Unremitting Suicidality in Borderline Personality Disorder: A Single Case Study and Discussion of Technology in Clinical Care

CASE HISTORY AND TREATMENT This report presents the case of a young woman, “Jane.” The case is followed by commentary from three experts in suicidal or self-destructive behavior and the use of technology in clinical care. Jane is a college-aged female with a diagnosis of borderline personality disorder (BPD) and a history of repetitive self-harm, including head banging and cutting, as well as chronic, unremitting suicidal ideation. Jane had been in an intensive residential program for many months, and during this time she was stepped up to inpatient psychiatric hospitalization multiple times due to concerns about safety. Jane had a history of multiple suicide attempts while in outpatient, residential, and inpatient settings. Most of these attempts occurred on hospital units using methods like self-strangulation with available objects, self-suffocation, and self-starvation. Upon discharge from her last inpatient psychiatric admission, Jane was admitted to an intensive specialized residential program for adult women with borderline and other severe personality disorders. This unit integrates evidence-based treatments including dialectical behavior therapy (DBT),1 mentalization based treatment (MBT),2 and good psychiatric management (GPM).3 The standard treatment protocol includes daily DBT diary cards, check-ins with the program’s 24-hour counselors, and regular assessment of patient risk for behavioral dyscontrol, specifically self-harm and suicidal action. Jane continued to have multiple episodes of cutting and head-banging. On standard assessment methods such as diary cards, Jane’s reported consistently high levels of suicidality, with very little variability. This resulted in hypervigilance among the staff, as it was difficult to discern when Jane’s level of immediate risk was spiking. The need to provide one-to-one attention to Jane was also causing burnout amongst the staff. After two months, Jane was again stepped up to inpatient care due to a lack of reduction in her level of suicidality, despite extensive coaching from interdisciplinary staff. During hospitalization, the clinical team regrouped with two novel ways of assessing Jane’s risk for suicidal action. The first adjustment was for Jane to send daily emails to her primary therapist. This choice was driven by Hooley et al.’s finding that daily journaling improves deliberate self-injury, regardless of the topic of the journal entry.4 The team also implemented daily ecological momentary assessment (EMA) prompts assessing momentary stress and enjoyment, and they purchased an over-the-counter wearable device to track Jane’s activity and sleep levels. In order to understand the data collected, the team gathered retrospective information from Jane about her subjective experiences during this phase of treatment. Data was collected over a period of 44 days. The wearable device used was a Garmin Vivofit 4 smartwatch, and data regarding the patient’s activity and sleep were exported daily. Online surveys were sent to the patient three times a day at randomly generated times. The survey asked three questions: (1) What are you doing right now? (2) How enjoyable is it (1–5)? and (3) How stressful is it (1–5)? In total, 133 EMA responses were recorded, with 99% compliance. Jane granted the first author and the research team access to her daily emails, her diary cards, and her medical chart. We conducted a literature review of narrative expressions of suicidality and constructs related to suicide risk and constructed a list of relevant terms. The term list included words and concepts related to (1) passive suicidal thinking, (2) active suicidal thinking and urges, (3) nonsuicidal self-injury urges, and 4) passive thoughts of nonsuicidal self-injury. (See Table 1 for a list of terms included.) Two research assistants read and manually coded each email’s count of words/concepts from each category. Table 1 - Email Coding Terms Passive suicidal thinking - Burnap et al. (2015), 5 O’Dea et al. (2017) 6 • Sleep and never wake or sleep forever • Disappear or don't want to exist • Can't go on • Not worth living/nothing matters Active suicidal thinking • Overdose • Hurt/kill myself • End it all/end my life • Description of suicidal urge NSSI urges • Description of an explicit urge to self-harm NSSI-related thoughts • Descriptions of thoughts regarding self-harm, with no indication of a specific urge Joiner’s Interpersonal Theory of Suicide - Joiner et al. (2002) 7 • Feeling like a burden • Thwarted belongingness - coded as descriptions of incidents in which the patient tries to connect with a group and reports it to be unsuccessful • Hopelessness - coded as the literal word “hopeless-” or an expressed feeling that things will “never” get better Uncategorized - O’Dea et al. (2017), 6 Van de Nest et al. (2018) 8 • Death/die/dying • Suicide/suicidal • Aloneness • A general “other” category - used rarely, when content clearly referenced self-harm or suicidality but did not fit any other phrase or idea in this list With the EMA data, we visualized the ratings of momentary stress and enjoyment levels, taking into account the patient’s retrospective report of her experiences, as well as the highly structured daily schedule of the program. Her enjoyment stayed low throughout the EMA period, ranging only from 1–2 out of 5 (Mean = 1.13, SD = 0.34). Isolated spikes in enjoyment ratings typically occurred during solitary activities, consistent with her retrospective report that less stimulation was “comforting” and that she often lay in bed or watched television to “distract herself” from her distress. Other spikes in enjoyment were related to structured interactions with the milieu. The patient’s stress was more variable, ranging from 1–5 out of 5 (Mean = 2.57, SD = 0.83), though her ratings were consistently greater than or equal to 2 in the three weeks before she was again hospitalized. Spikes in stress tended to accompany regular treatment-related activities, which the patient retrospectively described as “overwhelming” due to both the required homework and the fact that she “was around other people” (Figure 1).Figure 1: Stress and Enjoyment Ratings in Context of Patient Experience. Visual representation of EMA data on momentary stress and enjoyment levels. Enjoyment is low throughout, never exceeding 2 out of 5 (M = 1.13, SD = 0.34). Isolated spikes in enjoyment ratings mostly occurred during solitary activities. Stress ratings show greater variability, ranging from 1–5 out of 5 (M = 2.57, SD = 0.83). After day 19, stress was consistently rated 2 or higher.Using a Hidden Markov Model, a computational model for identifying hidden discrete states in dynamic data,9 we fit a two-state model with the R package depmix4.10 The two-state model captured two underlying states: high versus low stress. The model (Figure 2) indicates that Jane tended to oscillate between these states relatively quickly at the beginning of data collection, but that over time, duration of the high stress states increased. We additionally used a vector autoregression model to evaluate contemporaneous (i.e., same assessment) and lag relationships (i.e., across assessments) between stress and enjoyment (Figure 3). We fit the model with the R package graphicalVAR.11 As expected, there was a negative contemporaneous correlation between stress and enjoyment. There was no significant lagged relationship between the two variables, meaning stress at Time 1 did not predict enjoyment at Time 2. There was a positive significant autocorrelation for stress, but not for enjoyment. This finding indicates that if Jane was already feeling stressed, she was likely to continue feeling stressed, but if she was already feeling some enjoyment, she was not significantly more likely to continue feeling enjoyment.Figure 2: Stress States Over Time. The top graph shows the patient’s raw stress rating at each observation. The bottom graph is a representation of a Hidden Markhov model of two proposed underlying states: high versus low stress. In the beginning of data collection, the patient oscillated relatively quickly between states. Over time the duration of high stress states increased.Figure 3: Vector Auto Regression Models of Stress and Enjoyment. Visual representation of a vector autoregression model evaluating contemporaneous and lag relationships between stress and enjoyment. There was a negative contemporaneous correlation between stress and enjoyment as shown on the left. Stress at Time 1 did not predict enjoyment at Time 2, as indicated by the lack of significant lagged relationship between the two variables.Jane’s diary card showed no variability in her daily ratings of suicidal ideation throughout her stay. However, her daily emails included detailed and explicit mentions of suicidality. A simple visualization of the data by day showed repeated ebbs and flows of suicidality over time, reflected in active suicidality, passive suicidality, aloneness, and hopelessness. The active suicidality seems to increase over time, with the peaks steadily increasing, even while separated by periods of low acuity. Interestingly, diary card ratings of suicidal urges during the same time period remained relatively stable at a high level, with little variation (Figure 4).Figure 4: Suicidality in Daily Emails and Diary Cards. Graph of daily data on active suicidality, passive suicidality, aloneness, hopelessness, and diary card ratings of suicidal urges over 45 days. Peaks of active suicidality increase over time, even while separated by periods of low acuity. Diary card ratings of suicidal urges remain stably high with little variation.To understand how the emails reflect changing levels of suicidality over time, we ran a nonlinear growth curve model with the predictor variable as day at the RTP and the outcome variable as the number of references to active suicidality in the daily emails (Figure 5). The result indicates an increase over time, as well as a bump prior to Jane’s hospitalization, which may represent some warning signs of Jane’s increasing risk.Figure 5: Nonlinear Growth Curve Model of Active Suicidality. Graph of nonlinear growth curve model of number of references to active suicidality in emails by day. Active suicidality by this measure increases over time, with an uptick prior to hospitalization.There was high variability in both hours of sleep (Mean = 8.29, SD = 1.19, Min. = 6, Max. = 11.3) and activity level (i.e., steps; Mean = 7,745.35, SD = 2,893.15, Min. = 1,519, Max. = 13,523). Linear models of each variable indicate an overall decrease in activity level, paired with an increase in hours of sleep (Figure 6).Figure 6: Linear Trends in Stress and Enjoyment. Ratings of stress (left graph) and enjoyment (right graph) over 133 observations. Stress level shows high variability throughout each day while enjoyment level stays stably low, between 1 and 2, throughout.On the daily level, we looked at correlations between the EMA and wearable data (Figure 7). To do this, we took the daily averages of the EMA data. Aside from the negative correlation between stress and enjoyment, we do not see significant relationships, suggesting that these two types of data provide unique information. Lastly, interested in correlations across EMA and the emails, we took a daily average of their stress ratings, and we found no significant relationship between the momentary stress and the active suicidal thinking reported in the emails (Figure 8). This suggests that these sources of data are providing unique information and the importance of understanding both momentary stress and a global, daily experience of suicidal thinking.Figure 7: Correlations between Stress, Enjoyment, Sleep, and Activity. Visual representation of correlations between EMA and wearable data using daily averages of EMA data. The only significant relationship is a negative correlation between stress and enjoyment.Figure 8: Relationship between Active Suicidal Thoughts and Stress. Graph of daily average of stress ratings and active suicidal thinking as reported by email. No significant relationship was found.While there were no instances of self-harm or suicide attempts during data collection, Jane needed to be stepped up to inpatient treatment a second time, during which time clinical consultation was obtained regarding the effectiveness of residential care and treatment alternatives; the patient was ultimately discharged. Jane, like most people with BPD, especially at intensive care levels, reported a history of suicidal behavior.12,13 Despite the fatality of the disorder, with up to 10% completing suicide,14 predicting when and which individuals are at greatest risk for lethal attempts remains elusive.15 Severity of symptoms and risk of completed suicide in BPD vary within and between individuals. In Jane’s case, like many, reactive suicidal threats (i.e., verbal expressions of suicidal thoughts or nonlethal gestures without intent to die) can be a communication of distress in close relationships, in contrast to suicidal ideation with significant hopelessness and despair, which is thought to occur when the individual is and suicidal of distress are with care and the that may be of some of self-destructive suicide risk level is especially to this for with BPD, but to suicide risk in suicidal is by of retrospective which and To these an of literature on EMA as a assessment of momentary suicidal ideation and which may be as of immediate Suicide risk can on individuals with BPD suggests that the relationship between momentary levels of negative and the of suicidality is in individuals with BPD than in without With research and EMA data may suicide which can the level of based on the of the patient’s EMA A review indicated that many EMA lack for their and but some have such as significant in suicidal and that may EMA wearable that data such as variability and as well as activity level, and other data collected wearable were among the most in a model predicting of that (i.e., significantly BPD and with BPD from with disorder and The use of behavioral and data may be especially in or repeated are or not such as with with to the clinical of the the in the not be conducted in This also on the and of the and is more in a residential or other Despite these this may be used as a for of suicidality in residential or inpatient settings. - What is the of EMA in the care of suicidal and - What need to be before EMA and wearable can be used in clinical The treatment of suicidal with BPD is often by the in between suicide related to a of and aloneness, and which can be communication of distress. a may not be clearly to the The to and to may the need for from an an that may self-destructive The the patient to and models responses that The is that over time, this is and the patient is to these of distress and to without self-destructive EMA some to of expressions of their distress and may as of suicide Data obtained such methods to that the and ultimately the about of that may be to suicidal in this EMA can suicide by the patient episodes of A of using this technology in the residential is the for have a relationship with the patient to with the data in a more with suicidal distress are very difficult to as they to the of the treatment often to burnout or In the case reported severe suicidal distress and self-destructive behavior continued despite inpatient and residential treatment suicidality that is to the the treatment team to an the outcome in this case patient did not self-harm or suicide during this time period, but did the treatment novel suggests they were to to the patient that they had not up and her distress be The team out this and were to go to to the patient their to The residential treatment the clinical team to and the patient’s in multiple of data collection, which be more difficult in an or less clinical The was that people about her and to even though attempts had been This two suicide the importance of feeling to people or and the of with The clinical of subjective distress data the that the underlying can be and with are interested in the patient’s been to be However, in this of treatments and we are to that is not but better to can to and in their residential treatment may be an to the use of EMA and wearable some of the data collected is by the of the program. the sleep and activity levels are by group and the daily EMA and activity data may be more when collected during when the patient’s are and However, in any treatment communication to of the treatment team to in order to or The is that the may be over time, without the need for the of and the to subjective to a that can be suicide can be in this especially when the patient is in or no in the to responses to suicide attempts with and for periods of time the of stress, and about the effectiveness of from with of the treatment This in their to and their for which in and In this may be as distress can an in and of distress and suicidal that growth and the experience of However, research is needed to the of data that are most in predicting risk and in order to use of wearable and EMA data The of technology in the clinical care of self-destructive behavior thinking and a to suicide for with are needed to evaluate these and including the use of for of BPD is a disorder by in relationships, and the that are for BPD, dialectical behavior therapy is among the most and DBT around which BPD as a disorder of that due to a between a high and a In this individuals with BPD behavioral nonsuicidal self-injury to their DBT that or are due to these and with to more behavior to DBT treatment by the of their are in such as NSSI or other suicidal are and their are and In the case of Jane, she had episodes of suicidal including lethal suicide attempts and is the top of a DBT to and track the suicidal which is using a daily diary The DBT diary card two to that occur therapy and to each individual a reports that she in a as an behavior in the will typically both a and the the and to to understand the that to in the the the and to and more that have been and may have the of that the the DBT diary card and on In the case of Jane, her to out the diary card her was to in life and the reported content from the diary card was to treatment. The clinical team Jane’s treatment to a of and subjective data methods to better understand and more evaluate Jane’s and behavioral included a daily EMA to for activities, enjoyment, and stress level, and a Garmin Vivofit 4 to activity and as well as access to Jane’s emails and medical chart. data have the to provide time clinical data that can the of In EMA momentary urges, and that are an of an the diary card for a are to the diary card but the and it less to as Jane, it was that she tended to report more stress than enjoyment, and that her stress was while enjoyment was more and data provide a novel for evaluating behavioral data such as and In Jane’s case, her wearable data indicated that her sleep and activity were highly Jane’s emails highly relevant information about her and Jane’s emails explicit mentions of suicide that her Interestingly, there were no correlations between Jane’s explicit mentions of suicide and her EMA or activity data. of data are used to better understand and predict and other of behavioral such as and is needed to better these data in clinical in they are and are to daily are in DBT can quickly at and understand the and correlations and for the thoughts if more is The of and data their in time clinical this case were used to the data. The data be more if they be within a in order to with other individuals. data and occurred the clinical there is significant of these of data into clinical in assessment methods are or it is to that Jane to the treatment team using and these data to treatment. In an of data and be the especially in predicting and Jane’s case is an of how clinical care is in a of and to data. The and to care are by the and sources of data the clinical team gathered three of clinical history and 2) assessment of the patient’s to the evidence-based treatments on the and data gathered in time EMA and The first by Jane’s clinical remains for diagnosis and clinical across and care. on from or and by other information is to the of clinical and The second of clinical data data. In Jane’s case, this captured an understanding of a level of risk on to the specialized care and the use of daily diary The second information that in the of the first of data to treatment not in this case, care also a of of the second of data. these two of data have the of clinical care. However, their have been A significant is that these are not to behavior or responses to that in This is of and for such as Jane are in treatment for that get and to negative is only over the to that the and of such as or the of a of time data on such The case of Jane how such data can to and highly In her case, it to for suicidality. Other have that this can be for a of clinical including of behavioral and symptoms in and of symptoms in However, significant will have to be before such can the standard of care. is the and required to and the of data collected assessment which data than a high of The of data are not of clinical and it is and intensive to which data may clinical Most the effectiveness of such are case conducted in that have the to the To this at will levels of in the computational needed to provide time is not a that the literature on such than time of data, despite the for such data to time is likely that the effectiveness of these will need to be as and clinical the in such there are two concerns that may be more relevant in care The first is the risk of driven care care it the access and of to this that such as and such as access and data are it is well that this is not the This be and while care is in The second is of data and have that a high of and information can to and the for and to data is likely to To this risk will a that includes of that can data and and in relevant will also the of and that the importance of into care without Jane’s case how can clinical for the but in the narrative are the significant that will need to be before this can be

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.