below the extremes are the outliers in the data set. (Note that January – April data was from 2021 and October and November data was from 2020) Results for the retrospective data from the EHR indicate that mean total number of blood draws each shift was 89.8 (standard deviation [SD] = 20.4; range, 46–131). Total mean number of labs ordered each shift was 206.3 (SD = 43.6; range, 107.0–273.0). Mean number of admissions was 2.3 (SD = 1.5; range, 0–6) and mean number of discharges was 2.7 (SD = 1.9; range, 0–6). Mean total patient hours per shift was 336.2 (SD = 32; range, 236.6–437.3) and mean total medication infusions was 118.5 (SD = 22.8; range, 49–152). Data in supplementary Table 1. Figure 2 represents the relationship between the NASA TLX score and the number of patients who required MV/NIPPV, vasoactive medications, ECMO, or renal replacement therapy. The center line again represents mean TLX score when each variable on the X-axis was reported at the noted value. The shaded area above and below each line represents 1 SD as it relates to each mean. In this case, the number of patients requiring renal replacement therapy as well as MV/PPV seemed to correlate with increased TLX scores. Fig. 2Correlation between NASA Task Load Index total score and self-reported number of patients requiring renal replacement therapy (A), vasoactive medications (B), extracorporeal membrane oxygenation (ECMO) (C), or mechanical and positive pressure ventilation (PPV) (D). Plots show the mean as well as 1 SD above and below the mean Correlation between NASA Task Load Index total score and self-reported number of patients requiring renal replacement therapy (A), vasoactive medications (B), extracorporeal membrane oxygenation (ECMO) (C), or mechanical and positive pressure ventilation (PPV) (D). Plots show the mean as well as 1 SD above and below the mean Figure 3 shows the collected data from the EHR (lab orders, blood draws, admissions, discharges, patient hours, and continuous medication infusions) as each relates to the NASA TLX score collected during a particular shift. Like the single-item stress score data, TLX scores were increased when the number of patient hours, admissions, discharges, continuous infusions (after a certain level) and blood draws increased. The total number of lab orders did not appear to affect the TLX scores. Fig. 3Correlation between NASA Task Load Index (TLX) total score and number of lab orders (A), number of blood draws (B), number of patients admitted (C), number of patients discharged (D), number of patient hours (E), or number of continuous infusions (F). Red line represents the local mean for each data set. The gold arrows represent threshold points Correlation between NASA Task Load Index (TLX) total score and number of lab orders (A), number of blood draws (B), number of patients admitted (C), number of patients discharged (D), number of patient hours (E), or number of continuous infusions (F). Red line represents the local mean for each data set. The gold arrows represent threshold points Single-item stress score regression analysis In the regression analysis, we evaluated four models that assessed the relationship between EHR data and single-item stress score (Table 1). Model 1 included the number of admitted patients. Model 2 included the number discharged. Model 3 included patient-hours (surrogate for census and overall patient demand). Model 4 included the number of continuous medication infusions. All models were adjusted for role, shift, and month (role with team-lead nurses was excluded because that scenario included only 7 observations). Multiple variables were associated with higher perceived stress level (as measured by the single-stress item): day shifts, more than 350 patient-hours per shift, and more than 100 continuous drips per shift. The thresholds for patient-hours and 100 continuous infusions were derived from these results. For every 10 additional patient hours exceeding 350, the single-stress level rose by an estimated 0.2 points (95% CI: 0.1–0.4 points, p = 0.001), and for every 10 additional continuous drips exceeding 100, the single-stress level rose by estimated 0.2 points (95% CI: 0.02–0.3, p = 0.032). Table 1Regression models of the relationship between electronic health record data and single item stress scoreModel 2Model 3Model 4Estimated slope (95% CI)Number admittedRoleFellow vs. charge RN0.1550.1310.148[-0.725, 1.035][-0.742, 1.003][-0.728, 1.024]MonthFeb vs Jan0.140-0.4900.120[-0.643, 0.922][-1.396, 0.417][-0.979, 1.218]Mar vs. Jan-0.562-0.600-0.567[-1.234, 0.110][-1.382, 0.182][-1.640, 0.506]April vs. Jan-0.326-0.392-0.301[-1.109, 0.458][-1.205, 0.420][-1.472, 0.870]Oct vs. Jan-0.792-0.787-0.371[-1.837, 0.253][-1.787, 0.214][-1.518, 0.776]Nov vs. Jan-0.538-0.504-0.0900[-1.450, 0.374][-1.390, 0.382][-1.162, 0.982]Dec vs. Jan0.09000.2140.125[-0.984, 1.164][-0.807, 1.235][-1.189, 1.439]ShiftPM vs. AM-0.880*-0.969***-0.754*[-1.585, -0.174][-1.505, -0.433][-1.354, -0.155]Number discharged0.0186[-0.142, 0.179]Patient-hours (per hour)Less than 350 patient-hours0.0000766[-0.0107, 0.0109]350 patient-hours or more0.0244**[0.00962, 0.0391]Number of continuous dripsLess than 100-0.0181[-0.0480, 0.0118]100 and more0.0174*[0.00153, 0.0332]Intercept5.021***5.079**6.361***[3.936, 6.105][1.709, 8.448][3.909, 8.813]Random-effect parametersNatural log of random effect SD for respondent -0.0208-0.0207-0.0205[-0.359, 0.318][-0.355, 0.314][-0.357, 0.316]Natural log of random effect SD for shift -0.161-0.232*-0.198[-0.375, 0.0529][-0.460, -0.00319][-0.420, 0.0231]Natural log of residual SD 0.325***0.326***0.325***[0.259, 0.392][0.259, 0.392][0.259, 0.392]N594594594AIC2278.72270.32275.1Regression models looking at the relationship between electronic health record data vs. single stress item:Model 1 includes number admittedModel 2 includes number dischargedModel 3 includes patient-hoursModel 4 includes number of continuous dripsAll models are adjusted for role, shift timing, and month (and role “Team RN” is excluded, since there were only 7 observations)95% confidence intervals in brackets* p < 0.05, ** p < 0.01, *** p < 0.001Interpretations: a number of variables are associated with higher stress level (as measured by the single stress item): AM shifts as well as more than 350 patient-hours per shift and more than 100 continuous drips per shift. For example, for every 10 additional patient hours exceeding 350, the single stress level goes up by estimated 0.2 points (95%CI: 0.1 to 0.4 points, p-value = 0.001) and for every 10 additional continuous drips exceeding 100, the single stress level goes up by estimated 0.2 points (95%CI: 0.02 to 0.3, p-value = 0.032) Table 1Regression models of the relationship between electronic health record data and single item stress scoreModel 2Model 3Model 4Estimated slope (95% CI)Number admittedRoleFellow vs. charge RN0.1550.1310.148[-0.725, 1.035][-0.742, 1.003][-0.728, 1.024]MonthFeb vs Jan0.140-0.4900.120[-0.643, 0.922][-1.396, 0.417][-0.979, 1.218]Mar vs. Jan-0.562-0.600-0.567[-1.234, 0.110][-1.382, 0.182][-1.640, 0.506]April vs. Jan-0.326-0.392-0.301[-1.109, 0.458][-1.205, 0.420][-1.472, 0.870]Oct vs. Jan-0.792-0.787-0.371[-1.837, 0.253][-1.787, 0.214][-1.518, 0.776]Nov vs. Jan-0.538-0.504-0.0900[-1.450, 0.374][-1.390, 0.382][-1.162, 0.982]Dec vs. Jan0.09000.2140.125[-0.984, 1.164][-0.807, 1.235][-1.189, 1.439]ShiftPM vs. AM-0.880*-0.969***-0.754*[-1.585, -0.174][-1.505, -0.433][-1.354, -0.155]Number discharged0.0186[-0.142, 0.179]Patient-hours (per hour)Less than 350 patient-hours0.0000766[-0.0107, 0.0109]350 patient-hours or more0.0244**[0.00962, 0.0391]Number of continuous dripsLess than 100-0.0181[-0.0480, 0.0118]100 and more0.0174*[0.00153, 0.0332]Intercept5.021***5.079**6.361***[3.936, 6.105][1.709, 8.448][3.909, 8.813]Random-effect parametersNatural log of random effect SD for respondent -0.0208-0.0207-0.0205[-0.359, 0.318][-0.355, 0.314][-0.357, 0.316]Natural log of random effect SD for shift -0.161-0.232*-0.198[-0.375, 0.0529][-0.460, -0.00319][-0.420, 0.0231]Natural log of residual SD 0.325***0.326***0.325***[0.259, 0.392][0.259, 0.392][0.259, 0.392]N594594594AIC2278.72270.32275.1Regression models looking at the relationship between electronic health record data vs. single stress item:Model 1 includes number admittedModel 2 includes number dischargedModel 3 includes patient-hoursModel 4 includes number of continuous dripsAll models are adjusted for role, shift timing, and month (and role “Team RN” is excluded, since there were only 7 observations)95% confidence intervals in brackets* p < 0.05, ** p < 0.01, *** p < 0.001Interpretations: a number of variables are associated with higher stress level (as measured by the single stress item): AM shifts as well as more than 350 patient-hours per shift and more than 100 continuous drips per shift. For example, for every 10 additional patient hours exceeding 350, the single stress level goes up by estimated 0.2 points (95%CI: 0.1 to 0.4 points, p-value = 0.001) and for every 10 additional continuous drips exceeding 100, the single stress level goes up by estimated 0.2 points (95%CI: 0.02 to 0.3, p-value = 0.032) Regression models of the relationship between electronic health record data and single item stress score Regression models looking at the relationship between electronic health record data vs. single stress item: Model 1 includes number admitted Model 2 includes number discharged Model 3 includes patient-hours Model 4 includes number of continuous drips All models are adjusted for role, shift timing, and month (and role “Team RN” is excluded, since there were only 7 observations) 95% confidence intervals in brackets * p < 0.05, ** p < 0.01, *** p < 0.001 Interpretations: a number of variables are associated with higher stress level (as measured by the single stress item): AM shifts as well as more than 350 patient-hours per shift and more than 100 continuous drips per shift. For example, for every 10 additional patient hours exceeding 350, the single stress level goes up by estimated 0.2 points (95%CI: 0.1 to 0.4 points, p-value = 0.001) and for every 10 additional continuous drips exceeding 100, the single stress level goes up by estimated 0.2 points (95%CI: 0.02 to 0.3, p-value = 0.032) NASA TLX score regression analysis The same four models as noted above were used to evaluate the TLX scores when compared to variables extracted from the EHR (Table 2). Several variables were associated with higher stress level as measured by the NASA TLX: day shifts, certain months (March-April and October-November were associated with less stress), more than 350 patient-hours per shift, and more than 100 continuous drips per shift. For every 10 additional patient hours exceeding 350, the TLX score rose by an estimated 4.2 points (95% CI: 1.1–7.3 points, p = 0.009) and for every 10 additional continuous drips exceeding 100, the TLX score rose by an estimated 3.6 points (95% CI: 0.2–6.9, p = 0.036). Table 2Regression models of the relationship between electronic health record data and Task Load Index total scoreModel 1Model 2Model 3Model 4Estimated slope (95% CI)Number admitted1.968[-1.595, 5.532]RoleFellow vs. charge RN1.0461.2300.7870.941[-11.89, 13.98][-11.54, 14.00][-12.10, 13.67][-12.03, 13.91]MonthFeb vs. Jan3.7740.460-9.8063.902[-13.55, 21.10][-15.78, 16.70][-29.07, 9.462][-19.06, 26.87]Mar vs. Jan-14.25-16.06*-17.50*-13.31[-29.34, 0.836][-30.09, -2.028][-34.26, -0.732][-35.78, 9.161]April vs. Jan-23.15**-26.01**-26.30**-20.97[-40.21, -6.086][-42.38, -9.636][-43.71, -8.897][-45.60, 3.656]October vs. Jan-15.83-22.88*-21.66*-10.45[-40.00, 8.348][-44.78, -0.983][-43.06, -0.252][-34.44, 13.54]Nov vs. Jan-20.51*-22.71*-21.16*-9.493[-39.53, -1.485][-41.61, -3.821][-39.95, -2.366][-31.85, 12.87]Dec vs. Jan11.255.82910.1112.83[-11.04, 33.53][-16.33, 27.99][-11.44, 31.65][-14.46, 40.13]ShiftPM vs. AM-22.63***-19.10*-26.79***-23.05***[-35.57, -9.688][-33.73, -4.479][-37.89, -15.70][-35.38, -10.72]Number discharged2.529[-0.849, 5.908]Patient-hoursLess than 350-0.0107[-0.238, 0.216]350 +0.419**[0.107, 0.732]Less than 100-0.529[-1.152, 0.0947]100 +0.357*[0.0229, 0.692]Intercept123.9***123.2***135.6***170.2***[101.7, 146.1][103.0, 143.4][65.60, 205.7][120.2, 220.2]Random-effect parametersNatural log of random effect SD for respondent 2.478***2.456***2.488***2.501***[1.980, 2.976][1.946, 2.966][1.991, 2.986][2.017, 2.986]Natural log of random effect SD for shift 2.959***2.956***2.925***2.923***[2.770, 3.147][2.766, 3.145][2.730, 3.119][2.729, 3.117]Natural log of residual SD 3.128***3.129***3.129***3.129***[3.047, 3.210][3.047, 3.210][3.047, 3.210][3.047, 3.210]N442442442442AIC4238.84237.84235.04235.3Residual intraclass correlationsStudy member0.14 [0.05, 0.31]0.13 [0.05, 0.31]0.14 [0.06, 0.32]0.15 [0.06, 0.32]Shift within study member0.50 [0.40, 0.60]0.49 [0.39, 0.59]0.49 [0.38, 0.59]0.49 [0.38, 0.59]Regression models of the relationship between electronic health record data and Task Load Index total scoreModel 1 includes number admittedModel 2 includes number dischargedModel 3 includes patient-hoursModel 4 includes number of continuous infusions. All models are adjusted for role, shift timing, and month (role “Senior RN” is excluded because there were only 5 observations)95% CI in brackets* p < 0.05, ** p < 0.01, *** p < 0.001 Table 2Regression models of the relationship between electronic health record data and Task Load Index total scoreModel 1Model 2Model 3Model 4Estimated slope (95% CI)Number admitted1.968[-1.595, 5.532]RoleFellow vs. charge RN1.0461.2300.7870.941[-11.89, 13.98][-11.54, 14.00][-12.10, 13.67][-12.03, 13.91]MonthFeb vs. Jan3.7740.460-9.8063.902[-13.55, 21.10][-15.78, 16.70][-29.07, 9.462][-19.06, 26.87]Mar vs. Jan-14.25-16.06*-17.50*-13.31[-29.34, 0.836][-30.09, -2.028][-34.26, -0.732][-35.78, 9.161]April vs. Jan-23.15**-26.01**-26.30**-20.97[-40.21, -6.086][-42.38, -9.636][-43.71, -8.897][-45.60, 3.656]October vs. Jan-15.83-22.88*-21.66*-10.45[-40.00, 8.348][-44.78, -0.983][-43.06, -0.252][-34.44, 13.54]Nov vs. Jan-20.51*-22.71*-21.16*-9.493[-39.53, -1.485][-41.61, -3.821][-39.95, -2.366][-31.85, 12.87]Dec vs. Jan11.255.82910.1112.83[-11.04, 33.53][-16.33, 27.99][-11.44, 31.65][-14.46, 40.13]ShiftPM vs. AM-22.63***-19.10*-26.79***-23.05***[-35.57, -9.688][-33.73, -4.479][-37.89, -15.70][-35.38, -10.72]Number discharged2.529[-0.849, 5.908]Patient-hoursLess than 350-0.0107[-0.238, 0.216]350 +0.419**[0.107, 0.732]Less than 100-0.529[-1.152, 0.0947]100 +0.357*[0.0229, 0.692]Intercept123.9***123.2***135.6***170.2***[101.7, 146.1][103.0, 143.4][65.60, 205.7][120.2, 220.2]Random-effect parametersNatural log of random effect SD for respondent 2.478***2.456***2.488***2.501***[1.980, 2.976][1.946, 2.966][1.991, 2.986][2.017, 2.986]Natural log of random effect SD for shift 2.959***2.956***2.925***2.923***[2.770, 3.147][2.766, 3.145][2.730, 3.119][2.729, 3.117]Natural log of residual SD 3.128***3.129***3.129***3.129***[3.047, 3.210][3.047, 3.210][3.047, 3.210][3.047, 3.210]N442442442442AIC4238.84237.84235.04235.3Residual intraclass correlationsStudy member0.14 [0.05, 0.31]0.13 [0.05, 0.31]0.14 [0.06, 0.32]0.15 [0.06, 0.32]Shift within study member0.50 [0.40, 0.60]0.49 [0.39, 0.59]0.49 [0.38, 0.59]0.49 [0.38, 0.59]Regression models of the relationship between electronic health record data and Task Load Index total scoreModel 1 includes number admittedModel 2 includes number dischargedModel 3 includes patient-hoursModel 4 includes number of continuous infusions. All models are adjusted for role, shift timing, and month (role “Senior RN” is excluded because there were only 5 observations)95% CI in brackets* p < 0.05, ** p < 0.01, *** p < 0.001 Regression models of the relationship between electronic health record data and Task Load Index total score Regression models of the relationship between electronic health record data and Task Load Index total score Model 1 includes number admitted Model 2 includes number discharged Model 3 includes patient-hours Model 4 includes number of continuous infusions. All models are adjusted for role, shift timing, and month (role “Senior RN” is excluded because there were only 5 observations) 95% CI in brackets * p < 0.05, ** p < 0.01, *** p < 0.001 Discussion Our results showed that perceived workload strain was higher during day shifts than during night shifts and was also higher for physician fellows than for charge nurses in the PICU. This finding could be related to the various levels of training of PICU fellows and that fellows are more responsible for the direct management of each patient. We found significant variation in median single-item stress and TLX scores over the course of each month that the data were collected, supporting the notion of variable workload strain in the PICU. Winter months exhibited the highest strain of the months where data was collected, which is consistent with previous trends of seeing higher peaks of viral illnesses during the winter months [12, 13]. Pediatric hospital admissions for respiratory failure, specifically due to bronchiolitis and pneumonia, also typically increase during winter months [14]. It is important to mention that this study took place during the “second wave” of COVID-19 in the winter 2021. Pediatric hospital admissions decreased during the COVID19 pandemic compared to pre-pandemic rates of hospital admission [14]. Pediatric healthcare utilization did increase during this study time period (October 2020-April 2021), but still remained below pre-pandemic levels [15]. Therefore, it is possible that patient volume and acuity are underestimated in this study. There are likely other seasonal factors that we did not clearly characterize that may have also contributed to our findings. In addition, retrospective data extracted from the EHR over the 43 shifts showed notable variation in all categories, including total number of blood draws and lab orders, admissions, discharges, total patient hours per shift, and medication infusions. The data confirm that the PICU is a dynamic and fluctuating environment, with frequently changing patient and environmental demands. The relationship between these factors and the patient-level characteristics (surrogates for acuity) support an overall trend toward increased workload strain when greater numbers of patients require renal replacement therapy and MV/NIPPV. These therapies require frequent evaluations by physician fellows for changes to therapies, such as electrolyte repletions, fluid balance for renal replacement and settings for mechanical ventilation. Charge nurses in our unit are often assisting bedside nurses with various tasks for mechanically ventilated patients, including sedation management and facilitating patient movement. The number of patients who required vasoactive medications or ECMO did not appear to influence perceived workload strain. In our unit, many vasoactive orders are nurse-titratable without constant clinician intervention. Furthermore, ECMO care in our PICU is determined by a thoroughly designed series of protocols and procedures carried out by an “ECMO specialist” at the bedside. Therefore, it is possible that the members of the care team we surveyed (fellows and charge nurses) were removed from the immediate stressor of caring for these patients. The unit-level factors of admissions, discharges and patient hours were associated with increased perceived workload strain, which were expected given the role of charge nurses and fellows in overseeing multiple patients and bed management. Total continuous infusions and number of blood draws were also associated with increased perceived workload strain, which may be due to the higher cognitive burden of interpreting and addressing labs as well as medication changes more frequently. Orders for lab tests were not associated with higher strain, but these orders are mostly performed by frontline providers, who were not included in this study. This study did not include personal factors that may influence perceived workload strain, such as coping styles or previous high-acuity experience, but understanding these factors will be important for future directions for managing workload strain. Our data are consistent with published findings from Tibby et al., which showed that perceived workload strain was increased when census and overall patient demand (acuity) were higher [5]. In our case, strain increased when patient-hours per shift were greater than 350. These results may indicate that a “breaking point” exists at which perceived strain starts to increase and that unit systems have a maximal capacity in which they can provide care before workload strain increases. A similar breaking point was noted for the overall number of medication infusions, which can serve as a possible surrogate for patient acuity, as patients who require more infusions are typically sicker than those who do not require continuous infusions. Limitations Our study had several limitations that are important to review. First and foremost, the study was conducted over multiple time periods during the COVID-19 pandemic. During this time healthcare workers experienced many unprecedented changes to their patient populations, workflows, and environments, among many other stressors. In addition, pediatric critical care is particularly susceptible to changes in patient populations based on seasonality, which affects patient makeup, volume, and acuity. This has been observed as an increase in the number of patients admitted to hospitals with respiratory diseases during particular seasons depending on geographic location [12, 16]. Furthermore, the generalizability of these findings is limited given the study was performed at a single institution and organizational differences in roles, such as ECMO specialists and nursing ratios, may vary by institution. Shift hours for nurses and physicians also vary per institution and may play a role in workload strain. Also, perceived workload strain is a somewhat subjective marker for unit strain that is susceptible to individual variation based on personal/professional experience and role within the unit. Previous work experience was not specifically analyzed in this study for nurses and physicians, which may affect perceived workload strain. Finally, during our data collection period our PICU was split into separate pediatric cardiac and pediatric non-cardiac ICUs, which required several adaptations from nurses, physicians, and other healthcare workers as they adjusted to the new system. While their roles were unchanged, it is unclear if this transition may have influenced the data. Conclusions Our study adds depth and granularity to the literature about variability in perceived workload strain in the PICU as well as factors contributing to strain felt by physicians and nursing leaders working in the PICU. In addition, our data showed that specific unit- and patient-level factors individually contributed to increased workload strain while others did not. Ideally, as we learn more about ICU strain, we can use this information to design systems that allow clinical resources to meet demand with a greater level of precision to improve patient care. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1