Introduction
Severe traumatic brain injury (TBI) is a leading cause of death and disability worldwide and across all ages [
1], and the mortality rate seems to be unchanged over the past 25 years [
2]. TBI affects 50 to 60 millions of new cases per year, with 2.5 million occurring in Europe, and this is the reason why it has been declared a priority for public health policy [
3]. Efforts should be put into investments in research and other disciplines, such as clinical management and prevention policies. The clinical challenge is represented by the fact that the patient group is extremely heterogeneous and the pathology is highly dynamic [
4]. Therefore, treatment protocols and prediction models are difficult to assess.
The critical care management in the acute phase is focused on reducing the probability and impact of secondary insults, which develop over time as a consequence of raised intracranial pressure (ICP) and/or reduced cerebral perfusion pressure, among other mechanisms [
4]. Therefore, continuous ICP monitoring is recommended by international guidelines as a standard of care in all surviving patients with severe TBI to provide information for ICP-directed therapy [
5,
6].
Along with ICP, a great deal of neuromonitoring techniques and imaging modalities can be used to improve the understanding of intracranial pathophysiology, which, as mentioned before, is highly heterogeneous and dynamic [
4]. Ultimately, more precise targets for therapies could be suggested with this integrated approach. In this perspective, the interaction between the brain and other organs has been suggested as one of the mechanisms that could potentially explain the complexity of this pathology.
In particular, much attention has been given in recent years to the study of interactions existing between the brain and heart [
7‐
9]. The dynamical interplay between the two organs is thought to ensure physiological functions and to be involved in pathological conditions, too [
7]. For example, in the work by Valenza et al. [
7], the authors describe episodes of paroxysmal sympathetic hyperactivity, which often happen when there is a severe axonal injury. In the work by Valenza et al. [
7], the authors describe paroxysmal sympathetic hyperactivity in the postresuscitation syndromes after serious anoxic-ischemic brain insults. An interesting pattern of frequency of these paroxysms has been noticed, as described in detail in the work by Valenza et al. [
7]. These observations show the importance that the brain–heart coupling has in pathological events. As a consequence, efforts have been made toward exploring analytical methodologies that could tackle this phenomenon, with the final goal of developing metrics describing the brain–heart interaction [
8,
10].
Nevertheless, this area is still wide open for investigation, particularly in TBI. In the work by Gao et al. [
9], the authors presented an interesting analysis of interaction between brain and heart measures, showing the presence of interaction and Granger causality between ICP, mean arterial pressure and heart rate (HR). In our previous studies, we derived an HR–ICP measure that we denominated “brain–heart crosstalks,” defined as transient elevations of ICP and HR that occur simultaneously [
11]. In that work, we presented multiple novelties. For the first time, to the best of our knowledge, the measure of brain–heart crosstalks was defined. Moreover, we presented a novel sliding window method to detect the presence of these events. We studied the crosstalks defined as so in a pediatric population and subsequently in a single-center study with an adult cohort [
12], where we also conducted a pilot analysis of the relationship of our novel interaction metric with mortality.
In a further work, we modeled the coupled HR–ICP system as a multilayer network [
13] that can be imagined as a framework in which different channels of the same overall modeled structure are included [
14]. In this framework, each channel is represented by a layer and each node can maintain different neighbors and characteristics across different domains. In recent years, multilayer networks have been fruitfully applied to a variety of fields such as, for example politics, medicine, economics, social interaction, and time series [
14]. Its strength is the capability of modeling relationships across variables (i.e., layers), and therefore their interaction and integration. In our past work [
13], to the best of our knowledge, we used for the first time a multilayer network approach to model the relationship between ICP and HR during brain–heart crosstalks events.
In this work, we aimed to integrate several brain–heart crosstalks measures and examine their role in the context of mortality prediction models, also taking into account clinical, demographics, and monitored features that collectively reflect the severity of primary injury. For this purpose, we used the Collaborative European Neurotrauma Effectiveness Research in TBI (CENTER-TBI) high-resolution intensive care unit (ICU) cohort of patients with TBI [
15,
16]. Our main hypothesis was that the integrated brain–heart crosstalk measures are statistically related to outcome in patients with TBI.
Discussion
In this article, we investigated the relationship between mortality and events of interaction between brain and heart in patients with TBI, named brain–heart crosstalks.
Brain–heart crosstalks were introduced in our previous work [
11]. This novel metric detects events of simultaneous increases of ICP and HR.
Brain–heart crosstalks measures were further extended, through a complex network modeling [
13], which led to the computation of two measures for the ICP-HR system: average edge overlap (
\(\omega_{ct} )\) and mutual information (
\(mi_{ct}\)), which give indication of the behavior of the system during brain–heart crosstalks, as described more in details in the methods section.
From the current study, we could see that the results suggested an inverse relationship between brain–heart crosstalks measures and mortality. The point biserial correlation coefficient was always negative, for the three network measures and mortality. This was true for the case in which we computed it using the 226 patients and when segmenting the population in four age groups. In fact, for our mortality model we divided the population in age groups in the attempt of decreasing the heterogeneity brought by this clinical variable. Important differences in terms of comorbidities, type of lesions and outcome prospective are related to age, particularly in the face of the TBI population aging in western countries.
The first indication of a negative relationship between mortality and brain–heart crosstalk measure, was further confirmed by the Kruskal–Wallis tests between survivors and nonsurvivors distribution of those measures in the whole dataset. In addition, when segmenting the original population, at least one of the brain–heart crosstalks measures was statistically significantly different between survivors and nonsurvivors for each age group, except for the age group 50–64. However, for this age group in the logistic regression model, the coefficient associated to average edge overlap, and the normalized numbers of crosstalks \({ct}_{np}\), was statistically significant and with a negative sign. Logistic elastic-net prediction models appeared to have a good fitting to the data, as the percentage of deviance ratio explained is sufficiently high. This was true except for the age group 30–49. These age-related findings are difficult to explain based on the dataset studied and warrant further investigation in a larger group.
Another important aspect to be considered is that the sign of the \({ct}_{np}\) coefficient was consistently negative for all the four age groups. Given these results, we can confirm that the higher the number of brain–heart crosstalks, the lower is the probability of mortality. This result is confirmed independently of the age group to which the individuals belong to.
Similar behavior was exhibited by the average edge overlap, while for the mutual information two age groups showed positive relationship between mortality and the indicator. This evidence will need further investigation in the future.
Overall, our results suggest that the newly introduced variables of brain–heart crosstalks, might be considered as biomarkers of a healthy brain–heart interaction and therefore the assessment of these biomarkers could play a role in understanding the pathological mechanisms in TBI. The complexity of such a disease is indeed difficult to explain with simply measuring mean ICP values and multimodality physiological monitoring techniques have been developed to improve our understanding and are implemented in the clinical practice. In this perspective, considering brain–heart crosstalks along with other physiology measures in integrated protocols may be a promising approach.
For what concerns the other variables of the prediction models fitted, age came as a statistically significant element only for the two older age groups, but not for the two younger ones. This is in line with findings suggested by the literature, where there is evidence that older age might be a factor worsening the outcome of TBI, because of the comorbidities associated along with other factors [
22]. Moreover, from the fitted models, as we can see in Table
3 for the older populations, almost all the variables have statistically significant coefficients, unlike younger population groups. This is consistent with the fact that the model for the two older population groups showed a better fitting than the younger population groups. Possibly, the higher prevalence of cardiologic preexisting disease may explain the relevance of the brain–heart interplay in the elderly.
As a final consideration our analysis shed light on the possibility of integrating brain–heart crosstalks measures into the existing TBI prognostic models. A very well-known prognostic model is the IMPACT score [
23], where admission features as for instance age, motor score, pupils, hypoxia, hypotension, CT variables, glucose, and hemoglobin are used to evaluate patient’s condition at the time of hospitalization. Some work has been done for what concerns more complex prognostic models development, but to the best of our knowledge, none has ever taken into account variables expressing the interaction between brain and heart [
24]. Therefore, our analysis opens the possibility for implementing the TBI prognostic models with brain–heart interaction measures, and hopefully increasing their precision.
Acknowledgements
The Collaborative European Neurotrauma Effectiveness Research in Traumatic Brain Injury (CENTER-TBI) High-resolution substudy participants and investigators: Audny Anke1, Ronny Beer2, BoMichael Bellander3, Erta Beqiri4, Andras Buki5, Manuel Cabeleira
6, Marco Carbonara7, Arturo Chieregato4, Giuseppe Citerio8, 9, Hans Clusmann10, Endre Czeiter11, Marek Czosnyka6, Bart Depreitere12, Ari Ercole13, Shirin Frisvold14, Raimund Helbok2, Stefan Jankowski15, Daniel Kondziella16, Lars-Owe Koskinen17, Ana Kowark18, David K. Menon13
, Geert Meyfroidt19, Kirsten Moeller20, David Nelson3, Anna Piippo-Karjalainen21, Andreea Radoi22, Arminas Ragauskas23, Rahul Raj21, Jonathan Rhodes24, Saulius Rocka23, Rolf Rossaint18, Juan Sahuquillo22, Oliver Sakowitz25,26, Peter Smielewski
6
, Nino Stocchetti
27
, Nina Sundström28, Riikka Takala29, Tomas Tamosuitis30, Olli Tenovuo31, Andreas Unterberg26, Peter Vajkoczy32, Alessia Vargiolu8, Rimantas Vilcinis33, Stefan Wolf34, Alexander Younsi26, Frederick A. Zeiler13,35.
1Department of Physical Medicine and Rehabilitation, University hospital Northern Norway
2Department of Neurology, Neurological Intensive Care Unit, Medical University of Innsbruck, Innsbruck, Austria
3Department of Neurosurgery & Anesthesia & intensive care medicine, Karolinska University Hospital, Stockholm, Sweden
4NeuroIntensive Care, Niguarda Hospital, Milan, Italy
5Department of Neurosurgery, Medical School, University of Pécs, Hungary and Neurotrauma Research Group, János Szentágothai Research Centre, University of Pécs, Hungary
6Brain Physics Lab, Division of Neurosurgery, Dept of Clinical Neurosciences, University of Cambridge, Addenbrooke’s Hospital, Cambridge, UK
7Neuro ICU, Fondazione IRCCS Cà Granda Ospedale Maggiore Policlinico, Milan, Italy
8NeuroIntensive Care Unit, Department of Anesthesia & Intensive Care, ASST di Monza, Monza, Italy
9School of Medicine and Surgery, Università Milano Bicocca, Milano, Italy
10Department of Neurosurgery, Medical Faculty RWTH Aachen University, Aachen, Germany
11Department of Neurosurgery, University of Pécs and MTA-PTE Clinical Neuroscience MR Research Group and János Szentágothai Research Centre, University of Pécs, Hungarian Brain Research Program (Grant No. KTIA 13 NAP-A-II/8), Pécs, Hungary
12Department of Neurosurgery, University Hospitals Leuven, Leuven, Belgium
13Division of Anaesthesia, University of Cambridge, Addenbrooke’s Hospital, Cambridge, UK
14Department of Anesthesiology and Intensive care, University Hospital Northern Norway, Tromso, Norway
15Neurointensive Care, Sheffield Teaching Hospitals NHS Foundation Trust, Sheffield, UK
16Departments of Neurology, Clinical Neurophysiology and Neuroanesthesiology, Region Hovedstaden Rigshospitalet, Copenhagen, Denmark
17Department of Clinical Neuroscience, Neurosurgery, Umeå University, Umeå, Sweden
18Department of Anaesthesiology, University Hospital of Aachen, Aachen, Germany
19Intensive Care Medicine, University Hospitals Leuven, Leuven, Belgium
20Department Neuroanesthesiology, Region Hovedstaden Rigshospitalet, Copenhagen, Denmark
21Helsinki University Central Hospital, Helsinki, Finland
22Department of Neurosurgery, Vall d'Hebron University Hospital, Barcelona, Spain
23Department of Neurosurgery, Kaunas University of technology and Vilnius University, Vilnius, Lithuania
24Department of Anaesthesia, Critical Care & Pain Medicine NHS Lothian & University of Edinburg, Edinburgh, UK
25Klinik für Neurochirurgie, Klinikum Ludwigsburg, Ludwigsburg, Germany
26Department of Neurosurgery, University Hospital Heidelberg, Heidelberg, Germany
27Department of Pathophysiology and Transplantation, Milan University, and Neuroscience ICU, Fondazione IRCCS Cà Granda Ospedale Maggiore Policlinico, Milano, Italy
28Department of Radiation Sciences, Biomedical Engineering, Umeå University, Umeå, Sweden
29Perioperative Services, Intensive Care Medicine, and Pain Management, Turku University Central Hospital and University of Turku, Turku, Finland
30Neuro-intensive Care Unit, Kaunas University of Health Sciences, Kaunas, Lithuania
31Rehabilitation and Brain Trauma, Turku University Central Hospital and University of Turku, Turku, Finland
32Neurologie, Neurochirurgie und Psychiatrie, Charité – Universitätsmedizin Berlin, Berlin, Germany
33Department of Neurosurgery, Kaunas University of Health Sciences, Kaunas, Lithuania
34Department of Neurosurgery, Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health, Berlin, Germany
35Section of Neurosurgery, Department of Surgery, Rady Faculty of Health Sciences, University of Manitoba, Winnipeg, MB, Canada