This longitudinal study investigated callers to the Beijing Psychological Support Hotline, established in 2002. Although the hotline is located in Beijing, it serves callers across China [14]. The hotline is one of the largest crisis lines for suicide prevention in China and delivers free 24/7 psychological services nationwide. It receives 20,000–30,000 calls annually. All callers identified as high-risk are promptly administered a semi-structured psychological intervention lasting up to 90 min and are followed up for up to 12 months, which aims to decrease the caller’s risk for suicide [18]. In follow-up calls, if the callers presented high suicide risk, psychological interventions including a safety plan would be delivered by hotline counsellors. Details of the intervention were described in a previous study [18].

From January 2017 to December 2018, AYA callers assessed as being high suicide risk were recruited. High suicide risk was defined as meeting any one of the three conditions: (1) having a suicide plan and will make suicidal acts within the next 72 h, (2) reporting a suicide attempt in the past two weeks, and (3) ongoing suicidal acts several minutes before or during the call. The participants were followed up with 12 months after the call. The inclusion criteria were (1) being 12–25 years old at the index call (the first call included during the study period), (2) the index call being answered between 1 January 2017 and 31 December 2018, and (3) meeting the criteria of high suicide risk. The exclusion criteria were (1) a lack of baseline assessment for any reason and (2) refusal to follow-up at the end of the index call.

The primary outcome of this study was subsequent suicidal act including suicide death or suicide attempt. The assumed primary risk factor was low hopefulness at the beginning of the index call. Previous studies have indicated that approximately 10% of high-risk AYA callers exhibit subsequent suicidal acts [18, 21]. The estimated relative risk of low hopefulness for subsequent suicidal acts was 2 [26]. A sample size estimation formula for cohort studies was used [29]. Considering a power of 0.9, α = 0.05, and a two-sided test, the required sample size was 494. Considering the potential loss of 20% of participants during follow-up, the expected sample size was 593.

Baseline and follow-up data were collected by qualified psychological counsellors who served as operators at the Beijing Psychological Support Hotline. The counsellors underwent rigorous and systematic training provided by the hotline [30].

The hotline’s “call-on” system [18] was developed to assist operators with taking calls and data collection and storage. While receiving a call, questions or prompts containing one or more blank spaces are automatically presented on a screen in front of the operator. The operator fills in the blanks by gathering information from the caller. The collected data is automatically saved in a predefined format. The baseline measurement included several factors, such as demographic characteristics, hopefulness, psychological distress, suicide intent, depression, suicide attempt history, relatives or acquaintances’ history of suicidal acts, alcohol or substance misuse, acute life events, chronic life events, history of being abused, fear of being attacked, and physical illness.

To assess hopefulness, psychological distress, and suicide intent, callers were asked to rate their experiences on a scale of 0 to 100 at the beginning and end of the index call. A score of 0 indicated no hope, distress, or suicide intent, whereas a score of 100 represented the maximum level of hope, extreme distress, or suicide intent. To assess depression, a structured depression screening questionnaire was adapted from a previous study on suicidality for use on hotlines [31]. The questionnaire evaluated the caller’s depressive symptoms and duration. The system automatically calculated a score to evaluate the caller’s depressive mood within the past two weeks. The total score ranged from 0 to 100, with higher scores indicating more severe depression. Relatives or acquaintances’ history of suicidal acts referred to whether any blood relatives, non-blood relatives, or friends of the caller had a history of suicide or suicide attempts. When enquiring about alcohol and substance misuse, operators asked callers if they had experienced repeated episodes of excessive alcohol consumption in the past year and if they had been using hypnotics, anti-anxiety drugs, anaesthetics, stimulants, or any drugs excessively, casually, or continuously for more than three months in the past year. Alcohol or substance misuse was determined if the answer to any of these questions was ‘yes’ and whether this had a moderate or significant impact on their daily life, social interactions, or workability in the past month. Callers’ experiences of chronic and acute life events, history of abuse, and severe physical illness were recorded if these factors had moderate, severe, or significant impacts. Fear of being attacked referred to callers frequently worrying about being attacked by others during the past month.

The short-term effects of the hotline intervention were assessed by measuring changes in hopefulness, psychological distress, and suicide intent during the index call. To calculate these changes, the scores reported by callers at the end of the call were subtracted from those reported at the beginning of the call. An increased score for hopefulness and decreased scores for psychological distress and suicide intent indicated improvements in these aspects.

The primary outcome was the occurrence of suicidal acts, including both suicide deaths and attempts, during the follow-up period. The occurrence of subsequent suicidal acts was assessed at each follow-up call. Information on death by suicide was obtained from the family members of deceased callers, and data on suicide attempts were provided by the callers themselves. Operators enquired whether the callers had any suicidal acts since the index call or last follow-up. In cases where callers could not be reached, family members who answered the follow-up call were asked if the caller had died by suicide.

Hotline operators informed high-risk callers of follow-up arrangements before ending the call. Operators clearly explained the purpose of the follow-up and informed them of the specific follow-up time to minimise the loss of follow-up. In cases where the caller could not be reached at the agreed-upon time, the operators were required to make three consecutive calls at different times on the same day. Callers were considered ‘lost’ after three unsuccessful calls, with the follow-up process continuing to the next scheduled follow-up. Follow-ups were conducted one day, one week, one month, three months, six months, and 12 months after the index call, totalling six follow-ups via telephone.

The study endpoint was defined as the date of the first suicidal behaviour, including suicide attempts and death by suicide. The censoring dates were determined based on the date of last contact and interview, death from causes other than suicide, or 12 months after the index call, whichever occurred first. The entry date was defined as the date of the index call.

Continuous variables, such as hopefulness, psychological distress, suicide intent, and depression, were transformed into categorical variables based on their medians. Age was categorised into two groups: adolescents (aged 12–17) and young adults (aged 18–25).

Baseline characteristics (i.e. sex, age, hopefulness, psychological distress, and suicide intent) were compared between callers who did and did not follow up using χ2 tests. Kaplan–Meier survival curves and log-rank tests were used to investigate the potential suicide risk factors collected in the baseline assessment. Survival analysis was conducted using the Cox proportional hazards model under the assumption of proportional hazards to identify risk factors associated with suicidal acts during follow-up. The variables incorporated into the Cox regression analysis were those that exhibited p-values less than 0.30 [32] in the log-rank tests. A stratified Cox regression analysis was performed if the data did not meet the proportional hazards assumption. P-values < 0.05 were considered statistically significant.

Two Cox regression models were used. This population exhibited a high collinearity between marital status, education level, and age. Therefore, in the Cox regression, we adjusted for age and gender, without including marital status and education, to ensure stability and reliability. In Model 1, we adjusted for sex and age; alcohol or substance misuse, acute life events, severe physical illness, severe depression, suicide attempt history, low hopefulness, and high suicide intent at the beginning of the index call were included and screened. In Model 2, we adjusted for gender and age; low hopefulness and high suicide intent at the beginning of the index call were replaced with improvements in hopefulness, psychological distress, and suicide intent. The remaining variables in Model 1 were retained and screened in Model 2. The callers were de-identified before data analysis. Data analyses were performed using SPSS 18.0.