Sónia Gouveia

QT variability and HRV interactions in ECG: quantification and reliability

In this paper, a dynamic linear approach was used over QT and RR series measured by an automatic delineator, to explore the interactions between QT interval variability (QTV) and heart rate variability (HRV). A low-order linear autoregressive model allowed to separate and quantify the QTV fractions correlated and not correlated with HRV, estimating their power spectral density measures. Simulated series and artificial ECG signals were used to assess the performance of the methods, considering a respiratory-like electrical axis rotation effect and noise contamination with a signal-to-noise ratio (SNR) from 30 to 10 dB. The errors found in the estimation of the QTV fraction related to HRV showed a nonrelevant performance decrease from automatic delineation. The joint performance of delineation plus variability analysis achieved less than 20% error in over 75% of cases for records presenting SNRs higher than 15 dB and QT standard deviation higher than 10 ms. The methods were also applied to real ECG records from healthy subjects where it was found a relevant QTV fraction not correlated with HRV (over 40% in 19 out of 23 segments analyzed), indicating that an important part of QTV is not linearly driven by HRV and may contain complementary information.

Thinning-based models in the analysis of integer-valued time series: a review:

This article aims at providing a comprehensive survey of recent developments in the field of integer-valued time series modelling, paying particular attention to models obtained as discrete counterparts of conventional autoregressive moving average and bilinear models, and based on the concept of thinning. Such models have proven to be useful in the analysis of many real-world applications ranging from economy and finance to medicine. We review the literature of the most relevant thinning operators proposed in the analysis of univariate and multivariate integer-valued time series with either finite or infinite support. Finally, we also outline and discuss possible directions of future research.

Mathematical biomarkers for the autonomic regulation of cardiovascular system

Heart rate and blood pressure are the most important vital signs in diagnosing disease. Both heart rate and blood pressure are characterized by a high degree of short term variability from moment to moment, medium term over the normal day and night as well as in the very long term over months to years. The study of new mathematical algorithms to evaluate the variability of these cardiovascular parameters has a high potential in the development of new methods for early detection of cardiovascular disease, to establish differential diagnosis with possible therapeutic consequences. The autonomic nervous system is a major player in the general adaptive reaction to stress and disease. The quantitative prediction of the autonomic interactions in multiple control loops pathways of cardiovascular system is directly applicable to clinical situations. Exploration of new multimodal analytical techniques for the variability of cardiovascular system may detect new approaches for deterministic parameter identification. A multimodal analysis of cardiovascular signals can be studied by evaluating their amplitudes, phases, time domain patterns, and sensitivity to imposed stimuli, i.e., drugs blocking the autonomic system. The causal effects, gains, and dynamic relationships may be studied through dynamical fuzzy logic models, such as the discrete-time model and discrete-event model. We expect an increase in accuracy of modeling and a better estimation of the heart rate and blood pressure time series, which could be of benefit for intelligent patient monitoring. We foresee that identifying quantitative mathematical biomarkers for autonomic nervous system will allow individual therapy adjustments to aim at the most favorable sympathetic-parasympathetic balance.

Time domain baroreflex sensitivity assessment by joint analysis of spontaneous SBP and RR series

Abstract The sequences technique is frequently used for time domain assessment of the arterial-cardiac baroreceptor reflex sensitivity (BRS). The BRS is estimated by the slope between systolic blood pressure and RR interval values in baroreflex sequences (BSs) and an overall estimate is obtained by slope averaging. However, only 25% of all beats are in BSs with 60% of those located in 3-beat length segments. Also, in cases of BSs absence (usually associated with poor BRS function), the BRS cannot be quantified. Here, baroreflex events (BEs) are introduced and used with global/total slope estimators to improve BRS assessment. The performance of the novel method is evaluated using the EuroBaVar dataset. The events technique benefits from a higher number of beats: 50% of all beats are in BEs with more than 70% exceeding 3-beat length. It always provides a BRS estimate, even when BSs cannot be identified. When BSs are available, estimates from BEs and BSs are highly correlated. The estimates from BEs for the cases without BSs are lower than the estimates for the remaining cases, indicating poorer BRS function. The events technique also offers superior ability to discriminate lying from standing position in the EuroBaVar dataset (23/23 versus 18/23 for the sequences technique).

Long-range dependence in heart rate variability data: ARFIMA modelling vs detrended fluctuation analysis

Heart rate variability (HRV) data display non-stationary characteristics and exhibit long-range correlation (memory). Detrended fluctuation analysis (DFA) has become a widely-used technique for long memory estimation in non-stationary HRV data. Recently, we have proposed an alternative approach based on fractional integrated autoregressive moving average (ARFIMA) models. ARFIMA models, combined with selective adaptive segmentation may be used to capture and remove long-range correlation, leading to an improved description and interpretation of the components in 24 hour HRV recordings. In this work estimation of long memory by DFA and selective adaptive ARFIMA modelling is carried out in 24 hour HRV recordings of 17 healthy subjects of two age groups. The two methods give similar information on long-range global characteristics. However, ARFIMA modelling is advantageous, allowing the description of long-range correlation in reduced length segments.

BRS analysis from baroreflex sequences and baroreflex events compared using spontaneous and drug induced data

Spontaneous time domain BRS estimation is based on the SBP-RR slope, which can be computed from either baroreflex sequences (BS) or baroreflex events (BE). BRS analysis from BEs was recently shown be advantageous particularly in the cases of reduced BRS or when BS are not identified. Also, it offers a superior discrimination between lying and standing positions. In this work, the methods developed for spontaneous BRS analysis are further compared using spontaneous and drug induced data. The results corroborate that spontaneous and drug induced estimates are different although correlated. In particular, if BEs are used the differences and the correlation between the estimates is higher. No precision improvement is achieved if the BRS is estimated from drug induced data. In spontaneous, the higher number of beats in BEs in comparison with BSs (at the expense of a lower SBP-RR correlation) allows a higher BRS estimate precision using recordings of the same length.

Fuzzy logic model to describe anesthetic effect and muscular influence on EEG Cerebral State Index.

The well-known Cerebral State Index (CSI) quantifies depth of anesthesia and is traditionally modeled with Hill equation and propofol effect-site concentration (Ce). This work brings out two novelties: introduction of electromyogram (EMG) and use of fuzzy logic models with ANFIS optimized parameters. The data were collected from dogs (n=27) during routine surgery considering two propofol administration protocols: constant infusion (G1, n=14) and bolus (G2, n=13). The median modeling error of the fuzzy logic model with Ce and EMG was lower or similar than that of the Hill with Ce (p=0.012-G1, p=0.522-G2). Furthermore, there was no significant performance impact due to model structure alteration (p=0.288-G1, p=0.330-G2) and EMG introduction increased or maintained the performance (p=0.036-G1, p=0.798-G2). Therefore, the new model can achieve higher performance than Hill model, mostly due to EMG information and not due to changes in the model structure. In conclusion, the fuzzy models adequately describe CSI data with advantages over traditional Hill models.

Improved time domain BRS assessment with the use of baroreflex events

The sequences technique is based on a linear regression of SBP and RR values in baroreflex sequences (BSs). In normal conditions most of BSs have 3 beats length and the estimated slope is potentially inaccurate. Also, the number of beats in BSs is approximately 25% of the total number of beats, evidencing that a large part of the data is discarded from BRS analysis. In this work, global BRS estimators combined with baroreflex events (BEs) are proposed as a way to improve BRS assessment in normal conditions and to allow its quantification in cases of BS absence. The results in the EuroBaVar dataset showed that the number of beats in BEs gets up to 50% of the total, with correlation between the corresponding SBP and RR values close to 0.8. Also, BRS assessment based on global estimators combined with BEs distinguishes Lying from Standing positions in all the subjects of that dataset, including those without identified BSs.

Assessing baroreflex sensitivity in the sequences technique: local versus global approach

The sequences technique is the most used time domain technique for the assessment of arterial baroreceptor reflex sensitivity (BRS) and is based in the analysis of the beat-to-beat spontaneous variability of systolic blood pressure and heart period. Although a common used method, the sequences technique imposes the setting of several parameters to determine what is a valid baroreflex event and no consensual opinion about these parameters is found in the literature. The theory is overlooked, and linear regression over three values is the usual procedure. The methodology itself can be questionable and the estimator has not been clearly examined regarding its statistical properties, namely bias and variance. In this work, an alternative estimator that we have been using is evaluated and compared with the traditional approach, considering real and simulated data. The results obtained show that the proposed estimator is less biased and presents lower variance than the traditional approach

Area burned in Portugal over recent decades: an extreme value analysis

Forest fires are a major concern in Europe, particularly in Portugal where large forest fires are responsible for negative environmental, social and economic effects. In this work, a long time series of daily area burned in 18 Portuguese districts (north, coastal areas and inner–south) from 1980 to 2010 are analysed to characterise extreme area burned and regional variability. The analysis combines the peak-over-threshold method and classification techniques to cluster the time series on the basis either of their corresponding tail indices or their predictive distributions for 5- and 15-year return values, that is, the level that is exceeded on average once every 5 or 15 years. As previously reported in other wildfire studies, the results show that the distributions of area burned (1980–2010) are heavy tailed for all Portuguese districts, with considerable density in the tail, indicating a non-negligible probability of occurrence of days with very large area burned. Moreover, clustering based on tail indices identified three distinct groups with spatial pattern closely related to the percentage of shrub cover within each district. Finally, clustering based on return values shows that the largest return levels of area burned are expected to occur in districts located in the centre and south of Portugal.