Objective To develop a model of heart rate variations in preterm infants for detecting false alarms.

Methods Infants <32 weeks gestation were recruited. Heart rate over the first 24 h of stay was recorded from cotside monitors networked by Infinity Gateway (Draeger Medical Canada Inc). Sampling interval was one minute. The heart rate was classified into two states that will trigger alarms (A1 80–100 bpm and A2 >150 bpm) and a normal state 100–150 bpm. The probabilities of these infants remaining in each state and changing from states were calculated as a Markov chain probability model using MATLAB 7.5 (Mathworks, Natick, Massachusetts, USA). The Kolmogorov entropy principle (the probability of error increasing the longer it is from an event) was used to calculate the false alarm rate. The probability of the infant’s heart rate reaching alarm states (A1, A2) was computed. These probabilities were compared with the actual changes in the heart rate after 2, 4, 6 and 8 minutes have elapsed from the initial time.

Results 45 infants with mean (SD) gestational age of 28.7 (2) weeks, mean (SD) birthweight of 1250 (346) g were studied. The average probabilities of error for predicting state A1 and state A2 are summarised in the table.

Conclusions It is feasible to utilise statistical techniques for calculating the probabilities of false alarm rate of heart rate signals in preterm infants using statistical techniques.