Suicide is the 2nd leading cause of death among adolescents, exponentially rising over the past decade and accounting for more than all other non-mental health diagnoses combined, excluding trauma [1]. Parents often utilize emergency departments (EDs) in an effort to help keep their adolescent safe when suicidal symptoms escalate. A recent nationally-representative study found that the number of ED visits for pediatric mental health complaints more than doubled in the most recent decade, now accounting for approximately 1.2 million annual visits [2]. This is a growing problem for EDs, as studies have shown that pediatric mental health patients require more resources than non-mental health patients and have lengths of stay counted in days, not hours [3,4]. Recidivism is a significant challenge, as up to 20% of unsuccessful adolescent suicide attempts will have a second attempt within 12 months of discharge, and 45% of pediatric mental health ED patients will have a subsequent visit for a similar complaint [5,6].

Given the significant public health burden, there is a critical need for improved outpatient monitoring that allows for early detection and intervention before an acute mental-health crisis results in a suicide attempt, associated ED visit, or hospitalization. Previous studies have identified physiologic markers that may help identify suicidal patients, including changes in heart rate or breathing related to stress [7-9]. Expanding on that work, it has been postulated that a dysregulation of the interplay of the parasympathetic and sympathetic branches of the autonomous nervous system is present in suicidal patients [10]. One physiologic marker for autonomic nervous system dysregulation is heart rate variability (HRV). There exist metrics focused on the parasympathetic branch, namely the high frequency component (HF) and separate metrics focused on the sympathetic branch. Prior research has found that patients with a history of suicidality demonstrate a decrease in their HF when undergoing stressful tasks [11,12]. This suggests that patients with suicidality may have impaired parasympathetic nervous system regulation. However, studies to date have been limited by measurements at a single time point and confined to laboratory settings where patients are connected to electrocardiographic (ECG) monitors. More recently, it has become possible to measure HRV through wrist-worn devices using photoplethysmography. This method has been shown to be accurate when compared to the gold standard of ECG and may be highly feasible in clinical practice among at-risk adolescents given the popularity of smartwatches [13]. A critical gap in current research is to identify changes in HRV during acute suicidal episodes in adolescents.

With that in mind, the objective of this study was to identify correlations between the HF component of HRV and severity of suicidality in a cohort of acutely suicidal, inpatient adolescents. Based on evidence from prior literature [11,12], we hypothesized that the primary predictor of HF metric would be inversely correlated with the primary outcome of Columbia Suicide Severity Score (CSSRS). Measuring HRV using wearable technology has the potential to be widely implemented and facilitate earlier detection of suicidal symptoms before they reach a crisis level among a large number of atrisk adolescents. This creates the ability to provide early detection and intervention, potentially reducing ED recidivism, acute hospitalizations and suicide attempts.

This was a prospective, observational study from September 2017 thru November 2019 in a tertiary care pediatric emergency department and an inpatient pediatric psychiatric unit. These 2 locations are within different hospital systems, but located in close geographic proximity. The tertiary care children’s hospital where the emergency department is housed does not have an inpatient psychiatric unit, so many patients requiring admission are transferred to the enrolling inpatient pediatric psychiatric unit. Patients were eligible if they were between the ages of 13-19 years and presented to the ED or were admitted to the inpatient psychiatric unit for acute suicidality. Admission to the psychiatric facility was determined by the treating psychiatry team unrelated to the study. The indication for admission in this population is generally risk of self-harm and need for acute psychiatric interventions. Patients were excluded if they were in the Oregon Department of Human Services custody, acutely agitated as deemed by the primary medical team, were pregnant, had an initial CSSRS <4 (not suicidal), or had insufficient data quality for analysis. Acutely agitated patients were excluded due to the concern for self-harm with a study device and danger to the study team. The study was approved by the Institutional Review Board at both sites and registered with clinicaltrials.gov (NCT03030924). Informed consent was obtained from the patient or legal guardian and assent was obtained from patients <18 years.

There were a total of 104 acutely suicidal adolescents enrolled in this study. As detailed in Figure 2, after exclusions, the final cohort consisted of 51 patients. The average age was 16.7 years (stdev 1.5 years) with a predominance of females (72.6%). Almost half of patients (47.1%) had a previous ED visit for a mental health complaint or inpatient psychiatric hospitalization. The majority (60.8%) were enrolled after presenting to the hospital following a suicide attempt. The patients were enrolled equally between the ED and inpatient psychiatric unit (45.1% vs. 54.9% respectively) and approximately half remained inpatients at the time of study completion 7 days after enrollment (Table 1). The majority of patients had CSSRS (69.4%) greater than or equal to 15. Figure 3 visualizes the percentage change of CSSRS from time of initial enrollment with the majority showing improvement as evidenced by a negative change. A decrease in CSSRS indicates improvement in their suicide severity.

There were a total 1,215 HRV segments across 51 patients available for analysis across all days, and 697 HRV segments available across 51 patients on day 1. Model fit was acceptable for all regression models. Due to heteroskedastisicity, we used a sandwich estimator for variance in all models. No outliers or leverage values were identified. To accomplish the study objective, HRV measurements were focused on the transformed sqrt (HF). Tables 2 and 3 detail the association between HF and CSSRS as assessed by the regression models including both the initial CSSRS and HF and change in CSSRS and HF HRV during the study period (a decrease of at least 25% during the study period). Among patients who had a second CSSRS (29 of the 51 patients), approximately one quarter of patients (n=7) had a 25% decrease in the CSSRS after the initial enrollment. This resulted in 478 HRV measurements during all hours and 177 HRV measurements at nighttime. The sqrt (HF) component significantly increased over time in patients that had at least a 25% decrease in their CSSRS (nighttime mean difference 11.89 ms/HZ; 95% CI: 3.70 to 20.10; p=0.005; all hours mean difference 2.165 ms/HZ; 95% CI: -5.18 to 9.51; p=0.564). Of the approximately 697 HRV measurements on day 1, 506 were found with a CSSRS≥15. Nighttime data consisted of 233 HRV measurements in which 166 were found with a CSSRS≥15. Patients with a CSSRS≥15 had a lower sqrt(HF) component compared to patient with CSSRS<15 (nighttime mean difference -8.34 ms/HZ; 95% CI: -17.11 to 0.43; p=0.071; all hours mean difference -1.69 ms/HZ; 95% CI: -11.19 to 7.82; p=0.73), although these findings did not reach statistical significance.

This novel study was consistent with our hypothesis and found a significant negative association between the HF power, which measures the parasympathetic branch of the autonomous nervous system, and the CSSRS in a population of acutely suicidal adolescents using a wearable technology. This finding indicates that increasing suicidality is associated with decreased parasympathetic tone. The lifetime prevalence of a suicide attempt in adolescence is currently over 20% [17]. For this reason, suicide prevention is a critical national public health priority especially among adolescents [18]. This study addresses the large unmet need and uses an innovative approach to meet this challenge. A socially-acceptable, easy-to-use technology that is continuously available to measure HRV at various points in the day, such as smartwatches, should be evaluated as a tool to identify worsening parasympathetic function, and associated suicidality, in real time. This would allow for preventive measures to be implemented when needed such as biofeedback or coping skills to decrease suicide attempts, suicide-related ED visits, and suicide-related mortality. The smartwatch is a wearable biosensor platform that has high social acceptability among adolescents, allowing discreet monitoring that would not be embarrassing or associated with increased stigma from peers.

The rationale for this work stems from previous cross-sectional studies evaluating HRV in patients with mental health problems as a biomarker of the autonomic nervous system; in particular parasympathetic activity through the HF power of HRV. Across multiple study designs and research teams, the findings have consistently found an association between the HF and suicidality [12,19,20]. These studies have found that the HF power was lower in patients with worse depression and suicidality or a history of suicide attempts compared to normal patients. Our study is unique in that it was conducted among acutely suicidal patients, used photoplethysmography to measure HRV, and included longitudinal measurements.

One significant clinical impact wearable technology that monitors physiologic parameters, such as HRV, can have is to identify dynamic changes in suicidality beyond single timepoint measurements. Previous studies have investigated more dynamic changes in HRV by using stress-inducing tasks in the laboratory setting while monitoring HRV. A large systematic review in adults included studies measuring HRV during stressful situations in patients with clinical depression [21]. These studies focused on well-validated stress tasks and generally found that among depressed individuals, there was hypo-reactivity of the parasympathetic system during stressors, as evidenced by lower HF power. One study attempted to evaluate both resting HRV and stress-induced HRV’s relation to active suicidal ideation [22]. They found decreased HF power was predictive of higher suicide severity scores. Another study examined changes in the HF HRV power during a stress-induced task and differentiated suicide attempters from suicide ideators without an attempt [11]. This type of research is pivotal to the clinical hope of identifying the latter group before progressing to a suicide attempt. In that study, while there were no differences between groups in anger and stress measures at baseline, following a social stressor, they found that attempters had a lower HF during the stress test than non-attempters. These studies corroborate our findings of lower HF HRV being associated with increased severity of suicidality.

While several studies have shown a significant inverse correlation between the HF and depression/suicidality in laboratory settings, this study is significant in that it found similar results in acutely suicidal adolescents admitted to the hospital or ED. The next phase to this area of research would be to intermittently measure HRV longitudinally and as part of a system to alert patients and their loved ones to autonomic dysregulation and higher risk of increasing suicidality. There are several potential interventions patients could implement upon receiving an alert. One area of research has focused on the impact of biofeedback, such as breathing, targeted to change HRV in patients with mental health disorders [23-25]. This could be employed using a mobile application in conjunction with an alert for worsening HF HRV. Previous studies have shown the ability to utilize biofeedback guided by dynamic changes in HRV resulting in improved overall symptoms [24,26,27]. Wearable technology, especially in conjunction with smartphones, can also collect other biosignals and monitor other elements of behavior that could be incorporated into machine learning applications to improve the predictive ability of an algorithm [28,29].

This pilot study has similar findings to past literature, but does have a number of limitations. As detailed in Figure 1, we excluded a number of patients due to poor data quality. The majority of excluded patients were at the beginning of the study while the technical aspects of the study PPG device were evaluated in adolescent suicidal patients. After identification of technical issues, the remaining enrollments resulted in improved data quality. Accurate measurements of HRV need relatively long segments of artifact-free data. Notably, the majority of data collection problems were during the initial phase of the study, while the usability and other device features were being optimized such as how tight the wristband needed to be secured to the wrist. There was a higher proportion of available nighttime data (63.3% vs. 43.0% of hourly segments on day 1) compared to daytime, which highlights the impact of motion. The inverse correlation was found in both nighttime only and all hours, but stronger at night. This may be secondary to better quality data or inherent circadian rhythm, although this was controlled for. This study focused on the HF HRV data, but other HRV metrics that focus on vagal tone as demonstrated in prior studies should be evaluated in a larger sample size to confirm or refute the findings of this pilot study [30,31]. Another limitation is that we limited our HRV-analysis segments to one-minute intervals of artifact-free data. This is different from past studies that have collected data over longer periods of time, but was necessary to obtain precise HRV measurements in a real-world scenario using PPG. Our study design used a validated, clinical standard (Columbia Suicide Severity Score) as a measure of suicide severity. However, there is inherent subjectivity to this score, and it begs the question if some patients answer falsely to try and leave the hospital sooner with low scores. We believe this is lower likelihood as patients were explicitly told by the research team that their results would not directly change clinical care. It is noteworthy that we excluded a small number of patients who had extremely low scores on their baseline CSSRS, even though they were admitted for acute suicidality. The dichotomous clinical cutoffs we chose for analysis were to detect relatively large clinical changes. However, it may be of interest in the future to detect smaller changes in suicidality that can be accomplished with larger studies. Lastly, we did not enroll a cohort of healthy controls in our study, as each patient served as their own baseline to compare their changes to over the week study period. Future studies evaluating a single baseline measurement should include healthy controls or patients with a mental health history who are not currently suicidal.

This pilot study in acutely suicidal adolescents showed a statistically significant inverse correlation between the HF HRV and CSSRS measured by photoplethysmography using a wrist-worn device. This association was found both in comparisons using a single point in time, as well as when analyzed as a change over time. Given that many adolescent patients have a repeat visit to the ED or suicide attempt after an initial visit, the ability to detect worsening suicidality in this high-risk population may have positive impacts on patient outcomes and the healthcare system. Further research will need to confirm this relationship in larger cohorts of acutely suicidal adolescents.