Article Text
PDF
Abstract
Introduction We aimed to develop and validate a prediction table for a simplified measure of rightward shift of the fetal oxyhaemoglobin saturation (SpO2) versus inspired oxygen pressure (PIO2) curve as an objective marker of lung disease severity in very preterm infants, independent of unit altitude or oxygen prescribing policies.
Methods Very preterm infants (n=219) had an oxygen reduction test at median (IQR) test age of 354 (345–360) weeks’ postmenstrual age (PMA). Shift was derived from at least three paired SpO2 versus PIO2 measurements using a computer algorithm, using the fetal oxyhaemoglobin dissociation curve as the reference. Linear regression of resultant shift values enabled construction of a table to predict shift using a single paired SpO2 versus PIO2 measurement, validated subsequently in a separate infant cohort using Bland-Altman analysis. Receiver operating curve analysis provided threshold values equating to a clinical diagnosis of mild bronchopulmonary dysplasia (BPD) or moderate to severe BPD.
Results The median (IQR) age of 63 infants in the validation cohort was 360 (356–362) weeks’ PMA. Mean difference (95% CI) between predicted and measured shift was 2.1 (−0.8% to 4.9%) with wide limits of agreement (−20.7% to 24.8%). Predicted shift >10.1 kPa identified mild BPD with 71% sensitivity and 88% specificity while values>13.0 kPa identified moderate to severe BPD with 81% sensitivity and 100% specificity.
Discussion Shift predicted from a single paired SpO2 versus PIO2 measurement using our validated table enables objective bedside screening of lung disease severity in very preterm infant cohorts at 36 weeks’ PMA.
- paediatric lung disaese
- lung physiology
Statistics from Altmetric.com
Key messages
What is the key question?
Can a single paired oxyhaemoglobin saturation (SpO2) versus inspired oxygen pressure (PIO2) measurement predict the rightward shift of the SpO2 versus PIO2 curve as a severity measure for bronchopulmonary dysplasia (BPD).
What is the bottom line?
Our observational study established and validated a table for predicting right-shift of the SpO2 versus PIO2 curve as an indicator of severity of BPD that agrees with the clinical semi-quantitative NIH definition of BPD severity, although with wide limits of agreement.
Why read on?
The simple, globally applicable prediction table for rightward shift of the SpO2 versus PIO2 curve facilitates simple and equitable bedside assessment of BPD severity, and a sensitive and specific clinical measure for comparing pulmonary outcomes of infant cohorts.
Introduction
Bronchopulmonary dysplasia (BPD) is characterised by impaired development of alveoli and pulmonary capillaries. The cardinal feature of BPD is impaired pulmonary gas exchange. A continuous measure of pulmonary gas exchange impairment is essential for benchmarking purposes and as an outcome for clinical research studies.1 2 However, in infants, the traditional measures of pulmonary gas exchange are either technically difficult (A-a nitrogen difference), or impossible (eg, the multiple inert gas elimination technique).3–5 Assessment of rightward shift of the oxyhaemoglobin saturation (SpO2) versus inspired oxygen pressure (PIO2) curve relative to the oxyhaemoglobin dissociation curve will transformthe assessment of BPD severity by providing a sensitive and non-invasive measure of impaired pulmonary gas exchange across the full spectrum of BPD severity.6 Moreover, shift of the SpO2 versus PIO2 can be assessed daily to monitor pulmonary gas exchange.7
The PIO2 versus SpO2 curve is constructed using a comprehensive test by a simultaneous measurement of SpO2 and PIO2 while the PIO2 is reduced in graded fashion using a headbox.8–11 Using the oxyhaemoglobin dissociation curve as a reference, a reduced ventilation to perfusion ratio (VA/Q) is reflected by rightward shift of the SpO2 versus PIO2 curve as the PIO2 – PaO2 gradient increases.8 12 Increased right-to-left shunt displaces the curve downwards. Rightward shift is highly correlated with the PIO2 required to achieve an SpO2 of 90%, whereas shunt is less well correlated with this endpoint.8 Healthy adults have a rightward shift of approximately 6 kPa due to the carbon dioxide in the alveolus and R, the gas exchange ratio (alveolar carbon dioxide tension/R).10 Term infants at day 3 of life have a shift of 5.5 kPa.13
The comprehensive shift test involves multiple paired measurements of SpO2 and PIO2, which often necessitates use of hypoxic gas mixtures to assess infants breathing room air. As such, the comprehensive test is not suitable for wide-scale implementation in routine clinical practice8 9 12 14 15 (figure 1). Rightward shift of the SpO2 versus PIO2 curve was successfully derived from a single paired measurement of SpO2 and PIO2 in infants previously, but the analysis was restricted to infants with moderate to severe BPD and used the adult oxyhaemoglobin dissociation curve as a reference.8
Comprehensive shift test enables assessment of pulmonary gas exchange and monitoring of changes in pulmonary gas exchange. Assessment of shift at day 4 and at day 8 of life in contrast to a predicted normal shift of 7.1 kPa. Foetal oxyhaemoglobin dissociation curve is measured as SaO2/PO2 and predicted shift curves are measured as SpO2/PIO2. ODC, oxyhaemoglobin dissociation curve; PIO2, oxygen pressure; SpO2, oxyhaemoglobin saturation; VA/Q, ventilation to perfusion ratio (adapted from14).
Therefore, we aimed to develop a table and an online calculator to derive shift using a single paired SpO2 and PIO2 measurement obtained from infants across the spectrum of BPD severity and with reference to the fetal oxyhaemoglobin dissociation curve.16 17 The table should include hypobaric PIO2 to enable assessment of shift in infants at high altitudes, and extend to 100% SpO2 to predict shift in preterm infants with minimal lung disease. We hypothesised that the shift value obtained from this prediction table enables assessment of BPD severity and provides an assessment of pulmonary gas exchange comparable to that using multiple paired SpO2 and PIO2 measurements. Some of the results of these studies were reported as abstracts.18 19
Methods
The study was conducted in two phases. First, we derived the prediction table for shift of the SpO2 versus PIO2 curve based on data analysed from infants enrolled in an observational study (ACTRN12613001062718).6 Second, the prediction table was validated in a separate cohort of infants, enrolled in a randomised controlled trial (ACTRN12616000408482). The only difference in the enrolment criteria of the two studies was the gestational age. Infants enrolled in the observational study were born <32 weeks’ gestation, whereas infants enrolled in the randomised controlled trial were born <28 weeks’ gestation. Infants with congenital malformations were excluded from the studies. Written informed consent for assessment of pulmonary gas exchange was obtained for each infant as part of the respective study.
Development of the prediction table
Infants (n=219) contributing gas exchange data to the development of the prediction table were recruited between the 21 July 2013 and the 8 January 2017 as described.6 We assessed rightward shift of the SpO2 versus PIO2 curve by the comprehensive shift test recording multiple paired SpO2 versus PIO2 measurements (online supplemental figure 1s and table 2s).6 Shift of the SpO2 versus PIO2 curve relative to the fetal oxyhaemoglobin dissociation curve was derived using an algorithm.11 The algorithm used a two-compartment model of pulmonary gas exchange (shunt and one perfused compartments) to analyse the relationship between fractional inspired oxygen (FiO2) and SpO2 in terms of pulmonary shunt and VA/Q. The algorithm enables estimation of VA/Q and shunt based on paired values of SpO2 and PIO2 at known haemoglobin content.11
Supplemental material
Rightward shift in the prediction table was calculated using a series of linear regressions. The equation derived from shift (dependent) versus the required PIO2 to achieve a SpO2 range of 86%–94% (independent) was used to calculate shift at a given PIO2 (online supplemental figure 2s and table 2s). Subsequently, the shift values in the table were calculated from the equation of shift at a given PIO2 (dependent) versus the measured SpO2 (independent) (online supplemental figure 3s and table 3s, 4s). The methodology is described in more detail in online supplemental.
Supplemental material
Validation of the prediction table
Shift was calculated for all infants using the prediction table and categorised based on the National Institutes of Health (NIH) BPD classification published in 2001.20 Next, threshold shift values to distinguish between infants with any BPD (mild, moderate or severe) versus infants without BPD were calculated using receiver operating curve (ROC) analyses. Similar thresholds were defined to identify infants with only moderate to severe BPD.
Finally, the prediction table was validated using data collected from a separate group of extremely preterm infants. These infants had both the 15-min bedside assessment (single paired average SpO2 and PIO2) with shift estimated from our prediction table, and a comprehensive shift test (multiple paired SpO2 and PIO2) with shift of the SpO2 versus PIO2 curve estimated using the computer algorithm as described above. Preterm infants were eligible for the validation study if they were (1) born at King Edward Memorial Hospital; (2) enrolled in a randomised controlled trial using shift as an outcome variable; (3) not enrolled in the cohort of infants used to calculate the table and (4) had a 15-min bedside single paired SpO2 and PIO2 measurement within 7 days of the comprehensive shift test without a change in the respiratory support or PIO2 in the intervening period. Infants participating in the validation study were recruited between 7 January 2016 and 5 April 2018.
A Bland-Altman plot was used to quantify the agreement between the 15-min bedside assessment and the comprehensive shift test. The data are presented as a percentage difference plot with 95% CIs of the mean bias and the limits of agreement. Acceptable limits of agreement were defined a priori as ±20% of the mean shift value.
Study size and potential source of bias
The prediction table for shift is based on 219 very preterm infants enrolled in a prospective observational study. For the validation study, we aimed to recruit a minimum of 60 extremely preterm infants who were also enrolled in a randomised controlled trial (ACTRN12616000408482). The data from both studies were analysed twice by two independent investigators, and reanalysed a third time by both investigators together in case of discrepancy in the results.
Statistical methods
Study data were collected and managed using Research Electronic Data Capture software hosted at the University of Western Australia.21 The prediction table for shift was calculated from linear regression of data obtained from the observational cohort (n=219). Threshold values for predicted shift for the classification of infants according to the BDD definition derived from the prediction table were calculated using receiver operating curve (ROC) analyses.
Shift values calculated from the prediction table using the 15-min single paired SpO2 versus PIO2 assessment was compared against the results from the comprehensive shift test using a Bland-Altman plot. Data were analysed statistically using SPSS software (V.25, IBM). A p<0.05 was considered significant. Descriptive statistics are reported as mean and SD for parametric data and median and IQR for non-parametric data.
Results
The recruitment process and the demographic and clinical characteristics of the study population used for the development of the prediction table were described previously.6 Infants were tested at a median (IQR) age of 354 (345–360) weeks postmenstrual age (PMA). The baseline characteristics of the 219 infants enrolled in the observational study are reported in table 1 alongside the characteristics of the infants enrolled in the validation study. None of the infants required mechanical ventilation at the time of the test. However, some infants were measured on continuous positive airway pressure.
Summary characteristics of both cohorts of infants
The accuracy of predicted shift depends on the position of the single SpO2 versus PIO2 measurement on the oxyhaemoglobin dissociation curve. Measurements with SpO2 values >94% are located close to the plateau of the curve, whereas measurements with SpO2 values<94% are located on the steep part of the curve. The coefficient of determination (R2) between shift and SpO2 decreases at higher SpO2 levels, leading to an inherent fall in the prediction accuracy of shift. All values in table 2 are derived from infants breathing room air at sea level (PIO2=20 kPa). Since PIO2=FiO2 x (barometric pressure – saturated water vapour pressure), an FiO2 of 0.21 is equal to 20 kPa PIO2.
Shift dependent on SpO2 in infants breathing room air at sea level (PIO2=20 kPa)
Prediction table for shift
The prediction table for the estimation of shift derived from the fetal oxyhaemoglobin dissociation curve is reported in table 3. An individual shift calculator and a bulk calculator are provided in the supplementary material of this manuscript.
Prediction table for shift from a single paired SpO2 versusPIO2 measurement
Bedside assessment and NIH BPD definition
The differences in shift between infants with none or mild, mild or moderate and moderate or severe BPD were statistically significant (figure 2). Threshold values for shift derived from the 15-min bedside assessment for infants with any BPD (mild, moderate or severe—representing the transition from no BPD to at least mild BPD), and infants with moderate or severe BPD (vs no BPD) are shown in table 4.
Thresholds for shift (kPa) for any BPD and moderate-to-severe BPD
Shift assessed with the 15-min bedside assessment in relation with the NIH BPD definition published in 2001. Infants breathing room air on continuous positive pressure respiratory support (severe BPD) are not included in the graph. BPD, bronchopulmonary dysplasia.
Validation of the prediction table
A total of 63 infants in the validation cohort underwent both the 15-min bedside assessment and the comprehensive shift test. Median (range) date difference between the two tests was 0 (−5.0, 3.0) days. Summary characteristics of preterm infants enrolled in the validation study and significant differences between infants enrolled in the observational study compared with the validation study are shown in table 1. The linear correlation between shift derived from the 15-min bedside assessment and shift assessed by the comprehensive test was statistically significant (R2=0.83, p<0.0001) (figure 3). Bland-Altman analysis showed a non-significant mean bias (95% CI) between the 15-min bedside assessment and the comprehensive shift test of 2.1 (−0.8 to 4.9)%. The limits of agreement extended from −20.7% to 24.8%, which just exceeded our prespecified acceptable limits of agreement of ±20% (figure 4).
Scatter plot: shift assessed by the 15-min bedside assessment dependent on shift assessed by the comprehensive test. R2=0.83; p<0.0001.
Plot of differences between the 15-min bedside assessment and the comprehensive shift test, expressed as percentage difference from the mean of two tests. Each open circle denotes a single infant. Solid black line=mean; dotted line=95% CI of the mean; dashed line=limits of agreement (1.96 SD).
Discussion
Impaired pulmonary gas exchange is the fundamental physiological disorder in infants with BPD. Gas exchange can be assessed on a daily basis in a non-invasive and quantitative manner.6 8 10 22 In contrast, BPD assessments based on prescribed oxygen dependency provide a qualitative assessment of the disease, are highly subjective due to differing oxygen prescription behaviours, and are not usually adjusted for altitude. Furthermore, different qualitative BPD definitions are used in neonatal units and as endpoints in clinical trials.1 Together, these factors make it difficult to use conventional clinical data to provide meaningful and equitable benchmarking of respiratory outcomes between neonatal units, to compare interventional clinical trial outcomes and to interpret study results.1 23
Our study validates a viable, simple, non-invasive and quantitative assessment of BPD severity for use in preterm infants prior to discharge home. We showed that VA/Q measured from multiple paired samples of SpO2 and PIO2 is reflected by the degree of rightward shift of the SpO2 versus PIO2 curve and can be predicted from a single paired SpO2 versus PIO2 measurement.6 We developed a prediction table and an online calculator for shift of the SpO2 versus PIO2 curve. The prediction table enables bedside assessment of pulmonary gas exchange in preterm infants. The table is applicable across the full spectrum of BPD severity, can be used equitably in neonatal units at differing altitudes and is based on fetal haemoglobin. The PIO2 range of the table extends from as low as 14 kPa to 95 kPa. The extension to as low as 14 kPa enables estimation of shift from the prediction table in infants breathing room air up to an altitude of 3 000 m above sea level.23 24 For an example, 0.21 FIO2 (room air) is equal to 15.6 kPa PIO2 in a neonatal unit in Mexico City at 2 250 m altitude. Conversely, a FIO2 of 1.0 is equal to 95 kPa PIO2 at sea level, hence the upper limit of 95 kPa PIO2 in the prediction table.
The shift values in our table are based on the fetal oxyhaemoglobin dissociation curve as a reference. The switch from fetal to adult haemoglobin occurs around 42–44 weeks independent of the infant’s gestational age16 or number of prior blood transfusions.25–27 The use of fetal haemoglobin as the reference is justified further in the online supplement.
A predicted shift value of 10.1 kPa enables the clinician to classify an infant as having BPD according to the 2001 NIH classification of BPD with a sensitivity of 71% and a specificity of 88%. Moreover, a shift value of 13.0 kPa allows assessment of moderate or severe BPD with a sensitivity of 81% and a specificity of 100%. The shift threshold values are similar to the threshold values reported in our previous manuscript.6 Therefore, shift predicted from a single paired SpO2 versus PIO2 measurement enables objective physiologic assessment of BPD severity according to the current widely used clinical classification for severity of BPD in an individual infant.
Validation of the table showed no bias and no mean difference between shift values estimated from the prediction table and shift values derived from the comprehensive shift test, indicating that the prediction table is useful for assessing pulmonary gas exchange at a population level. However, the limits of agreement exceeded the predefined criterion of ±20%. The ±20% acceptability criterion for the Bland-Altman analysis was based on the fall in prediction accuracy of shift at SpO2 levels>94% (as shown in table 2), and on the accuracy of the SpO2 probe (±3%) and the portable oxygen sensor (±2%). Bland-Altman limits of agreement exceeding the prespecified ±20% acceptability criterion indicates that a single measure of shift derived from the prediction table is not sufficiently accurate to identify changes in pulmonary gas exchange at the individual level. Even though the 15-min bedside assessment of shift does not allow assessment of right-to-left shunt, we could rule out shunt as an explanatory factor for the wide limits of agreement against the more comprehensive and lengthier assessment of right shift (data available on request). Nevertheless, as the limits of agreement were close to the predefined acceptable limits, future refinements to the method might improve the prediction of shift at an individual level. The 15-min test remains useful at an individual level for an objective screening and identification of infants who may benefit from subsequently undergoing a comprehensive shift test. We also showed that the 15-min test has high sensitivity and specificity at an individual level for identification of infants with a clinical diagnosis of moderate-to-severe BPD.
The recorded SpO2 used to predict shift of the SpO2 versus PIO2 curve from the table is key to understanding these wide limits of agreement. A single point measurement on the plateau of the curve (ie, high SpO2) can have multiple solutions at the steep part of the curve8 (figure 1). Predicted shift values from SpO2 >94% when breathing room air are therefore only estimates, and require a comprehensive shift test incorporating hypoxic gas mixtures to more accurately assess shift.
Preterm infants with shift values>20 kPa may also benefit from the comprehensive measurement of shift using a headbox.6 The comprehensive measurement of shift using a headbox provides additional information including assessment of right-to-left shunt, VA/Q and the possibility of identifying the presence of multiple gas exchange compartments in the lung.14 Hence, undertaking a comprehensive shift test would provide more accurate evaluation of pulmonary gas exchange for the effect of treatment, future prognosis and planning of clinical follow-up at an individual level.6
The strengths of our study include the large sample size, use of an infant cohort that reflected the whole spectrum of BPD severity, a table of predicted shift values that allows for adjustment for altitude and the use of developmentally appropriate fetal haemoglobin as a reference. In contrast to the comprehensive test, our prediction table for shift allows clinicians to assess pulmonary gas exchange impairment from a simple bedside assessment using widely accessible SpO2 monitoring, without the need of hypobaric testing in preterm infants at 36 w PMA. Paired measurements of SpO2 versus PIO2 can be taken bedside by any healthcare provider at any time of the day to promote benchmarking of neonatal respiratory outcomes after preterm birth. Moreover, predicted shift values from a single paired SpO2 versus PIO2 enable classification into BPD severity grades at an individual level.
A limitation of our study is that shift values derived from a single paired SpO2 versus PIO2 measurement using the prediction table are less accurate at an individual level. This reduced accuracy of individual shift values limit utility for using the 15-min shift test to assess short-term changes in clinical condition. The accuracy decreases especially in infants with SpO2 >94%. We also validated our prediction table with a slightly more immature group of preterm infants than used for the development of the prediction table. However, the rate of infants with moderate or severe BPD was equal between groups. The infant cohort for the validation study was 4 days older at test compared with the infants measured for the prediction table development (36 w PMA versus 35.4 w PMA). However, shift does not decrease significantly between 35 w and 36 w PMA.6
The use of different BPD definitions in clinical trials makes the interpretation and the comparison between the trials difficult. In contrast to the qualitative assessments of BPD, we and other groups showed previously how pulmonary gas exchange and BPD severity can be assessed in a quantitative manner using a continuous outcome measure. Simultaneous measurements of SpO2 and PIO2 enable quantification of pulmonary gas exchange. However, the comprehensive test requires multiple measurements of SpO2 and PIO2 and also hypoxic gas mixtures for infants with mild BPD not requiring supplemental oxygen and is therefore not suitable for daily clinical practice.
Conclusion and future directions
Rightward shift of the SpO2 versus PIO2 curve relative to the oxyhaemoglobin dissociation curve provides a non-invasive, and continuous population level outcome measure for pulmonary gas exchange in infants across the spectrum of BPD severity. Moreover, the bedside assessment is suitable for the classification of individual infants into BPD severity grades according to the NIH BPD definition.20 Our prediction table for shift from a single paired measurement of SpO2 and PIO2 is based on fetal haemoglobin as appropriate for <42 w GA and is therefore suited for the assessment of pulmonary gas exchange at 36 w PMA. The prediction table is deliberately limited to 85%–94% SpO2 (except for measurements at 20 kPa), reflecting a fall in accuracy of shift predictions at SpO2 above 94%. Further research is required to evaluate whether our method can be refined. Paired measurements of SpO2 and PIO2 from a 15-min bedside assessment used with our prediction table allows accurate estimation of shift in preterm infants at a population level without requirement for hypoxic gas mixtures for the assessment. Therefore, the bedside assessment is suitable as a benchmarking tool for assessing pulmonary gas exchange in infants with BPD between neonatal intensive care units, as well as for assessing treatment outcomes of clinical interventions in large randomised controlled trials.
Acknowledgments
The authors would like to acknowledge the research nurses Natasha Mackay-Coghill and Amanda Woods, as well as all the included infants and their parents. We acknowledge Dr Patricia Woods for the coordination of the ANZNN shift test at the King Edward Memorial Hospital.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
BS and YJC are joint first authors.
Twitter @jane_pillow
Correction notice This article has been corrected since it was published Online First. Author last name for GJ has been amended and first initial for JC has been added. A duplicate reference has also been removed.
Contributors BS collected and analysed the data, interpreted the data, assisted with the REDCap database design and development, drafted the initial manuscript, performed literature search, drafted the figures and approved the final manuscript as submitted. JC collected and analysed the data, interpreted the data, drafted the figures, reviewed and revised the manuscript, and approved the final manuscript as submitted. AR collected the data, reviewed and approved the final version of the manuscript as submitted. JS collected the data, analysed the data, reviewed and approved the final version of the manuscript as submitted. GJ developed the algorithms used for the calculation of shunt, shift and VA/Q, reviewed and revised the manuscript, and approved the final manuscript as submitted. JP was the principal investigator obtaining funding, leading study design including development of the REDCap database, obtained the study funding, verified all statistical calculations, interpreted the data, critically reviewed and revised the manuscript and approved the final manuscript as submitted.
Funding All phases of this study were supported by the University of Western Australia and the Women and Newborn Health Service of Western Australia. Funded by National Health and Medical Research Council (NHMRC) of Australia (GNT1047689, GNT1057514) and the Metropolitan Health Research Infrastructure Fund (MHRIF). BS was supported by the Swiss National Science Foundation (P2BSP3_158837) and a Research Training Program scholarship from The University of Western Australia. JS was supported by The Swedish Society of Medicine, The Samaritan Foundation, The Princess Lovisa Memory Foundation, The Society of Child Care (Sällskapet Barnavård), and The Fernstrom Foundation. JP was supported by an NHMRC Senior Research Fellowship (RF1077691).
Competing interests None declared.
Patient consent for publication Not required.
Ethics approval Each clinical study received ethics approval from the Women and Newborn Health Service Human Research Ethics Committee (HREC: 1883EW and 20 130 193EW respectively) and the University of Western Australia (RA/3/1/5942 and RA/4/1/426, respectively). Studies were conducted in the Neonatal Clinical Care Unit at the King Edward Memorial Hospital in Perth, Western Australia.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement All data relevant to the study are included in the article or uploaded as online supplemental information. We have collected deidentified participant data and we will make our data available on reasonable request. ORCID-ID: 0000-0003-0021-0262.