Assessing mood symptoms through heartbeat dynamics: An HRV study on cardiosurgical patients

Abstract

Background

Heart Rate Variability (HRV) is reduced both in depression and in coronary heart disease (CHD) suggesting common pathophysiological mechanisms for the two disorders. Within CHD, cardiac surgery patients (CSP) with postoperative depression are at greater risk of adverse cardiac events. Therefore, CSP would especially benefit from depression early diagnosis. Here we tested whether HRV-multi-feature analysis discriminates CSP with or without depression and provides an effective estimation of symptoms severity.

Methods

Thirty-one patients admitted to cardiac rehabilitation after first-time cardiac surgery were recruited. Depressive symptoms were assessed with the Center for Epidemiologic Studies Depression Scale (CES-D). HRV features in time, frequency, and nonlinear domains were extracted from 5-min-ECG recordings at rest and used as predictors of “least absolute shrinkage and selection” (LASSO) operator regression model to estimate patients' CES-D score and to predict depressive state.

Results

The model significantly predicted the CES-D score in all subjects (the total explained variance of CES-D score was 89.93%). Also it discriminated depressed and non-depressed CSP with 86.75% accuracy. Seven of the ten most informative metrics belonged to non-linear-domain.

Limitations

A higher number of patients evaluated also with a structured clinical interview would help to generalize the present findings.

Discussion

To our knowledge this is the first study using a multi-feature approach to evaluate depression in CSP. The high informative power of HRV-nonlinear metrics suggests their possible pathophysiological role both in depression and in CHD. The high-accuracy of the algorithm at single-subject level opens to its translational use as screening tool in clinical practice.

Introduction

Depression and cardiac disorders are very often presented together (Chauvet-Gelinier et al., 2013, Rovai et al., 2015). Several epidemiological studies have suggested that patients with coronary heart disease (CHD) are more prone to develop depressive symptoms or even a full-blown major depressive disorder (MDD) (Konrad et al., 2016a, Konrad et al., 2016b). Moreover, MDD increases the risk of developing CHD (Gan et al., 2014, Wu and Kling, 2016). Furthermore, the presence of depressive symptoms predicts a worsening of the prognosis of CHD (van Melle et al., 2004). Accordingly, targeting depressive symptoms has been shown to not only improve mood, but also positively impact the CHD outcome (Angermann et al., 2016, Freedland et al., 2015, Pizzi et al., 2011, Taylor et al., 2009, Whalley et al., 2015).

Of particular interest is the study of patients who underwent cardiac surgery such as heart valve surgery or a coronary artery bypass graft (CABG). Cardiac surgery is a significant life event with a relevant impact on the patients' mood. Indeed, 20%–60% of patients who underwent cardiac surgery reported depressive symptoms, whereas up to 23% of them fulfilled the criteria for MDD (Blumenthal et al., 2003, Glassman and Shapiro, 1998, Gu et al., 2016). Importantly, patients with postoperative depression are at a greater risk of cardiac morbidity and mortality (Blumenthal et al., 2003, Connerney et al., 2001, Penninx, 2016, Scheier et al., 1999).

Researchers have proposed several pathophysiological mechanisms to explain the association between depression and cardiac diseases. Among others, autonomic nervous system (ANS) dysfunctions are considered as one the most important (Rovai et al., 2015). ANS activity is very often investigated through the analysis of the series extracted from the time intervals between two consecutive R-waves detected from the Electrocardiogram (i.e. the R-R intervals), whose variability is defined as Heart Rate Variability (HRV) (Acharya et al., 2006).

HRV alterations have been found in cardiac diseases, including CHD and heart failure, when compared to healthy, gender/age matched populations (Feng et al., 2015, Harris et al., 2014, Peterson et al., 2014, Radaelli et al., 2014, Sandercock and Brodie, 2006, Wendt et al., 2014). Typically, cardiac patients are characterized by a reduced vagally-mediated HRV, as estimated by the high-frequency (HF) power of HRV. From a clinical perspective, a reduced HRV has been shown to represent an independent risk factor for cardiovascular mortality in patients after a myocardial infarction and in patients with a stable CHD (Bigger et al., 1993, Ho et al., 2011, Karimi Moridani et al., 2016, Kleiger et al., 1987, Molon et al., 2010, Rich et al., 1988). In general, HRV-derived measures have been considered a way to measure the state of cardiac health (Sandercock and Brodie, 2006).

In mood disorders, particularly in MDD, consistent findings have reported a reduced HRV. Specifically, depressed patients showed a significantly decreased HRV and parasympathetic activity compared to healthy individuals (Kemp and Quintana, 2013, Kemp et al., 2010, Liang et al., 2015, Stapelberg et al., 2012, Wang et al., 2013). Although MDD and depressive symptoms in CHD seem to possess clinical specificity in terms of symptomatology and course (Chauvet-Gelinier et al., 2013, Chavez et al., 2014, Foxwell et al., 2013), the same relationship between the HRV and clinical courses can be found. There is an abundance of evidence that depressed patients with CHD or acute myocardial infarctions have a reduced HRV compared to those without depression (Carney et al., 2001, Carney et al., 2002, Krittayaphong et al., 1997). It has also been consistently shown that depression is associated with reduced HRV in patients who underwent cardiac surgery (Patron et al., 2012, Patron et al., 2014a).

Notwithstanding the consistency of these pieces of evidence, HRV measures still are not routinely used in clinical practices of cardiology or in clinical psychology. This is rather surprising because HRV measures may serve as an endophenotype for psychiatric disorders in the general population and particularly in cardiac patients (Sgoifo et al., 2015). Indeed, the presence of effective and objective indices of depressive symptoms in patients with cardiac diseases may be very useful for screening and clinical management.

One limitation that may have hindered the clinical use of the HRV measurement is the lack of specificity and sensitivity of a single HRV measure. Most of the studies dealing with depressed patients (either with or without CHD) rely on one or a few HRV measures. The lack of specificity is clearly represented by the fact that alterations of a single metric can be found in several psychiatric and somatic conditions, including stroke (Tang et al., 2015), epilepsy (Lotufo et al., 2012), multi-organ failures (Liu et al., 2013, Zhang et al., 2015), cancer fatigue (Crosswell et al., 2014) and autoimmune diseases (Adlan et al., 2014, Holman and Ng, 2008), in addition to depression. On the other hand, since depression, and mood disorders in general, are highly heterogeneous in terms of clinical presentations and endophenotypes (Alhajji and Nemeroff, 2015), one single metric may lack the sufficient sensitivity to be altered in all of the patients. This could account for some of the negative or partially inconsistent results (Brunoni et al., 2013, Gehi et al., 2005, Liang et al., 2015, Moon et al., 2013, Wang et al., 2013). Finally, and most relevant to the issue of discriminant efficacy, group differences in HRV metrics, even when consistent, could not have enough sensitivity and specificity if used to discriminate among individuals with or without depression at a single-subject level.

While these limitations are regarded with the univariate analysis of one or a few HRV metrics, it has been recently shown that a multi-feature analysis combining multiple HRV measures is highly accurate in detecting pathological mood states in patients with bipolar disorders at a single-subject level (Gentili et al., 2016, Valenza et al., 2014, Valenza et al., 2013a, Valenza et al., 2013b). A similar approach could bridge the actual gap between the research and the clinical practice by providing a valuable tool for the screening and an assessment of depressive symptoms of cardiac diseases. To be valuable, the tool should be able to discriminate at a single-subject level with patients with or without depression and provide a fair estimation of depression severity.

Therefore, the primary aim of this study is to test whether a multi-feature analysis can provide an automated distinction between depressed and nondepressed patients at a single-subject level and whether it can predict the score of the Center for Epidemiological Studies of Depression (CES-D) scale for each patient. Particularly, we derived a simple, nontrivial mathematical equation that properly combines HRV measures to predict clinical severity as measured by CES-D.