Improving Ventilation Rates During Pediatric Cardiopulmonary Resuscitation

We conducted this study in the 3 PICUs of NewYork-Presbyterian Morgan Stanley Children’s Hospital, an urban, quaternary academic medical center with 41 PICU beds, including both medical-surgical and cardiothoracic patients. This project received a “Not Human Subjects Research” designation from the Columbia University Institutional Review Board.

We performed a retrospective review of CPR events between April 2016 and December 2018 to determine baseline rates of hyperventilation. Central monitoring and electronic medical record documentation, including continuous pulse oximetry, telemetry, arterial pressure, and end-tidal carbon dioxide (ETCO₂) monitor output, provided most CPR metrics. Chest compression (CC) fraction and percent CC rate were obtained from ZOLL R-series monitor-defibrillator (ZOLL Medical, Chelmsford, MA) dual sensor defibrillators.¹⁷  Each CPR event was reviewed by a single author (J.C.), with 20% of events reviewed independently by a second author (A.S.). Manual capnography review is the standard for determining CPR ventilation rate,^18,19  mitigating difficulties with automated ventilation detection.^20,21  Interrater reliability was assessed via Cohen’s κ and the intraclass correlation coefficient.^22,23  CPR events with fewer than 2 minutes of chest compressions and without ETCO₂ data were excluded.

We planned a QI initiative to reduce the frequency of clinically significant hyperventilation during CPR from our baseline of 51% to 30% of CPR time. The 30% target was based on study team consensus that a 20% decrease would represent a clinically significant decrease from baseline, while also taking into account that some clinical situations may warrant a higher ventilatory rate. Clinically significant hyperventilation was defined as a rate ≥30 bpm. We chose this target, as opposed to strict adherence to the 2010 AHA guidelines recommending 6 to 12 bpm,¹²  because: (1) pediatric cardiac arrests more commonly occur because of respiratory deterioration and may require higher ventilation rates³ ; (2) the hemodynamic impact of hyperventilation during CPR is most clinically relevant with severe hyperventilation (≥30 beats per minute)^{13,14,16,19,24–26} ; and (3) evidence suggests a survival benefit in pediatric CPR with higher ventilation rates.¹⁹ 

After confirming high rates of hyperventilation during pediatric CPR at our institution, we interviewed key stakeholders, including PICU physicians, nurses, and respiratory therapists (RTs). We identified several factors leading to severe hyperventilation during CPR, including: (1) knowledge deficit (many bedside providers unaware of ventilation guidelines; CPR leaders unaware of the prevalence of hyperventilation), (2) cognitive overload (memory, cognitive function, and performance worsened in high-stress environments²⁷  such as CPR, corroborated during interviews with stakeholders), and (3) lack of resuscitation team buy-in. Areas for improvement, outlined in a key driver diagram (Fig 1) included: (1) bedside provider education, (2) CPR leader education and feedback, and (3) ventilation-focused CPR tools. Potential solutions were introduced using the Plan-Do-Study-Act model.²⁸ 

Education was conducted in 3 phases. Phase 1 (January 2019 to February 2019) consisted of structured sessions for respiratory therapists (RTs) and bedside PICU nurses. The sessions were led by a single author (J.C.), consisted of 1 to 2 learners, and occurred twice weekly. Sessions began with a survey assessing knowledge of AHA ventilation guidelines during CPR, (asking learners to identify the correct compression-ventilation ratio for a natural airway, the correct ventilatory rate for an advanced airway, and knowledge of compression-ventilation synchrony for a natural versus advanced airway.) A review of both the current AHA guidelines and our institution’s baseline rate of clinically significant hyperventilation followed. We identified barriers to appropriate ventilation and introduced ventilation tools to avoid hyperventilation during CPR. Time was allotted to providers to familiarize themselves with the ventilation tools during these sessions.

Phase 2 (March 2019 to July 2019) consisted of monthly meetings with the PICU nursing unit practice councils, and monthly presentations at the RT huddles. The most recent CPR ventilation data and the QI intervention aim and components were presented.

In Phase 3 (August 2019 to March 2020), presentations to the PICU nursing unit practice councils and RT huddles occurred every other month. The same survey assessing CPR ventilation knowledge was readministered to PICU nurses and RTs.

During phase 1, PICU fellows (designated CPR leaders) and PICU attending physicians underwent CPR ventilation education at monthly PICU quality meetings. CPR leaders were encouraged to provide real-time feedback using ventilation-focused tools to achieve patient-appropriate ventilation rates. Training sessions emphasized that if CPR leaders felt that the clinical scenario warranted a faster ventilation rate, they should direct the arrest team accordingly. Monthly quality meetings assessed CPR ventilatory practices via review of primary outcome run charts.

CPR review sessions are routinely conducted by our quality director (A.S.) and include review of a standardized hot debriefing tool,²⁹  the electronic medical record, telemetry, and ZOLL CC data. ETCO₂ tracing review was added to these sessions to include ventilation-focused feedback.

Three ventilation rate tools were introduced to address cognitive overload in CPR. (1) A mobile phone-based CPR metronome with separate tones for CC and ventilation rates (http://www.omnimedicalsolutions.com) was introduced, as tone guidance has been shown to improve CPR performance.^15,30–32  (2) RTs were tasked with ensuring consistent use of ETCO₂ during CPR to provide real-time feedback for CPR ventilatory rate^18,19  as well as data on CPR quality and changes in physiology.^33–35  (3) Ventilation reminder cards (Fig 2) were distributed to PICU providers, as cognitive aids such as preprocedural checklists,³⁶  pocket cards and handbooks³⁷  are routine in high stakes clinical settings. Ventilation tools were incorporated into ongoing PICU simulation programs to further solidify their use.

The first 10 minutes of each CPR event were divided into 20-second epochs. Each epoch’s ventilation rate was measured in bpm using ETCO₂ waveforms. For aggregate data analysis, individual epochs were classified as within the target ventilation range (<30 bpm) or clinically significant hyperventilation (≥30 bpm). For time series data, CPR epoch ventilation rates were grouped into 3-month quarters. The primary outcome was the proportion of CPR epochs with clinically significant hyperventilation (≥30 bpm) pre and postintervention. Secondary outcomes were the overall median ventilatory rate across all epochs pre and postintervention, and the median ventilation rate per quarter.

Process measures included phase 1 education completion, change in provider baseline ventilation knowledge (defined by pre- and postintervention surveys), ventilation specific instruction provided by CPR leaders, individual CPR leader feedback, and use of ventilation rate tools. Balancing measures included the proportion of CPR with hypoventilation (<6 bpm), CPR quality (CC fraction and rate), failure to achieve return of spontaneous circulation (ROSC), and death before hospital discharge.

Characteristics included age, event time of day, event location (medical-surgical versus cardiothoracic PICU), cardiac arrest etiology (respiratory versus cardiac as determined by the CPR event leader), and shockable rhythm at CPR initiation. Patient characteristics and outcomes were compared pre and postintervention using χ² or Fisher exact tests for categorical variables, and Mann-Whitney U test for continuous variables, with a level of significance of 0.05 for all tests.

A time series analysis was completed using run charts to analyze the primary outcomes. Because of the relative rarity of CPR events, CPR data were pooled together by quarter. Baseline median values for the center lines were obtained from the 8 quarters of CPR data preceding the intervention and the first quarter the intervention roll-out began based on lack of signal and the absence of any established practice changes. Unfortunately, data collection was stopped 1 year after the QI bundle implementation because of the unexpected redirection of clinical and quality resources in response to the coronavirus disease 2019 pandemic, yielding 4 quarters of data post implementation.

Risk differences, risk ratios (RR), and 95% confidence intervals (CI) were estimated for outcomes and balancing measures, with exposed patients defined as the postintervention group. Primary (presence of clinically significant hyperventilation) and secondary outcomes (ventilation rate per epoch) were adjusted for the patient and event characteristics described above using logistic and linear regression, respectively. Covariates were chosen per previously described associations between patient characteristics and CPR outcomes.^38–40  Odds ratios were converted to risk ratios using relative risk regression with maximum likelihood estimation via the epitools package in R,^41,42  and data were visualized using the ggplot2 package.⁴³  All statistical analyses were performed using R software (R version 4.0.3; R Foundation for Statistical Computing, Vienna Austria).⁴⁴ 

In the 32-month preintervention period, 64 CPR events were screened and 37 (699 20-second epochs) met inclusion criteria. In the 15-month post-intervention period, 34 CPR events were screened and 24 (426 epochs) were included (Fig 3). Patient and event demographics are shown in Table 1. Compared with postintervention, the preintervention cohort had a smaller percentage of patients less than 1 year old (22% versus 50%, P = .02). No other significant differences were observed between the 2 cohorts. Interrater reliability measures for presence of clinically significant hyperventilation and for respiratory rate during CPR epochs were excellent (κ = .97, intraclass correlation coefficient = .98).

An annotated run chart (Fig 4A) displays the proportion of CPR time with clinically significant hyperventilation. Quarters with no qualifying CPR events were omitted. Nine data points total (8 data points before implementation of the intervention and the first quarter the bundle was introduced) were used to form the center line. Following the first quarter after implementation of this improvement bundle the proportion of CPR epochs with clinically significant hyperventilation decreased from a baseline median of 53% below the goal of 30% for all subsequent quarters. Although this data did not meet criteria for special cause variation,⁴⁵  given the strong likelihood of signal had data collection continued, we felt comfortable shifting the center line to reflect the new process post-implementation.

The aggregate proportion of all 20-second CPR epochs with clinically significant hyperventilation decreased from 51% to 29% in the preintervention versus postintervention period, for an absolute risk reduction of 22% (95% CI 16% to 28%, P <.001). Distributions of ventilation rates during each 20-second CPR epoch in the pre and postintervention periods are shown as smoothed density plots (Fig 4B) and further demonstrate lower ventilatory rates in the postintervention period.

An annotated run charge (Fig 5A) displays the median ventilation rate per quarter. Following the first quarter of bundle implementation all subsequent data points (quarterly median ventilation rates) tracked below the baseline median ventilation rate of 27 bpm. Similar to Fig 4A, the center line was shifted to reflect this new process postimplementation, based on confidence in the change despite lack of adequate data to show signal.

The aggregate median ventilation rate across all epochs decreased from 30 bpm preintervention (interquartile range [IQR] 21 to 36) to 21 bpm postintervention (IQR 12 to 30) (P <.001). (Fig 5B). On a per-patient level, the proportion of patients who had median ventilatory rates with clinically significant hyperventilation decreased from 57% to 29% (absolute risk reduction 28%, 95% CI 3% to 52% P = .03).

We modeled the association between clinically significant hyperventilation and the QI intervention using logistic regression to adjust for potential confounders. The intervention remained significant, with an adjusted RR of 0.56 (95% CI 0.47 to 0.65, P<.001) (Supplemental Table 2). Next, we modeled the association between ventilatory rate per epoch and the QI intervention using linear regression. After adjusting for confounders, the average effect of the QI intervention was to reduce the ventilatory rate by 6.3 bpm (95% CI 4.8 to 7.8, P<.001) (Supplemental Table 3).

Eighty-eight percent of bedside providers (142 of 164 [87%] PICU nurses and 62 of 68 [91%] RTs) completed phase I of education. Bedside provider knowledge of AHA ventilation guidelines showed modest improvement on follow up knowledge assessment surveys with 89% of RTS able to correctly identify the recommended ventilation rate during CPR (compared with 76% preintervention) and 76% of RNs (versus 74% preintervention).

100% of PICU fellows and attending physicians completed the CPR leader training. Postintervention, 79% (19 of 24) of CPR leaders provided ventilation-specific instruction to the resuscitation team (Fig 6A), 96% (23 of 24) of the CPR leaders received post-CPR ventilation rate feedback, and the CPR-ventilation metronome was used in 88% (21 of 24) of CPR events (Fig 6B).

Hypoventilation (ventilation rate <6 bpm) did not occur in any postintervention CPR epoch. In the pre versus postintervention periods, 10 (27%) and 7 (29%) patients failed to achieve ROSC, and 22 (60%) and 16 (67%) patients died before hospital discharge, respectively. Differences in failure to achieve ROSC (RR 1.08, 95% CI 0.48 to 2.45, P = .86) and death before hospital discharge (RR 1.12, 95% CI 0.76 to 1.65, P = .57) were not statistically significant. The CPR quality of 42 events (24 preintervention, 18 postintervention) with ZOLL R-series monitor-defibrillator data were assessed. Pre versus postintervention, the median CPR fraction improved from 87% (IQR 83% to 91%) to 94% (IQR 92% to 95%), P = .02. The median percentage of time with CC rate within AHA guideline range also improved from 60% (IQR 40% to 69%) to 87% (IQR 77% to 96%), P<.001.

A QI initiative focused on education of CPR team members, CPR leader feedback, and use of CPR ventilation rate tools decreased the proportion of CPR time with clinically significant hyperventilation rates (defined as $≥$30 bpm) from 51% to 29%. Median ventilation rates decreased from a baseline of 30 bpm to 21 bpm postintervention. These changes remained statistically significant after adjusting for potential confounders. While the run charts did not meet standard rules for signal, based on the patterns seen as well as the multiple other analyses conducted, we believe real improvement was seen. To our knowledge, this is the first study to address modifying ventilation rates during pediatric CPR.

At the time of study implementation, the 2010 AHA guidelines recommended a ventilation rate of 6 to 12 bpm during pediatric CPR.¹²  Based on the hemodynamic risks of hyperventilation, as well as emerging evidence that pediatric patients may benefit from higher ventilation rates than stated in the 2010 AHA guidelines,¹⁹  we designed our initiative to reduce clinically significant hyperventilation. After the completion of data collection for this study, the 2020 Pediatric Advanced Life Support update changed the recommended assisted ventilation rate for infants and children from 6 to 12 bpm to 20 to 30 bpm.⁴⁶  This change agrees with our group’s belief that avoiding clinically significant hyperventilation is appropriate in most pediatric CPR events. Going forward, this QI model can be used to disseminate and promote adherence to the 2020 Pediatric Advanced Life Support ventilatory rate guidelines.

Our preintervention survey of bedside nurses and RTs revealed a large knowledge gap, specifically that many bedside providers were not familiar with either the risks of hyperventilation or the AHA recommendations for ventilation during CPR. CPR education was a major part of our QI initiative. Survey results of pre and postintervention knowledge of AHA guidelines showed only modest improvement in knowledge of AHA ventilation guidelines. However, rates of significant hyperventilation decreased almost immediately after initiation of CPR education, suggesting that our overall QI program was effective.

In addition to education, level 2 reliability strategies, using intentionally designed tools aimed to error-proof and standardize systems,⁴⁷  were incorporated in the Plan-Do-Study-Act cycles of this study. We used a CPR metronome with tone guidance for ventilation rates as a major intervention and level 2 reliability strategy. Our hospital’s PICU healthcare team embraced this tool, and it has become a standard PICU resuscitation component. Finally, feedback to healthcare providers—both individually to CPR leaders postevent, as well as on a quarterly basis to bedside providers—allowed us to build on our results over time.

Balancing measures were tracked to ensure that our interventions did not have a negative clinical impact. By reducing hyperventilation, it was possible that our strategies might lead to hypoventilation (defined as <6 bpm). Postintervention, no patient was hypoventilated during any CPR epoch. Additionally, we wanted to ensure that this intervention would not lead to worse cardiac arrest outcomes. In this small cohort, no differences in failure to achieve ROSC or death before hospital discharge were observed when comparing pre and postintervention periods, and CPR quality metrics (CC fraction and rate) both improved postintervention.

Feasibility is a key component of any QI intervention. This study did not require significant additional resources or personnel. Educational sessions and feedback were easily incorporated into daily huddles. Additionally, this initiative offered an opportunity for maintenance of RT skills, which was positively received by RT leadership. Educational outreach, feedback, and data collection were performed by a single PICU physician and could easily be transitioned to multiple PICU providers going forward. Institutions with limited resources could feasibly implement and maintain this QI initiative.

This study has several limitations. First, study inclusion was dependent on good quality ETCO₂ measurement. CPR events without analyzable ETCO₂ were excluded (∼30% of events in both cohorts), possibly including a bias to our results. Improving use of ETCO₂ during CPR requires ongoing improvement. Second, there was a significant difference in patient age pre and postintervention, with more infants in the postintervention period. Infants tend to receive higher ventilation rates during CPR.^19,25  If infants were intentionally ventilated at higher rates during CPR at our institution and were overrepresented in the postintervention period, the impact of the QI intervention may have been more dramatic had there not been an age disparity between the groups. Our findings were also robust to adjustment for age in both linear and logistic regression models. Third, a trend toward improvement may be evident in the run charts for both clinically significant hyperventilation and median ventilation rate during CPR before the implementation of our QI project. A hot debriefing program introduced in 2016 identified excessively fast CC during CPR as a potential improvement area.²⁹  The focus on CC rate starting in 2016 may also have led to an increased focus on ventilation rate. However, despite this possible preintervention improvement, we felt that a programmatic effort was necessary for more significant improvement. Fourth, the potential confounders included in multivariable models were limited by small sample size. We chose variables based upon literature indicating important variables that were available in our dataset.^38–40  Similar to all nonrandomized studies, residual confounding is still possible. Finally, further study on sustainability of this project is necessary. Data collection ended in March 2020 when the coronavirus disease 2019 pandemic unexpectedly redirected clinical and quality resources and introduced multiple confounders.

Clinically significant hyperventilation during pediatric CPR should be avoided. A QI initiative grounded in improved AHA guideline education, CPR team member feedback, and tools focused on CPR ventilation rates was effective at reducing rates of clinically significant hyperventilation. Future studies should focus on the sustainability of such a QI initiative, as well as its utility in other hospital settings.

AHA

American Heart Association

BPM

breaths per minute

CPR

cardiopulmonary resuscitation

CI

confidence interval

CC

chest compression

EMR

electronic medical record

ETCO₂

end-tidal carbon dioxide

IQR

interquartile range

RR

risk ratio

ROSC

return of spontaneous circulation

RT

respiratory therapist