Despite great improvements in the survival of infants born prematurely, there continues to be a large number of infants who develop bronchopulmonary dysplasia (BPD). This causes them to remain in hospital longer, prolongs their requirement for supplemental oxygen, and is associated with long-term morbidity and an increased risk of mortality. Reducing BPD remains a major focus of clinical and research activity.

An objective definition of BPD is required to enable reliable interpretation of clinical trial outcomes and to serve as a baseline in prognostic studies. Yet, an ideal definition has been elusive. Definitions based on the infant having a requirement for supplemental oxygen at 28 days of life¹ or at 36 weeks’ gestation² have been used widely, but their usefulness is severely limited by the marked variation among clinicians in their criteria for oxygen supplementation.³ A definition based on the use of oxygen treatment alone gives wide variations in the incidence of disease, which reflect little more than clinician variation and have little relevance to the severity of any underlying pathology.⁴ Recently, a physiological definition has been proposed that aims to remove this bias by defining BPD as a requirement for supplemental oxygen, to maintain an oxygen saturation of 90% at 36 weeks’ gestation.⁴ This is undoubtedly an advance. However, healthy preterm and term infants have saturations around 97% in air,^5,6 and saturations lower than this in air must reflect a degree of gas exchange impairment, even if supplemental oxygen is not always deemed necessary. Present approaches to defining BPD classify these infants as disease free.

By non-invasive measurements of PIo₂ and Spo₂, it is possible to quantify the severity of gas exchange impairment in a graded fashion and to partition this between the contribution made by reduced ventilation:perfusion ratio (V_A:Q) and that made by right to left shunt.^7,8,9,10,11 We have applied this method to analyse the gas exchange abnormalities in infants with BPD and used these observations to model an improved approach to the definition of BPD, which measures the severity of gas exchange impairment.

METHODS

Underlying physiology

A reduced V_A:Q ratio and an increased shunt have different effects on the relationship between inspired oxygen pressure (PIo₂) and arterial oxygen saturation (Spo₂). A reduced V_A:Q ratio decreases alveolar and arterial oxygen tension (Po₂) and raises alveolar and arterial carbon dioxide tension (Pco₂). Increasing PIo₂ restores the alveolar Po₂ and Spo₂ to normal, overcoming the effect of the reduced V_A:Q ratio. Increased shunt does not raise Pco₂ but reduces Spo₂ because the shunted blood is not exposed to alveolar oxygen. Increasing PIo₂ can compensate for only a small amount of shunt, because the non-shunted blood is already almost fully saturated and does not carry much more oxygen than small amounts in solution when PIo₂ is increased. These independent effects on gas exchange can be represented in the form of plots of Spo₂ against PIo₂ (fig 1).^8,9,10,11,12

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

 Plots of oxyhaemoglobin saturation (Spo₂, %) against inspired oxygen pressure (PIo₂, kPa).¹⁰ (A) Increasing shunt from 0% to 40% lowers the position of the upper part of the curve. (B) Reducing ventilation:perfusion ratio (V_A:Q) from 0.8 to 0.1 shifts the curve to the right. The right shift of each PIo₂–Spo₂ curve from the position of the dissociation curve (dashed line) is the PIo₂–Pao₂ difference (kPa), which includes Paco₂/R. The 0.8 curve represents the normal adult curve, which intercepts a PIo₂ of 21 kPa (vertical line) at 97% Spo₂. R, respiratory gas exchange rate.

At sea level (1 atm), PIo₂ (kPa) is the same as the inspired oxygen percentage. The curve relating alveolar (mixed capillary) Po₂ to oxygen saturation in the ideal lung represents the shape of the oxyhaemoglobin dissociation curve. Increasing shunt displaces the top part of the curve downwards (fig 1A) as the maximum Spo₂ obtainable falls. In contrast, reducing the V_A:Q ratio shifts the whole curve to the right (fig 1B). The degree of the right shift, using the oxygen dissociation curve as the reference, is determined by the reduction in the V_A:Q ratio and a rise in alveolar PCo₂. The normal curve is shifted to the right of the haemoglobin–oxygen dissociation curve by 6 kPa, which is largely Pco₂/R (where R is the respiratory gas exchange ratio). An additional right shift compared with the normal represents the increase in PIo₂ that will be required to restore the mixed capillary Po₂ to normal levels and thereby permit normal arterial saturation. If multiple pairs of PIo₂ and Spo₂ values are obtained from the same patient, then a single pair of shunt and shift values can be derived. This can be done graphically^7–9 by moving a set of shunt curves such as those in fig 1A laterally over the plot of PIo₂–Spo₂ data points until one of the shunt curves superimposes the data points. The degree of shift can then be read off the axis of the graph, and the shunt can be determined by selecting the shunt curve that most closely fits the data. Alternatively, shift and shunt can be calculated using a computer algorithm that derives confidence intervals for the shunt and shift values, and coefficients of determination (R²) for the fit of the data to the shunt and shift model.^8,9,10

Procedure

We studied 21 preterm infants who were considered to have BPD on the basis of a continuing requirement for supplemental oxygen at 36 weeks’ gestation. The study was approved by the institutional ethics advisory committee and written informed consent was obtained from the parents in all cases. This was a convenience sample. Infants were included if they were receiving supplemental oxygen but not other support at the time of study. According to our unit policy, oxygen was being given as required to maintain oxygen saturation in the target range 86–94%. All infants were being cared for in the neonatal unit of the Simpson Centre for Reproductive Health, Royal Infirmary of Edinburgh, Edinburgh, UK, between January 2002 and March 2005. All studies were carried out within 2 weeks of the infant reaching 36 weeks’ gestation.

At the time of study all infants were receiving oxygen via nasal cannulae. To enable measurement and control of PIo₂, the infants were placed in a neonatal intensive care incubator with oxygen under servo control (Draeger 8000IC, Draeger Medical AG & Co KGaA, Lübeck, Germany). The oxygen analyser was calibrated at the start of each study. Functional Spo₂ was measured using either a Siemens SC7800 (Siemens Medical Systems Inc, Danvers, MA, USA) or a Philips M3046A multiparameter patient monitor (Agilent Technologies, Boblingen, Germany), depending on which nursery the infant was being nursed in. Studies began at least 30 min after a feed and were conducted with the infant lying supine. PIo₂ was reduced in increments of 1–2% (1–2 kPa) and then increased again to vary Spo₂ between 94% and 86%. At each PIo₂ value, the infant was allowed to stabilise for about 5 min before a pair of Spo₂ and PIo₂ values was recorded. Values were recorded only if there was a good pulse waveform on the oximeter and the infant was not displaying gross body movements.

Sleep state was not standardised. PIo₂ was never reduced below 21 kPa, and oxygen administration to saturation >94% was minimised as per unit policy.

The pairs of PIo₂ and Spo₂ data points obtained from each infant were plotted and analysed using a computer algorithm that gave a curve representing a single solution for shunt and shift (the difference between PIo₂ and mixed capillary Po₂) for each infant’s dataset.^8,9,10 If the data pairs obtained did not describe the plateau of the oxyhaemoglobin dissociation curve, the algorithm was sometimes unable to calculate the shunt and shift values. Under these circumstances the data were plotted manually, and shunt and shift were determined from the graphs directly using the method described above.

As right shift is due to the combined effects of raised Pco₂ and reduced V_A:Q ratio, we determined the contribution of these two variables using a mathematical model of gas exchange described by Olszowka and Wagner.¹² PIo₂–Spo₂ curves were constructed for V_A:Q values from 0.9 to 0.15, and the predicted Paco₂ value was calculated from the model. From this we related the shift value in each infant to a particular V_A:Q and compared the child’s most recent Pco₂ measurement with the predicted Pco₂.

RESULTS

Table 1 describes the clinical characteristics of the 21 infants. All infants were receiving supplemental oxygen at the time of study. There was wide variation among the infants in the length of time that each had been on a ventilator or supported with continuous positive airway pressure. Some were still cycling on and off continuous positive airway pressure.

The main finding was that in every infant the PIo₂ versus Spo₂ curve was shifted to the right of the oxygen dissociation curve. The mean (standard deviation (SD)) shift was 16.5 (4.7) kPa (table 2). The normal PIo₂–Spo₂ curve is shifted 6 kPa to the right of the dissociation curve and is identical to the V_A:Q 0.8 curve shown in fig 1B. This intercepts the vertical line at 21 kPa PIo₂ at an Spo₂ of about 97%. Figure 2 shows the data plots of infants 2, 10, 11 and 19 in relation to the oxyhaemoglobin dissociation (ideal) curve and the normal adult PIo₂ versus Spo₂ curve. Table 2 shows the values for shunt and right shift for all 21 infants, with 95% confidence interval (CI) and R². In three infants (5, 8 and 18), the range of paired values of PIo₂ and Spo₂ obtained did not describe enough of the plateau of their PIo₂ versus Spo₂ relationship to enable the algorithm to calculate the shunt and shift values. Their data points were plotted, and the shunt and shift values were determined manually from the graphs as described above; data of these infants do not therefore include CI or R². All three infants had Spo₂ ⩾90% breathing air.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

 Plots of oxyhaemoglobin (OxyHb) saturation (Spo₂, %) versus inspired oxygen (PIo₂, kPa) for infants 2, 10, 11 and 19. Reference grid lines are added at an Spo₂ of 90% and a PIo₂ of 21 kPa. The haemoglobin oxygen dissociation curve (ideal lung) is the reference point for derivation of shift. The normal Spo₂–PIO₂ curve is also included for comparison.

As a significant right shift of the PIo₂ versus Spo₂ curve occurred in every case, and was a sensitive measure of a reduced V_A:Q ratio, we looked for an index of right shift that might obviate the need to produce a range of different PIo₂ values in each case. In every infant, the 90% Spo₂ value fell on the steep part of the dissociation curve so that the PIo₂ needed to produce 90% Spo₂ might be a candidate marker of right shift. Figure 3 shows plots of the degree of shunt and shift in each infant against the PIo₂ required to achieve 90% Spo₂. At this Spo₂, the relationship between PIo₂ and shift was highly siginificant and linear, whereas that between PIo₂ and shunt was weak, indicating that shift was the main determinant of reduced Spo₂ in these infants.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

 (A) Partial pressure of inspired oxygen (PIo₂) required to achieve an arterial oxygen saturation (Spo₂) of 90% versus shunt. (B) PIo₂ required to achieve an Spo₂ of 90% versus shift.

We constructed plots of shift against the PIo₂ required to achieve each saturation in the range 86–94% Spo₂, and found linear relationships between shift and PIo₂ at all of the Spo₂ values in this range, with all values of R²>0.9. The consistency of the relationship between shift and PIo₂ in this range was such that multiple data pairs in each infant were not required to derive shift. A single pair of PIo₂ and Spo₂ values in the Spo₂ range 86–94% was sufficient to predict the shift value that would be derived from the whole data series. The mean (SD) difference between shift calculated from individual data pairs and that calculated from the entire data series for each infant was 0.24 (1.25) kPa.

To enable others to derive shift from a single pair of Spo₂ and PIo₂ values, shift was calculated from the results of this study for a range of PIo₂ and Spo₂ values that are likely to be observed in infants with BPD at 36 weeks’ gestation. Table 3 shows the results.

Pco₂ predicted by the model of Olszowka and Wagner¹² for the degree of a reduced V_A:Q ratio in each infant was linearly related to measured Pco₂ (y = 0.68x+3.2, R² = 0.6). This is only a consistency check on the model, because in many cases the most recent Pco₂ value available was obtained several days or more from the time of the study. When this analysis was restricted to the six infants with Pco₂ values obtained within 1 day of the study, there was a close correlation (y = 0.88x+1.5, R² = 0.99).

DISCUSSION

Series of paired values of Spo₂ and PIo₂ were used in preterm infants with BPD to show, non-invasively, an increase in shunt and a reduction in the V_A:Q ratio. The reduced V_A:Q ratio was the dominant gas exchange defect causing a major right shift of the Spo₂ versus PIo₂ curve in all these infants. The right shift of the curve explained the need for an increased PIo₂, and there was a strong linear relationship between the degree of shift and the PIo₂ required to achieve any chosen saturation in the range 86–94%. The consistency of this finding was such that a single measurement of the PIo₂ required to achieve any Spo₂ in the range 86–94% was sufficient to derive shift. Presently, no definition of BPD exists that provides both a robust physiological threshold for diagnosis and a continuous scale of severity that is based on the degree of gas exchange impairment. We believe that the degree of right shift of the relationship between Spo₂ and PIo₂ provides such a measure. This could be expressed either as the shift in kPa or as the PIo₂ required to maintain a saturation of 90%. Such information has previously been derived by more invasive methods.

This non-invasive method of partitioning gas exchange impairment has been applied in sick infants,¹⁰ and in healthy and sick adults.^8,9,11 The results showed a good fit between the model and the clinical data in all age groups and disease states studied. Kjaergaard et al,¹¹ using saturation measurements non-invasively, obtained results almost identical to those obtained from simultaneous more invasive measurements. Iles and Edmunds¹³ showed that measures of gas exchange impairment derived from arterial blood gases sampled around term predict the prognosis of BPD, and that infants with low saturation are at higher risk of acute life-threatening events.¹⁴

The recent National Institutes of Health workshop consensus definition of BPD¹⁵ categorises infants as having no BPD, or with mild, moderate or severe BPD according to the amount of oxygen supplementation and ventilatory support required up to 36 weeks’ gestation. These categories predict later respiratory morbidity,¹⁶ but they are not physiologically based; and respiratory problems are also seen later in life in a substantial number of infants not identified to have BPD by this definition. Infants with BPD have reduced numbers of alveoli, enlarged air spaces, interstitial fibrosis and variable degrees of small airway narrowing.¹⁷ Defining the degree of physiological impairment associated with this pathology is likely to be more informative than quantifying the preceding exposure to treatment. Similar to the Walsh test,⁴ the National Institutes of Health network definition¹⁵ is likely to permit infants with saturations lower than those observed in healthy infants to be categorised as having no BPD. Measurement of the reduced V_A:Q ratio in terms of shift enables the determination of the degree of gas exchange impairment in all infants, including those who are stable in air, and may provide a more detailed description of study outcomes and serve as a more informative baseline for studies on the prognosis of BPD.

The Walsh test⁴ is not really a threshold test, because many infants fail it during the 30-min observation period after having adequate saturation to begin with. In a recent study, infants with a saturation of >96% with an effective fraction of inspired oxygen <23% had a positive predictive value of 66% for passing the Walsh test.¹⁸ This reflects the fact that an oxygen saturation around 90% lies on the steep part of the dissociation curve, where reducing the PIo₂ by 1% is associated with changes in saturation of ⩾2–3%, and small changes in alveolar ventilation will therefore cause considerable desaturation (fig 2). Using our method, a mean saturation value obtained at a fixed PIo₂ during a brief observation period while an infant was lying supine and free from gross body movements could be used to determine whether the infant would have a saturation of 90% in air.

We have not measured the day-to-day variability of our test within patients. The accuracy of the test is dependent on the accuracy of the measurement of PIo₂ and Spo₂. The issue of PIo₂ measurement can be eliminated if Spo₂ is determined with the infant breathing air. Current Spo₂ monitors are accurate to 1–3% in the range studied.¹⁹ On the steep part of the dissociation curve, changes in Spo₂ of this magnitude have a small effect on shift.

The saturation values used to derive the models used in the methods described above were derived from the normal adult haemoglobin oxygen dissociation curve.²⁰ Curves of newborn infants may differ slightly in the presence of varying amounts of fetal haemoglobin. However, by the time of our study, the infants were around 11 weeks old. Beresford et al⁵ showed that healthy 1-week-old and 6-month-old preterm infants have mean saturation values of around 97% in air, and Parkins et al⁶ showed similar values in healthy term infants at 3 months of age; so there is probably no major bias from using the normal curve in adults as a reference point for calculating shift by the time these infants reach 36 weeks’ gestation. The model assumes a fixed arteriovenous oxygen gradient of 5 ml/100 ml. Any variation in this within or between individuals affects the position of the plateau of the oxygen dissociation curve but has little effect on its shape, and should not therefore have much effect on calculations of shift.⁸ Similarly, estimations of shunt are more sensitive to haemoglobin level than estimations of shift.⁸

In conclusion, we have shown that the contribution of reduced V_A:Q (shift) and shunt to impaired gas exchange can be derived easily from non-invasive measurements of Spo₂ and PIo₂ in infants with BPD. The degree of right shift of the Spo₂ versus PIo₂ curve provides a readily accessible continuous measure of the severity of gas exchange impairment due to a reduced V_A:Q ratio, which is a more dominant contributor than shunt. This measure could be used to define BPD as an outcome in studies and to categorise infants in prospective studies on BPD prognosis.