Artikel
Precursors of ICP changes: The analysis of causal relationships
Suche in Medline nach
Autoren
Veröffentlicht: | 9. Juni 2017 |
---|
Gliederung
Text
Objective: It is common practice in signal processing to use statistical methods to infer causal relations from data if the theoretical background is insufficient or experimentation impossible. A typical biomedical application is the calculation of Granger-causality from a set of precursors and the average ICP of long-term epidural ICP recordings in patients with idiopathic intracranial hypertension (IIH).
Methods: In five ICP episodes of 50 min from five different patients with marked ICP dynamics an offline beat-to-beat analysis of pulse pressure waveforms was carried out following the MOCAIP concept (DOI: 10.1109/TBME.2009.2037607). Nine precursors were chosen reflecting both the pulse pressure magnitude and time interval related metrics. Also, the waveform’s curvature at discrete times was taken into account. The working hypothesis was that a precursor „Granger-causes“ mean ICP (mICP). Additionally, the inverse assumption, i.e. mICP Granger-causes precursors, was considered. To ensure results second order stationarity (DOI: 10.1111/rssb.12015) of the time series analysed was evaluated.
Results: Direct Granger-causality (accepted for significance levels p<0.01, F-statistic) was computed in a total of 90 analyses. 22/45 analyses showed that precursors cause mICP, however, 38/45 proved that mICP causes precursors. Three precursors did not cause mICP in all patients, however mICP caused these precursors. Pronounced bidirectional causality was found in amplitude related precursors whereas time intervals and curvatures showed more variable results. Local stationarity was tested (accepted for significance levels p<0.01). Non-stationarities were found in all data.
Conclusion: Results were dominated by bidirectional causality for precursors and mean ICP. The meaning and consequences of bidirectional causality are not quite understood. It appears reasonable to attribute this phenomenon, primarily, to the overall non-stationarity of real-world clinical/biomedical data. Moreover, common cause fallacy or insufficient sampling might be involved. Set against this background, the complexity of causal relationships should be considered when evaluating ICP predicting studies.