Results

Data were analysed from all 27 patients (14 girls) successfully monitored in the three study centres. Median (range) birthweight and gestation were 1170 (560–3200) g and 28 (24–40) weeks, respectively. Median (range) time for MPIAS monitoring was 93.9 (24.1–428.8) hours. Nine patients were monitored for more than 7 days. Median (range) of complete datasets per patient was 25 (12–75). There was a total of 3260 hours of monitoring time, generating 753 complete datasets. Bland and Altman analyses8 were performed on the data from all patients. Bland and Altman plots are shown for pH (fig 2A), PaCO₂ (fig 2B), and PaO₂ (fig 2C). Overall bias (mean difference: MPIAS–ABG) and precision (standard deviation of differences) for pH, PaCO₂, and PaO₂ are shown in table 1.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Bland and Altman plots for combined data, for each variable. Overall bias is shown as bold horizontal line. 95% upper and lower limits of agreement are shown as thinner horizontal lines.

There was a tendency for the MPIAS to over read the PaCO₂. Indeed, the bias was positive in 23 out of 27 patients, although the range of mean bias for individual patients was −0.3 to +1.0 kPa. This wide range in bias is apparent for the other variables and has an important effect on interpreting overall precision values when data from different patients are combined. This is of particular significance in continuous ABG monitoring, where bias can be reduced or eliminated by recalibration against a true ABG and trend analysis is clinically important. To address this difficulty in interpreting the overall precision figure, we have produced bias/precision data for individual patients. These are shown for pH (fig 3A), PaCO₂(fig 3B), and PaO₂ (fig 3C). For each variable, the precision value for most individual patients is below that of the overall precision figure: pH (23 patients), PaCO₂ (24 patients), and PaO₂ (19 patients). This clearly shows that precision is much better within individual patients than in the overall data, for all three measurements.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Bias/precision plots for 27 patients, for each variable. Overall precision (1 SD) for each variable is shown as bold horizontal line.

Recalibrations were kept to a minimum: 17 patients did not require any recalibration. The variables recalibrated in the remaining 10 patients were: pH (four patients), PaCO₂ (five patients), and PaO₂ (eight patients). All were single recalibrations. Thus a total of 17 recalibrations (any variable) were made over 3280 hours of monitoring. This indicates that there was a minimal tendency for the MPIAS to drift with time. The most common scenarios leading to a need for calibration were: a drift down in PaO₂ over the first 24 hours (then stopping) or a very slow drift up in PaCO₂after 5 to 7 days of monitoring. Clearly, these were not consistent findings and affected only a few patients.

Any adverse events during monitoring were recorded, as were the reasons for removal of the umbilical arterial catheter. No problems with blood pressure transducer monitoring were observed and there was no damping of the trace with the current SIS design. In 16 patients the catheter was removed because it was no longer required; in eight the catheter stopped sampling. Two patients died because of severe lung disease. The remaining patient had the catheter removed after 25 hours because of haematuria. This was related to Escherichia coli septicaemia, disseminated intravascular coagulation (DIC), and a catheter that had migrated from T10 to L1 onx-ray picture. Renal and Doppler ultrasonography remained normal and the patient made a complete recovery. One patient (24 weeks gestation, 770 g) had an extremely difficult delivery, with cord prolapse and the need for rotation before assisted breech delivery. Examination of the placenta revealed a large retro-placental clot. At delivery there was a severe metabolic acidosis. She was very ill with evidence of a coagulopathy. This infant developed signs of paraplegia. Magnetic resonance imaging revealed three areas of spinal cord haemorrhage between the mid thoracic and lumbar regions consistent with trauma rather than vascular accident. This was confirmed by an independent neurological and neuroradiological opinion.

During the trial, a further 19 patients underwent attempts to insert an MPIAS. The rate of successful placement was specificallynot an aim of this study, because it was anticipated that the SIS would require modification when clinical use started. We have reported it because it is clearly clinically important. In 11 patients the insertion was either unsuccessful (n=8) or resulted in an incorrectly placed sensor (n=3). As indicated in the methods, this did not result from difficulties in passing the MPIAS down the lumen of the umbilical arterial catheter, but related to difficulties with the seal mechanism (O ring) between the catheter and the SIS. The proportion of successful insertions increased as the SIS was modified. In the remaining eight patients the catheter had to be removed before generating enough data (all underwent two or more ABG samples). This group included patients who developed transient lower limb duskiness (n=2), or who had catheters that stopped sampling (n=2), or were no longer required (4 samples). No other complications were seen in this group.

Discussion

There are several reasons why continuous ABG monitoring in neonatal intensive care may be clinically beneficial. The effects of changes in PaCO₂ on cerebral blood flow in preterm infants have been well described.9 Both hypercapnia and hypocarbia are associated with the development of intraventricular haemorrhage and periventricular leucomalacia.10-13 Therefore, in preterm infants PaCO₂ has to be kept within strict limits, particularly in the first 72 hours of intensive care. The need to avoid hyperoxia as well as hypoxia has similar implications for PaO₂. Moreover, rapid changes in ABG can occur after surfactant administration, especially natural surfactants,14 and during high frequency oscillatory ventilation. Indeed, some workers have suggested patients receiving the latter should always have continuous PCO₂monitoring.15

The Neotrend MPIAS provides pH sensor readings that agree closely with those of ABG analysis, with negligible bias and good precision. The PaCO₂ sensor showed positive bias in most individuals with acceptable precision. Finally, the PaO₂ sensor showed an inconsistent bias with acceptable precision. The fourth variable, aortic temperature, was not studied, as there is no standard for comparison available, although it is a potentially useful clinical indicator in sepsis. The biases shown for individual sensors are similar to the pattern reported for the Paratrend MPIAS in adult1 16 17 and paediatric7 intensive care patients. However, the precision is better than demonstrated in most of these studies, particularly for PaO₂. This is partly because the earlier Paratrend sensor monitored the PaO₂with a Clark electrode. Furthermore, there are difficulties with peripheral arterial lines because of artefacts resulting from limb movement and changing blood flow.16 The neonatal descending aorta does not create these difficulties.

It is also important to consider the precision of ABG analysers. A recent study18 found negligible variation within instruments, where precision (SD of the differences) values of 0.03 and 0.06 kPa were reported for PaCO₂ and PaO₂, respectively. Comparison of 10 different analysers gave precision values of 0.13 and 0.18 kPa for PaCO₂ and PaO₂, respectively, although the full clinical range of PaCO₂ and PaO₂ values was not tested. Other comparisons between two ABG analysers, using the whole ABG range, have shown less precision, with 0.011, 0.18, and 2.8 kPa reported for pH, PaCO_2, and PaO₂, respectively.19 These figures are similar to those found for portable ABG analysers, although PaCO₂ precision varied from 0.26–0.48 kPa in recent studies.19-21

Clearly, it is useful to consider the variation between two ABG analysers (two gold standards) when interpreting the ABG analyser/Neotrend MPIAS precision figures. The level of overall precision for the Neotrend MPIAS is not sufficient to eliminate the need for ABG sampling. Nevertheless, the precision of the MPIAS will considerably reduce the frequency of ABG sampling and therefore the need for blood transfusion. Our clinical experience also suggests that MPIAS monitoring facilitates earlier detection of endotracheal tube obstruction and pneumothorax than current monitoring methods. Similarly, MPIAS monitoring seemed to have beneficial effects on the speed of ventilator weaning and the avoidance/reduction of periods of hypo/hypercapnia. Such anecdotal findings point to the need for controlled trials to assess both the potential short and long term benefits.

MPIAS monitoring also has to be considered in the light of non-invasive alternative techniques. Clearly, non-invasive methods of continuous monitoring are still appropriate in infants in whom umbilical arterial catheter insertion is unsuccessful or not clinically indicated. However, MPIAS monitoring is a highly attractive option in those who do merit a catheter. Intra-arterial PaO₂ monitoring via the Clark electrode is well established in neonatal intensive care.4 5 Precision of the Clark electrode has not been subject to Bland and Altman analysis. The reported correlation coefficients are similar to that of the MPIAS, particularly if an allowance for the Clark electrode response time is made when it is calibrated.22 Moreover, the use of a fluorescence sensor instead of a Clark electrode for PaO₂ monitoring, seems to avoid the drift with time, associated with the latter.5

With intra-arterial PaO₂ monitoring already available, the justification for MPIAS monitoring has to come from monitoring the other variables. This essentially asks: how well does PaCO₂monitoring compare with transcutaneous PCO₂(TcPCO₂) monitoring? This comparison is not straightforward. The initial validation studies for TcPCO₂monitoring were performed before the limitations of using correlation coefficients for such purposes were evident.8 These studies indicated a positive bias (over reading) and reasonable precision.23 24 While it was suggested that a correction factor could be used,23 it became apparent that the error increased with higher PaCO₂ 25 and electrode temperature.26 Some studies also experienced large drifts.27 28 Most importantly, TcPCO₂monitoring seemed to be least reliable in the sickest patients, particularly in the presence of hypotension.1 More recently, the validation of TcPCO₂ has been reappraised using the methods of Bland and Altman. Carter et al 29 reviewed two newer devices, demonstrating bias (precision) values (kPa) of +0.05 (0.33) and −0.13 (0.47). These values seem impressive, but both groups lost patients to analysis, either as outliers or because of excessive drift. Moreover, while over a quarter of the patients received inotropes, the demographic details (median birthweight 2.3 kg, gestation 34 weeks) are not those of the sickest intensive care infants. The greatest disadvantage of TcPCO₂ monitoring, however, remains the need to recalibrate and resite at least 4 hourly, more so in the smallest infants.30 The MPIAS demonstrated minimal drift and so rarely needed recalibration. MPIAS monitoring avoids excess handling, loss of monitoring time and retains precision, even in sick infants.

As well as providing an advantageous alternative to current continuous PaO₂ and PaCO₂ techniques, the MPIAS has two novel features. It permits continuous arterial pH monitoring. This advance has the potential to provide early warning of deteriorations in the metabolic component of arterial blood gas status. The method of introducing the MPIAS after umbilical arterial catheter placement also allows the clinician to defer MPIAS insertion. Thus in those infants in whom it is not clear whether continuous intra-arterial blood gas monitoring is justified, the decision can be deferred (for the first few hours of intensive care) without losing the opportunity to establish umbilical arterial access.

It is important to understand that the purpose of this study was to validate the sensor (MPIAS) in vivo and not the mechanism used to introduce it (SIS). The SIS was modified five times during the study and this substantially improved the success rate of insertion. Thus the failure rate in this study does not reflect the reliability of the current design. In those infants in whom the catheter had to be removed “before clinically indicated,” seven out of eight had been in place for more than 7 days. We did not find evidence of thrombus formation in the catheter tips. We did not find any complications associated with using the MPIAS. The complications following the breech delivery were unusual, but so is the combination of tractional and rotational forces on the spinal column in association with intrapartum asphyxia in a preterm infant. We were sufficiently concerned to arrange for an independent opinion which confirmed that this was not an event associated with use of the catheter.

In conclusion, the Neotrend MPIAS represents an exciting development in the field of continuous monitoring in neonatal intensive care. Bias/precision data indicate that it is a valuable clinical tool. It will not eliminate the need for ABG sampling, but the potential benefits of reduced blood transfusions and earlier intervention in adverse events are clear. Whether this translates into long term patient benefits, needs to be examined in randomised controlled trials.