Case.

Individuals aged 18 years and over admitted to THP for medical management of non-fatal self-poisoning, during July-December 2018, were invited to participate in the study. Patients who were judged by the research team to be physically unable or too unwell to participate prior to discharge from the hospital (those who were unconscious or had a lowered level of consciousness at the time of transfer to another hospital), and those who had previously received a diagnosis from a clinician as being cognitively impaired, were excluded from the study. In cases of uncertainty, the data collectors discussed the issue of study inclusion with the investigators (PB or TR) who referred to the inclusion/exclusion protocol in the decision-making process [15]. Cases of self-poisoning were initially identified from the patient admission record and verbally reconfirmed by the patient through self-report. The degree of suicidal intent was not recorded as part of the interview schedule due to time constraints.

Controls.

Hospital based controls. Sex and age frequency matched controls were recruited from outpatients and accompanying visitors presenting to the THP outpatient department (OPD) and specialist clinics in the same premises between July 2018 to January 2019. Patients present to the OPD for a variety of medical and surgical complaints, such as hypertension, chest infections or cough, which are largely unrelated to the outcome of interest.

Controls from the community. The potential for selection bias within the hospital-based control group was identified as a limitation in the study protocol [15]. To address this, an additional control group was recruited from the community. Controls were recruited from the general population corresponding to the population catchment of cases between January 2019 to April 2019. Two administrative divisions where most cases resided were selected for sampling. Within each of the administrative divisions, all Grama Niladhari (GN) sub-divisions (‘villages’) were stratified by sex, age and ethnicity using population census data, and a list of GN subdivisions that were most demographically similar to the wider administrative division was created. All households within the 12 selected GN subdivisions were initially included in the study, but due to difficulties in access, such as hilly terrain, all households within the each subdivision could not be reached. Within each recruited household, the resident who most closely matched the cases according to age (plus or minus five years) and sex, was invited to participate in the interview.

As with cases, individuals who were judged by the research team to be physically unable or too unwell to participate in the study, or those who had previously been diagnosed as being cognitively impaired were excluded as either hospital or community controls. Controls with a self-reported previous self-harm episode were excluded in the analysis.

Procedure.

All participants took part in an interview, lasting 30–40 minutes. The interview was conducted by a trained interviewer, and was made up of a structured pretested questionnaire in the local language (Sinhala, Tamil or English), and several validated tools (described below). The trained interviewers were recently qualified graduates, with either a basic sciences or nursing degree. In order to reduce bias, interviewers (n = 4) were given a standard script to follow, with regular checks conducted by the supervisor (PB). Further details of the data collection can be found in the protocol paper but are outlined here in brief [15].

Adverse childhood experience (ACE- the main exposure) was measured using the World Health Organisation’s Childhood Experiences International Questionnaire (ACE-IQ) [16–18]. This scale is intended to measure ACEs in all countries and is designed for administration to people aged 18 years and older. Questions cover family dysfunction; physical, sexual and emotional abuse and neglect by parents or caregivers; peer violence; witnessing community violence, and exposure to collective violence. The scale was translated into the local languages (Sinhala and Tamil), and back translated to English, by science or nursing graduates. The scale was then piloted among a small group of outpatients at Teaching Hospital Peradeniya. Any difficulties noted in using the translated scale were discussed by the research team (including PB and TR) and minor adaptations were made where needed. In this study the Cronbach’s alpha for the ACE-IQ was 0.71, indicating acceptable internal reliability.

We also collected data on known confounding factors (age, sex, religion and ethnicity) and a range of potentially confounding/mediating variables. These included the following information, collected via an interviewer administered questionnaire: knowing someone who had previously self-harmed; having a young child (≤11 year old); family size; marital status; occupation; education; asset (vehicle) ownership; and being chronically ill. In addition, we collected data on current depression, using the locally validated Patient Health Questionaire-9 (PHQ9)); the PHQ-9 is an internationally validated, 10-item scale for the screening of depression, which has been translated and validated for use in Sinhala, with a sensitivity and specificity of 0.75 and 0.97 respectively (at a cut off of ≥10) [19, 20]. We screened for alcohol use disorders (using the locally validated Alcohol Use Disorders Identification Test (AUDIT)) and social capital (using questions from a large social capital community survey in Sri Lanka) [21, 22]. The AUDIT is a ten-item questionnaire which has been validated for use in Sinhala in Sri Lanka, and the area under the ROC curve to differentiate hazardous use and alcohol use disorder from low risk drinking is 0.96 [21]. The questions on social capital asked about the individual’s current interpersonal relationships with, and sense of support from, their immediate household and neighbourhood. We explored exposure to domestic violence in the previous 12-months using the four-item Humiliation, Afraid, Rape, Kick (HARK) questionnaire; it explores experience of four types of abuse–physical, sexual, emotional abuse and fear of an intimate partner. For the purpose of this study, the questionnaire was broadened to include abuse by any household member, not just an intimate partner. At an optimal cut-off mark of ≥1, the HARK has a sensitivity of 81% and specificity of 95% [23]. This tool was translated, back translated and piloted in the same manner as described for WHO Childhood Adversity Scale above. The HARK did not require modification after piloting.

Ethics.

Ethical approval for the study was granted by the Ethical Review Committee of the Faculty of Medicine, University of Peradeniya, Sri Lanka. All data collectors received detailed training in the ethical aspects of the study. All participants were given a verbal explanation of the study with a written information sheet, with a clear explanation that declining to participate would have no impact on medical care. All participants gave written informed consent. All interviews were conducted confidentially in a separate area or room. All data were stored confidentially and in a de-identified manner. Full details of the ethical and safeguarding protocols related to this study have been previously published [15].

Analysis.

The primary analysis included cases and hospital controls with no missing data. Hospital controls were considered in preference to community controls for purpose of analysis, because the hospital controls more closely matched the residential areas of all cases. Control participants who reported a history of self-harm were excluded (n = 26, 5% of hospital controls).

Following the WHO guidance for calculating an ACE score, we calculated two scores [24]. Each question in the ACE-IQ assesses an ACE. The frequency method counted an individual as having been exposed to a certain ACE if they met a certain frequency threshold (e.g. experiencing the adversity “many times”) whereas the binary method counted individuals regardless of severity (i.e., a simple presence or absence of each ACE). The ACE IQ records frequency of each ACE on a 4-point scale: never; once; a few times; many times. For our frequency analysis we summed the number of ACEs each individual had experienced “many times” (giving a score 0–13). For the binary analysis, participants score 1 if they have experienced each ACE (regardless of frequency), and the total again ranges from 0–13. We used the frequency score in the primary analysis and present the binary score (based on the sum of exposure to each ACE, regardless of frequency) as supplementary analysis. In addition to the continuous score, we also dichotomised the frequency score to indicate presence (continuous score>0) or absence (continuous score = 0) of exposure. We extended this to further explore associations by the type of ACE.

The ACE-IQ questions have been grouped into 13 categories of ACEs. We therefore created 13 categorical variables to indicate presence of one of the ACE sub-categories based on the frequency scoring method: emotional abuse; physical abuse; sexual abuse; violence against household members; living with household members who were substance abusers; living with household members who were mentally ill or suicidal; living with household members who were imprisoned; parental death, separation or divorce; emotional neglect; physical neglect; bullying; community violence; collective violence.

Individuals were categorised as experiencing domestic violence if they answered positively to any of the HARK questions [23]. Based on the local validation of the PHQ-9, a score of 10 or more on the PHQ-9 was used to indicate the presence of depression, and AUDIT scores of 8 or more indicated hazardous drinking behaviour [20, 21]. A continuous score was created for social capital, with higher scores indicating lower social capital.

Descriptive statistics for the study variables were generated. All variables were coded with the hypothesised lowest-risk category as the reference group. Bivariate relationships between the study variables and the outcome of self-poisoning were assessed using logistic regression models, adjusted for age and sex. The relationship between ACE and self-poisoning was evaluated using a series of logistic regression models (Fig 1). Participant age and sex were included as covariates in all models. We explored the relationship between ACE and self-poisoning adjusting for age and sex only (model 1). We then adjusted for the key confounders of ethnicity and religion (model 2), and this model was used for our main assessment of the association between ACEs and self-poisoning. As an additional analysis we explored the relationship between ACE and self-poisoning adjusting for age, sex, ethnicity, religion and all potentially confounding/mediating variables (model 3). We present this as supplementary results as many of the variables could be mediating variables. The inclusion of mediating variables in our adjusted models can inappropriately suggest an association when one does not exist or give a biased estimate due to over adjustment. In the original protocol we indicated that childhood socioeconomic position was being considered as a confounder, but on reflection we felt that this was unlikely, and that it was more likely to capture aspects of the early environment akin to our other measures of childhood adversity. Therefore, we included childhood socioeconomic position (parental education) in model 3. We also present the confounder adjusted model (model 2) stratified by sex for the overall ACE score, and subcategories. We formally tested to see whether sex modified associations observed for the overall ACE frequency score and presence of any ACE.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Graphical presentation of the conceptual models used to inform statistical analysis.

https://doi.org/10.1371/journal.pone.0242437.g001

We also conducted a series of sensitivity analyses to explore whether our results were robust. We have indicated which of these were defined a priori. First, models were repeated using all available data (i.e., including participants with missing data–a priori analysis). Second, analyses were repeated using the ACE binary score as the exposure. Finally, we explored associations with the community control group as opposed to hospital controls.

Sensitivity analyses.

In the analysis using all available data (i.e., including all cases and controls) we found very similar associations between ACE score and self-poisoning to the initial models (data not shown). We repeated the analysis using the ACE binary score (Table 2) and whilst the associations were weaker in these models, the overall findings remain unaltered. S2 and S3 Tables show demographic differences between the hospital and community control groups and model results for the community control group analysis, which were consistent with those using the hospital control group.

Possible mechanisms.

Childhood adversity may lead to adult suicidal behaviour via multiple pathways—exposure to childhood abuse has been shown to be associated with increased risk of adult depression in HIC, which is associated with increased risk of self-harm and suicidal behaviour [36–39]. Whilst the association between depression and suicidal behaviour is less consistently observed in LMIC in the general population, it may be that individuals who experience childhood adversity are at higher risk of developing depression as adults and engaging in suicidal behaviour [40, 41]. In keeping with this, in this study we observed that cases with self-poisoning were more likely to screen positive for depression, and adjusting for current depression reduced the association observed between ACE and self-poisoning. Adversity can have physiological impacts on individuals through biological embedding, with repeated activation of the ‘fight or flight’ response causing lasting changes in hypothalamic-pituitary-adrenal (HPA) axis functioning, which in turn impacts on multiple biological systems such as inflammation [42]. This may result in altered levels of neurotransmitters such as dopamine and serotonin, implicated in the aetiology of suicidal behaviour [42].

Childhood adversities do not occur in isolation, and previous work has shown that exposure to childhood abuse often co-occurs with other indicators of household dysfunction such as domestic violence, parental unemployment, substance misuse and poverty [43, 44]. Exposure to both childhood abuse and growing up in a household with domestic violence has a ‘double whammy’ effect, in causing a greater risk of psychological problems in adulthood [43]. As adults they are also likely to be more vulnerable to psychological problems, and externalizing behaviours such as suicidal behaviours [43]. Furthermore, as adults they are more likely to have increased emotional dysregulation and poor attachment patterns, and are vulnerable to further victimization in adult relationships, all of which may contribute towards increased risk of suicidal behaviours [43].

Exposure to community or collective violence can lead directly or indirectly to self-harm and suicide, possibly through increasing mental distress, reduced social cohesion, reducing sensitivity to violence, or by increased perception that violence is a solution. Therefore self-harm may be more cognitively available to those exposed to such violence [31, 43]. Qualitative research is needed to elucidate the possible mechanisms by which community/collective violence are associated with increased self-harm.