The Allocation of Intensivists’ Rounding Time under Conditions of Intensive Care Unit Capacity Strain

To the Editor:

With rising demand for critical care, intensivists’ time must increasingly be divided among patients (1–6). Recent studies suggest that increased strain at intensive care unit (ICU) admission leads to higher mortality in closed ICUs (7) and that increased strain at discharge leads to increases in ICU readmissions (8). These relationships between strain and outcomes could be mediated by strain-induced changes in the time intensivists devote to patients during patient care rounds (7–9). We therefore examined how the allocation of intensivists’ time during rounds changes at times of low versus high ICU strain and whether intensivists preferentially allocate time away from certain patient groups as strain increases. Some results have been previously reported in the form of an abstract (10).

Methods

We conducted a prospective study of patient care rounds in the 24-bed medical ICU of the Hospital of the University of Pennsylvania in 2012. Time spent performing various rounding activities was recorded in real time by trained data collectors, using a tablet computer. Methods for assessing interrater reliability can be found in the online supplement. Data collection was randomly assigned to one of two intensivist-led medical ICU teams each day and was not performed on weekends. Variables describing patient characteristics, staffing, and ICU strain were obtained from the electronic medical record and as part of a separate clinical trial (11).

Our analysis focused on “cognitive rounding time” (time spent on the patient’s assessment and plan) and on total rounding time (presentation of events and data, assessment and plan, and teaching related to that patient). Three validated strain variables (5) were considered: team census (“census”), representing the number of patients rounded on by the observed team each day; number of new admissions (“admissions”) since the end of rounds the previous day; and average severity of illness (“acuity”) of patients on the team, using Acute Physiology and Chronic Health Evaluation III (APACHE III) scores (12).

We constructed explanatory linear mixed-effects models for cognitive and total rounding time for each patient-day. Patients were treated as random clusters, cumulative days hospitalized in the ICU were included as a random slope and as a linear term, and attending was treated as an indicator variable (13). Patient race was obtained from the electronic medical record and could be reported by either patient or provider. We considered three race categories: black, nonblack (white or Asian), and unknown. Table 1 and Table E1 in the online supplement describe all evaluated covariates.

Table 1.

Descriptive Statistics

Variables	Percentage or Median (IQR) (N = 566)
Patient variables
Patient age, yr	60.5 (48.5–70.1)
Sex
Male	43.1
Female	56.9
Race
White	50.5
Black	38.5
Asian	2.2
Race indeterminate	8.8
ICU admission source
Emergency department	50.7
General floor	24.2
Another ICU in same hospital	2.8
Another hospital	16.8
Direct admission	5.5
Night admission (4:00 p.m. to 7:00 a.m.)
No	39.1
Yes	60.9
Is a readmission
No	90.1
Yes	4.9
ICU admission Acute Physiology and Chronic Health Evaluation III score	74 (51–99)
Day variables
Team census	11 (10–13)
Number of new admissions	2 (1–3)
Average severity of illness	85.5 (78.1–91.3)
Cognitive rounding time, min	91.9 (77.9–107.3)
Total rounding time, min	188.6 (164.8–212.6)
Total teaching time, min	7.3 (3.9–12.9)
Maximum team size, number of people	12 (10.5–13)
Patient-day variables
Day in ICU stay	4 (1–9)
Family on rounds
No	69.3
Yes	30.7
Patient located outside of medical ICU
No	91.7
Yes	8.3
Outcome variables
ICU mortality	15.6
Hospital mortality	20.7
ICU readmission*	4.1
Hospital readmission†	32.0
ICU length of stay, d	3.7 (1.8–8.5)

Definition of abbreviations: ICU = intensive care unit; IQR = interquartile range.

^*Within 2 calendar days of ICU discharge.

^†Within 90 days of hospital admission.

We constructed separate models to determine whether time was allocated away from specific patient groups as strain increased by exploring interactions between the three strain variables and the following six patient variables: admission status (new admission vs. follow-up), patient race (black vs. nonblack), age (continuous), severity of illness (continuous), family presence on rounds, and patient sex. We then constructed a fully adjusted model with all interactions having a P value < 0.2 and used backward selection, removing nonsignificant terms (14). We used Holm tests of conditional significance given the multiple comparisons made (15). Additional details regarding the statistical analyses are available in the online supplement.

Results

Rounds were observed for 566 patients over the course of 114 noncontiguous weekdays, for a total of 1,295 patient-days. Intensivists rounded on a median of 11 patients (interquartile range [IQR], 10–13) each day, including two new admissions (IQR, 1–3). Median daily rounding time was 188.6 minutes (IQR, 164.8–212.6 min); 91.9 minutes (IQR, 77.9–107.3 min) were spent on cognitive rounding time (Table 1). Daily rounding time increased as census (6.3 min; 95% confidence interval [CI], 2.4–10.1 min; P = 0.002) and admissions (6.0 min; 95% CI, 0.6–11.4 min; P = 0.031) increased (Figure E1 in the online supplement); cognitive rounding time increased as census increased (2.5 min; 95% CI, 0.1–4.9 min; P = 0.045).

In fully adjusted models, with increasing daily admissions, newly admitted patients received 1.38 fewer minutes (95% CI, −2.43 to −0.33 min; P_Holm = 0.002) total rounding time (interaction P value = 0.01) and 0.73 fewer minutes (95% CI, −1.42 to −0.07 min; P_Holm = 0.0113) cognitive rounding time per additional admission. No significant changes occurred among follow-up patients (interaction P value = 0.030; Figure E2). As census increased, each unit increase led to a 0.5-minute (95% CI, 0.87–0.13 min; P_Holm = 0.0135) decrease in cognitive rounding time among new admissions, with no decrement among follow-ups (interaction P value = 0.028).

The effect of census on total rounding time was modified by new admission status and race. A three-way interaction (P = 0.04) revealed that among follow-ups, nonblack patients received 3.4 minutes (95% CI, −5.6 to −1.2 min; P_Holm = 0.02) more than blacks at low census (eight patients); however, the excess time spent with nonblacks disappeared as census increased (P < 0.01). In contrast, no significant differences in strain-induced decrements in rounding time were observed between black and nonblack new admissions (P = 0.22; Figure 1). These relationships persisted in two sensitivity analyses, excluding patients of indeterminate race or excluding Asians from the nonblack group (Table E2).

Figure 1.

Total rounding time. Models are adjusted for acuity, severity of illness measured on Day 1 of the first intensive care unit (ICU) stay, day number in ICU course, attending, data collector, order patient was rounded on (inverse), maximum team size, attending’s first day (yes/no), patient age, and presence of family on rounds. Mixed-effects model with patient identification treated as a random intercept, day in ICU course treated as a random slope, unstructured covariance matrix specified, and variance residuals. Model with three-way interaction excludes patients of indeterminate race.

Neither patient age, sex, acuity, and severity of illness nor the presence of family on rounds affected the allocation of rounding time.

Discussion

This study provides the first description of how ICU physicians allocate rounding time among patients and how this allocation changes as ICUs become strained. Daily rounding time increased with increases in census and admissions, but less time was spent per patient, primarily affecting new admissions and nonblack follow-up patients. These findings are consistent with studies showing that clinicians perceive their time to be highly constrained (1, 5, 6). The observation that strain preferentially affected new admissions and nonblack follow-ups may reflect the fact that these patients received more time in general, such that further reductions were challenging.

Importantly, we found that increases in ICU strain did not result in disproportionate decreases in the time allocated to other patient subgroups, suggesting that ICU physicians generally ration their time equitably. Although total rounding time was allocated away from nonblack follow-ups as census increased, the facts that nonblacks received more time overall and that similar patterns were not observed among new admissions casts doubt on this finding, representing a true racial disparity.

This study had several limitations. First, data were not collected outside of morning rounds, and therefore we could not assess how strain affected time allocation at other times. Second, although a differential effect of census on time allocation was not found between newly admitted black and nonblack patients, future research should determine whether a larger sample would reveal a significant disparity. Third, severity of illness was assessed only at ICU admission, limiting severity adjustment on subsequent days; however, bias introduced by inadequate severity adjustment is unlikely to be differential across different levels of census, and therefore it is unlikely to have affected the results. Residual confounding could still be present, as severity of illness may be indirectly affected by census. Finally, data capture was not formally evaluated; however, interrater reliability was excellent.

In summary, this study provides a description of how intensivists allocate their time among patients as their workloads increase, providing objective confirmation of the common perception that time is a scarce resource. However, as a single-center study, these results may not generalize to other ICUs. In addition, because we often lacked data on rounding time on the same patient over contiguous days, we could not address whether observed decreases in rounding time mediated previously observed relationships between strain and outcomes or whether they represent improved efficiency. Future research is needed to explore these questions (7, 8).