Increasing Volume-Targeted Ventilation Use in the NICU

We performed this project in the 98-bed, level IV,¹³  academic NICU of the Vanderbilt University Medical Center. This NICU includes 2 geographically distinct units in separate but contiguous hospitals: a 78-bed unit with single-patient rooms primarily housing outborn infants and infants with congenital anomalies and/or surgical conditions, and a 20-bed, open-bay unit primarily housing inborn preterm infants. The NICU staff includes ∼280 neonatal nurses, 50 pediatric respiratory therapists (RTs), 45 advanced practice providers (APPs), 25 neonatologists, and 11 neonatal-perinatal medicine fellows.

Our NICU has nearly 1500 admissions annually and is supported by a full range of subspecialists. Patients include preterm infants (≥22 weeks' gestation), infants with congenital anomalies, including complex surgical and cardiac conditions, and term infants with respiratory or other medical conditions. Approximately 25% of infants admitted to the NICU receive MV, including both conventional and high-frequency ventilation. The Dräger Evita Infinity V500 (Drägerwerk AG and Co, Lübeck, Germany) ventilator is used for conventional MV. Before project interventions, conventional ventilation consisted of a mixture of 9 VTV and PLV modes (Supplemental Information). Although only physicians (neonatologists and fellows) and APPs can write orders to change ventilator settings, physicians, APPs, and RTs are permitted to physically adjust ventilator settings.

All infants admitted to the NICU who received conventional MV were eligible for the interventions. We hypothesized that including all ventilated infants would be more easily implemented than limiting interventions to only infants at highest risk for BPD (<1500 g at birth). We excluded infants while they were receiving high-frequency ventilation or extracorporeal membrane oxygenation and those with congenital lung anomalies (eg, congenital diaphragmatic hernia), a tracheostomy, or an endotracheal tube leak >80% when a VTV approach could not be reliably used¹⁴  and those infants receiving end-of-life care.

We followed 2 main project outcomes: (1) percentage of conventional MV hours when a synchronized VTV mode was used and (2) percentage of conventional MV hours when a synchronized, exclusively VTV mode was used. We considered ventilator modes in which the majority of breaths were volume-targeted as a VTV mode. Ventilator modes in which all breaths, whether initiated by the ventilator or the infant, were volume-targeted and potentially less injurious to the lung¹⁰  were classified as exclusively VTV modes (Supplemental Information). Secondary outcomes included mortality, BPD, duration of MV, supplemental oxygen at discharge, and severe intraventricular hemorrhage. High-frequency ventilation use was included as a balancing measure. We prospectively collected data through medical records review and, beginning in September 2016, augmented with data downloaded directly from the ventilators. During the project, concerns arose that increased VTV use was resulting in greater ventilator alarm frequency. We therefore calculated ventilator alarm rates for a convenience sample of patients during each period as a further balancing measure. No value metrics were followed.

Our multidisciplinary QI team consisted of neonatal nurses, RTs, APPs, neonatal fellows, neonatologists, medical and nursing leadership, a systems engineer, and a data analyst. We evaluated outcomes in 3 periods: baseline, no interventions (May 1, 2016–December 18, 2016); epoch 1, improving the use of VTV modes and decreasing PLV use (December 19, 2016–August 31, 2018); and epoch 2, improving the use of exclusively VTV modes (September 1, 2018–August 31, 2019). These dates were chosen because the main outcome of interest during epoch 1 was any-VTV use, and the main outcome during epoch 2 was exclusively VTV use. Before testing interventions, we evaluated baseline data; performed structured observations of the MV workflow, including clinician interviews and simple cognitive walkthroughs (structured technique for evaluating software and computer interfaces)¹⁵  for choosing ventilator modes; generated a process map with failure modes and effect analysis for MV; identified SMART aims for each epoch; and generated a key driver diagram for iterative testing (Fig 1). We used the Influencer Change Model¹⁶  as our framework to develop interventions for behavioral change, attempting to target all 6 potential sources of influence, and the Model for Improvement¹⁷  as our overall improvement methodology. Interventions were iteratively tested through plan-do-study-act (PDSA) cycles and ramps. Goals of our interventions comprised 4 general categories: (1) standardization of MV processes and practices, (2) improved team communication, (3) system optimization to facilitate use of VTV, and (4) education regarding the evidence for superiority of VTV and the capabilities of our ventilators. In Table 1, we list interventions, their components, Influencer Change Model framework, and timing of testing and implementation for each intervention.

Clinical variables, outcomes, and balancing measures were compared between baseline and the 2 epochs by using analysis of variance for continuous parametric data, Kruskal–Wallis test for continuous nonparametric data, and χ² tests for dichotomous data.

We used time series graphs to evaluate the effectiveness of our interventions and tests of change in real time as the project was implemented and XmR charts¹⁸  to analyze full project data. During active improvement periods, we used the Western Electric rules¹⁹  to identify special-cause variation.²⁰  During sustainment (after active testing had ceased), we used only the rule for shift to adjust our center line and investigated all points outside of the control limits. Because our primary unit of measurement was a ventilator hour, data both within and between subgroups were potentially autocorrelated (nonindependent) because an infant may have multiple ventilator hours within a week and be ventilated across multiple weeks. We therefore calculated the autocorrelation coefficient²¹  to ensure the ability to use statistical process control (SPC) methods for monitoring. The autocorrelation coefficient early during the baseline period was 0.08, below the recommended threshold of 0.5, confirming our ability to use SPC charts for monitoring.²¹ 

To ensure our main analyses were robust to autocorrelation, we completed 2 interrupted time series (ITS) (autoregressive integrated moving average) analyses.²²  The first ITS analysis used any-VTV mode as the outcome and introduced interruptions at the beginning of each epoch. The second ITS analysis used exclusively VTV mode use as the outcome and introduced interruptions at the times of ventilator interface reformatting, the 2 interventions temporally related to the largest increases during real-time SPC monitoring.

During the baseline period, we noted episodes of MV when an unsynchronized ventilator mode (the ventilator delivers a breath whether an infant is breathing or not) was inadvertently used. Thereafter, we also followed a g-chart^20,23  of the number of conventional MV hours between events in which an infant received an unsynchronized mode.

Finally, to assess the overall effect of our project on clinical outcomes in very low birth weight (VLBW) infants, we developed multivariable logistic or negative binomial regression models with a dummy variable denoting the project period for all secondary outcomes. These analyses were limited to VLBW infants inborn or transferred in by 72 hours of age and adjusted for multiple potential confounders. Analyses were performed in Stata/MP 15.1 (StataCorp, College Station, TX), and SPC charts were constructed by using QI Macros for Excel v. 2014.1 (KnowWare International Inc, Denver, CO). Our institutional review board reviewed and exempted the project under a waiver for QI.

During the project period, 1269 infants received MV in the NICU for a total of 275 202 ventilator hours. Of these, 1225 of 1269 (97%) infants met project inclusion criteria for 213 751 (78%) ventilator hours. Reasons for exclusion from the project and the breakdown of ventilator time across the periods are shown in Figure 2. Patient characteristics (Table 2) were similar across the periods except that VLBW infants constituted a higher percentage of admissions during the baseline period (P = .04).

Both main outcomes increased during the project in the entire eligible population and VLBW infants (P < .001 for all comparisons) (Table 3). Use of any-VTV mode increased from 18 694 of 22 387 (83%) conventional MV hours during the baseline period to 72 846 of 77 264 (94%) and 58 174 of 60 605 (96%) of conventional MV hours during epochs 1 and 2, respectively (P < .001). Use of an exclusively VTV mode increased from 5967 of 22 387 (27%) during baseline to 47 364 of 77 264 (61%) and 46 091 of 60 605 (76%) of all conventional MV hours during epochs 1 and 2, respectively (P < .001).

Using XmR charts, we observed special-cause variation multiple times during the project for both outcomes. For the use of any-VTV mode (Fig 3), we noted a shift with 8 points above the center line temporally after the following: (1) beginning to test a standardized process for initiating MV (PDSA ramp no. 1), (2) implementing diagnosis-based MV “pathways” unit-wide (PDSA ramp no. 2), and (3) reformatting the ventilator interface to remove all PLV choices from the home screen. After improvements and tightening of our control limits, we also noted multiple episodes of points below the lower control limit. We investigated these in real time, and, in nearly all cases, this variation was caused by 1 or more older infants with BPD who were being ventilated with a PLV mode. We noted 4 shifts in the use of an exclusively VTV mode (Fig 4) during the project. Improvements were temporally related to (1) reformatting the ventilator interface screens to include all available VTV modes used at that time, (2) beginning to test and implement a new exclusively VTV mode (pressure control–assist control with volume guarantee [PC-AC/VG]), and (3) reformatting the ventilator interface to remove all PLV modes from the home screen and include PC-AC/VG as a choice. A decrease in exclusively VTV mode use was temporally associated with the beginning of a new academic year and an influx of new faculty, staff, and trainees.

Using ITS analyses, we noted an immediate and sustained 9.5% (95% confidence interval [CI]: 2.1% to 16.8%, P = .01) increase in any-VTV mode use after beginning interventions in epoch 1 without a change in the slope (Supplemental Information). We also observed an immediate 39% (95% CI: 26.5 to 51.3, P < .001) increase in exclusively VTV mode use after reformatting the ventilator interface screens in January 2017, with no change in the slope of use across these periods (Supplemental Information). Although the second ventilator screen reformatting in April 2019 was associated with special-cause variation in SPC analyses, this change was not significant in the ITS analysis (30.8%, 95% CI: −38% to 100%, P = .38).

We unexpectedly noted use of an unsynchronized mode, pressure control–continuous mandatory ventilation with volume guarantee (PC-CMV/VG), during the baseline period. Using clinical workflow observations, we found that these events mostly occurred while clinicians were attempting to navigate the ventilator screens to choose a commonly used exclusively VTV mode, pressure control–pressure support ventilation with volume guarantee (PC-PSV/VG). Using this finding as the impetus to modify the ventilator interface home screens, we noted 2 episodes of special-cause variation on the g-chart corresponding to prolonged periods of time when PC-CMV/VG was not used, both occurring after reformatting the ventilator user interface (Fig 5).

The frequency of ventilator alarms per hour increased during the project periods from 12.53 (95% CI: 12.45 to 12.6) during baseline to 16.96 (95% CI: 16.88 to 16.99) and 16.43 (95% CI: 16.39 to 16.47) during epochs 1 and 2, respectively (P < .001). The majority of alarms during all periods were related to peak inspiratory pressure limits (data not shown).

Despite increasing the use of VTV, clinical outcomes in VLBW infants in our NICU, including mortality, BPD, severe intraventricular hemorrhage, and duration of MV, remained unchanged after adjusting for patient characteristics (Table 4).

We have shown that focused QI interventions over a 3-year period improved the evidence-based use of MV in our NICU, with a 16% relative increase (13% absolute) in the use of any-VTV mode, 280% relative increase (49% absolute) in the use of exclusively VTV modes, and nearly complete elimination of the use of unsynchronized modes of MV. The greatest improvements were temporally related to standardization of MV processes and practices and reformatting the ventilator interface screens so that it was easier for clinicians to choose a preferred VTV mode and more difficult to choose a PLV mode. To our knowledge, this is the first QI report to describe interventions to improve evidence-based use of neonatal MV. Disappointingly, these improvements in MV were not associated with improved clinical outcomes, after adjusting for patient characteristics.

The 2 largest improvements in VTV use occurred after changes to the ventilator interface home screens. We made the first adjustment after noting that clinicians were inadvertently selecting an unsynchronized mode while trying to choose a VTV mode. Before adjusting the screens, clinicians navigated through 4 menus, with multiple options at each to choose the preferred VTV mode. We found that at the second menu, clinicians were inadvertently choosing an unsynchronized mode with an abbreviation similar to the preferred VTV mode (PC-CMV/VG versus PC-PSV/VG). After moving PC-PSV/VG to the home screen, we observed special-cause variation on the g-chart with a prolonged period of time between use of PC-CMV/VG. Unexpectedly, we also saw the use of PC-PSV/VG greatly increase. During postimplementation interviews, several clinicians stated that they did not use PC-PSV/VG before reformatting the screens because they were concerned they would make an error while navigating the multiple menus. Our findings underscore the importance of user-centered design (UCD)²⁴  methods in implementing technology-intensive interventions such as MV. To date, no studies have described UCD methods to optimize neonatal MV. With our findings, we suggest that optimizing local systems using human factors engineering methods²⁵  such as UCD could greatly improve the quality of neonatal MV.

Several potential reasons may explain why improved use of VTV was not associated with improved clinical outcomes for VLBW infants. First, our NICU had relatively high VTV use in VLBW infants at baseline (81% of MV time). Researchers demonstrating the efficacy of VTV have compared exclusive use of VTV to exclusive use of PLV.^10,11  Before our project, many infants already received VTV for most of their MV course and thus may have already gained its benefits. Second, multiple factors contribute to the clinical outcomes we assessed. For example, potentially modifiable risk factors for BPD include the type and duration of MV,^9–11  postnatal growth,²⁶  and fluid and sodium intake early in life.^27,28  Although we adjusted for many factors in our analyses, we could not adjust for all potential confounders. Third, although we saw improvements in the use of MV modes, we were unable to determine if the ventilation strategies used by clinicians were appropriate for each patient, with tidal volumes selected on the basis of patient pathology.²⁹  Future work in our NICU will focus on assessing the VTV strategies used and evaluating whether these agree with available evidence.

In addition to improving VTV use, we observed an increase in ventilator alarm frequency during our project. This increase was driven mainly by peak inspiratory pressure alarms, in which the set tidal volume was not delivered because of a set pressure limit being met. Because limited data exist on MV alarms in the NICU, we are not able to say how much of the increase in alarms was expected³⁰  and how much could have been mitigated. This increasing alarm frequency has the potential to result in alarm fatigue and increased workload and stress for bedside providers and parents. Although some expert recommendations for alarm limits during VTV exist,³¹  future work should define the epidemiology of ventilator alarms in the NICU and develop strategies to mitigate alarms commonly seen with the use of VTV.

Although successful in our NICU, our interventions may not be generalizable to all NICUs. Specifically, our strategy to reformat the ventilator interfaces may not be as easily implemented in NICUs using ventilators in which the screens cannot be modified. In addition, our high baseline rate of VTV suggests that clinicians were already aware of the evidence regarding efficacy of VTV compared with PLV. In NICUs with lower baseline use of VTV, more attention, effort, and education will likely be required to target the drivers of knowledge and buy-in to the available evidence for choosing a ventilation mode.

We have shown that specific QI interventions in our NICU were associated with improved use of VTV and decreased use of unsynchronized and PLV modes but no change in clinical outcomes. Our work can be used as a blueprint for QI teams who desire to implement VTV use in their unit.

We thank the RTs, bedside nurses, nurse practitioners, and physicians who took part in the improvement initiatives, generated ideas for improvement, and provided feedback and testing of interventions in clinical practice.

APP

advanced practice provider

CI

confidence interval

BPD

bronchopulmonary dysplasia

ITS

interrupted time series

MV

mechanical ventilation

PC-AC/VG

pressure control–assist control with volume guarantee

PC-CMV/VG

pressure control–continuous mandatory ventilation with volume guarantee

PC-PSV/VG

pressure control–pressure support ventilation with volume guarantee

PDSA

plan-do-study-act

PLV

pressure-limited ventilation

QI

quality improvement

RT

respiratory therapist

SPC

statistical process control

UCD

user-centered design

VLBW

very low birth weight

VTV

volume-targeted ventilation