In response: Heart rate differential method simple but inefficient method for seizure detection

To the Editors, We wish to thank Widman and colleagues for their interest in our recent publication on seizure detection based on heart rate variability.1 We agree that the equation of HR-diff can be rewritten as spelled out in the comment. However, we see no quandary in the original equation as it is written in both this and in an earlier publication,1,2 since it clearly defined the result. The HR-diff equation can be written in the way we originally presented it in the papers1,2 or rewritten as Widman and colleagues have done. However, we do not believe that one way is “better” than the other because each is simply just another way of writing the formula. Thus, we see no reason why the HR-diff parameter should not be written as we originally defined it. The HR-diff is a parameter that we stated in2: “simply computes the positive rate of heart rate change within the given window.” We used this parameter in both papers to compare this simplest form of computing heart rate change (HR-diff) against more advanced heart rate variability methods (ModCSI and CSI). In the comments by Widman and colleagues, they state that: “As could then be expected, this (HR-diff) parameter scores badly (visible in figure 2) while this parameter scores better after prefiltering, which enables the possibility to place information from more than the four remaining arbitrarily chosen R-R intervals (−1;0;k;k + 1) into the parameter.” First, seizure detection based on heart rate changes as the simple change in pulse-rate or R-R time-length within a given timeor R-R interval window length has been the method of choice in several papers analyzing seizure detection. Therefore, we opted to assess this aspect too. The explanation for the poor performance of this parameter is probably not related to any physiologic issues but rather to some few misplaced R-peak detections (noise) during baseline data of the patients, which can create too high a threshold for seizure detection in the test data. The algorithm with prefiltering before calculating HR-diff (HR-diff-filtered) confirm this suggestion, as it filters out outliers (noise) and preforms much better than HR-diff without pre-filtering. However, still the more advanced ModCSI and CSI algorithms performs better than any stand-alone HR-diff algorithm with or without prefiltering in our analyses of the test data. Second, the four R-R intervals in the equation are not “arbitrarily” chosen. They have k-length difference (window length: 50 or 100 R-R interval lengths), which, as stated, simply computes the positive rate of heart rate change within the given window. We have never stated that the HR-diff method is using averaging/smoothing as suggested in the comments by Widman and colleagues.