Goals of Care Assessment during Hospitalization for Sepsis

Abstract

Rationale

Sepsis is a common cause of hospital mortality, as well as new morbidity among survivors. Clinical practice guidelines recommend assessing goals of care (GoCs) during sepsis hospitalization to ensure goal-concordant care is provided.

Objectives

To determine how often GoCs are assessed during sepsis hospitalization in routine practice.

Methods

We conducted a cohort study of adult patients hospitalized with community-onset sepsis at 66 hospitals (2020–2023) participating in the Michigan Hospital Medicine Safety Consortium’s sepsis initiative. The primary outcomes were GoC discussion documented in the health record and GoC assessment inferred to have occurred on the basis of 1) documented GoC discussion, 2) treatment limitations on admission, 3) treatment limitations initiated during hospitalization, 4) palliative care consultation, or 5) discharge to hospice. We examined the incidence of GoC discussion and GoC assessment among all sepsis hospitalizations, as well as in three subgroups defined by advanced age or health impairment; admission to intensive care; and presentation with shock or respiratory failure. We also evaluated factors associated with GoC discussion/assessment and quantified variation across hospitals using multilevel logistic regression.

Results

Among 18,711 patients in the Michigan Hospital Medicine Safety Consortium sepsis initiative registry, 54.0% had advanced age or health impairment, 22.5% were admitted to intensive care, and 10.2% presented with shock or respiratory failure. GoC discussion and assessment occurred in 35.7% and 45.3% of patients, respectively, in the overall cohort. GoC discussion occurred within 3 days of presentation in 23.8%. Age, race, chronic conditions, preexisting cognitive impairment, preexisting functional limitation, admission to intensive care, and receipt of life support were each associated with GoC assessment. Median odds ratios for GoC discussion and assessment were 2.30 and 2.09, respectively, indicating substantial cross-hospital variation after accounting for patient characteristics. GoC discussion and assessment were more common among patients with advanced age or health impairment, admission to intensive care, or presentation with shock or respiratory failure, but cross-hospital variation was similar.

Conclusions

GoC assessment did not occur consistently during sepsis hospitalization in this multihospital cohort, even among higher-risk patients, and there was marked variation in practice across hospitals. Future work is needed to better understand what drives high performance in assessing GoC.

Sepsis is a leading cause of hospital mortality, contributing to up to half of all inpatient deaths in the United States (1, 2). Moreover, patients who survive sepsis often experience new or worsened physical, cognitive, or emotional morbidity, as well as heightened risk of hospital readmission and death (3). Although default medical practice in the United States maximizes longevity, many patients prefer to optimize function or comfort (4). For this reason, sepsis clinical practice guidelines recommend discussing goals of care (GoCs) in all patients hospitalized for sepsis and suggest discussing GoCs within the first 3 days of hospitalization (5). However, it is unclear how often this occurs in routine clinical practice. A recent study suggests that GoCs may be assessed in only a minority of sepsis hospitalizations, but the study was limited by the relatively small sample size and focus on patients enrolled in a clinical trial (6). Thus, it remains unclear how often GoCs are assessed during hospitalization for sepsis in routine practice.

In this study, we sought to determine how commonly GoCs were assessed in a large, multihospital cohort of sepsis hospitalizations. We also identified factors associated with GoC assessment and evaluated variation in practice across hospitals. Because GoC discussions are not always documented in the health record, we used a composite measure to infer whether GoCs were assessed during hospitalization.

Methods

Study Design and Setting

This is a multihospital cohort study of adult patients hospitalized for community-onset sepsis at 66 hospitals participating in the Michigan Hospital Medicine Safety Consortium (HMS). HMS is a collaborative quality initiative sponsored by Blue Cross Blue Shield of Michigan with the goal of improving management and outcomes of hospitalized medical patients (7, 8). The HMS sepsis initiative (HMS-sepsis) launched in 2021 and expanded to all HMS hospitals by March 2023. Professional abstractors at each hospital enter data on sepsis hospitalizations into the central HMS-sepsis registry using structured abstraction forms and standard definitions. Data collection focuses on patient characteristics, sepsis management, and patient outcomes. Baseline functional status is evaluated on the basis of clinical documentation of limitations in the following activities and instrumental activities of daily living: bathing, dressing, toileting, transferring, walking, and managing medications independently. HMS uses a variety of processes to ensure the accuracy of abstracted data, including centralized abstractor training, automated data checks, and routine auditing of abstracted data.

A random sample of sepsis hospitalizations from each hospital is selected for inclusion into the HMS-sepsis registry on the basis of a two-step process. Hospitalizations are first identified by primary discharge diagnoses of sepsis, infection, or respiratory failure, then screened in random order for clinical evidence of infection and acute organ dysfunction during the first 2 days of the encounter, consistent with the Centers for Disease Control and Prevention’s Adult Sepsis Event surveillance criteria (9). This two-step approach ensures a consistent threshold for inclusion in the HMS-sepsis registry despite any differences in recognition or coding of sepsis across hospitals or over time (10, 11). Additional information on the HMS-sepsis inclusion criteria, exclusion criteria, and sampling strategy are provided in the data supplement.

Outcomes

The primary outcomes of this study were 1) documented GoC discussion during sepsis hospitalization and 2) GoC assessment during sepsis hospitalization. GoCs have been defined as “the overarching aims of medical care for a patient that are informed by the patients’ underlying values and priorities, established within the existing clinical context, and used to guide decisions about the use of or limitation(s) on specific medical interventions” (12).

GoC discussions were abstracted from the health record. Specifically, abstractors were directed to identify “any clinical documentation that a conversation occurred between the care team and the patient and/or family about the patient’s overall focus of care (e.g., curative or life-prolonging versus comfort-focused)” and to record the date that GoCs were first discussed. However, because GoC discussions are not always documented (13), we inferred that GoCs were assessed during hospitalization if any of the following criteria occurred: 1) documented GoC discussion, 2) treatment limitations present at time of admission to the hospital, 3) new treatment limitations initiated during hospitalization, 4) palliative care consultation during hospitalization, or 5) discharge to hospice services. We reasoned that each of these criteria indicate that GoCs may have been assessed during the hospitalization, so they provide a liberal assessment of whether GoCs were assessed. Treatment limitations present on hospital admission could have been long-standing treatment limitations or, alternatively, limitations initiated in the emergency department or during the hospital admission process. However, because our data do not distinguish between new versus long-standing treatment limitations at hospital admission, we included treatment limitations present on hospital admission in our composite outcome. From here, we use “GoC discussion” to refer to a documented GoC discussion and “GoC assessment” to refer to the five-part composite outcome. In sensitivity analyses, we considered alternative, more stringent composite definitions for GoC assessment using two to four of the five criteria.

Statistical Analysis and Study Cohorts

We evaluated study outcomes among the entire HMS-sepsis cohort and in three subcohorts of interest: 1) patients with advanced age or health impairment, defined as age ⩾80 years, select advanced chronic health conditions, or baseline functional impairment, as defined by two or more of the following limitations: inability to bathe, dress, toilet, transfer, walk, or manage medications independently; 2) patients admitted to intensive care; and 3) patients presenting with shock or respiratory failure (see the data supplement for further details on subcohort definitions). We reasoned that GoC assessment may be particularly important in patients with advanced age or health impairment because these patients may be more likely to prioritize function or comfort over survival. We also reasoned that GoC assessment is warranted for all patients admitted to an intensive care unit or initiated on life support.

We also tested for differences in GoC assessment by specific patient characteristics, including age, sex, race, and specific comorbid conditions, using chi-square tests. We assessed independent associations between patient characteristics and outcomes using multivariable logistic regression. In our primary analysis, race was categorized as White, Black, and other because of sample size. However, in post hoc exploratory analyses, we considered more granular categorizations of race/ethnicity (White/Caucasian, Black/African American, Asian, Native American, Hispanic/Latino, Middle Eastern, Native Hawaiian or Pacific Islander, other, unknown) and explored associations with GoC assessment using both single-level and multilevel regression models.

To quantify variation across hospitals, we fit multilevel logistic regression models with hospitalizations nested within hospitals, adjusting for patient characteristics (14, 15). We excluded a single hospital with fewer than 10 hospitalizations from just this analysis. We calculated the intraclass correlations (ICCs) to quantify the proportion of variation in GoC discussion and GoC assessment attributable to the hospital where the patient was treated. An ICC value close to 0 indicates a low proportion of overall variance explained at the hospital level, and a value close to 1 indicates nearly all variation is explained at the hospital level (16). We also calculated median odds ratios (MORs) for GoC discussion and assessment (15). A MOR of 1.0 indicates that odds of GoC discussion or GoC assessment are equivalent across hospitals; the larger the MOR, the greater the importance of hospital-level effects to differences in GoC discussion or GoC assessment. Finally, in a post hoc exploratory analysis, we examined differences in hospital characteristics among hospitals in the top quintile, middle three quintiles, and bottom quintile of GoC assessment.

Data management and analysis were performed in SAS 14.1 (SAS Institute) and Stata/SE 16.1 (StataCorp). We considered a two-sided P < 0.05 to be significant. The study was reviewed by the University of Michigan Institutional Review Board and deemed exempt because HMS is a quality improvement initiative.

Results

Study Cohort

During the study period (June 2020 through September 2023), there were 18,711 hospitalizations for community-onset sepsis in the HMS-sepsis registry. The median age was 71 years (interquartile range, 60–80), and 9,500 (51.0%) were male. A total of 2,664 (14.3%) were admitted from a healthcare facility (long-term acute care, skilled nursing, or subacute rehabilitation), and 7,147 (38.2%) had baseline functional impairment present on admission to the hospital. In-hospital mortality occurred in 2,344 (12.5%) and 90-day mortality in 4,743 (25.3%). A total of 10,095 (54.0%) had advanced age or health impairment, 4,204 (22.5%) were admitted to intensive care, and 1,905 (10.2%) presented with shock or respiratory failure. Characteristics of the overall cohort and subcohorts are presented in Table 1 and Table E1, respectively.

Table 1.

Patient and hospitalization characteristics

Characteristics	Overall Cohort (N = 18,711)
Age, yr, median (IQR)	71 (60–80)
Male sex, n (%)	9,500 (51.0)
Race, n (%)
White	14,995 (80.1)
Black	2,777 (14.8)
Other	385 (2.1)
Unknown	554 (3.0)
Place of residence prior to hospitalization, n (%)
Home	14,544 (78.1)
Healthcare facility	2,664 (14.3)
Unknown	1,329 (7.1)
Other	86 (0.5)
Chronic conditions, n (%)
Chronic kidney disease	7,312 (39.1)
Chronic pulmonary disease	5,706 (30.5)
Congestive heart failure	4,701 (25.1)
Baseline cognitive impairment	3,561 (19.0)
Malignancy (metastatic, nonmetastatic, or hematologic)	4,055 (21.7)
Dementia	2,461 (13.2)
Chronic liver disease	1,397 (7.5)
Left ventricular ejection fraction ⩽39%	776 (4.1)
Severe/critical aortic stenosis	242 (1.3)
Baseline functional limitations*, n (%)
0 limitations	11,564 (61.8)
1 limitation	865 (4.6)
2 or more limitations	6,282 (33.6)
Severity of illness during hospitalization
Receipt of renal replacement therapy	1,070 (5.7)
Receipt of advanced respiratory support	2,686 (14.4)
Receipt of vasopressor therapy	4,621 (24.7)
Receipt of cardiopulmonary resuscitation	623 (3.3)
Intensive care unit use during hospitalization
Admitted to intensive care from emergency department	4,204 (22.5)
Admitted to intensive care later in hospitalization	5,977 (31.9)
Never admitted to intensive care	12,734 (68.1)
Outcomes
In-hospital mortality	2,344 (12.5)
90-d mortality	4,743 (25.3)
Discharge to healthcare facility	3,083 (16.5)
Discharge to hospice care	1,324 (7.1)

Definition of abbreviation: IQR = interquartile range.

^*Baseline functional limitations include inability to bathe, dress, toilet, transfer, walk, and manage medications independently.

Incidence of GoC Assessment

Among the overall cohort, 8,477 (45.3%) had a GoC assessment during hospitalization (37.0–39.6% by alternative, more stringent definitions), including 6,684 (35.7%) with a GoC discussion and 4,458 (23.8%) with a GoC discussion during the first 3 calendar days of presentation (Table 2). Among 10,095 patients with advanced age or health impairment, 57.9% had GoC assessment (46.5–49.9% by alternative definitions), including 45.0% with a GoC discussion and 30.5% with a GoC discussion within 3 days. Among 4,204 patients admitted to intensive care, 59.8% had GoC assessment (53.9–56.8% by alternative definitions), including 52.3% with GoC discussion and 35.7% with GoC discussion within 3 days. Among 1,905 patients admitted with shock or respiratory failure, 65.8% had GoC assessment (58.5–62.2% by alternative criteria), including 57.3% with GoC discussion and 41.3% with GoC discussion within 3 days.

Table 2.

Goals of care assessment among overall and high risk cohorts

	Overall Cohort (N = 18,711)	Advanced Age or Health Impairment (n = 10,095)	Admitted to Intensive Care (n = 4,204)	Presented with Shock or Respiratory Failure (n = 1,905)
Primary analysis
GoC assessment* (criteria 1, 2, 3, 4, or 5), n (%)	8,477 (45.3)	5,841 (57.9)	2,514 (59.8)	1,254 (65.8)
(1) Documented GoC discussion*, n (%)	6,684 (35.7)	4,544 (45.0)	2,198 (52.3)	1,091 (57.3)
(2) Treatment limitations on presentation, n (%)	2,907 (15.5)	2,317 (23.0)	597 (14.2)	287 (15.1)
(3) Treatment limitation initiated during hospitalization, n (%)	4,310 (23.0)	3,040 (30.1)	1,602 (38.1)	828 (43.5)
(4) Palliative care consultation, n (%)	3,634 (19.4)	2,670 (26.4)	1,251 (29.8)	588 (30.9)
(5) Discharged to hospice, n (%)	1,324 (7.1)	1,058 (10.5)	357 (8.5)	153 (8)
Sensitivity analyses with alternate composite outcomes for composite GoC assessment
GoC assessment (criteria 1, 3, 4, or 5), n (%)	7,414 (39.6)	5,038 (49.9)	2,388 (56.8)	1,184 (62.2)
GoC assessment (criteria 1, 4, or 5), n (%)	6,948 (37.1)	4,722 (46.8)	2,269 (54)	1,118 (58.7)
GoC assessment (criteria 1 or 4), n (%)	6,916 (37.0)	4,695 (46.5)	2,264 (53.9)	1,115 (58.5)

Definition of abbreviation: GoC = goal of care.

Documented GoC discussions occurred within the first 3 calendar days of the presentation in 23.8% of the overall cohort, 30.5% of patients with advanced age or health impairment, 35.7% of patients admitted to intensive care, and 41.4% presenting with shock or respiratory failure.

^*Primary study outcomes.

Variation in GoC Assessment by Patient Characteristics

GoC discussion and GoC assessment varied across patient subgroups defined by age, sex, comorbid conditions, functional impairment, admission to intensive care, and illness severity (Tables E1 and E3). In multivariable models, older age, non-White race, chronic conditions, baseline cognitive impairment, baseline functional impairment, admission to intensive care, and receipt of life support were each independently associated with GoC discussion and GoC assessment (Table 3). In post hoc exploratory analysis using more granular race/ethnicity categories, only Hispanic/Latino ethnicity was associated with lower odds of GoC assessment (adjusted odds ratio, 0.76; 95% confidence interval, 0.62–0.93; P = 0.008) (Table E4). However, in multilevel models adjusting for treating hospital, this association was no longer significant (Table E4).

Table 3.

Patient and hospitalization characteristics associated with goals of care assessment

	aOR for GoC assessment	95% CI	P Value
Age, yr
18–39	0.41	(0.34, 0.50)	<0.001
40–59	0.72	(0.65, 0.81)	<0.001
60–69	Ref
70–79	1.59	(1.44, 1.75)	<0.001
80+	3.95	(3.56, 4.38)	<0.001
Male sex	1.08	(1.01, 1.16)	0.025
Race
White	Ref
Black	0.78	(0.70, 0.86)	<0.001
Other/unknown	0.79	(0.67, 0.93)	0.004
Chronic conditions
Moderate or severe kidney disease	1.09	(1.01, 1.17)	0.023
Congestive heart failure	1.06	(0.97, 1.16)	0.171
Baseline cognition abnormality	1.76	(1.57, 1.97)	<0.001
Chronic pulmonary disease	1.10	(1.04, 1.21)	0.004
Dementia	1.68	(1.47, 1.92)	<0.001
Malignancy	1.96	(1.80, 2.13)	<0.001
Left ventricular ejection fraction <39%	1.37	(1.14, 1.64)	0.001
Liver disease	1.33	(1.17, 1.52)	<0.001
Severe/critical aortic stenosis	1.26	(0.92, 1.71)	0.147
Baseline functional limitations, n (%)
No limitations	Ref
One limitation	1.02	(0.86, 1.20)	0.858
Two or more limitations	1.73	(1.59, 1.87)	<0.001
Severity of illness during hospitalization
Receipt of renal replacement therapy	1.31	(1.12, 1.53)	<0.001
Receipt of advanced respiratory support	2.2	(1.97, 2.56)	<0.001
Receipt of vasopressor therapy	1.89	(1.70, 2.12)	<0.001
Receipt of cardiopulmonary resuscitation	1.57	(1.26, 1.96)	<0.001
Intensive care unit use during hospitalization
Admitted to intensive care from emergency department	0.77	(0.68, 0.88)	<0.001
Admitted to intensive care later in hospitalization	1.91	(1.67, 2.19)	<0.001
Never admitted to intensive care	Ref

Definition of abbreviations: aOR = adjusted odds ratio; CI = confidence interval; GoC = goal of care; Ref = reference.

Baseline functional limitations include inability to bathe, dress, toilet, transfer, walk, or manage medications independently.

Variation in GoC Assessment by Hospital

GoC discussion varied more than eightfold across hospitals, ranging from 7.7% to 87.9% (median, 32.5%), whereas GoC assessment varied more than fourfold, ranging from 19.2% to 91.1% (median, 41.8%) (Figure 1). Palliative care consultation ranged from 0.0% to 52.2% (median, 16.9%), with seven (10.6%) hospitals having no palliative care consults. The median odds ratio for GoC discussion in the adjusted model was 2.30 (ICC, 0.189) overall and 2.09–2.39 (ICC, 0.154–0.202) across subgroups. The median odds ratio for GoC assessment was 2.09 (ICC, 0.153) overall, and 1.86–2.06 (ICC, 0.114–0.148) across subgroups (Table 4). None of the measured hospital characteristics (bed size, urbanicity, ownership, teaching status, and self-reported structures/processes for documenting and communicating GoC) were associated with GoC assessment (Table E5).

Figure 1.

Variation in goals of care (GoCs) assessment by hospital. This figure depicts the percentage of sepsis hospitalizations with GoC discussion during the first 3 encounter days (dark blue), GoC discussion at a later point (medium blue), or other evidence that GoC assessment occurred (light blue). Each bar represents one hospital, ranked from highest (left) to lowest (right) percentage of hospitalizations with GoC assessment.

Table 4.

Variability in goals of care discussion and assessment attributable to treating hospital

Study Cohort	GoC Discussion		GoC Assessment
	MOR	ICC	MOR	ICC
Overall cohort	2.30	0.189	2.09	0.153
Advanced age or health impairment	2.09	0.154	1.86	0.114
Admitted to intensive care	2.25	0.179	2.06	0.148
Presented with shock or respiratory failure	2.39	0.202	1.93	0.127

Definition of abbreviations: GoC = goals of care; ICC = interclass correlation coefficient; MOR = median odds ratio.

ICC is calculated as Covariance Parametric Estimate/(Covariance Parametric Estimates + 3.29). MOR is calculated as exp(sqrt[2 × Covariance Parametric Estimate *0.4549]).

Discussion

In this multihospital cohort of over 18,000 adult hospitalizations for community-onset sepsis at 66 hospitals, GoC discussion was documented in less than 40% of hospitalizations for community-onset sepsis. Even with the most liberal measurement definition, GoC assessment occurred in less than half of sepsis hospitalizations. Among patients with stronger indications for GoC assessment (patients with advanced age or health impairment, admitted to intensive care, or presenting with shock or respiratory failure), incidence of GoC discussion and assessment were higher (45–57% for GoC discussion; 55–66% GoC for assessment)—but still not universal.

Multiple patient characteristics were associated with GoC assessment and GoC discussion, including older age, White race, preexisting chronic conditions, functional impairment, cognitive impairment, admission to intensive care, and receipt of life support. However, in adjusted analyses accounting for differences in patient characteristics, there was residual variation in GoC discussion and GoC assessment across hospitals. Odds of GoC discussion and GoC assessment were, on average, about twofold higher among higher-assessing versus lower-assessing hospitals, and approximately 15% of the variation in GoC discussion and GoC assessment was explained by the treating hospital.

Our findings are consistent with two prior studies showing inconsistent assessment of GoCs among patients hospitalized for sepsis. In the Improving Morbidity and Post-Acute Care Transitions for Sepsis (IMPACTS) trial cohort, which enrolled approximately 700 patients with community-onset sepsis at three U.S. hospitals, only 19% had GoCs documented in the standardized care alignment tool used at these hospitals (6). Second, in a prospective cohort of 1,024 patients referred to intensive care at a tertiary care hospital in Australia during 2012–2013, 63% had life-limiting illnesses; yet, only 34% of patients with life-limiting illnesses had GoCs documented (36% among patients admitted for sepsis) (17).

The inconsistent assessment of GoCs has important implications. Early elicitation of GoC has been associated with better quality of life, lower use of nonbeneficial medical care near death, reduced costs, and positive family outcomes (18). The 2021 Surviving Sepsis Campaign Guidelines recommended discussing goals of care and prognosis with all patients and their families, acknowledging it may improve concordance between medical treatment and patient values (5). Finally, multicomponent sepsis transitional care programs prioritizing care alignment, including palliative care referral when appropriate, have been associated with lower postdischarge mortality and readmissions (19, 20). Our study adds to the literature suggesting that GoC assessment occurs less commonly than is recommended.

An additional important finding of our study was the marked variation in GoC discussion and assessment across hospitals. The treating hospital was associated with a twofold difference in the odds of GoC assessment and GoC discussion during sepsis hospitalization in adjusted models that accounted for differences in patient characteristics across hospitals. Furthermore, there was also potential unwarranted variation in GoC discussion and assessment by patient characteristics, including lower odds of GoC assessment in patients of non-White race. In exploratory analyses using more granular race/ethnicity data, this finding was driven by lower GoC assessment among Hispanic/Latino patients. The finding was no longer significant after adjusting for treatment hospital, indicating the disparity was driven by hospital-level differences in GoC assessment (i.e., Hispanic/Latino patients more likely to be treated at a hospital with lower GoC assessment). This place-of-care disparity is a form of systemic racism (21) and is consistent with prior work showing that racial disparities in hospital care are driven largely by where minority patients receive care (22). It is also consistent with prior work showing racial disparities in sepsis care specifically driven by place of treatment (23). Future work to improve GoC assessment should focus not only on improving overall rates of GoC assessment but also on reducing cross-hospital disparities.

Inclusion of patients and families in GoC discussions was identified as the top priority among patients and family stakeholders in a recent prioritization of research topics for hospital medicine (24), underscoring the importance of mitigating the observed cross-hospital disparities on GoC assessment. We hypothesize that there may be specific hospital structures and resources that explain some of the observed variation. For example, availability of palliative care consultation may increase GoC assessment. Palliative care consultation varied markedly across hospitals in our cohort and was not used at all in seven hospitals. Future work is needed to identify scalable hospital structures and resources that facilitate consistent GoC assessment.

Limitations

Our study should be interpreted in the context of some limitations. First, we measured GoC assessment on the basis of data abstracted from the electronic health record, which may not perfectly reflect GoC assessment. Electronic health record documentation of GoC discussions is often incomplete (25), GoCs may be recorded in variable locations within the health record (26, 27), and failure to document important GoC discussions has been identified as a key concern among primary care providers assuming care of patients after critical illness (13). Indeed, for approximately 10% of patients in our study, GoC assessment was inferred in the absence of a documented GoC discussion on the basis of new treatment limitations, palliative care consultation, or discharge to hospice. Second, our study assessed the incidence of documented GoC discussions but did not consider the quality of GoC discussions or reasons why GoCs were not assessed during hospitalization. Many patients in our cohort were admitted from postacute care facilities or were hospitalized in the months before the index sepsis hospitalization. GoCs may have been discussed in the weeks preceding sepsis hospitalization, such that repeat discussion was not required, but data on recent GoC discussion was not available. Third, although we noted variable use of palliative care consultation across hospitals, we do not have information on palliative care availability across hospitals to understand the drivers of observed variation in palliative care use.

Conclusions

In this multihospital cohort, we found that GoCs were assessed in less than half of patients hospitalized for community-onset sepsis. Several patient and hospitalization characteristics were associated with GoC assessment, but, even after adjustment for these patient characteristics, there remained wide variation in GoC assessment across hospitals. Our findings suggest a need for interventions to improve GoC assessment among patients hospitalized with sepsis.