An Observational Study of Decision Making by Medical Intensivists

Abstract

Objectives:

The ICU is a place of frequent, high-stakes decision making. However, the number and types of decisions made by intensivists have not been well characterized. We sought to describe intensivist decision making and determine how the number and types of decisions are affected by patient, provider, and systems factors.

Design:

Direct observation of intensivist decision making during patient rounds.

Setting:

Twenty-four-bed academic medical ICU.

Subjects:

Medical intensivists leading patient care rounds.

Intervention:

None.

Measurements and Main Results:

During 920 observed patient rounds on 374 unique patients, intensivists made 8,174 critical care decisions (mean, 8.9 decisions per patient daily, 102.2 total decisions daily) over a mean of 3.7 hours. Patient factors associated with increased numbers of decisions included a shorter time since ICU admission and an earlier slot in rounding order (both p < 0.05). Intensivist identity explained the greatest proportion of variance in number of decisions per patient even when controlling for all other factors significant in bivariable regression. A given intensivist made more decisions per patient during days later in the 14-day rotation (p < 0.05). Female intensivists made significantly more decisions than male intensivists (p < 0.05).

Conclusions:

Intensivists made over 100 daily critical care decisions during rounds. The number of decisions was influenced by a variety of patient- and system-related factors and was highly variable among intensivists. Future work is needed to explore effects of the decision-making burden on providers’ choices and on patient outcomes.

Medical decision making is an inherently complex process, particularly among critically ill patients in whom multiple organ systems and social circumstances must be considered simultaneously (1–4). To date, only one study has characterized the frequency with which decisions are made in the ICU (5). Although this study suggests intensivists experience a significant decision-making burden, it did not use standardized assessments across respondents and did not assess factors associated with this burden.

It is possible that patient, provider, and systems variables influence the number of decisions made daily by intensivists. For example, recent studies have documented relationships between ICU strain or occupancy and decisions to admit or discharge patients (6–9). Similar relationships have not been explored across the broader realm of ICU decision making.

ICU decision making is also a key contributor to quality of care. Several studies identify decision making as central to patient and family satisfaction (10, 11). In a survey of family members of ICU patients, good decision making was a key determinant of family satisfaction (12). Following discharge from the ICU, disappointment with clinical decision making was a predictor of dissatisfaction with ICU care (13). Although these studies focused on the broad family experience and not individual physician decision making, they highlight the importance of physician decision making in patient- and family-centered outcomes.

It is possible that the decision-making burden among intensivists could have untoward implications for the quality of care. Nonmedical evidence indicates that repeated decision making alters the decision-making process and influences subsequent choices in subconscious ways (14, 15). As a first step toward understanding how intensivist decision making may impact patient care, we sought to describe the intensivists’ decisions, including their frequency, type, and variability among patients and physicians. Patient outcomes and quality of decision making were not assessed.

Preliminary results of this study were previously reported in abstract form (16).

MATERIALS AND METHODS

This study was conducted in the Hospital of the University of Pennsylvania’s Medical ICU (MICU), a 24-bed closed unit in which critical care physicians (intensivists) manage all patients in 14-day rotation blocks. Patients are divided between two care teams based on admitting schedule. Each team comprises one intensivist, one critical care fellow, six medical residents, and one advanced nurse practitioner or physician’s assistant. During morning rounds, the intensivist-led team establishes each patient’s daily care plan. During this period, patients are discussed in a nonrandom order based on patient and work-flow influences.

We directly observed morning rounds for 80 days between August 27, 2012, and January 12, 2013. We observed 85% of intensivists who attended in the MICU that academic year, excluding the study’s senior author. All observed intensivists provided written informed consent prior to observation. No intensivists declined participation. We restricted observation to days 2–14 of each service block and excluded days staffed by a covering provider not scheduled for the entire 14-day block. Finally, we selected 4–5 days of observation for each of 17 attendings based on investigator availability.

Study data were collected and managed using Research Electronic Data Capture (REDCap) software (Vanderbilt University, Nashville, TN) (17). All data were collected by one of three investigators. We established satisfactory interrater reliability (κ > 0.7) prior to study initiation during which all three investigators observed rounds simultaneously and compared coding after rounds completion. Investigators recorded the number and types of critical care decisions, rounding duration, patient order, and family presence on rounds. A decision was recorded when an action plan was verbalized following a clearly stated problem. Confirming objective data or stating adherence to ICU policies was not recorded as decisions. Because ultimate responsibility falls to the attending, all decisions verbalized by the team were attributed to the intensivists. This included decisions made exclusively by the attending and decisions verbalized by other team members while the intensivist was present. Observation was restricted to verbal decisions without inclusion of electronic orders.

All decisions were recorded for each patient rounded on during an observation day. Decisions were sorted into 18 categories with 126 subcategories (Table 1). The decision matrix was created during observations prior to the pilot phase by author M.S.M. The decision matrix was finalized prior to any data collection and not modified during the study period. The timing of each decision was recorded, and each decision was coded as either a decision to continue or change the current therapy.

TABLE 1.

Decisions Were Classified Into 18 Categories and 126 Subcategories

Decision Category	Decision Subcategory
Ancillary management	Occupational therapy Physical therapy Swallowing evaluation
Code management	Cardioversion Chest compressions Initiate or terminate code Intubation Line placement Needle decompression Thrombolysis
Code status	Code status decision (not family discussion)
Diagnostic: imaging studies	CT scan Echocardiogram Electroencephalogram Electrocardiogram Electromyogram MRI Nuclear study Transcranial Doppler ultrasound Ultrasound Radiograph
Disposition decision	End-of-life care provisions (not code status) Transfer out of ICU
Family meeting	Family meeting: goals of care Family meeting: information sharing
Hematology management	Cryoprecipitate transfusion Fresh-frozen plasma transfusion Platelet transfusion RBC transfusion
Hemodynamic management	Bicarbonate administration Cardiac pacing Cardioversion Fluid management: colloid Fluid management: crystalloid Goal: fluid status Goal: physiologic variable
Infection management	Source control: drain management Source control: line management Source control: surgery
Laboratory testing	Blood panel: one time Blood panel: repeated measurement Blood single test: one time Blood single test: repeated measurement Fecal study Fluid (other) test Urine single test Urine panel test
Medications	Antacid medication Antiseizure medication Anticoagulation: prophylactic Anticoagulation: therapeutic Antiemetics Antihypertensive Antimicrobial management Antipsychotic/delirium management Anxiolytic Chronotropic medication Code medication Diabetic management Diuretic therapy Electrolyte management Hepatorenal syndrome treatment Hypnotic Immunosuppression Inhaler therapy Inotropic medication Other Pain management Renal medication Sedation Steroids Tachycardia: rate control Tachycardia: rhythm control Vaccine Vasopressor Vitamin
Nursing/general care	Activity level Bladder pressure Drain management Foley tube Ostomy management Rectal tube Wound care
Nutrition/electrolytes	Enteral Free water Nil per os Oral Total parenteral nutrition
Obtain information	Consultant Family/friend Nonphysician (ethics, legal) Outpatient provider Outside hospital clinician
Other	Other (free text)
Procedure	Arterial line Biopsy: other tissue Bronchoscopy Bronchoscopy Central venous line Central venous line Chest tube Cooling protocol Desensitization Endoscopy Hemodialysis line Hyperbaric therapy Inferior vena cava filter placement Lumbar puncture Nasogastric tube/lavage Orogastric tube Other Paracentesis Percutaneous endoscopic gastronomy tube Pulmonary artery catheter Thoracentesis Tracheostomy
Pulmonary management	Extracorporeal membrane oxygenation Extubate Intubate Invasive ventilator management Noninvasive ventilator management Oxygen supplementation Paralysis Positioning to improve oxygenation Pulmonary vasodilator Respiratory therapy: mechanical intervention
Renal replacement therapy	Continuous renal replacement Hemodialysis: intermittent

We recorded physician characteristics, including age, race, gender, and years since fellowship training, and patient-level factors, including Acute Physiology and Chronic Health Evaluation (APACHE) III score (18) and ICU admission source. One patient without an APACHE III score was excluded from all analyses.

During our study period, a randomized trial of intensivist nighttime staffing was simultaneously ongoing (19). Given the potential impact on daytime intensivists’ decision making, we recorded overnight coverage status for the night prior to all observed rounds and included it as a covariate in all models. After approximately one third of observations, we began collecting whether a family member was present on rounds, hypothesizing that family presence may influence intensivist decision making.

Physician and patient demographic data are described as number (percent) for categorical data and median (interquartile range, IQR) or mean (SD) for nonnormally and normally distributed continuous data, respectively. Decisions were evaluated and analyzed for each patient-day (decisions made for a single patient on 1 d) and total day (including all patients the intensivist rounded on that day).

Rounding order was coded based on the timing of the first decision made for each patient. This included rare occasions (n = 23 patient-days) when a patient was briefly discussed by the intensivist with another provider (resident or consultant) but was not formally rounded on until later. In a sensitivity analysis in which these 23 patient-days were instead coded as when the full rounding actually occurred, the results were unchanged (20,21).

APACHE III scores reflect the patient’s predicted probability of dying during the hospitalization based on data available on the day of ICU admission (18). Because our patient-day observations were not limited to patients’ admission days, we created a time variable equal to the number of days since admission. To provide a more discrete analysis of how length of stay might influence decisions, we divided this variable into five intervals: 1) day 1, 2) days 2–3, 3) days 4–5, 4) days 6–9, and 5) day 10 and beyond. This categorical variable was entered as a covariate in all patient-day models. Patients who had been discharged from the ICU previously and were readmitted without leaving the hospital had APACHE III adjustment based on their index ICU stay. These patients were flagged with an additional “readmission” variable to provide further risk adjustment

The primary outcome for day-level (n = 80) analyses was total decisions per day per intensivist. The primary outcome for patient-day analyses (n = 920) was decisions per patient per day. We specifically sought to evaluate how the total number of new patients affected decision making. Measures of severity of illness were included in all models. Other independent variables were included if they were clinically relevant and associated with the outcome in bivariate analyses at p value less than or equal to 0.2. These variables were added in a step-wise fashion based on association strength and removed if their addition reduced the explained variance (20, 21).

After confirming appropriate normality linear regression was used for day-level and patient-level analyses. Physicians were modeled as a fixed effect in patient-and day-level analyses to adjust for within physician correlation of the number of decisions made per patient and to prevent confounding by practice differences among physicians. Because some patients who were observed multiple times contributed multiple patient-days, patient-level clustered analyses were performed but did not alter results (data not presented). All models use robust variance estimators (20,21). In day-level analyses, physicians were modeled as random effects to enable assessments of how measured characteristics influenced the number of decisions made each day. In secondary analyses, we modeled physician as a fixed effect in day-level analyses. Restricted analyses were performed for observations when family presence was recorded (620 of 920 patient observations). Reported p values are two sided. Statistical analyses were performed using Stata 12.1 software (Stata Datacorp, College Station, TX). The institutional review board at the University of Pennsylvania approved this study.

RESULTS

Seventeen intensivists were observed on 80 days (Table 2). Intensivists were observed a median of 4 days (IQR, 4–4.5) each. Intensivists were a median of 40.1 years old (IQR, 37–52) and had a median of 6 years (IQR, 3–22) post-fellowship experience. Four (23.5%) intensivists were females.

TABLE 2.

Intensivist Characteristics

Characteristica	All Intensivists (n = 17)
Age	40.1 (37–52)
Race
White	14 (82)
Asian	2 (12)
Black	1 (6)
Male	13 (76)
Years since training completion	6 (3–22)
Weeks of medical ICU service in year prior	4 (2–5)

^aCategorical data presented as number (%); continuous data presented as median (interquartile range).

We observed 920 patient rounds representing 374 unique patients (Table 3), such that each patient was observed a median of three times (IQR, 1–4). Patients were a median of 62 years old (IQR, 50–69). The majority of patients were either white (49.0%) or black (40.6%). Patients had a median APACHE III score of 75 (IQR, 56–99) and were observed a median of 4 days (IQR, 1–12) since initial ICU admission. ICU and hospital mortality were 17.3% and 20.2%, respectively.

TABLE 3.

Patient Characteristics

Characteristica	All Patients (n = 374)
Age	61.8 (49.8–69.0)
Race
White	183 (49)
Black	152 (40.6)
Asian	6 (1.6)
Unknown	22 (5.9)
Other	11 (2.9)
Male	220 (58)
Acute Physiology and Chronic Health Evaluation III Score	75 (56–99)
Unit prior to medical ICU
Emergency room	166 (44.5)
General floor	161 (43.0)
Other ICU	22 (6.0)
Outside hospital	22 (6.0)
Direct admission	3 (0.5)
ICU mortality	64 (17.3)
Hospital mortality	75 (20.2)
ICU length of stay (d)	3.4 (1.7–76)

^aCategorical data presented as number (%); continuous data presented as median (interquartile range).

Intensivists provided nocturnal coverage the night prior to observed rounds for 39 of observed days (49%). The first observation day for each intensivist was a median of 2 days (IQR, 2–3) into the 14-day block. Family presence was assessed on 55 days, representing 620 of 920 patient rounds (67%).

Decisions

We recorded 8,174 critical care decisions. On average, intensivists made 102.3 ± 25.9 decisions daily with mean rounding duration of 3.7 ± 0.92 hours. Number of daily decisions and rounding duration were weakly correlated (r = 0.29). Mean daily census included 11.5 ± 1.5 patients, resulting in roughly nine decisions made per patient per day. There were 2.0 ± 1.5 new patients per day. The most frequent decision category was medication management, representing 36.4% of decisions (Fig. 1). Other common decisions included obtaining input from another provider (9.7%), laboratory testing (9.4%), hemodynamic management (7.7%), and pulmonary management (7.4%). Of 18 decision categories, 10 accounted for fewer than 5% of decisions each (Fig. 1). Heat map depiction of decision frequency by category and physician is available in Supplement Figure 1 (Supplemental Digital Content 1, http://links.lww.com/CCM/B302).

Figure 1.

Types and frequency of all decisions. Other = decisions marked as “other” during initial coding or categories with less than 1% of all decisions: hematologic management, infection management, code status decision, and code management.

Decision Analysis

Patient and Day Level.

In bivariable linear analyses, patient factors independently associated with increased numbers of decisions included being a new admission, a shorter interval since initial admission, and an earlier location in rounding order (all p < 0.05). Patients with higher APACHE III scores had slightly but nonsignificantly more decisions (p = 0.08). Physician factors independently associated with increased decisions per patient in bivariable analyses included being younger and being female (all p < 0.05). Physician age and years after fellowship experience were highly correlated; thus, only age was included in analyses. Physicians made more decisions per patient later in their rounding block (p < 0.05). A figure depicting decisions per patient by rounding day is available in Supplemental Figure 2 (Supplemental Digital Content 2, http://links.lww.com/CCM/B303). Total patient census, nocturnist presence the night prior, and family presence did not significantly alter the number of decisions made per patient.

In multivariable linear models, after including physician as a fixed effect, each additional variable improved the amount of explained variance (Table 4). The complete model accounted for 24% of observed variance as measured by R². Intensivist identity explained a greater proportion of the variance in the number of decisions per patient than did any other measured factor. The predicted number of decisions per patient varied greatly among physicians when all other variables included in the multivariable linear model were held at their mean values (Fig. 2). Seven of 17 intensivists (41%) made at least four additional decisions per patient compared with the reference physician. In modeling the physician as a random effect, physician gender was the only factor that remained significant. Female intensivists made significantly more decisions per patient (β-coefficient, 2.77; p < 0.001).

TABLE 4.

Patient- and Day-Level Multivariable Linear Predictive Model (n = 920)

Variable	β-Coefficient (95% CI)	p
Intensivista
1	Reference	Reference
2	0.56 (−0.45 to 1.57)	0.27
3	0.43 (−0.74 to 1.61)	0.47
4	0.77 (−0.45 to 2.00)	0.21
5	2.29 (1.21–3.38)	0.00
6	2.31 (0.98–3.63)	0.00
7	2.30 (1.01–3.58)	0.00
8	2.66 (1.75–3.57)	0.00
9	3.31 (1.94–4.67)	0.00
10	3.62 (2.45–4.78)	0.00
11	3.70 (2.27–5.12)	0.00
12	4.09 (2.72–5.46)	0.00
13	4.16 (2.71–5.61)	0.00
14	4.71 (3.41–6.00)	0.00
15	5.30 (3.89–6.71)	0.00
16	4.92 (3.47–6.39)	0.00
17	6.23 (4.65–7.82)	0.00
Acute Physiology and Chronic Health Evaluation IIIb	0.02 (0.01–0.02)	0.00
Days since initial ICU admissionc
0–1	Reference	Reference
2–3	−1.79 (−2.55 to −1.03)	0.00
4–5	−1.51 (−2.46 to −0.56)	0.00
6–10	−1.66 (−2.50 to −0.82)	0.00
>10	−2.96 (−3.75 to −2.17)	0.00
Day of service blockd	0.08 (0.00–0.16)	0.03
Location in rounding ordere	−0.15 (−0.22 to −0.07)	0.00
Total new patientsf	0.02 (−0.22 to 0.22)	0.02

^aIntensivist 1 is reference intensivist, β-coefficient represents change in number of decisions made per patient comparing each intensivist to intensivist 1.

^bFor every 10-point increase in Acute Physiology and Chronic Health Evaluation III score, intensivists made significantly more decisions.

^cIntensivists make fewer decisions daily for patients who have been in the medical ICU the longest.

^dIntensivists make more decisions per patient for patients later in their rotation block.

^ePatients rounded on later in the day have fewer decisions per patient.

^fFor each additional increase in total new patients, the number of decisions per patient increases.

Figure 2.

Variation in decisions per patient by intensivist for the average patient Each point represents the intensivis’s predicted number of decisions per patient for the average patient. All included variables are held at their sample means. The model includes factors from bivariable analysis that had a p value of less than 0.2 and variables that were hypothesized to increase the number of decisions per patient-day: physician identity, Acute Physiology and Chronic Health Evaluation III score, time since admission, day in the intensivis’s rounding block, location in the daily rounding order, and total new patients on the team.

Day Level.

Intensivists made an average of 102.3 ± 25.9 decisions daily during 80 observation days. In bivariable analyses, physician identity was significantly associated with the number of decisions (p < 0.001) (Table 5) and explained a greater proportion of the variance in the number of decisions per day than did any other measured factor. Additional physician factors independently associated with increased total daily decisions included younger age and female gender (both p < 0.05). The intensivists weeks of MICU service, years of post-fellowship experience, and nocturnal coverage the night prior were not significantly associated with the number of decisions per day.

TABLE 5.

Day-Level Multivariable Linear Predictive Model (n = 80)

Variable	β-Coefficient (95% CI)	p
Intensivista
1	Reference	Reference
2	8.44 (−12.17 to 29.05)	0.42
3	−0.73 (−22.89 to 21.42)	0.95
4	5.99 (−19.03 to 31.01)	0.63
5	22.14 (6.87–37.41)	0.01
6	20.41 (0.26–40.57)	0.05
7	21.31 (7.33–34.28)	0.01
8	23.68 (9.23–38.12)	0.01
9	33.28 (15.69–50.87)	0.01
10	29.56 (13.12–45.99)	0.01
11	35.09 (17.76–52.41)	0.01
12	34.92 (18.15–51.69)	0.01
13	39.76 (19.19–60.33)	0.01
14	43.79 (16.64–70.94)	0.01
15	53.61 (28.10–79.12)	0.00
16	49.80 (22.32–77.27)	0.00
17	62.32 (48.58–76.06)	0.51
Mean Acute Physiology and Chronic Health Evaluation IIIb	−1.08 (−7.21 to 5.05)	0.72
Day of service blockc	0.67 (−0.52 to 1.86)	0.27
Total new patientsd	7.62 (4.22–11.03)	0.00

^aIntensivist 1 is reference intensivist, β-coefficient represents change in number of decisions daily comparing each intensivist to intensivist 1.

^bThe number of decisions made daily does not significantly change for every 10-point increase in the team’s mean Acute Physiology and Chronic Health Evaluation III score.

^cThe number of decisions daily does not vary significantly on different days of the service block when adjusting for other variables.

^dThe total daily decisions significantly increase for each additional increase in total new patients.

The complete model at the day level accounted for a greater proportion of variance in decisions made (R² = 0.63) than the patient-day model. Physicians made more decisions when there was a larger census (p < 0.05). The team’s mean APACHE III score was not significantly associated with the number of decisions per day. In modeling the physician as a random effect, female physicians made significantly more decisions (β-coefficient, 28.01; p = 0.002).

Impact of Family Presence.

Family presence was assessed for 620 of all observed patient rounds (67%). In a restricted patient-level model of only these observations, the explained variance was unchanged (R² = 0.24). Family presence was associated with a small and not statistically significant increase in the number of decisions made per patient (0.64 decisions per patient per day; p = 0.06). Inclusion of this variable did not significantly alter the other model coefficients.

Decisions by Clinical Impact.

Decisions were also divided into high and low clinical impact. Code status, disposition, family meetings, invasive hemodynamics, procedures, pulmonary management, and the use of continuous renal replacement therapy were all considered high impact decisions. These decisions represented 985 of the total decisions observed (12%). In day-level analysis (n = 80), the number of high impact decisions significantly decreased with more years of experience (coefficient, −0.10; p = 0.04) and significantly increased with a larger patient census (coefficient, 1.02; p = 0.001). All other day-level analyses were insignificant.

DISCUSSION

Through direct observation of decision making, this study suggests that intensivists make more than 100 critical care decisions daily during morning rounds. This number underestimates the true decision-making burden, given that critical care decisions are routinely made at other times of the day. Although it is difficult to compare this decision-making burden with those of other professionals, recent decision research reveals that even modest levels of repeated decision making impair self-control and alter subsequent choices (14, 15).

This study captures decision-making quantity but cannot comment on decision quality. Decisions were not linked to patient outcomes. The most frequent decisions included managing medications, obtaining further expert opinion, and laboratory monitoring. Decision categories that represented smaller proportions of the total may nonetheless be important and may even signal areas for potential improvement. For example, of 8,174 recorded decisions, intensivists made only 157 family meeting decisions and 145 ancillary support (occupational therapy and physical therapy) decisions. It is unlikely that these specific decision types were uniformly made during parts of the day not observed, and the ICU under study did not have protocols addressing these issues during the study period. It is possible that these categories were crowded out by seemingly more time-sensitive decisions being made in large volume for patients.

A key finding of this study is the dramatic variability among physicians in the same ICU in the numbers of decisions made daily and per patient. Indeed, the physician’s identity explained a greater proportion of the variance in decisions made per patient than any patient characteristic. Prior studies have also noted variation in physician decision making and practice patterns with regard to such decisions as ICU admission, intubation, and withdrawal of life-sustaining therapy (22–24). Other single-center studies have shown that physician-attributable differences in care provision account for significant variation in ICU spending and in ICU-based limitations on life support (24, 25). Similar to our study, these data confirm physician variation in important decision-making tasks, which may manifest as variable rounding durations and, perhaps, the number of decisions vocalized for any given patient.

Of note, female intensivists made significantly more decisions than males, even when controlling for other provider-specific variables. Research in other fields has noted that women differ in their approach to decision making (26, 27). Beyond quantity of decisions, our study was not designed to further elucidate this difference, though it is an interesting point for future study.

A second important finding was that the presence of intensivist coverage during the night prior to morning rounds did not significantly influence the number of decisions made either per patient or per day. This study cannot assess differences in decision quality with nocturnal coverage, but similar number of decisions may help explain why the presence of nighttime intensivists also did not lead to changes in patient outcomes (18).

Third, many patient characteristics influenced the number of decisions made by intensivists. More time since admission, likely a marker of decreased acuity, was associated with fewer decisions. Interestingly, however, the first rounding encounter with a new patient was also associated with fewer decisions. This somewhat surprising result may reflect an initial lack of familiarity with the patient or a tendency to focus on the acute care issues. Future research is needed to determine whether intensivists may intentionally defer specific types of decisions (such as end-of-life decisions) until later in a patient’s course when trust and familiarity have been developed.

Fourth, we found that intensivists tended to make more decisions for patients when a family member was present during rounds, though this finding was not significant (p = 0.07). As family presence on rounds is increasingly encouraged, it is important to determine whether this finding is replicable and, if so, what the consequences of this additional decision making maybe for patients and family members.

A key strength of this study is the use of direct observation of multiple providers on multiple days. We recorded and cataloged a large repository of decisions with very detailed categorization. Alternative approaches, such as capturing recorded decisions in the electronic medical record, would miss a large number of decisions, particularly those for which physicians considered an issue but actively chose to not alter therapy. Although direct observation is resource intensive and subject to observer interpretation, we established good interrater reliability (κ > 0.7) during a pilot study prior to data collection.

There are several important limitations to this study. First, intensivists were observed in a single academic medical center in the United States. Intensivist decision making may differ in other settings. Second, because observations were limited to patient care rounds, we missed decision making at other points in the day. However, ICU team rounds represent the majority of patient care discussion. This study likely quantifies a realistic lower bound of the decision-making burden. Third, the order of rounding was not random. New admissions, unstable patients, and potential discharges influenced this order. Caution is needed in interpreting the finding that patients later in the rounding order had fewer decisions. Fourth, as with many studies of patients evaluated after their ICU admission day, risk-adjustment scores including APACHE III may be less calibrated or discriminant than desired. We attempted to overcome this by adjusting for the interval between the rounding day and the ICU admission day, but residual confounding may persist.

Fifth, although this study was designed to provide a quantitative lower bound of intensivists’ decision-making burden, we cannot determine whether this burden is too great. Intensivists were not asked to provide self-evaluations of whether they found the decision making to be particularly burdensome, as this would have revealed our study hypotheses. Finally, each recorded decision required attention and evaluation by the observers, a cognitive burden that may itself have altered the data.

Future studies are needed to confirm our findings and explore their clinical implications. In addition to decision number, researching decision-making characteristics such as time to change in management for a particular therapy may help to understand differences in provider approaches to critical care (24, 25). Additionally, understanding how care protocols affect and potentially alleviate provider decision-making burdens would be beneficial.

CONCLUSIONS

Intensivists make more than 100 critical care decisions each day. This decision-making burden is influenced by a variety of patient- and service-related factors and is highly variable among intensivists. Future work is needed to replicate these findings and determine the consequences of this decisional workload for the types of choices made and the outcomes of patients, family members, and the physicians themselves.