Oxygenation Index Predicts Outcome in Children with Acute Hypoxemic Respiratory Failure

Acute hypoxemic respiratory failure (AHRF) causes significant morbidity and mortality both in adults and in children. Over the past 15 years, better understanding of the structural and biochemical effects of mechanical ventilation on the lungs and consequent changes of ventilation strategies, together with advances in critical care, have contributed to better survival of acute respiratory failure in adults (1-3). In pediatric AHRF, a number of observational studies regarding specific treatment modalities as well as a few interventional studies have similarly reported better survival in the treatment arms compared with either historical data or with conventionally managed control subjects. As such, survival rates of 62 to 89% have been reported in children with AHRF who were managed with high-frequency ventilation (4,5) and high-frequency pressure-controlled ventilation, respectively (6), and survival rates of 71 to 88% in children who were placed on extracorporeal membrane oxygenation (ECMO) (7-9). Most observational studies that did not target a specific treatment modality, however, still revealed high mortality rates in pediatric AHRF and pediatric acute respiratory distress syndrome (ARDS), with survival rates from 28 to 58% (10-12). A higher survival rate of greater than 70% in AHRF has been observed only recently (13, 14).

Various attempts have been made to characterize children with AHRF who likely will have an unfavorable outcome. Significant associations between indicators of severity of AHRF and mortality have been reported (7, 15), but their predictive power still remains controversial (13, 16). Differences in study populations, time of measurements, treatment policies, and methods of statistical analysis contributed to variable study results. Nonetheless, there is wide agreement that such predictors would be a valuable aid for difficult-to-make decisions, such as transferring a child to an ECMO center.

We hypothesized that the impact of AHRF severity on outcome may change during the course of disease. The objective of this study was to identify determinants of prognosis in pediatric AHRF and to analyze time-dependent risk of death and utility of respiratory parameters as outcome predictors. This would facilitate an appraisal of the risk of mortality and guide management in children with severe respiratory failure.

METHODS

Data were collected on all mechanically ventilated patients in the pediatric intensive care unit for 3 consecutive years. Severe AHRF was defined as oxygenation failure on continuous mechanical ventilation, necessitating a positive end-expiratory pressure of 6 cm H2O or more and an Fi^sub O2^ of 0.5 or more for adequate oxygenation for at least 12 consecutive hours. These criteria have been used previously to identify children with severe AHRF (7, 12,17). Exclusion criteria were congenital or acquired heart disease, age younger than 1 month, and severe neurologic compromise. The ventilation strategy included attempts to limit peak airway pressure to less than 35 cm H2O, with aggressive use of positive end-expiratory pressure to enable lung recruitment. Further details are provided in the online supplement.

Data were collected prospectively, including demographic data, diagnosis, ventilator settings, arterial blood gases, and markers of organ dysfunction. Pediatric Risk of Mortality (PRISM) scores were calculated once within the first 12 hours of intubation (18). Oxygenation failure was serially assessed by calculating the alveolar-arterial oxygen gradient (A-aDO^sub 2^) (19), the Pa^sub O2^/Fi^sub O2^ (P/F) ratio (20), and the oxygenation index (OI) (21). These ventilatory parameters[dagger] were documented at the time of intubation and after 6, 12, 24, 48 hours, and so forth to the study endpoints of death or extubation. Parameters of the initial 5 hours after intubation were not included in the analysis, to allow for early recruitment of atelectasis, airway suctioning, and cardiovascular stabilization. In cases with fatal outcome, death in refractory oxygenation failure -as opposed to death from an indeterminate or a nonhypoxemic cause -was arbitrarily defined by a final Pa^sub O2^ of 50 mm Hg or less.

Organ dysfunction rather than organ failure was documented. Criteria for organ involvement were therefore less stringent than criteria previously applied for organ failure (22, 23). Specific definitions are provided in an online supplement.

Data Analysis

Data are described as frequencies, medians with ranges, and means with standard deviations. Because both extubation and death without extubation are time-related events and are competing, parametric modeling of the phases of the hazard function was undertaken (24). Incremental risk factors for both events were sought using automated techniques. Possible factors tested included age, sex, worst ventilatory parameters, dysfunction of additional organs, and PRISM scores. Transformations of continuous variables were used to additionally calibrate the relationship of the variable to the risk. Bootstrap bagging was used to determine the reliability of variable selection (25). The multivariable hazard models were then combined in a competing risks analysis to give a more accurate depiction of the actual proportion of patients who had achieved each outcome over time. This type of analysis is useful when patients are simultaneously at risk for two independent and competing outcomes (26). Stratified graphs of actual risk-adjusted time-related mortality were predicted based on significant factors. To investigate whether the time when the measurement of the worst ventilatory parameter was made had an additional impact on prediction of death, multiple logistic regression analysis was used to predict mortality using serial measurements, entering the time of measurement after intubation and its interaction with the value of the ventilatory parameter at that time point. All analyses were performed using SAS statistical software, version 8 (SAS Institute, Inc., Cary, NC).

RESULTS

Over the 3-year study period, 5,766 patients were admitted to the pediatric intensive care unit. One hundred and thirty-five patients with noncardiogenic AHRF (2.3%) were eligible for study participation. Three patients had to be excluded because of incomplete data collection, and one patient was excluded in whom high ventilatory pressures were used because of a massively distended abdomen, but who did not have significant lung disease. The mean age of the remaining 131 patients was 4.9 years (range, 1 month to 18 years), including 59 (45%) female and 72 (55%) male patients.

Thirty-five patients did not survive to successful extubation, reflecting an overall mortality of 27%. Causes of AHRF and respective outcomes are shown in Table 1. Sixteen children (46%) died with refractory oxygenation failure (i.e., a last Pa^sub O2^ = 50 mm Hg). Evidence of dysfunction of at least two other organs was, however, present in 12 (75%) of these children. Of those 19 patients who died with final Pa^sub O2^ values of 50 mm Hg or greater, multiorgan dysfunction (MOD) was found in all but one (95%). Because severe neurologic impairment was an exclusion criterion, no child died because therapy was withdrawn before demise was imminent. Among the AHRF survivors, there was one death within 28 days, and 10 late deaths related to the underlying diseases.

One patient was placed on high frequency oscillatory ventilation (HFOV) from the beginning, and five additional patients who were not manageable by conventional mechanical ventilation were transferred to HFOV later in the course of AHRF, a number too small to allow subgroup analysis. Two patients were placed on ECMO, one of whom died. The second child, suffering from severe aspiration pneumonia, was placed on ECMO after 48 hours with an OI of 111, A-aDO^sub 2^ of 639 mm Hg, a P/F ratio of 27 mm Hg, and a Pa^sub O2^ of 27 mm Hg. For subsequent analysis of prediction, this patient was categorized as not surviving the AHRF episode, although eventually she could be weaned successfully from ECMO. The rationale for this was that the survival of this child required availability of ECMO, so that the preECMO phase, arguably, characterized the evolution of a patient with AHRF with dismal outlook rather than the course of a prospective survivor. Because one objective of this study was to assist physicians in deciding whether to transfer a child to an ECMO center, it seemed reasonable to regard this child as not survivable on conventional or high-frequency oscillatory ventilation.

Mean survival time on the ventilator before death was 9.8 days (range, 12 hours to 69 days). Survivors were ventilated for a mean of 7.8 days before extubation (range, 23 hours to 27 days). The parametric models for these mutually exclusive timerelated events are shown in Figure 1, together with their respective hazard function. Survival to extubation was characterized by a peak rate of extubation at approximately 1 week after intubation, with a more gradual decline thereafter. In contrast, death without extubation was a constant risk over the study time period. In this study, approximately 4% of all patients still on mechanical ventilation died during each 24-hour period. At approximately 4 weeks after the intubation, the risk of death exceeded the risk of survival to extubation. When the two hazard functions were combined in the competing risks analysis, it was noted that the great majority of outcomes had occurred by 2 weeks after intubation (Figure 2). At 10 weeks after intubation, 27% of patients had died without extubation, with the remainder having survived to extubation.

Incremental independent risk factors for the time-related events were sought. For time-related death without extubation, only two factors were significant: a higher peak OI (p < 0.001, 91% reliability in bootstrapping, after inverse transformation) and a higher PRISM score within the first 12 hours after intubation (p < 0.001, 63% reliability in bootstrapping, after square transformation). After controlling for these two factors, no other factor was significantly associated. Mean peak OI was 23 (range, 7-111), and the lowest P/F ratio was less than 150 in all but six patients (95%). For time-related survival to extubation, independent incremental factors significantly associated with a longer time to extubation included a higher peak OI (p < 0.001, 98% reliability in bootstrapping, after log transformation), younger age (p < 0.003,83% reliability, after square transformation), and need for renal replacement therapy (p < 0.001, 80% reliability). After controlling for these three factors, no other factor was significantly associated. It was noteworthy that, of the three ventilatory parameters measured (which were all somewhat correlated), OI had the greatest significance in the multivariable models. (A-a)DO^sub 2^ and P/F ratio were also significantly related to death without extubation and survival to extubation in univariate analysis; however, because these variables were highly correlated with OI, they were not all selected for inclusion in the multivariable model, with OI being preferred and implying that OI was more significantly related to the outcome. From the competing risks analysis, stratification of the actual proportion who died without extubation clearly shows the impact of maximum OI (Figure 3).

From serial measurements of OI over time since intubation, multiple logistic analysis as described previously showed that initial measurements were not as predictive as those at 24 hours and thereafter (Figure 4). No defined threshold of OI, but a continuously increasing gradient of risk of mortality was observed that was, from the second day onward, not dependent on the time the measurement was taken. Figures depicting the course of the ventilator settings and respiratory parameters during the first 48 hours stratified for survivors versus deaths are depicted in the online supplement.

DISCUSSION

The present study found that successful extubations in severe p iatrie AHRF peaked toward the end of the first week and showed a steady decline thereafter. A higher peak OI, younger age, and need for renal replacement therapy were found to be independently associated with longer duration of mechanical ventilation. For all patients still intubated at any point in time, death was a fairly constant risk throughout the entire period of mechanical ventilation. In this cohort, the risk of dying within the next time interval exceeded the chances of successful extubation after approximately 4 weeks. PRISM score within the first 12 hours of mechanical ventilation and peak OI at any point in time of AHRF were identified as independent predictors of outcome. The impact of the severity of AHRF on outcome as represented by OI was found to be less predictable within the first 24 hours after intubation but remained consistent throughout the observational period thereafter.

The heterogeneity of the study population and the fact that the severity of MOD other than respiratory failure was not analyzed were identified as potential weaknesses of this study. A majority of the children in this cohort had an underlying medical condition with a potential impact on survival. Particularly, immunosuppressed individuals, such as bone marrow transplant (BMT) patients, can distort survival data (11). Eleven BMT patients were included in this study, with a mortality of 64%, which is comparable to previously reported data from this institution, which included 39 patients requiring mechanical ventilation for various reasons (27), but lower than reported fatality rates of 89 to 95% in BMT patients with AHRF (11, 12, 28). Sepsis has been associated with higher mortality in AHRF (11, 15), but its independence from MOD as a predictor of mortality is unclear (22, 23, 29). Thus, mortality observed in AHRF cohorts is dependent on the variety and distribution of underlying conditions. In addition, with many possible variables tested as potential predictors of outcome, a number of significant factors identified in the univariate analysis may prove too closely correlated with a stronger predictor in the multivariate analysis. Because the calibration and discriminative power of a variable in a specific cohort affects its significance as predictor and thereby its rank in the multivariate analysis, differences in definitions of any given variable can contribute to contradictory findings between studies. It is argued that future studies in ARDS be conducted in well-defined subgroups to overcome these flaws inherent to the typical case mix of pediatric AHRF cohorts (30). However, this would require much larger case loads than have been so far within the scope of pediatric AHRF studies.

Peak OI was found to independently predict outcome in AHRF, which is in agreement with some recent studies of pediatric or adult AHRF and ARDS (7, 15, 31) and in contradiction to other studies, which could not confirm a predictive power of OI (13, 32). Again, differences in case mix, timing of measurements, resources, and other factors might account for these discrepancies. The other two evaluated ventilatory parameters, A-aDO^sub 2^ and P/F ratio, were not independently predictive for outcome. Conceivably, this might be related to their greater independence from ventilation strategy. Airway distending pressure is an important factor of oxygenation in AHRF and, therefore, influences the assessment of AHRF severity, if Pa^sub O2^ is used as a variable (32). Thus, ventilatory parameters that do not account for this important physician-directed variable can reflect AHRF severity inaccurately. This potential flaw of the A-aDO^sub 2^ may be attenuated by incorporating mean airway pressure to calculate a respiratory severity index (respiratory severity index = mean airway pressure *A-aDO^sub 2^/100), which was reported to improve specificity of prognostication (33).

We found OI to be less predictive of outcome within the first hours of invasive positive pressure ventilation than later in the course of disease. It is assumed that, in the early phase of AHRF, simple therapeutic interventions, such as airway suctioning and recruitment maneuvers, as well as response to specific treatment, will give room for significant improvement in a portion of children with high OI at presentation and good outcome. This might in part explain why some studies did not observe a significant association between outcome and early respiratory indices (10, 13). It has been held that the usefulness of respiratory indices is limited if significant correlation to outcome is established only with worst values measured at any time after intubation (16). In the present study, however, multiple logistic analysis of serial measurements showed that from the second day onward, the predictive power of OI remained intact, with a steadily increasing risk of dying with increasing OI. This provides justification for the use of OI as a principal predictor of time-related mortality without extubation. However, as seen with many continuous measurements, no clear cut-point, or threshold effect, was found that would predict mortality with high accuracy.

In this study, the number of failing organs was a significant factor associated with increased time-related mortality in univariate analysis, but this was not significant after controlling for PRISM score, because the two variables were highly correlated. Conceivably, the application of less stringent criteria for organ dysfunction than were used in studies focusing on MOD could have smoothed the discriminatory power of MOD in this study (i.e., 65% of patients in this cohort had three or more dysfunctional organs). The impact of MOD on survival in children is well described in the literature (22, 23, 34), and MOD has also been implicated as major cause of death in previous pediatric AHRF studies (11-13). In addition, MOD has also been associated with a significantly higher risk of death in children treated with HFOV for AHRF (35) and in children on ECMO (36, 37).

Forty-six percent of deaths were categorized as being chronologically related to severe refractory hypoxemia. However, three of four of these children had evidence of dysfunction of at least two other organs. This is in keeping with previous reports, which attributed mortality from respiratory failure to associated dis ease rather than the severity of gas exchange impairment per se in a majority of cases (13). In the large-scale study by Timmons and coworkers (12), progressive hypoxemia leading to arrest or respiratory acidosis was listed in 22% of subjects as the mechanism of death, excluding those dying of a nonspecified cause (5%) or after support was terminated because of cardiorespiratory futility (25%). Montgomery and others (38) defined irreversible respiratory failure as inability to achieve a Pa^sub O2^ of 40 mm Hg or more for a period of 2 hours or more on 1.0 Fi^sub O2^ or respiratory acidosis with a pH less than 7.1 on full support, and found this to be responsible for death in only 16% of adult patients with ARDS. Similar findings have been reported later in adult patients with ARDS (39, 40). So far, however, the assignment of deaths to respiratory and nonrespiratory origin remains arbitrary, because a threshold of hypoxemia incompatible with life has not been defined and likely varies independent of organ integrity. Additional noxious factors, such as inflammatory response and reduced organ perfusion, might render organs more vulnerable to hypoxemia. We found MOD to be quasi-omnipresent in dying patients. A similar statement has been made by Proulx and coworkers (22) who reported that all patients dying in the intensive care unit during the period of their MOD study fulfilled the criteria of MOD. MOD developing shortly before death after longstanding hypoxemia might merely reflect the initiated process of dying on sustained life support as a final common pathway and thus not be the primary cause of death, even in states of moderate hypoxemia. For the purpose of validation of a MOD score, this preterminal period has recently been arbitrarily limited to the last 2 hours of life (34). Therefore, we would argue that close attention to optimization of pulmonary gas exchange and protective ventilation strategy will facilitate recuperation of organ integrity in all patients with MOD, because moderate hypoxemia might contribute to mortality in patients with MOD.

In conclusion, severity of oxygenation failure at any point in time in the course of pediatric AHRF does have an impact both on the length of mechanical ventilation and survival, although less so within the first hours of intubation. This is best reflected by OI, presumably because of its greater independence from ventilation strategy compared with A-aDO^sub 2^ and P/F ratio. Death from ARHF occurs as a rather constant event from the beginning of ventilatory support on with no significant chronologic variation. Thus, for the individual patient, the risk of death exceeds chances of survival after a certain period.

Conflict of Interest Statement: D.T. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript; B.W.M. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript; S.N. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript; D.B. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript.

Acknowledgment: The authors thank Tara Karamlou, M.D., for her invaluable assistance with the statistical analyses and the creation of the graphs.

© 2005 American Thoracic Society Provided by ProQuest LLC. All Rights Reserved.