Association of Time of Day with Delays in Antimicrobial Initiation among Ward Patients with Hospital-Onset Sepsis

Abstract

Rationale

Although the mainstay of sepsis treatment is timely initiation of broad-spectrum antimicrobials, treatment delays are common, especially among patients who develop hospital-onset sepsis. The time of day has been associated with suboptimal clinical care in several contexts, but its association with treatment initiation among patients with hospital-onset sepsis is unknown.

Objectives

Assess the association of time of day with antimicrobial initiation among ward patients with hospital-onset sepsis.

Methods

This retrospective cohort study included ward patients who developed hospital-onset sepsis while admitted to five acute care hospitals in a single health system from July 2017 through December 2019. Hospital-onset sepsis was defined by the Centers for Disease Control and Prevention Adult Sepsis Event criteria. We estimated the association between the hour of day and antimicrobial initiation among patients with hospital-onset sepsis using a discrete-time time-to-event model, accounting for time elapsed from sepsis onset. In a secondary analysis, we fit a quantile regression model to estimate the association between the hour of day of sepsis onset and time to antimicrobial initiation.

Results

Among 1,672 patients with hospital-onset sepsis, the probability of antimicrobial initiation at any given hour varied nearly fivefold throughout the day, ranging from 3.0% (95% confidence interval [CI], 1.8–4.1%) at 7 a.m. to 13.9% (95% CI, 11.3–16.5%) at 6 p.m., with nadirs at 7 a.m. and 7 p.m. and progressive decline throughout the night shift (13.4% [95% CI, 10.7–16.0%] at 9 p.m. to 3.2% [95% CI, 2.0–4.0] at 6 a.m.). The standardized predicted median time to antimicrobial initiation was 3.2 hours (interquartile range [IQR], 2.5–3.8 h) for sepsis onset during the day shift (7 a.m.–7 p.m.) and 12.9 hours (IQR, 10.9–14.9 h) during the night shift (7 p.m.–7 a.m.).

Conclusions

The probability of antimicrobial initiation among patients with new hospital-onset sepsis declined at shift changes and overnight. Time to antimicrobial initiation for patients with sepsis onset overnight was four times longer than for patients with onset during the day. These findings indicate that time of day is associated with important care processes for ward patients with hospital-onset sepsis. Future work should validate these findings in other settings and elucidate underlying mechanisms to inform quality-enhancing interventions.

Sepsis is a leading cause of mortality and healthcare costs among hospitalized patients, accounting for approximately one in two in-hospital deaths and $7.4 billion in healthcare costs annually (1–3). To date, research and quality improvement efforts have primarily focused on patients presenting to the emergency department with community-onset sepsis. Among this population, delays in antimicrobial initiation are associated with increased mortality (4–9). Although hospital-onset sepsis accounts for 11–13% of sepsis cases and 22% of sepsis-related deaths (10) and is associated with longer antimicrobial delays and higher mortality than community-onset sepsis (10–13), the mechanisms underlying treatment delays for hospital-onset sepsis have not been explored.

One factor that may contribute to delays in antimicrobial initiation is the time of day. The time of day is associated with suboptimal care delivery in the outpatient setting, including for ordering and completion of influenza vaccination and age-appropriate cancer screening and appropriate statin, antibiotic, and opioid prescribing (14–22). In the surgical literature, nighttime procedure start times are associated with increased patient morbidity and mortality (23).

These time-of-day effects have several potential explanations, including decision fatigue, a decrement in the ability to make appropriate decisions resulting from the cumulative burden of prior decision making (24, 25), disruptions in normal workflows due to competing scheduled obligations, or staffing changes, among others. Although each of these mechanisms is relevant to time-sensitive care provided to hospitalized patients, time-of-day associations have not been studied among hospitalized patients with sepsis. Clinician identification and management of sepsis may be particularly prone to these influences, given that sepsis is a heterogeneous clinical syndrome with high diagnostic uncertainty and requires a rapid and coordinated multidisciplinary response.

We sought to assess the association of the time of day with antimicrobial initiation among ward patients with hospital-onset sepsis. We hypothesized that the probability of antimicrobial initiation would decrease over the course of clinicians’ shifts, similar to declines in optimal care practices observed in other clinical settings. Some of the results of this study have been previously reported in the form of abstracts (26, 27).

Methods

Study Design, Participants, and Setting

We performed a retrospective cohort study of all ward patients with a hospital-onset sepsis episode admitted to five acute care hospitals with different patient demographics and case mixes (quaternary referral center; tertiary referral center; urban, academic-affiliated community hospital; two rural/suburban community hospitals) in the University of Pennsylvania Health System between July 1, 2017, and December 31, 2019. Typical shift structure for ordering clinicians and nurses at the study hospitals includes a shift change at 7 a.m. and 7 p.m.

Data Source

We extracted electronic health record data from the Epic Clarity Database, a clinical data warehouse integrated across the University of Pennsylvania Health System hospitals.

Sepsis Definitions

For the primary analysis, we implemented the Centers for Disease Control and Prevention Adult Sepsis Event (ASE) criteria to identify sepsis cases in an approach previously implemented by our group (28), operationalized by: 1) a collected blood culture; 2) broad-spectrum antimicrobial initiation within 48 hours before or after the blood culture collection day and continued for ⩾96 hours; and 3) new organ dysfunction within the same ±48-hour window as measured by a simplified version of the Sequential Organ Failure Assessment optimized for electronic health record data score ⩾1 (see eMethods, Table E1, and Figure E1 in the data supplement) (29, 30). Sepsis onset was defined as the time that the earliest of these three criteria was met in accordance with prior definition implementation (29–33). To distinguish hospital-onset from community-onset sepsis, we excluded sepsis episodes with onset within the first 48 hours of hospitalization (30). We included the first episode of hospital-onset sepsis for any given hospitalization. We excluded patients with sepsis onset in nonward locations (e.g., procedural suites, intensive care units) to examine care delivery on units with similar operations and patient acuity and excluded patients with sepsis onset in wards with fewer than five cases of hospital-onset sepsis over the study period (n = 37) to minimize bias related to clinician inexperience with sepsis.

To assess the robustness of our findings given the heterogeneity in existing sepsis definitions (34), we conducted sensitivity analyses among a cohort of ward patients with hospital-onset sepsis based on the Third International Consensus Definition for Sepsis (Sepsis-3) criteria (eMethods) (31, 32).

Study Variables

In the primary analyses, we tested for an association between the time of day and initiation of antimicrobial treatment. The primary exposure was the hour of day (e.g., 7 a.m., 8 a.m., 9 a.m., etc.), modeled as a categorical variable, with 7 a.m. as the reference for the start of the workday. The primary dependent variable was a binary indicator for the receipt of a new broad-spectrum intravenous antimicrobial medication during that hour (eMethods and Table E1). We grouped antimicrobial initiation by hour (e.g., antimicrobial initiation between 7 a.m. and 7:59 a.m. was grouped to 7 a.m.) (14, 15).

In secondary analyses, we tested for associations between the time of sepsis onset with several measures of delays in antimicrobial initiation. The exposure variable for these analyses was the hour of day of sepsis onset. Secondary outcomes included a continuous variable for the number of hours from sepsis onset to antimicrobial initiation and binary variables for antimicrobial initiation within 1 and 3 hours of sepsis onset (a Surviving Sepsis Campaign recommendation [35] and a component of the Centers for Medicare and Medicaid Services SEP-1 quality measure [36], respectively) and in-hospital mortality. Exact antimicrobial administration times were used for secondary outcomes.

Covariates for multivariable model adjustment included hospital, year (to account for secular trends in sepsis care), quarter (to account for seasonal trends in sepsis epidemiology), service type (medicine vs. surgery), age, self-reported gender, race, and ethnicity, and admission diagnosis category (from International Classification of Diseases, Tenth Revision codes). For the in-hospital mortality model, we explicitly did not adjust for antimicrobial initiation, as we hypothesized this to be in the causal pathway of the relationship between time of day of sepsis onset and mortality.

Analyses

For the primary analysis, we fit a discrete-time time-to-event model using a generalized linear model with a binomial outcome distribution, complementary log-log link, and hours from sepsis onset as the time axis, to assess the association between the hour of day and antimicrobial initiation at each hour among patients with hospital-onset sepsis. The hour-of-day exposure included hours from the time of sepsis onset through the hour of first antimicrobial administration, and thus observations were censored after the outcome of antimicrobial initiation was met. We selected discrete-time time-to-event models for two reasons. First, this modeling strategy accounts for interval-censored data, which is necessary because we are modeling antimicrobial initiation at the hour level rather than continuously. Second, when the complementary log-log estimates are exponentiated, they can be interpreted as the relative risk of antimicrobial initiation given that the patient meets the criteria for sepsis and has not yet received antimicrobials (37). We estimated the standardized probability of antimicrobial initiation for patients with hospital-onset sepsis at each hour of day by marginalizing adjusted estimates over the distribution of covariates, holding time from sepsis onset constant at 1 hour. Clinically, this can be interpreted as the predicted probability of antimicrobial initiation at any given hour of day for patients with hospital-onset sepsis 1 hour after sepsis onset.

In secondary analyses, we first fit quantile regression models at the median time to antimicrobial initiation to assess the association of time of day of sepsis onset with time to antimicrobial initiation measured in hours, and multivariable logistic regression models to assess the association of time of day of sepsis onset with odds of antimicrobial initiation within 1 and 3 hours of sepsis onset and in-hospital mortality. We estimated standardized probabilities of initiation within 1 and 3 hours and mortality for each hour of day of sepsis onset by marginalizing adjusted estimates over the distribution of covariates.

Finally, to assess whether the observed associations persisted among patients with high severity of illness (who we hypothesized would receive rapid antimicrobial initiation regardless of time of day), we repeated all analyses stratified by SOFA score at sepsis onset <75th percentile or ⩾75th percentile of the study cohort (SOFA = 6).

In a post hoc secondary analysis, we repeated the logistic regression model estimating the association between time of day of sepsis onset and in-hospital mortality, adjusted for severity of illness, with the inclusion of SOFA score at sepsis onset as an additional covariate. To assess whether our findings were biased by the circularity of including antimicrobial initiation in both the definition of sepsis onset and the outcome, we performed additional post hoc subgroup analyses among the subset of patients with ASE hospital-onset sepsis with time zero defined by either blood culture collection or onset of organ dysfunction.

In sensitivity analyses, we repeated all analyses among the Sepsis-3 hospital-onset sepsis cohort.

We performed all analyses using Stata version 16 (StataCorp LP). The University of Pennsylvania Institutional Review Board approved this protocol as exempt (#832918).

Results

The primary cohort consisted of 1,672 patients with a hospital-onset sepsis episode while on a ward unit. Median time from hospital admission to sepsis onset was 7.3 days (interquartile range [IQR], 3.8–12.7 d). The incidence of hospital-onset sepsis onset was approximately equally distributed throughout the day. Corresponding with more frequent lab testing in the last hours of the night shift, a higher proportion of patients had sepsis onset due to new organ dysfunction near the end of the night shift than during the day shift (Figure 1). The SOFA score at sepsis onset did not vary by time of day regardless of time-zero–defining criterion (Figure E2). Unadjusted median time from sepsis onset to antimicrobial initiation was 4.1 hours (IQR, 0.4–23.3 h). A total of 320 (19%) patients died during their hospitalization (Table 1).

Figure 1.

Distribution of hospital-onset sepsis cases by time of day of sepsis onset among ward patients defined by the Centers for Disease Control and Prevention Adult Sepsis Event Criteria. Per Adult Sepsis Event criteria, the first of three sepsis criteria to occur was used to define the time of sepsis onset. MN = midnight.

Table 1.

Patient characteristics and unadjusted outcomes among ward patients with hospital-onset sepsis using the Centers for Disease Control and Prevention Adult Sepsis Event criteria

	CDC ASE
No. of subjects	1,672
Patient characteristics
Hospital of admission
1	1,221 (73)
2	215 (13)
3	122 (7)
4	69 (4)
5	45 (3)
Admission year
2017	314 (19)
2018	680 (41)
2019	678 (41)
Age, median (IQR)	63 (54–71)
Male	910 (54)
Race
White	1,095 (66)
Black	428 (26)
Other*	149 (9)
Latinx	45 (3)
Admission diagnosis category
Symptoms, signs, and abnormal clinical and laboratory findings, NOS	456 (27)
Neoplasms	291 (17)
Diseases of the circulatory system	175 (10)
Diseases of the digestive system	152 (9)
Certain infectious and parasitic diseases	119 (7)
Injury, poisoning, and certain other consequences of external causes	97 (6)
Factors influencing health status and contact with health services	86 (5)
Diseases of the respiratory system	62 (4)
Diseases of the blood and blood-forming organs and certain disorders involving the immune mechanism	57 (3)
Endocrine, nutritional, and metabolic diseases	50 (3)
Diseases of the musculoskeletal system and connective tissue	46 (3)
Diseases of the genitourinary system	36 (2)
Diseases of the nervous system	23 (1)
Other†	22 (1)
Medicine service at time of sepsis onset	1,372 (82)
Days from admission to sepsis onset, median (IQR)	7.3 (3.8–12.7)
Quarter of sepsis onset
January through March	355 (21)
April through June	341 (20)
July through September	495 (30)
October through December	481 (29)
SOFA score, median (IQR)
At sepsis onset	4 (3–6)
Maximum within 7 d of sepsis onset	8 (5–11)
Vasopressors
At sepsis onset‡	29 (2)
Any time within 7 d of sepsis onset	669 (40)
Mechanical ventilation
At sepsis onset‡	58 (3)
Any time within 7 d of sepsis onset	568 (34)
Unadjusted outcomes
Hours from sepsis onset to antimicrobial initiation, median (IQR)
Overall	4.1 (0.4–23.3)
Onset during day shift (7 a.m.–7 p.m.)	2.7 (0.0–14.0)
Onset during night shift (7 a.m.–7 p.m.)	8.3 (1.5–34.3)
Antimicrobials initiated within 1 h of sepsis onset	479 (29)
Antimicrobials initiated within 3 h of sepsis onset	751 (45)
In-hospital mortality	320 (19)

Definition of abbreviations: ASE = Adult Sepsis Event; CDC = Centers for Disease Control and Prevention; IQR = interquartile range; NOS = not otherwise specified; SOFA = Sequential Organ Failure Assessment.

All values are presented as n (%) unless otherwise specified.

^*Other: Asian, American Indian/Alaskan Native, Pacific Islander, mixed, unknown, other.

^† Other: mental, behavioral, and neurodevelopmental disorders; diseases of the eye and adnexa; diseases of the ear and mastoid process; pregnancy, childbirth, and the puerperium; diseases of the skin and subcutaneous tissue; congenital malformations, deformations, and chromosomal abnormalities. Each of these diagnoses represented <1% of the overall population and was derived from International Classification of Diseases-10 codes.

^‡ Includes new start of vasopressors or mechanical ventilation concurrent with sepsis onset.

Primary Analysis

The standardized predicted probabilities of initiating antimicrobials for hospital-onset sepsis at 1 hour from sepsis onset at any given hour throughout the day exhibited a nearly fivefold difference, ranging from 3.0% (95% confidence interval [CI], 1.8–4.1%) at 7 a.m. to 13.9% (95% CI, 11.3–16.5%) at 6 p.m. Relative nadirs were observed at 7 a.m. and 7 p.m., with subsequent relative peaks at 9 a.m. and 9 p.m., and probabilities of initiating antimicrobials declined throughout the night shift (13.4% [95% CI, 10.7–16.0%] at 9 p.m. to 3.2% [95% CI, 2.0–4.0] at 6 a.m.) (Figure 2).

Figure 2.

Standardized probability of antimicrobial initiation at any hour by time of day among ward patients with hospital-onset sepsis according to the Centers for Disease Control and Prevention Adult Sepsis Event criteria. Error bars denote 95% confidence intervals. Hour of day listed on 24-hour scale. Shift changes occur at 7:00 and 19:00. Dotted vertical line marks time of change from day shift to night shift. Analysis included hospital and admission year fixed effects and hours from sepsis onset as the time axis and were adjusted for age, self-reported gender, race, ethnicity, admission diagnosis category (from International Classification of Diseases, Tenth Revision codes), and service type (medicine vs. surgery). MN = midnight.

Secondary Analyses

The standardized predicted median time to antimicrobial initiation after sepsis onset exhibited a 26-fold difference, ranging from 1.3 hours (95% CI, 0.6–2.1 h) for sepsis onset at 9 p.m. to 33.2 hours (95% CI, 27.2–39.2 h) for sepsis onset at 6 a.m. (Figure 3A). It was fourfold longer for patients with sepsis onset during night shift hours (7 p.m.–7 a.m.) than for those with onset during day shift hours (7 a.m.–7 p.m.) (12.9 h [IQR, 10.9–14.9 h] vs. 3.2 h [IQR, 2.5–3.8 h]).

Figure 3.

Standardized time to antimicrobial initiation (A) and standardized probability of antimicrobial initiation within one and three hours (B), by time of day of sepsis onset among ward patients with hospital-onset sepsis according to the Centers for Disease Control and Prevention Adult Sepsis Event criteria. Error bars denote 95% confidence intervals. Hour of day listed on 24-hour scale. Shift changes occur at 7:00 and 19:00. Dotted vertical line marks time of change from day shift to night shift. Analysis included hospital, year, and quarter fixed effects and were adjusted for age, self-reported gender, race, and ethnicity, admission diagnosis category (from International 13 Classification of Diseases, Tenth Revision codes), and service type (medicine vs. surgery). MN = midnight.

The standardized predicted probability of antimicrobial initiation within 1 hour of sepsis onset exhibited a 14-fold difference, ranging from 3.9% (95% CI, 0.2–7.7%) for sepsis onset at 6 a.m. to 53.3% (95% CI, 41.5–65.1%) for sepsis onset at 2 p.m. The adjusted probability of antimicrobial initiation within 3 hours of sepsis onset exhibited a fivefold difference, ranging from 13.3% (95% CI, 6.6–20.1%) for sepsis onset at 6 a.m. to 72.4% (95% CI, 61.9–83.0%) for sepsis onset at 2 p.m.

The standardized predicted probability of initiating antimicrobials within 1 and 3 hours hour of sepsis onset generally declined for patients with sepsis onset after 9 p.m. until midnight, rebounding for sepsis onset around 1 a.m. It again declined for sepsis onset during the latest hours of the night shift, with a nadir at 6 a.m. and subsequent rebound over the first few hours of the day shift (Figure 3B).

Although the standardized predicted probability of in-hospital mortality by time of day of sepsis onset exhibited a threefold difference, ranging from 9.2% (95% CI, 1.6–16.7%) for sepsis onset at 3 a.m. to 30.3% (95% CI, 18.4–42.1%) for sepsis onset at 11 p.m., we did not observe a consistent pattern of increased mortality during times of day with longer times to antimicrobial initiation (Figure 4). These findings were consistent in post hoc analysis additionally adjusted for severity of illness as measured by SOFA score at the time of sepsis onset (Figure E3).

Figure 4.

Standardized probability of in-hospital mortality by time of day of sepsis onset among ward patients with hospital-onset sepsis according to the Centers for Disease Control and Prevention Adult Sepsis Event criteria. Error bars denote 95% confidence intervals. Hour of day listed on 24-hour scale. Shift changes occur at 7:00 and 19:00. Dotted vertical line marks time of change from day shift to night shift. Analysis included hospital, year, and quarter fixed effects and were adjusted for age, self-reported gender, race, ethnicity, admission diagnosis category (from International Classification of Diseases, Tenth Revision codes), and service type (medicine vs. surgery). MN = midnight.

The findings for the primary and secondary analyses were consistent among patients with the highest quartile of SOFA score at sepsis onset (Figures E4–E7).

Subgroup Analysis

Three-quarters of patients in the ASE cohort had sepsis onset defined by blood culture collection or onset of organ dysfunction (n = 1,272, 75%). Findings for this subgroup were also largely consistent with our findings for the full ASE cohort, with decreased probability of antimicrobial initiation at shift changes and overnight and longer times to antimicrobial initiation overnight (Figures E17–E24).

Sensitivity Analyses

The Sepsis-3 hospital-onset sepsis cohort included 3,883 patients, 1,330 (34%) of whom overlapped with the ASE cohort. Consistent with prior literature demonstrating greater sensitivity of Sepsis-3 criteria for sepsis cases with lower severity of illness (29, 38), this larger cohort had similar patient characteristics, but lower maximum SOFA scores, lower rates of need for vasopressors or mechanical ventilation, and lower mortality compared with the ASE cohort (Table E2). The findings for the primary, secondary, and post hoc analyses were largely consistent across both ASE and Sepsis-3 hospital-onset sepsis cohorts (Figures E8–E16).

Discussion

Among ward patients with hospital-onset sepsis, the probability of initiating antimicrobial treatment at any given hour was lower at 7 a.m. and 7 p.m. shift changes, was generally lower during the night shift than the day shift, and declined as the night shift progressed. In addition, the time from sepsis onset to antimicrobial initiation was relatively rapid for patients with sepsis onset during the day shift but was four times longer for patients with sepsis onset overnight. Adherence to guideline-recommended care bundles of antimicrobial initiation within 1 and 3 hours of sepsis onset followed similar trends of decline throughout the night shift, suggesting the presence of an accumulating barrier to prompt clinical care. These findings persisted in analyses restricted to patients with the highest severity of illness and were consistent across several sensitivity analyses.

This study expands our understanding of how the time of day may influence medical decision making and care processes for patients with sepsis. Our work builds on prior studies demonstrating shift-level differences in care delivery (39) by demonstrating variation in antimicrobial initiation at a more temporally granular level with hour-specific estimates. Three salient temporal patterns of antimicrobial delays emerged in this study: 1) at times of shift change; 2) during the night shift compared with the day shift; and 3) a cumulative hourly effect throughout the night shift. We propose several potential mechanisms that may explain these findings and that highlight opportunities to improve the timeliness and reliability of care for patients with hospital-onset sepsis throughout the workday.

First, handoffs, or transfers of patient care between clinicians (40), require time away from direct patient care to complete, frequently result in incomplete data transmission, may be prolonged by distractions and interruptions, and hinder timely communication between nurses and ordering clinicians (41–44). In addition, handoffs contribute to delayed clinical evaluation, medical errors, and adverse events (45–54). If sepsis onset occurs before or during a handoff, it may go unrecognized, and antimicrobials may be less likely to be administered given competing tasks. However, once handoffs are completed, nurses assess each patient, allowing for proactive management in the hours immediately following handoffs. Such a mechanism would fit with the observed pattern of significantly decreased probability of antimicrobial administration during morning and nighttime shift changes followed by a rebound in the ensuing hours.

Second, night shifts frequently have lower staffing ratios than the daytime (55, 56). Therefore, overnight clinicians practice under relatively increased capacity strain, which may lead to fewer bedside evaluations, shorter interactions with patients, and less chart review and follow-up of new diagnostic results (56–59). In these situations, sepsis may go unrecognized, and ordering, delivering, and administering antimicrobials may be delayed (28). These findings are consistent with prior work demonstrating decreased adherence to prescribed sepsis bundles for patients presenting at night (39).

Third, decision fatigue may also contribute to decreased cognitive capacity to interpret and act on novel findings (24, 25), impeding sepsis recognition and thus delaying the initiation of antimicrobials. The decline in the observed probability of antimicrobial initiation throughout the night shift may be explained by clinicians operating not just with scarce hospital resources but also under related accumulating cognitive strain.

Fourth, longer times to antimicrobial initiation for patients with sepsis onset during the late hours of the night shift may reflect delays in decision making until clinical rounds. Day shift clinicians, reviewing and examining many patients at shift start, may not recognize new findings on morning labs (typically collected in the last hours of the night shift) until they are evaluated and synthesized on rounds. Clinicians may also wait until team discussion on rounds before ordering new antimicrobial therapy based on morning data.

We observed several unexpected findings. First, although we observed a decline in the probability of antimicrobial initiation over the course of the night shift, we observed relative stability in antimicrobial initiation throughout the day shift following the rebound in antimicrobial initiation after completion of morning handoff. This is consistent with prior literature demonstrating higher sepsis bundle completion during daytime compared with nighttime hours (39), as well as lower nighttime staffing and worse outcomes among patients admitted to the hospital at night (60, 61). The effects of decreased wakefulness, lower staffing levels, decision fatigue, reduced team cognition and communication through formal or informal rounds, or other factors present overnight may make these hours particularly vulnerable to accumulating barriers to optimal care delivery.

In addition, despite the observed variation in the rapidity of antimicrobial initiation across time of day, we did not find an association between time of day of sepsis onset and in-hospital mortality. This finding is considered exploratory and has several potential explanations that warrant further study. First, it may be that although antimicrobial delays among patients with hospital-onset sepsis are present, they do not contribute significantly to the overall risk of in-hospital mortality. Patients with hospital-onset sepsis tend to have more comorbidities than those with community-onset sepsis, and therefore the degree to which sepsis-attributable mortality is preventable in this population may be diminished. An alternative explanation is that antimicrobial delays contribute to mortality only for certain subgroups, such as those presenting with more severe disease, and incomplete adjustment for clinical severity in our models biased these results toward a null effect.

Limitations

This study has limitations. First, this is an observational study and is susceptible to unmeasured confounding. However, we adjusted for patient, service, and hospital-level characteristics, and the findings of this study were consistent with prior studies assessing medical decision making (14–20). Second, although we adjusted for these factors, we did not evaluate their relative contributions or those of individual clinicians. Future work should explore these domains for potential mechanisms to explain our findings. Third, although we included five acute care hospitals with different patient demographics, case mixes, and practice settings, this is a single–health system study. Our findings may not generalize to other institutions, particularly those with different staffing structures. Fourth, our study cohort included patients retrospectively identified as having had sepsis. Thus, our findings may not be generalizable to patients suspected of having sepsis at the time of antimicrobial initiation who do not ultimately develop sepsis. Fifth, in the absence of a gold standard for sepsis onset, there are inherent limitations in current consensus methods for identifying sepsis cases and defining time zero, as these include antimicrobial initiation in the criteria, which is also an outcome of important clinical interest. However, our findings were robust to the exclusion of patients with time zero defined by antimicrobial initiation. More work is needed to explore the impact of sepsis criteria on treatment timing estimates. Although this study examined timing of antimicrobial administration, future studies differentiating antimicrobial orders from administration may circumvent some of the tautology of current definitions.

Last, it is possible that the variations in antimicrobial timing we observed reflect differences in severity of illness of patients meeting sepsis criteria at different times of day. For example, patients who meet sepsis criteria based on testing obtained on routine morning labs may be more stable than those who meet sepsis criteria at atypical lab collection times that might reflect a clinician’s suspicion of clinical deterioration. However, SOFA scores at sepsis onset did not vary with the time of day, suggesting that longer times to antimicrobial initiation for patients meeting sepsis criteria on morning labs are not explained by differences in severity of illness. Future work should include more granular severity of illness adjustment and causal inference methods to further account for confounding by indication.

Conclusions

Antimicrobial initiation among ward patients with hospital-onset sepsis was less likely at shift change and during the night compared with the day shift and was increasingly delayed as the night shift progressed. These findings indicate that the time of day is associated with important care processes for ward patients with hospital-onset sepsis. Between-shift handoff processes, staffing patterns, and/or decision fatigue may contribute to delays in treatment for ward patients with hospital-onset sepsis. Future work should validate these findings in other settings and distinguish between potential mechanisms to inform quality-enhancing interventions.