Introduction
Acute Hypoxemic Respiratory Failure (AHRF), Acute Lung Injury (ALI) and Acute Respiratory Distress Syndrome (ARDS) result in significant morbidity and mortality for patients admitted to Pediatric Intensive Care Units (PICUs). While changes in ventilation strategies have improved outcome, mortality is still between 30-40% for adults [1], and 8-22% for children [2-4] with ALI or ARDS.
Children with ALI or ARDS have longer lengths of mechanical ventilation than patients requiring mechanical ventilation for other reasons [5]. While systemic illness or lung disease severity explains much of this, Ventilator Induced Lung Injury (VILI) may contribute to longer lengths of mechanical ventilation and mortality. Low tidal volume studies by the ARDS Network (ARDSNet) and Amato et al. demonstrated significantly more ventilator free days and reduced mortality for adults using volume control ventilation with tidal volumes of 6ml/kg compared to 12ml/kg body weight [1, 6].
It is unclear, however, whether a tidal volume strategy of 6 ml/kgshould be applied universally to all patients with ALI or ARDS[7][8],and if this is better than tidal volumes such as 8-10 ml/kg. Moreover, it is unclear if this strategy should be employed for all patients with ALI, regardless of disease severity.Three randomized trials in adults have shown no difference between groups ventilated with mean tidal volumes of 10.2-10.6 ml/kg compared to 7.2-7.3 ml/kg [9-11]. Moreover, in the two beneficial trials mean tidal volume of the control group actually increased by 17-18% after randomization, with mean increases in airway plateau pressures of 4-6 cmH20 [1, 6, 12]. So, the perceived benefits in the ARDSnet trial could have been from increased mortality in the 12 ml/kg VT group.
Performing similar randomized controlled trials for pediatric populations poses several challenges. Smaller number of patients, lower mortality, heterogeneity of disease conditions, lack of uniform ventilator management strategies and protocols, and unclear surrogate outcomes all make multi-center randomized pediatric mechanical ventilation trials difficult [13, 14]. As a result, many pediatric intensivists extrapolate conclusions from adult studies regarding practices to minimize VILI. To this end, pediatric intensivists now target lower tidal volumes for children with ALI, and retrospectively it appears that tidal volumes in the 11-12 ml/kg range are harmful [15].
However, adult and pediatric practices differ in tidal volume measurement. First, adult intensivists typically target tidal volume to predicted body weight (PBW) while it is likely that pediatric intensivists use actual body weight. Second, exhaled tidal volume measured at the ventilator even after adjustment for tubing compliance and volume tends to be higher than that measured at the proximal airway using a flow sensor, particularly for small children [16][17].
Moreover, PICUs infrequently use volume control ventilation (particularly in assist control like most adult centers), with pressure limited modes being utilized nearly 75% of the time [5]. Given the decelerating flow characteristics of pressure control or pressure regulated volume control (PRVC) modes, attention is paid to the set or generated peak inspiratory pressure, rather than plateau pressure used in volume control ventilation. Proponents of pressure limited ventilation argue that by setting a reasonable peak inspiratory pressure, tidal volume will inherently be limited for patients with the most severe respiratory disease [18]. Finally, the developing lung and chest wall have different elastance and compliance properties than the adult, although these differences may only be clinically relevant for the premature infant. Because of these differencesone has to be cautious extrapolating adult evidence regarding tidal volume and outcome for ALI to children.
We sought to determine the effect of generated tidal volume on outcome for children with ALI or AHRF using historical data from greater than 90%of patients on pressure control ventilation. Additionally, we examined how well lung injury severity measures such as PaO2/FiO2 (PF) ratio, Oxygenation Index (OI), and a Modified Murray Pediatric Lung Injury Score (LIS) [19]reflect mortality at various time points during the initial days of mechanical ventilation [20].
Methods
Patient Selection
Retrospective review of all admissions to a tertiary care PICU between January 2000 and July 2007 was conducted. January 2000 was the beginning of the cohort because it was when lung protective ventilation was gaining more acceptance, with the knowledge of the Amato study, reinforced by the ARDSnet data and multiple abstracts on the effects of tidal volume and outcome[6, 15, 21-23]. Patients were screened for inclusion if at least two arterial blood gas values had a PF ratio less than 300. Patients were eligible for the study if endotracheally intubated and mechanically ventilated, and at least one PF ratio was less than 300 within twenty-four hours after intubation. Patients were excluded if they had evidence of left ventricular dysfunction (either by echocardiography or clinically), cyanotic congenital heart disease, cardiomyopathy, myocarditis, or primary pulmonary hypertension. Patients were also excluded for incomplete data on ventilator management or intubation and extubation times. All patients met three of the four diagnostic criteria for ALI (acute onset of disease, PF ratio <300, and lack of left ventricular dysfunction). The presence of bilateral infiltrates on chest radiograph (the fourth criteria for ALI) was handled as a separate variable, and analysis was made comparing those who met all four criteria to the entire cohort. Finally, to account for the effect a leak around the endotracheal tube may have on volume and compliance measurements and calculations, all patients with an endotracheal tube leak [(Inhaled Tidal Volume-Exhaled Tidal Volume)*100/(Inhaled Tidal Volume)] greater than 20% were excluded from analysis[24]. The data on leaks was available every 6 hours. During the study period, 60% of intubated patients were ventilated with cuffed endotracheal tubes.
Variable Selection
Two databases were used for data extraction. An electronic ICU flow sheet (Philips CareVue®, Waltham, MA) captures data in real-time from several integrated monitors and charting, and a separate patient database (Microsoft Access ©, Redmond, WA) containing diagnostic and demographic data is maintained in real-time by the ICU physicians providing clinical care. From these two databases, information on patient age, race, gender, weight, primary diagnosis, year of admission, chest x-rays, and a Pediatric Risk of Mortality (PRISM) Score were extracted [25]. PRISM was used instead of PRISM III because subjective data elements necessary for PRISM III were not available for all patients prior to 2002.Random samples of patient records from this database are routinely checked for accuracy, and elements of standardized mortality scores are checked monthly for all patients. Internal audits have demonstrated consistency in >98% of data measurements. All chest x-rays were read by an attending pediatric radiologist blinded to the study, and reports reviewed for the presence of diagnostic criteria for ALI. In addition, all arterial blood gas values and respiratory therapy records were reviewed and ventilator variables extracted.
Ventilator Strategy
While this study was non-interventional, there was a lung-protective strategy in place in our PICUduring the study period. In general, pressure limited modes of ventilation were used, with target Peak Inspiratory Pressure <40 cm H20, Ventilator Rate <35BPM, employing permissive hypercapnea. PEEP and FiO2were targeted to maintain SpO2 between 88-95% or PaO2 >60. Recruitment maneuvers were not commonly usedduring the study period. While no specific written protocol was in place, the strategy was widely accepted by all practitioners.
Ventilator and Arterial Blood Gas (ABG) Variable Definition
The time of first PF ratio < 300 after intubation was defined as “baseline,” where blood gas measurements of pH, PaCO2, PaO2, and base deficit were extracted. In addition, the closest charted ventilator settings at or prior to the ABG were extracted. This included mode of ventilation, ventilator rate, Peak Inspiratory Pressure (PIP), Positive End Expiratory Pressure (PEEP), Mean Airway Pressure (MAP), Fraction of inspired Oxygen (FiO2), Inspiratory Time (Ti), Inspired Tidal Volume (VTi), and Exhaled Tidal Volume (VTe). Tidal Volume (VT ml/kg) was calculated using VTe measured at the ventilator with appropriate compensation for tubing compliance and make of the ventilator. Three ventilators were used during this time period (Servo 300 (Siemens, Solna, Sweden), Avea (Viasys Healthcare, Yorba Linda, CA), and Servo i (Maquet Medical, Solna, Sweden)), and tidal volume measurements were adjusted as per the published guidelines of the ventilator and tubing manufacturers. This value was then divided by actual body weight to report tidal volume in ml/kg.
Next, aggregate variables were created to express time-weighted averages of ventilator settings and blood gas values over the first three days after “baseline.” For example, PEEP1 was defined as the patient’s time-weighted average value for PEEP for the first 24 hours after baseline. Each charted PEEP in the first 24 hours was multiplied by the hours the patient remained at that PEEP, summed with the other time-weighted PEEP values, and divided by 24 hours. The same methodology was used for the second (PEEP2) and third (PEEP3) 24-hour intervals, each representing independent 24-hour periods, not aggregate 48 or 72-hour values. For intervals shortened by death, extubation, or conversion to high frequency oscillatory ventilation, the denominator was appropriately adjusted.
Lung Injury Severity Markers
PF ratio, Oxygenation Index [OI= (Mean Airway Pressure*FiO2)/(PaO2) *100], Dynamic Compliance of the Respiratory System [Crs = VT (ml/kg)/(PIP-PEEP)], and a Modified Murray Lung Injury Score (LIS) for pediatric application [19] were determined at “baseline” and then again aggregated over the first three days after baseline, similar to above. The Modified Lung Injury Score is a composite of PF ratio, PEEP, Crs, and involved quadrants on chest radiograph, with integer values from zero to four assigned for each component. Specific data on number of quadrants of alveolar consolidation on chest x-ray were not included in this analysis. As a result, the total Modified Lung Injury Score was the average of three, rather than four components.
Outcome Measures
The primary outcome measure of the study was ICU mortality. Secondary outcome was 28-day ventilator free days, defined as the total number of days in the 28 days after intubation for which the patient was alive and free from mechanical ventilator support [26]. Data on both outcome measures were complete for all patients; as such, no censoring was necessary.
Statistics
Statistical analysis was performed using Statistica v. 5.5 (StatSoft, Tulsa, OK) and Stata v. 10 (StataCorp, College Station, TX). Descriptive statistics regarding distribution of variables and population characteristics were first performed. Next, univariate analysis was performed examining the variables of interest against the outcome of mortality. Continuous variables were analyzed with a Wilcoxon rank-sum test, as assumptions of normality could not always be satisfied. Dichotomous outcomes were analyzed using a Yates-corrected chi-squared test. Kruskall-Wallis ANOVA was used to examine differences in median tidal volume stratified by categories. Next, logistic regression analysis was performed to examine the impact of the variables of interest on the outcome of mortality, and control for potential confounding variables, or effect modifiers. A multivariate logistic regression model was built incorporating variables with univariate associations with mortality (p<0.2), with care taken to avoid terms that were collinear. Terms were deemed collinear if standard errors for point estimates in the regression model became large, or if terms were thought to measure similar clinical parameters (e.g. Mean Airway Pressure and Peak Inspiratory Pressure) [27]. Analysis was limited to subjects with complete data for all the variables included in the multivariate model. Assumptions of linearity of the dependent variables in the logistic model were examined and satisfied, and no transformations were necessary. Overall goodness of fit was assessed using the Hosmer-Lemeshow test, as well as graphical evaluation for influential points. ROC plots were created and overall discrimination ability of the various predictive models was assessed using empirical estimates of the overall Area Under the Curve (AUC).
The outcome of 28-day ventilator free days had an almost bimodal distribution, as noted in previous studies. Hence, for the purpose of logistic regression analysis and controlling for confounding variables, it was examined as a dichotomous outcome of <14 days, or ≥ 14 days, as previously described [4]. However, when appropriate, comparisons of median number of ventilator free days between two groups were analyzed using a Wilcoxon rank-sum test [26].
Sample Size Estimates
With an alpha level of 0.05 and a power of 0.8, an anticipated mortality of 17% at a mean tidal volume of 7.5 ml/kg, 398 patients could detect an effect size of 4.5% at one standard deviation above or below the mean. For the secondary outcome measure of <14 VFD, with an anticipated outcome of 40% (i.e. 40% of patients will have <14 VFD) at the mean tidal volume of 7.5 ml/kg,398 patients could detect an effect size of 6.5% at one standard deviation above or below the mean.
Results
Developing Lung and Chest Wall
To examine the hypothesis that characteristics of the developing lung and chest wall may impact VT choice and outcome, patients were divided into neonate (<2 mo), infant (2mo - 2 yr), child (2 yr - 12 yr), and adolescent (12 -18 yr). Median VT was no different between the four groups (7 ml/kg, 7.4 ml/kg, 7.7 ml/kg, 7.5 ml/kg, K-W ANOVA p=0.94). Overall, VT had no impact on mortality for neonates (n=24, OR = 0.93 (0.63, 1.37)), infants (n=132, OR = 0.98 (0.82, 1.17)), or adolescents (n=89, OR= 1.07 (0.92, 1.25)). For children (2 yr - 12 yr), higher VT was associated with lower mortality (n=153, OR=0.79 (0.66, 0.95)).
In the multivariate model which included VT, age category, Delta P (PIP-PEEP), PEEP, admission diagnostic category, and severity of lung disease using LIS, PF Ratio, or OI, there was an interaction between the age category of child and VT with respect to mortality (p=0.02). However, this was no longer present when controlling for severity of illness and lung disease with the full multivariate model (p=0.17). As such, the lower mortality seen with higher VT in the child group could be explained by less severe lung disease or systemic illness in these children.
Definition of Refractory Respiratory Failure
As used by Montgomery et al, refractory respiratory failure is defined as the inability to attain PaO2 > 50 torr or SpO2 >80% for two hours on 1.0 FiO2, or respiratory acidosis with pH 7.1 on full support [28].
1