Sensitivity, Specificity, Receiver-Operating Characteristic (ROC) Curves and Likelihood Ratios for Electronic Foetal Heart Rate Monitoring using New Evaluation Techniques

Abstract

Hypoxia and acidosis adversely influence many foetal organ functions. We wanted to know how foetal heart rate (FHR) patterns are mirrored by the fetal acid-base status and if they could serve for predicting the actual pH in umbilical artery (UA) blood. For this purpose we condensed the FHR phenomena into one figure which was to be used as a testing variable and to analyse the performance of the new testing procedure. The direct FHR signals of 475 foetuses were stored in a computer and analysed offline (MATLAB). All foetuses were delivered by the vaginal route thus without a significant loss of signals. The last 30 min of each recording were used. Acid-base variables and blood gases were determined in cord blood (UA and UV) using RADIOMETER equipments. Three variables of the foetal heart rate (FHR) were computed for each minute: oscillation amplitude [oza (bpm)], microfluctuation [ozm (N/min)] and mean frequency [fhm (bpm)]. These variables were combined to a new index, which we call the WAS index: WAS(T)=FHM(T) x OZF(T)/OZA(T). Using optimisation programmes this index was tailored to actual pH, UA leading to the novel, adapted index: WAS(T)=[FHM(T) x GFHM] x [OZF(T) x GOZF] x [OZA(T) x GOZA]-1,where GFHM, GOZF and GOZA denote three mathematical functions comparable to boundaries in discontinuous scoring-procedures, e. g., the APGAR score. The mean of the WAS index for the last 30 min of delivery is called the WAS score and is used as a discriminator in the testing procedure. WAS score and measured pHUA-values were submitted to correlation and linear regression analysis. Sensitivity, specificity, likelihood ratios, and post-test probabilities were computed including their 95% confidence intervals (CI). A ROC analysis was performed by applying different thresholds for pHUA. pH and WAS score are normally distributed in this sample. The correlation coefficient (r) for pHUA and the WAS score amounts to 0.657, P<<10 (-4). Using ROC plots the area under the curve (AUC) is steadily increased with decreasing pHUA reaching 1.0 for pH 7.0 indicating excellent test accuracy. The AUC for pHUA=7.100 is already 0.963+/-0.066, 95% CI (0.942-0.978), P<0.001. The positive likelihood ratios (+LR) far exceed 10.0 when lowering the threshold for pHUA. Aiming at a sensitivity of 100% the discriminatory power of the test becomes clinically an optimum when using a discriminator of 1.816 and a threshold pH of 7.122: Sensitivity=100%, specificity=89.3%, FNR=0%, FPR(%)=10.7% and AUC=0.958+/-0.049, 95% CI (0.936-0.974), P<0.001. Computer-aided evaluation of FHR patterns leads to a novel index (WAS score) which predicts foetal acidaemia with a high level of accuracy. Therefore online WAS scoring is proposed as an ancillary test procedure for future evaluation of FHR patterns. The conventional EFM remains untouched.