Background: There is little evidence of important improvement in outcome for infants when time trended physiological data is available to clinical staff. We believe it is important to improve modern monitoring technologies in the neonatal ICU in order to aid the decision making process.

Hypothesis: It is possible to recognise pneumothoraces at a stage earlier than is clinically evident with continuous physiological monitoring, by the use of repeatable patterns provided by the derivatives of real time physiological data (mean, standard deviation, slope)

Methods: On a monitoring database of over 1000 sick infants where physiological data had been recorded as a time series at 1 Herz, 21 babies were identified who developed a PT. Channels of raw data (tcPCO2, heart rate and blood pressure) were examined, and derivatives of this data were calculated (slope tcPCO2 over 15 minutes, slopes over 15 minutes of the standard deviation over 2 minutes of the mean blood pressure and heart rate). A macro (simple computer program) was then developed combining the slope of the tcPCO2 and the slopes for the heart rate SD and blood pressure SD and used to predict the PT. The macro was applied to all data from these infants for - 1) the 12 hours before and 2) the 72 hours before the PT developed to calculate the false positive detection rate.

Results: The true positive detection rate (sensitivity) of the macro was 90.5% (19/21) and detection was often some hours before clinical diagnosis The false positive rate increased with the duration of the trend analysis before the PT occurred - see table.

Table 1 Analysis Duration prior to PT
Full size table

Conclusion: This study attempted to use time series (trends) of physiological data to give early detection of PT. The sensitivity of the “intelligent” computer macro was good, the specificity less good and dependent on the quantity of preceding physiological data analysed.