Current aortic SBP estimation methods require recording of a peripheral pressure waveform, a step with no consensus on method. This study investigates the possibility of aortic SBP estimation from radial SBP and DBP using artificial neural networks (ANN) with [ANNSBP.DBP.heart rate (HR)] and without HR (ANNSBP.DBP).Methods:
Ten-fold cross validation was applied to invasive, simultaneously recorded aortic and radial pressure during rest and nitroglycerin infusion (n = 62 patients). The results of the ANN models were compared with an ANN model using additional waveform features (ANNwaveform), to an N-point moving average method (NPMA) and to existing, validated generalized transfer function (GTF).Results:
Estimated aortic SBP for all methods was on average less than 1 mmHg away from measured aortic SBP with the exception of NPMA (difference 2.0 ± 3.5 mmHg, P = 0.62). Variability of the difference was significantly greater in ANNSBP.DBP.HR and ANNSBP.DBP (both SD of ± 5.9 mmHg, P < 0.001 compared with GTF, ± 4.0 mmHg, P < 0.001). Inclusion of waveform features decreased the variability (ANNwaveform ± 3.9 mmHg, P = 0.264). Estimated aortic SBP in all models was correlated with measured SBP, with ANN models providing statistically similar results to the GTF method, only the NPMA being statistically different (P = 0.031).Conclusion:
These findings indicate that use of radial SBP, DBP, and HR alone can provide aortic SBP estimation comparable with the GTF, albeit with slightly greater variance. Pending noninvasive validation, the technique provides plausible aortic SBP estimation without waveform analysis.