This study used a cross-sectional online survey of 342 undergraduate students in Thailand between November and December 2019. To participate in the online survey, the participant had to read the protocol and accept an informed consent document on the first page of the questionnaire. Participants who objected to the informed consent document were directed to the end of the survey. No respondent was forced to participate and he/she could withdraw at any time. For the interview and video recording of the interview, both the interviewer and the participant were informed that the proceedings would be private and confidential. The study was approved by the Ethics Committee of the Faculty of Medicine at Chiang Mai University, Thailand.
Participants
The participants were undergraduate students located all over Thailand, aged18–25, who spoke Thai. The exclusion criteria included being diagnosed with schizophrenia, bipolar disorder, drug or alcohol use disorder, and being intoxicated with alcohol within 24 hours prior to participating in the study. Participants were asked to complete the questionnaires on the Internet via personal computers, laptops, smartphones, or tablets.
Those who scored ≥ 1 on the Short-Bord were randomly selected to participate in a long-distance interview (using Zoom video conferencing) by one of the psychiatrist investigators (NK, AO, NW, and TW) to confirm the diagnosis using the Structured Clinical Interview for DSM-IV Axis II PDs (SCID II). Video of the interviews was recorded for later review.
Sample size estimation was calculated for a prevalence study, which the estimated prevalence of BPD in the university sample from previous study was 30%, the precision was set at 5%, and the confidence level was 0.95. This yielded an estimated sample size of 323. The dropout rate was determined to be 10%. The estimated total sample was 355 for the survey.
To calculate a sample size for ROC analysis, we expected an AUC of 0.725 for the SI-Bord, which was significant compared to the null hypothesis value 0.5 (no discriminating power). We expected to include thrice as many negative cases as positive cases based on the prevalence. An α-level was set to 0.05, and a β-level was set to 0.20 (power is 80%). This yielded a sample size of 68 (17 for positive cases, and 51 for negative cases) [25]. (See Fig. 1).
Inter-rater reliability
Four psychiatrist investigators, with 2, 4, 18 and 28 years of working experience, administered the SCID-II BPD for each interview. None of them knew the participants they had to interview. They were randomized to interview the respective participant arranged by research assistants. Before conducting an interview, an inter-rater agreement was examined using joint video. By that method, 2–3 interview videos of each interviewer, regardless of having BPD or not, were rated by another three interviewers using SCID-II BPD. No SCID-II Personality Questionnaire was used before any interview. This totaled 6 pairs of interviewers for each video, and 60 pairs for all comparisons. The intraclass correlation analysis yielded a coefficient of 0.925, which was considered excellent.
Instruments
In addition to the sociodemographic data, e.g., age, sex, academic year, income, etc., the participants were asked to complete the following measurements.
1) Screening instrument for borderline personality disorder (SI-Bord)
SI-Bord was modified from the Short-Bord [24], consisting of 5 questions representing the DSM-5 criteria of BPD for 1) abandonment avoidance, 2) interpersonal relationships instability, 3) identity disturbance, 4) suicidal and self-harm behaviors and 5) affective instability. SI-Bord has 4-point Likert scales, ranging from never (0) to very often (3), while the original version of the Short-Bord uses a true-false response. This extended response category was intended to increase reliability [26]. In addition, the content of each item was also modified to be more understandable. For example, item 3 (identity disturbance) read “My feelings suddenly changed, such as "I don't know who I am," "I don't know where I am going" or "I feel lonely deep down” or “I have no goal in life ” The total score ranges from 0 to 15. The higher score represents more BPD symptoms or traits. The study sample yielded a Cronbach’s alpha of 0.76.
2) Revised Thai Multi-Dimensional Scales of Perceived Social Support (r-MSPSS)
This tool measures the extent to which an individual has experienced being support by significant others (SO), friends (FR), and family (FA) [27]. It has 12 questions with 7-point Likert scales ranging from very strongly disagree (0) to very strongly agree (6). The higher the score, the higher the level is attained of perceived social support. The revised-Thai version demonstrated good psychometric properties [28]. The study sample yielded a Cronbach’s alpha of 0.91.
3) Thai version of Perceived Stress Scales (T-PSS-10)
The T-PSS-10 is a 10-item self-reporting questionnaire measuring the extent to which an individual perceived stress over the past four weeks. It uses a 5-point Likert scale ranging from never (0) to very often (4). The higher the total score, the higher the level is attained of feeling stress. T-PSS-10 demonstrates good psychometric properties [29]. The study sample yielded a Cronbach’s alpha of 0.85.
4) Patient-Health Questionaire-9 (PHQ-9)
The PHQ-9 is a 9-item self-reporting questionnaire measuring the extent to which an individual has experienced depressive symptoms over the past two weeks [30]. The 4-response Likert scale ranges from 0 (not at all) to 3 (nearly every day). The higher the total score, the higher the level is attained of depressive symptoms. The Thai version PHQ-9 showed a Cronbach’s alpha of 0.79 and a significant association between the PHQ-9 and the HAM-D[31]. The study sample yielded a Cronbach’s alpha of 0.89.
5) Thai version of Structured Clinical Interview for DSM-IV Axis II PDs (T-SCID II)
T-SCID-II is a structured clinical interview for personality disorder that has been used to diagnose personality disorders based on DSM-IV and DSM-IV TR[32]. The Thai version demonstrated a promising interrater reliability of ≥ 0.8. [33, 34]. In this study, only the BPD section was used.
Statistical analysis
For sociodemographics and scores of psychological measurement, descriptive statistics, e.g., frequency, percentage, mean and SD were used. In evaluating the diagnostic validity of the SI-Bord, receiver-operating characteristics (ROC) analyses were used to calculate the values of sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV). We calculated the Youden index J [35] which is defined as: J = max (sensitivityc + specificityc − 1), where c ranges over all possible criterion values. MedCalc Statistical Software, Version 19.2.1 (MedCalc Software Ltd, Ostend, Belgium) was used for ROC analysis. MedCalc was used to calculate the optimal criterion and associated sensitivity and specificity, and the optimal criterion value which takes into account not only sensitivity and specificity, but also disease prevalence, and costs of various decisions. Cost, as defined by MedCalc, is the average cost resulting from the use of the diagnostic test at that decision level such as the cost of doing the test, which is constant at all decision levels. ROC analyses were also used for the area under the curve calculation (AUC) to determine the performance of the test.
Interrater reliability calculated by intra-class correlation coefficient (ICC) was used for rater’s agreement. An ICC value of > 0.8 was acceptable. Internal consistency of the instrument was determined by Cronbach’s alpha, for which a value of > 0.7 was considered acceptable. Concurrent validity of the SI-Bord was examined using Pearson’s correlation coefficients between SI-Bord and with other measurements, i.e., rMSPSS, T-PSS-10, and PHQ-9. All analyses, except for ROC, were performed using IBM SPSS, Version 22.