Bundled Care to Reduce Sepsis Mortality: The Improving Pediatric Sepsis Outcomes (IPSO) Collaborative Free

Severe sepsis is a leading cause of worldwide pediatric mortality, with 8 000 000 deaths annually.^1–6  Care recommendations for sepsis patients are provided in guidelines from the National Institute for Health and Care Excellence, American College of Critical Care Medicine, the Surviving Sepsis Guidelines, and the European Resuscitation Council.^7–12  Despite these resources and disease burden, suboptimal adherence to recognition guidelines and timely administration of intravenous (IV) fluids, antibiotics, and vasoactive agents remains.^13–21  A recent New York state regulatory initiative demonstrated that despite improvement in several sepsis-related measures, only 25% of patient encounters adhered to all components of a 1-hour sepsis treatment bundle.²²  Bundled care improves outcomes in pediatric septic shock, but the optimal compilation of bundle elements is not known.^8,9,21–28 

Closing the performance gap between expected and actual care for pediatric sepsis has been the goal of several multicenter quality improvement (QI) collaboratives.^29,30  Comparing outcomes among disparate patient groups in these collaboratives was confounded as cohort and time-zero definitions were not standardized.

In 2016, Children’s Hospital Association sponsored the Improving Pediatric Sepsis Outcomes (IPSO) QI Collaborative with initial results reported here. Multimodal QI science was employed to improve outcomes across the acute care clinical continuum by improving adherence to pediatric sepsis recognition and treatment processes. Specifically, we aimed to improve utilization of a sepsis care bundle, decrease 3- and 30-day sepsis-attributable mortality and decrease the progression of hospital-onset sepsis. Additionally, we sought to determine which care elements of a sepsis bundle are associated with improved outcomes. The collaborative developed and used standardized sepsis definitions, time-zero derivations, metrics, and care bundles as described in previously published work (Fig 1 and Appendix 1).^31–33 

Through a QI learning collaborative based on Learning Healthcare Systems and the Model for Improvement, the IPSO Collaborative aimed to: (1) decrease sepsis-attributable mortality and (2) reduce the progression of hospital-onset sepsis with the goal of detecting cases earlier to prevent decompensation into severe sepsis or septic shock.^34,35  We also aimed to use our data to determine the optimal sepsis care bundle components associated with improved mortality. This report details results from the first wave of 40 pediatric hospitals, from January 2017 through March 2020. Data beyond March 2020 are not presented to mitigate fluctuations associated with the onset of the Coronavirus disease 2019 pandemic. Patients originating from another hospital (within 24 hour of time zero) or with a missing time-zero variable were excluded (<10% of all episodes). A multidisciplinary steering committee incorporated evidence and expert consensus into a Key Driver Diagram, which informed all metrics, described in our previous work.^31–33,36 

IPSO suspected sepsis (ISS) and IPSO critical sepsis (ICS) are populations described in our previous work and delineated retrospectively (Fig 1).^32,33  IPSO suspected sepsis (ISS) patients were those without frank organ dysfunction, where the provider intended to treat sepsis. IPSO critical sepsis (ICS) patients approximated septic shock as described by Goldstein et al.^37,38  Data abstraction evolved (from primarily manual toward predominantly automated), ultimately resulting in 67% with full automation.

Time-zero for time-bound metrics was defined as the time of earliest possible sepsis recognition documented within the electronic health record (EHR), denoted as functional time-zero (FTZ), which could occur in the emergency department (ED), ICU general care, or hematology and oncology units (Appendix 1).³³  Since there is no singular biomarker to identify sepsis, we used EHR surrogates, such as sepsis screen time, order set time, or bedside huddle time, to automate and standardize time-zero identification.

We derived 28 process, outcome, and balancing metrics from 44 variables.³³  Main outcomes included 3- and 30- day all-cause and sepsis-attributable mortality (Appendix 2). To address the secondary aim of reducing hospital-onset critical sepsis (ICS), we derived the incidence over time of hospital-onset ICS occurring after 12 hours from time-zero per 1000 hospital admissions. Secondary outcomes included hospital length of stay (LOS) and ICU-, vasoactive-, and ventilator-free days. Delineation of all-cause and sepsis-attributable mortality required a manual chart review of all deaths to determine whether clinically related to sepsis (Appendix 2).³³  Process metrics included compliance with recommended sepsis care bundle elements. As a balancing measure, we monitored total IV antibiotic days per sepsis episode to quantify possible overtreatment. Special cause variation was denoted on statistical process control charts for process and outcome measures using standard rules.³⁹ 

We retrospectively evaluated the association between bundle compliance and mortality. Analysis of bundles was delineated after the granular data from the entire time period had been submitted, irrespective of when that patient was treated during the collaborative. The differing bundles were not introduced to various hospitals in a prescriptive, controlled manner. The originally defined bundle included a structured sepsis screen, bedside clinician huddle or sepsis-specific order set (recognition method), fluid bolus delivery within 20 minutes (targeting 40–60 mL/kg within 60 minutes), and broad-spectrum antibiotics within 60 minutes of septic shock identification. As the IPSO collaborative progressed, additional collaborative data became available and newly published recommendations emerged suggesting different time goals for care processes.⁹  In unadjusted analysis, we evaluated sepsis outcomes associated with bundle element compliance across several timeframes using a χ-square test for categorical outcomes and a Kruskal-Wallis test for continuous outcomes. This resulted in an outcome-driven optimally modified bundle, which included any sepsis recognition method (screen, huddle, or order set), fluid bolus delivery within 60 minutes, and antibiotics within 180 minutes.

In adjusted analysis, we identified factors associated with primary 3-and 30-day mortality outcomes using a generalized linear mixed modeling approach, assuming a binomial distribution and logit link function and a γ distribution and natural log link function for secondary outcomes. Model covariates included age at FTZ, presence of comorbid high-risk conditions (immunocompromised, malignancy, stem cell transplant, solid organ transplant, asplenia, technology dependency, indwelling line, and intellectual disability), positive blood culture, blood lactate > 4 mmol/L, time to surgical source control in 24 hours or less, care setting in which FTZ occurred, ICS versus ISS delineation, and modified bundle compliance (Appendix 3).

All models (for primary and secondary outcomes) included a random IPSO center effect to account for the clustering of IPSO sepsis episodes within the center. We conducted analyses in SAS 9.4 (SAS, Cary, NC). P values < .05 were considered statistically significant. Each site obtained Institutional Review Board approval or QI waiver and a participation agreement. This study protocol was approved as Human Subjects Research, with a waiver of consent by Case Western Reserve University (Cleveland, OH).

During the 39-month study interval we analyzed 37 339 cases from an initial cohort of 40 hospitals. All were children’s hospitals, with 22 (55%) freestanding and a median bed count of 238 (interquartile range [IQR] 113–362). Total reported episodes of ISS and ICS were 24 518 and 12 821, respectively. Figure 2 displays a stable incidence of ICS over time. ISS cases increased the first year of the collaborative as case identification was optimized, however was stable thereafter. Table 1 summarizes characteristics of the total cohort and modified bundle cases by compliance. Patients with bundle compliance had higher rates of high-risk conditions and longer times to surgical source control. Both groups had similar lactate values and positive blood cultures. Most cases were identified in the ED, with the highest mortality observed in hematology and oncology and ICU settings. Those with bundle compliance had their FTZ defined more frequently with a screen, huddle or order-set time compared with those where bundle compliance was not achieved.

Table 2 outlines mortality outcomes associated with original and modified bundle compliance. Other bundle cutoffs are analyzed in Appendix 4. In the adjusted models, for both cohorts (ICS and ISS), there was a significant reduction in all-cause and sepsis attributable, 3- and 30-day mortality with compliance to the modified bundle. Specifically, in the ICS cohort, compliance with the original bundle was not associated with a decrease in 30-day sepsis-attributable mortality, whereas compliance with the modified bundle was from 4.75% to 2.4% (P < .01); adjusted odds ratio 0.982 (confidence interval 0.986 to 0.995). Modified bundle compliance was also associated with improvement in secondary outcomes (Appendix 5). The association between modified bundle compliance and mortality was the same despite presence or absence of a high-risk condition, with greatest reductions with bundle compliance seen in the acute care units. Individual bundle elements and their association with mortality and secondary outcomes are presented in Appendix 6.

Figure 3 displays process measures for the ISS and ICS groups. Significant improvements occurred in structured recognition (use of a screen, huddle, or order set), time to first antibiotic, and all-or-nothing compliance for both the original and modified bundles. The center line shifts demonstrate improved compliance with the modified bundle from 52% to 57% for ICS and 40% to 46% for ISS. By March of 2020, only 37% of inpatients (ICU, general floor, and hematology and oncology) had a screen, huddle, or orderset used, thus only marginal improvements in fluid and antibiotic delivery compliance were noted (62% and 72% respectively). By contrast 75% of ED patients had a screen, huddle, or orderset with 87% and 90% of ED patients adhering to fluid and antibiotic goals respectively. Among the 40 hospitals, 30 demonstrated significant improvement in at least 1 care process.

Figure 4 illustrates outcome measures for the 2 main collaborative aims. For the ISS cohort, 30-day sepsis-attributable mortality decrease occurred over time, with a relative reduction of 35.7% (absolute 1.4% to 0.9%, P < .001). There was no change in mortality for the ICS cohort. There was no reduction in hospital-onset sepsis, with the mean incidence remaining stable at 14.5%. Total IV antibiotic days over time (balancing metric, not displayed) dropped from 11.6 days to 9.5 days in the total cohort during the study period (slope [95% confidence interval]: −0.05 [−0.08 to −0.03], P < .001).

The IPSO Collaborative results demonstrate several important and novel findings from the largest cohort of pediatric sepsis patients described to date. First, pediatric sepsis episodes compliant with key evidence-based bundle elements (recognition methods, fluids, antibiotics- change strategies within our Key Driver Diagram³¹ ), have better outcomes. Second, compliance with numerous pediatric sepsis care processes can be improved over time utilizing QI methodologies without increasing antibiotic utilization. Although aggregate compliance data shifts are modest, 75% of sites achieved substantial improvements in at least 1 care process. Third, even modestly improved care processes meaningfully reduced sepsis-attributable mortality among the ISS cohort. Importantly, this is the first analysis stratifying deaths as all-cause versus sepsis-attributable, which is vital in understanding the true quality-related drivers of sepsis mortality.

This IPSO Collaborative demonstrates improved recognition of patients with suspected sepsis over time through screening tools (manual and automated), prompt clinical bedside sepsis-specific evaluations (huddles), and education campaigns. The most likely etiology of the significant rise in ISS incidence early in the collaborative is the identification of patients earlier in the sepsis continuum, with potentially lower acuity. Earlier recognition of lower acuity patients has confounded analysis of previous sepsis improvement literature by progressive incidence inflation and severity dilution of the cohort, resulting in improved time-series outcomes that cannot be clearly attributed to changes in care processes.^29,30  In contrast, we included an analysis of the more stable ICS cohort and the ISS cohort once the incidence stabilized in 2018, allowing for valid outcome interpretation.

Those with use of a structured recognition method (screen, huddle, or orderset) were more likely to achieve bundle compliance, which we have shown is associated with decreased mortality, demonstrating that earlier recognition is similarly associated with improved outcomes. This is despite these FTZ timestamps occurring earlier in the process than other time zero delineations, theoretically making bundle compliance even harder to achieve.

Our novel, large-scale analysis, demonstrates that adherence to individual bundle elements and all-or-nothing compliance with evidence-based, time-dependent practices is associated with improved mortality, ventilator-, ICU-, vasoactive-free days, and reduced hospital LOS.^12–30, Furthermore, we explored different time cut-offs for compliant care, as suggested by new evidence recommendations from the 2020 Pediatric Surviving Sepsis Campaign.⁹  A modified bundle that was more liberal in the targeted time window for first fluid bolus (within 60 vs 20 minutes) and time to antibiotics (within 180 vs 60 minutes), was associated with greater reductions in sepsis-attributable mortality when compared with the original bundle, which has been suggested in other populations as well.^40,41  Bundle delivery beyond the modified bundle time cut-points is associated with the worst outcomes. Finally, high bundle compliance was associated with improved mortality even in sicker patients, such as those with high-risk conditions.

Although better outcomes with a more time-liberalized bundle may seem counterintuitive, it is important to recognize the marked heterogeneity of sepsis presentations and the ways care is improved by clinical assessments to tailor the targeting, timing, and intensity of interventions. The time taken to obtain more diagnostic data, such as laboratories or imaging, can ensure therapy for individual patients is appropriate and proportional, preventing patients from receiving over-treatment with attendant untoward side effects. There is likely a trade-off between timeliness and precision in the treatment of sepsis. There are also inherent risks to crystalloid toxicity for certain populations that may already be fluid overloaded. We believe our results suggest a data-driven optimized bundle timeframe where it appears safe to focus on precision without subjecting patients to undue risk from excessive delays in therapies.

For this QI study, overall outcomes are different among the ICS and ISS cohorts. Improvements in mortality occurred for the ISS cohort over time, but not for the ICS cohort; in the ISS cohort, mortality was almost halved in the last 9 months reported during a period of incidence stability. This is possibly because of a relatively larger magnitude of improvement in care process improvements for ISS than for ICS, including time to antibiotics and time to first fluid bolus, likely driving improved mortality outcomes. Further, in the ICS group, small improvements in sepsis care process compliance were possibly offset by greater exposure to hospital and ICU-level interventions that are themselves not risk-free. More aggressive fluid resuscitation in the ICS group may have contributed to untoward effects associated with crystalloid toxicity and fluid overload.^40,41  Vasoactive agents, ventilators, and ICU utilization increased over time in a different-than-anticipated direction (more use versus less use), likely reflecting that recent guidelines recommend earlier initiation of vasoactive agents in severe sepsis patients at lower volumes of fluid delivery (30–40 ml/kg).⁹  Vasoactive, ventilator, and ICU use were delineated in this study as secondary outcome measures, yet can also be interpreted as therapeutic interventions, so their increase may reflect more aggressive early care as opposed to worse clinical outcomes. Additionally, although ICU settings differ from other hospital settings, given the safety net of ongoing close monitoring, the ICU is also a setting where early subtle signs of sepsis can be masked.^42,43  As sepsis progresses into a more severe physiologic status with increasing numbers of failing organ systems, poor outcomes may become increasingly inevitable, contributing to the observed decreased mortality in ISS but not ICS. Regardless, in the adjusted model, bundle compliance is still associated with reduced mortality in both cohorts.

Higher acuity patients (the ICS cohort) received more timely fluids, antibiotics, and had greater compliance with both the original and modified bundles, even at the start of the collaborative (Fig 3). This made improvements in these time-sensitive therapeutic interventions more challenging to achieve in the ICS cohort compared with the ISS cohort. However, both had a similar absolute improvement in the magnitude of compliance with the modified bundle, demonstrating improved reliability of care, regardless of illness severity.

Numerous large-scale efforts to improve sepsis care, such as the Surviving Sepsis Campaign and the New York State Sepsis Care Improvement Initiative, have demonstrated low compliance rates with evidence-based practices.^9,22  Similarly, although the IPSO collaborative has shown significant improvement in adherence to the original and modified bundle, there remains substantial room for further improvement, challenges partially related to the heterogeneous disease process such as sepsis. This collaborative focused on a population where there was “intent to treat” sepsis, as laboratory results and evidence of organ dysfunction are often not yet known initially. As such, a large percentage of patients eventually did not receive a full treatment course of antibiotics, nor had positive blood cultures. These patients, nonetheless, clinically presented with concern for infection and impending decompensation, and thus the collaborative directives were to implement the bundle if clinically relevant.

We did not see improvement in mortality for the ICS cohort, nor reduction in hospital-onset sepsis. The collaborative’s first 18 months prioritized building the “data pipeline,” with actual implementation science beginning later.³³  Therefore, we have just begun to realize measurable improvements in the mortality of some sepsis patients. Specifically, preventing the rapid decline of sepsis patients requires early recognition and screening. When specific care setting data are analyzed, we only start to see recognition tools being used in the inpatient setting at the beginning of 2020. We anticipate that more reliable recognition implementation moving forward will result in a decrease in hospital-onset ICS. Despite this, most collaborative hospitals have achieved high reliability in at least 1 care process, allowing an opportunity for exemplar hospitals in an all-teach or all-learn QI collaborative to spread effective implementation innovations to achieve bundle compliance at scale.

Systematic identification of cases through automation and structured tools has imperfect sensitivity and specificity for actual cases of sepsis, and electronic time-zero determinations vary based on physiologic changes and clinical assessments. Given the disparate implementation of screening tools and resources in each care unit, this also poses challenges for uniform recognition and subsequent implementation of care bundles. Exclusion of vital sign values from our reported variables limited our ability to conduct in-depth aggregate analyses of true physiologic time zero. Additionally, using antibiotic days as a balancing measure has its limitations; it is confounded by increased identification of less sick patients at the collaborative onset and its potential interpretation as an outcome measure, given better treatment, might reduce antibiotic days. Finally, outside hospital data can pose challenges for data clarity and thus were excluded here; it can however provide valuable insight into the global care of sepsis patients in future study.

We have demonstrated through the IPSO collaborative, the largest of its kind, that patients with sepsis who receive care compliant with a more liberalized, evidence-based bundle for recognition and timely administration of fluids and antibiotics have better outcomes, including improved mortality, hospital LOS, and days free from the ICU, ventilator, vasoactive agents, and antibiotics. With modest bundle compliance improvement over time, sepsis-attributable mortality among suspected sepsis patients decreased significantly. Ample opportunities to further improve outcomes exist by optimizing adherence to evidence based bundled care.

We thank Jayne Stuart MPH for her leadership throughout as the Director of the IPSO Collaborative and Director of Quality for the Children’s Hospital Association.

ED

emergency department

EHR

Electronic Health Record

FTZ

Functional Time Zero

ICS

IPSO Critical Sepsis

ICU

Intensive Care Unit

IPSO

Improving Pediatric Sepsis Outcomes

ISS

IPSO Suspected Sepsis

IV

intravenous access

QI

quality improvement