A methodological framework for conducting scoping reviews by Arksey and O’Malley and Levac and colleagues was followed in the preparation of this study (27, 28). Therefore, we (1) identified the research questions, (2) identified relevant studies, (3) selected final studies to be included in the review, (4) charted the data, and (5) collated, summarized, and reported results. Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping reviews (PRISMA-ScR, see Figure 1 ) was followed to ensure that the process was systematic, complete, and transparent (29).

Our research question aimed to identify which observable cues, spontaneously expressed by individuals with BPD, can be objectively measured during a clinical interview or in a home setting. We focused on identifying cues such as language use, speech patterns, facial expressions, other forms of nonverbal communication, and physiological measurements. To guide this investigation, we established specific inclusion and exclusion criteria, which are explained in the following section.

Scopus and Web of Science, two large and extensively used databases were used to identify the relevant papers (30). These databases complement each other well (31), and are known for their large overlap with other sources, such as MEDLINE and EMBASE (32). After several rounds of preliminary searches in both databases, which helped us refine our search strategy, we carried out the main search on March 31, 2023. The final version of our search strategy combined different terms related to borderline personality disorder and observable cues ( Table 1 ). All terms were searched for in their singular and plural forms, as well as British and American English. The terms were later combined into a nested format using Boolean operators (AND, OR) and searched for in titles, abstracts, and keywords. The exact search string was (“borderline personality disorder” OR “emotionally unstable personality disorder” OR “emotional intensity disorder”) AND [((sign OR cue OR express* OR feature OR indicator OR marker OR indices OR property OR reaction OR synchrony OR characteristic OR pattern] AND [text OR video OR image OR audio OR speech OR language OR paralinguistic OR prosodic OR semantic OR acoustic OR lexical OR facial OR visual OR appearance-based OR vocal OR written OR verbal OR nonverbal OR conversational OR behavioral OR behavioural OR movement OR soft)] OR [“respiratory sinus arrhythmia” OR “blood pressure” OR pulse OR “heart rate” OR “respiratory frequency” OR “respiration rate” OR “emotional recovery” OR communication OR terminology)].”

To be considered for inclusion, the records had to: 1) be available in English and 2) be published in scientific journals or conference proceedings. Records were then excluded if: 1) they employed samples of patients with disorders other than BPD (e.g., depression), 2) focused on cues that cannot be measured objectively (e.g., reliance on self-report), 3) focused on cues that need to be monitored over a longer period (e.g., monitoring patients for 24 hours), 4) focused on cues that need specialized equipment to be assessed (e.g., brain imaging techniques), 5) focused on cues that are responses to specific stimuli (e.g., recording eye movements during cognitively demanding tasks), 6) they were reviews or meta-analyses, and 7) if they were single case studies (i.e., with N = 1). The methodological quality of studies was not a reason for exclusion.

The main search, performed in SCOPUS and Web of Science, resulted in 3223 English language articles (SCOPUS: 1570, Web of Science: 1653) published in scientific journals and conference proceedings. An additional search in Google Scholar was also performed to ensure we didn’t miss any relevant records, but it did not lead to the identification of any previously unidentified articles. The two chosen databases overlapped considerably, which led to the removal of 1064 records. After screening the titles and abstracts, 185 records (8·6% of identified unique records) still fit the criteria. However, further assessment revealed that only 24 articles (1·1% of identified unique records) met the inclusion criteria. These articles were included in the final synthesis.

Following the research question, a spreadsheet form was developed and used for the extraction of the following information from the reviewed papers: (1) authors, (2) year of publication, (3) country where the study was carried out, (4) sample characteristics, (5) BPD diagnosis, (6) (potential) psychiatric comorbidity, (7) comparison group, (8) observed cues, and (9) methods of observation. The data were extracted from the reviewed papers by two researchers (SM and (NP)) and refined in an iterative process while reviewing the papers.

We followed the aim of scoping reviews, i.e., mapping of the existing findings and providing their overview in a descriptive manner (27). The results were analyzed by three authors (SM, HGK and NP) via thematic analysis (33). Main categories of cues of borderline personality disorders were predefined, and included features related to (1) language use, (2) speech, (3) facial expressions, (4) other features of nonverbal communication, and (5) features related to physiological measurements. The process of collating and summarizing results was reviewed by two authors (NP and US).

While the publication year ranged from 1993 to 2022, more than 75·0% of included studies (n = 19; 79·2%) were published in the last ten years, and precisely 50·0% (n = 12) were published in the last five years. Geographically, most of the studies (n = 17; 70·8%) were conducted in Europe, followed by studies conducted in North America (n = 5; 20·8%), Australia (n = 1; 4·2%), and Asia (n = 1; 4·2%).

The reviewed studies involving human participants had sample sizes ranging from 10 to 143 (average 58·1, SD = 39·0), but BPD patients were only a subset of the sample. Additionally, the three studies without human participants analyzed from 225 to 149,798 text posts. In studies reporting demographic data, the samples were entirely or predominantly female, whereas the average age of participants was mostly (73·9%) in the 25-35 range. Most studies used DSM (either DSM-IV or DSM-V) criteria to diagnose BPD (n = 17; 70·8%), while others used different ways of diagnosing BPD (n = 2; 8·3%) or did not attempt a diagnosis (n = 5; 20·8%). Nearly half (47·4%) reported significant comorbidity in their samples.

Regarding observed cues, half of the studies (n = 12; 50·0%) focused on physiological cues like heart-rate variability, blood pressure, skin conductance, and respiratory sinus arrhythmia. The other studies primarily examined language use (n = 6; 31·6%), facial expressions (n = 2; 8·3%), speech (n = 1; 4·2%), and non-verbal communication (n = 1; 4·2%). Two studies (n = 2; 8·3%) observed cues from multiple categories. Various methods, including electrocardiogram, Ag/AgCl electrodes, electromyogram, videotaping, audio recording, and coding systems, were used to measure these cues objectively. Patient data were compared to healthy controls in 45·8% of studies (n = 11), other disorder patients in 12·5% (n = 3), both healthy controls and other disorder patients in 20·8% (n = 5), and in the remaining 20·8% (n = 5), there was no comparison group. For more details, refer to Table 2 .

Patients with BPD within borderline personality organization (BPO) tend to favor language that highlights external viewpoints and maintains an impersonal tone. This language conveys ideas neutrally, avoiding personal attribution. Multiple studies support the prevalence of third person singular pronoun use among patients with BPD (47, 49). Notably, Jeanneau and Armelius highlight the prominence of the pronoun “they” as characteristic of BPD (47). Singular pronoun use, including “you” is heightened in BPD (49). Another distinct feature is the frequent use of future tense (47, 52).

Numerous other formal grammatical features are distinct in BPD, including adaptor words (like “so”), time-space adverbs (like “then”), conjunctions (such as “because”), intent expressions (like “I mean”), capability expressions (“I can”), negative verbs, nouns, and adjectives (like “awful”, “disgusting”, “unhappy”, “crazy”), as well as frequent negation (“not”) (47, 52). Another observed linguistic feature is the combination of nonfluencies with conjunctions. A preference for absolute words, coupled with common adverbs and article words (such as “a”, “an”, and “the”), is also noticeable. Conjunctions combined with nonfluencies, and the pairing of “you” with verbs and nonfluencies, also emerge as notable markers (56). While patients with BPD use adjectives frequently, interjections are less common (54).

When comparing bipolar disorder (BD) and BPD, several distinctions encompassing quantitative measures like Brunet’s index (BI), moving average type-to-token ratio (MATTR), and mean length sentence (MLS) emerged in Wang et al., but they do not clarify further on the actual differences between both conditions. Other notable differences include “we” with prepositions and a mix of absolute words, common adverbs, and negations. Thematic content focuses on social processes and drives. The study emphasizes the significance of linguistic complexity in differentiating BPD from BD and controls. Linguistic features play a key role, while content features have limited impact on classification. Findings underscore linguistic features’ importance, the drawback of omitting linguistic and dialogue aspects, and the relative dispensability of dialogue and content features (56).

Dyson and Gorvin unveiled discourse dynamics about BPD, depicting a spectrum from BPD as a source of tension to its depiction as a distinct and unique existence. Each of these narrative repertoires exhibited its own set of linguistic patterns and frequently employed words and phrases that actively contributed to shaping these portrayals. The “BPD as a source of tension” repertoire highlights several preferred themes, encompassing struggles for control, reductionist and deterministic terminology, expressions of powerlessness, and a biomedical standpoint. Specific linguistic patterns and commonly used words/phrases, like “disorder”, “do everything right”, “disorder takes over”, “out of nowhere disorder takes over”, “stupid things”, etc. reinforce this view. In contrast, the “BPD as a different existence” repertoire emphasizes themes of acceptance, positive self-comparisons, and openness. Using linguistic strategies describing BPD as just one of many ways to engage with the world, and the consistent use of words and expressions like “passion”, “specialness”, “embrace”, “unstable”, “abnormal”, and “different”, a coherent alternative narrative emerged (41).

In BPO language use, words often carry aggressive and depressive connotations (52). Individuals with BPD show limited negativity in interviews (55). On the other hand, their writing contains numerous negative emotion words (49). This is consistent with heightened negative emotional valence (45). Further exploration of their online conversations revealed distinctive characteristics, including an emphasis on mood-related vocabulary, a distinct focus on dating experiences (both positive and negative), a notable intertwining of work-related concerns with discussions about medication and diagnoses, and an overall optimistic outlook regarding the efficacy of treatment through medication, despite acknowledged challenges (54).

No significant differences were found between patients with BPD and healthy controls in terms of the complexity of sentence structure, or the complexity of the content being communicated (39).

Although there were no significant overall speech differences between BPD patients and healthy controls, individuals with BPD displayed notably higher pause frequency during neutral speech conditions (39). Additionally, a distinctive dialogue-related aspect emerges, encompassing number of words per second, pause duration, and relative floor control (56). Compared to the control group, patients show more frequent correlations, especially positive ones, between elements of prosody and facial expressions of certain negative emotions. Specifically, patients exhibited a positive correlation between facial expressions of disgust and anger and the acoustical parameters of adjectives and interjections, both in terms of decibels and fundamental frequency, which was absent in controls. Additionally, some correlations exhibit opposite patterns: negative in controls and positive in patients, notably for disgust and adjectives (f0-dB), anger and adjectives (dB), and anger and interjections (f0). Vocal characteristics of adjectives and interjections showed no significant differences between patients with BPD and controls (54).

In contrast to healthy controls, individuals with BPD exhibited reduced affiliative behavior, concerning patterns of behavior that invite and positively reassure social interaction, including “head to side” movements; “bob”, a sharp upward movement of the head, similar to an inverted nod; “flash”, a quick raising and lowering of the eyebrows; “raise”, a movement in which the eyebrows are raised and kept up for some time; and “smile”, in which the lip corners are typically drawn back and up. Both groups demonstrated comparable inclinations to engage in flight behaviors, expressing the avoidance of social interaction and comprising behavioral features that lead to cutting off communication (38).

Regarding facial affective behavior, individuals with BPD demonstrated prominent negative emotions like disgust and contempt. However, they also exhibited social smiles, indicative of positive social interactions (40). Patients with BPD displayed less than half the amount of sadness compared to the control group, indicating a complex blend of emotional expressions (54). Cluster analysis divided patients into two groups: Cluster 1 showed higher overall facial activity and intense negative emotions (anger, contempt, disgust), combined with significant social smiles; Cluster 2 displayed lower levels of specific negative emotions while maintaining notable social smiles (40).

Nonverbal synchrony between patients with BPD and interviewers was significantly higher than chance (pseudosynchrony). Patients with BPD often led in nonverbal interactions, with interviewers mirroring their cues more than the opposite (51).

Several studies found no significant heart rate differences between individuals with BPD and control groups at baseline (37, 42, 50). On the contrary, some studies revealed patients with BPD having elevated rates (37, 48, 57).

In terms of ECG measurements encompassing various parameters such as respiratory sinus arrhythmia (RSA), square root of the mean squared differences of successive NN intervals (RMSSD), standard deviation of normal-to-normal intervals (SDNN), and vagal-mediated heart rate variability (HRV), Bortolla et al. found no significant differences in baseline measurements between individuals with BPD and control groups (37). Likewise, no substantial differences in ECG amplitudes were observed between patients with BPD and healthy controls (43). Patients with BPD exhibited reduced RSA levels with a detected negative correlation between BPD symptom severity and RSA, indicating decreased RSA as symptoms intensify (48, 53). Females with BPD demonstrated lower mean RMSSD values compared to healthy controls (35). Regarding HRV, heightened BPD symptom severity was linked to reduced HRV during rest (57). In another study individuals with BPD showed notable heart rate fluctuations compared to other groups, without significant distinctions in HRV measures (50). The study by Flasbeck et al. unveiled significant connections among cardiovascular measures, symptom severity, and the parasympathetic nervous system (PNS) index (43). Additionally, individuals with BPD displayed longer QTcd duration than controls, aligning with findings from Bomba et al. showing a mild yet statistically significant positive correlation between QTc and QTcd measurements within the BPD group (36).

Patients with BPD showed higher systolic blood pressure (BPsys) than controls, particularly during rest. Additionally, patients with BPD consistently displayed elevated diastolic blood pressure (BPdia) in comparison to controls (44). Notably, BPD symptomatology did not predict cardiovascular reactivity (CVR), computed from BPsys and BPdia values (46).

Resting skin conductance levels (SCL) in patients with BPD resembled those of healthy controls (42). Yet Aleknaviciute et al. revealed heightened overall SCL in patients with BPD, indicating intensified physiological responses associated with emotional experiences (34).

Patients with BPD showed significantly lower mean resting scores in left orbicularis oculi EMG compared to controls (42).

The findings in this review should be considered in light of certain limitations observed in the included articles. Most studies primarily focused on female participants, potentially limiting generalizability, despite more recent research challenging the perception that BPD mainly affects females. Diagnostic bias toward females results from their greater likelihood to seek early mental health assistance, while males may delay diagnosis due to substance abuse and incarceration. Diverse gender representation in BPD research is crucial (63, 64). Additionally, the concentration of studies on young adults aged 25-35 neglects other age groups with BPD. Geographic concentration in Europe could introduce regional biases. Sample sizes varied significantly, some being very small, and some studies lacked comparisons with healthy controls or other groups, making it difficult to draw clear inferences. Nearly half of the studies had comorbidity within their samples, potentially confounding observed cues specific to BPD. Diagnostic heterogeneity arose from varying diagnostic methods. Most studies investigated physiological cues, while only a few explored language use, facial expressions, speech, and non-verbal communication. Our review is limited to English language articles, potentially introducing publication bias. We acknowledge the limitation of not being able to include articles in other languages, which could have offered valuable insights into this subject. Similarly, some of our findings may be biased due to the fact that we included conference proceedings, which are not always refereed, and the fact that we did not assess the methodological quality of the included articles. Mentioned limitations underscore the need for further research addressing these constraints and exploring a broader range of observable cues in diverse BPD populations. Future studies should consider comorbid conditions and the impact of medication on individuals with BPD, as well as recognize diagnostic variations. Addressing these factors can offer a more nuanced and accurate portrayal of BPD, facilitating improved assessment and treatment strategies for this complex disorder.

Our in-depth study of BPD characteristics uncovers innovative possibilities for precision psychiatry using digital biomarkers. We identify distinct language patterns in individuals with BPD, emphasizing external viewpoints and an increased use of the future tense. Additionally, specific linguistic characteristics unique to individuals with BPD shed light on their thought processes and emotional experiences, enhancing diagnostic precision. These linguistic cues serve as essential building blocks for the development of precise diagnostic algorithms.

In the realm of nonverbal behaviors, our investigation unveils intriguing insights, including an increased frequency of pauses in BPD individuals’ speech. These findings suggest unique communication tendencies and emotional regulation challenges. Moreover, our research highlights significant correlations between prosody (the rhythm and melody of speech) and facial expressions, providing valuable information about emotional states and interpersonal dynamics. These nonverbal cues enhance our ability to recognize and understand the complexities of the disorder.

The integration of physiological measurements into our study represents a significant advance. By analyzing emotional responses and physiological arousal in individuals with BPD, we gain deeper insights into the psychophysiological aspects of the disorder. This knowledge not only aids in early identification but also opens the door to personalized interventions that address the unique needs of each patient.

Our approach aims to complement existing BPD management methods, ushering in an era of precision and customization in mental healthcare. By incorporating machine learning insights into the diagnostic process, we intend to improve the accuracy of BPD assessments and empower clinicians to make more informed decisions. These groundbreaking tools represent a transformative shift in BPD diagnosis and treatment, offering renewed hope for those dealing with this condition. As our research advances, we anticipate improved diagnostic precision and highly personalized treatment strategies, ultimately creating a brighter future for those facing the challenges of BPD. Digital biomarkers play a central role in this transformative journey.

In conclusion, this review underscores the pressing need for improved BPD diagnosis and management. The integration of AI and observable cues, including language patterns and nonverbal behaviors, charts a promising course towards enhancing diagnostic precision and personalized treatments for individuals with BPD. These unique cues, harnessed by AI-driven machine learning algorithms, stand as a beacon for the potential transformation of BPD management, facilitating early identification and timely interventions. However, this promising future must be approached with caution, addressing significant ethical concerns such as privacy, data security, and bias. Ensuring transparency and preventing the exacerbation of existing disparities are essential to realizing AI’s transformative potential in mental health care. By balancing innovation with ethical considerations, we can pave the way for more equitable and effective solutions, ultimately improving outcomes and quality of life for those contending with BPD.