Discounting: A practical guide to multilevel analysis of choice data.

Background: Most of the neonatal deaths that occur after 6-48 hours postpartum can be prevented with appropriate newborn care and initiated immediately after delivery through adequate and standardized first neonatal visit. In Indonesia, it is still not in accordance with the expected target. This study aimed to analyze the factors that influence the first neonatal visit from various levels through a multilevel analysis approach. Subjects and Method: This was a cross sectional study with samples of individual level (level 1) was 1014 mother babies, village Level (level 2) as many as 95 village managers, public health center (level 3) was 51 manager of child health programs, and district level (level 4) was 13 managers of health programs in eight provinces Indonesia. The dependent variable was first neonatal visits. The independent variables were birth attendant and pregnancy counselling (level 1), number of village midwives and community control in development (level 2), partnership and resources (level 3), and policy strategy (level 4). Data analysis was performed through univariate, bivariate, multivariate analysis with logistic regression and multilevel modeling using multilevel regression logistic random intercept analysis run on Stata 14.0. Results: At level 1, birth attendants increased first neonatal visits (OR = 3.21; 95% CI 95% = 1,984 to 5,182; p <0.001), pregnancy counselling (OR = 1,705; 95% CI 95% = 1,162 to 2,503; p = 0.007) significantly. At level 2 modelling, the number of village midwives increased the risk of the first neonatal visit (OR = 1,815; 95% CI = 0.950 to 3,467; p = 0.049), community control (OR = 2,659; 95% CI = 1.396 to 5.066; p = 0.009) significantly. At level 3 modelling, partnerships and resources significantly increased the first neonatal visit (OR = 2,131; 95% CI = 1,114 to 4,078; p = 0.012) significantly. At level 4 modelling, birth attendants significantly increased the first neonatal visits (OR = 3.056; 95% CI = 1.901 to 4,914; p = 0.029) significantly. Conclusion: Birth attendants, pregnancy counselling, the number of village midwives, community control, partnerships and resources, birth attendants increase the risk of first neonatal visit. Contextual village areas, public health center, and district health offices have contextual effects on the first neonatal visit. Keywords: first neonatal visit, multilevel analysis Correspondence: Triseu Setianingsih. Drg. Suherman Medical Institut. Jl. Industri Pasirgombong Jababeka Cikarang, Bekasi, West Java. Email: triseu.pantyarso@imds.ac.id. Mobile: +6281299192199. DOI: https://doi.org/10.26911/the7thicph.03.85

IntroductionUnmet need for family planning is the main obstacle to achieve healthy timing and desired number of children. Decreasing the unmet need for FP respects and protects human right and help to decrease the influence on biodiversity. Unmet need for family planning is the contributor and devastating issue of maternal health. Therefore, meeting the unmet need of contraceptive averts the maternal death and poverty. Therefore, determining the magnitude and its determinants is very important to intervene and design appropriate program umbrella.ObjectiveTo determine the magnitude and its determinants of unmet need for family planning among reproductive age women in East Africa.MethodThis study was analyzed secondary data from Demographic and Health Surveys (DHS) of which contained detailed family planning for all interviewed women aged 15 to 49 years. The data were weighted using sampling weight before any statistical analysis to account the sampling design. STATA version 15 was used for extracting, editing, recoding, and multilevel analysis. Median odds ratio (MOR), proportional change in Variance (PCV), Intraclass correlation coefficient (ICC), and Akaike Information Criteria (AIC) was analyzed. Four model was build and the best model was selected based on the smallest Akaike Information Criteria (AIC). Both bivariable and multivariable multilevel analysis was done. Variable with p-value< 0.25 were selected for multivariable multilevel logistic regression analysis. Variables with p-value ≤5% declared as statistical significant with outcome variable.ResultsThe magnitude of unmet need for family planning was 24.66% (95%CI: 24.1–25.2). The identified determinants of unmet need for family planning was 30–39 years (AOR = 0.7; 95% CI 0.54–0.91), age of 40–49 (AOR = 0.76; 95% CI 0.58–0.99), rural residence (AOR = 1.17; 95% CI 1.02–1.34), female household head (AOR = 0.66; 95% CI 0.61–0.73), women having 4–6 child (AOR = 1.76; 95% CI 1.55–1.99), women having 7–9 child (AOR = 2.77; 95% CI 2.34–3.28) women having ≥10 child (AOR = 3.51; 95% CI 2.58–4.78), women who give their first birth 19-25 years (AOR = 1.1; 95% CI 1.0–1.26), 26–34 years (AOR = 1.4; 95% CI 1.19–1.83) ≥35 years (AOR = 2.1; 95% CI 1.1–4.27) and no fertility desire (AOR = 1.52; 95% CI 1.36–1.67) were the determinants of unmet need for family planning in east Africa.ConclusionUnmet need in east Africa is high as compare to other previous study. Maternal age, residence, sex of household head, number of children, age at first birth and fertility desire were the determinants identified in this study. Therefore, health interventions that reduce unmet need which enhance family planning service delivery among rural, male-headed household, women having more than three children and women who had no fertility desire needed in advance. Policies and programs of unmet need should be tailored the rural, young and no fertility desire women as well as male headed households.

The Somatic Marker Hypothesis, an influential neurobiological account of decision-making, states that emotional somatic markers (e.g., skin conductance responses) influence decision-making processes. Despite its prominence, the hypothesis remains controversial partly because of inconsistent results stemming from inappropriate statistical methods. Tasks designed to assess decision-making often use repeated measures designs, such as the Iowa Gambling Task (IGT), which requires participants to maximize profits by selecting 100 cards among four decks offering varying win-loss contingencies. Researchers often aggregate repeated measures into a single averaged value to simplify analyses, potentially committing an ecological fallacy by erroneously generalizing results obtained from aggregated data (i.e., interindividual effects) to individual repeated measurements (i.e., intraindividual effects). This paper addresses this issue by demonstrating how to analyze concurrent repeated measures of both independent and dependent variables using multilevel logistic models. First, the principles of logistic multilevel models are explained. Then, simulated and empirical IGT data are analyzed to compare the performance of traditional statistical approaches (i.e., general linear models) with multilevel logistic models. Our proposed multilevel logistic analyses address critical methodological gaps in decision-making research, ensuring more accurate interpretations of repeated measures data. This approach not only advances the study of the Somatic Marker Hypothesis but also provides a robust framework for similar research protocols, ultimately enhancing the reliability and validity of findings.

BackgroundHuman papillomavirus (HPV) is one of the sexually transmitted diseases infections that causes cervical cancer, and it is the second-leading cause of infection-related cancer globally. HPV infection causes around 604,000 cervical cancer cases (342,000 deaths) globally each year. Therefore, this study aimed to assess Human Papillomavirus Vaccination uptake and associated factors among schools girls in Ethiopia.MethodPerformance Monitoring for Action Ethiopia (PMA Ethiopia) is a survey project designed to generate data on various reproductive, maternal, and newborn health (RMNH) indicators that can inform national and regional governments. The prevalence of HPV vaccine uptake with a 95% Confidence Interval (CI) was reported and presented in a forest plot for East Africa Countries using STATA version 14.1. Intra-class Correlation Coefficient (ICC), Likelihood Ratio (LR) test, Median Odds Ratio (MOR), and deviance (−2LLR) values were used for model comparison and fitness. Adjusted Odds Ratios (AOR) with a 95% Confidence Interval (CI) and p-value ≤0.05 in the multilevel logistic model were used to declare significant factors associated with HPV vaccine uptake.ResultIn Ethiopia, the prevalence of HPV vaccine uptake among schools girls was 30.82% (95% CI: 29.21, 32.45). In the multilevel logistic regression model, girls in age groups of 12–14 years were 2.44 [AOR = 2.44, 95% CI: 1.86–3.16] times more likely to take HPV vaccine as compared to girls aged 9–11 years. Similarly, girls who had received any health service and received sexual and reproductive health services had 7.75 [AOR = 7.75, 95% CI: 5.65–10.62], and 3.24 [AOR = 3.24, 95% CI: 2.33–4.51] were more likely to take HPV vaccine compared to their counterparts respectively.ConclusionThe study findings indicate that the proportion of girls reporting receipt of the HPV vaccine in this nationally representative survey is an alarmingly low 30.8%. The following critical factors have influenced this rate: age, access to sexual and reproductive health services, general health service utilization, and regional health disparity.

BackgroundIn 2010, more than six million children in sub-Saharan Africa did not receive the full series of three doses of the diphtheria-tetanus-pertussis vaccine by one year of age. An evidence-based approach to addressing this burden of un-immunised children requires accurate knowledge of the underlying factors. We therefore developed and tested a model of childhood immunisation that includes individual, community and country-level characteristics.Method and FindingsWe conducted multilevel logistic regression analysis of Demographic and Health Survey data for 27,094 children aged 12–23 months, nested within 8,546 communities from 24 countries in sub-Saharan Africa. According to the intra-country and intra-community correlation coefficient implied by the estimated intercept component variance, 21% and 32% of the variance in unimmunised children were attributable to country- and community-level factors respectively. Children born to mothers (OR 1.35, 95%CI 1.18 to 1.53) and fathers (OR 1.13, 95%CI 1.12 to 1.40) with no formal education were more likely to be unimmunised than those born to parents with secondary or higher education. Children from the poorest households were 36% more likely to be unimmunised than counterparts from the richest households. Maternal access to media significantly reduced the odds of children being unimmunised (OR 0.94, 95%CI 0.94 to 0.99). Mothers with health seeking behaviours were less likely to have unimmunised children (OR 0.56, 95%CI 0.54 to 0.58). However, children from urban areas (OR 1.12, 95% CI 1.01 to 1.23), communities with high illiteracy rates (OR 1.13, 95% CI 1.05 to 1.23), and countries with high fertility rates (OR 4.43, 95% CI 1.04 to 18.92) were more likely to be unimmunised.ConclusionWe found that individual and contextual factors were associated with childhood immunisation, suggesting that public health programmes designed to improve coverage of childhood immunisation should address people, and the communities and societies in which they live.

Does neighborhood violence lead to depression among caregivers of children with asthma?

Recent advances in statistical software1 have enabled public health researchers to fit multilevel models to a variety of outcome variables. Multilevel models facilitate inferences regarding unexplained variability among randomly sampled clusters of units (e.g., hospitals) in outcomes of interest and identify covariates that explain the variance in a given outcome at each level of a particular data hierarchy (e.g., patients within hospitals).2,3 Models with random intercepts enable researchers to accommodate correlations within higher-level units resulting from longitudinal or clustered study designs, and models with random coefficients enable researchers to identify higher-level covariates that explain between-cluster variance in relationships of interest.2,3 Public-use survey data sets collected from large national samples, such as the National Health and Nutrition Examination Survey, also have become widely available.4 The samples underlying these data sets are often “complex” in nature for 2 reasons: (1) the use of stratified multistage cluster sampling to increase sampling and cost efficiency and (2) unequal probabilities of selection from target populations for sampled elements, often as a result of oversampling of key subgroups (leading to the need to use weights for generating unbiased population estimates). Secondary analysts can accommodate these design complexities statistically by using “design-based” analyses, which ensure that population inferences are unbiased with respect to the sample design.4 However, these design-based approaches generally do not enable the types of cluster-specific inferences afforded by multilevel models,2,3 and researchers are now considering multilevel models for complex sample survey data. Multilevel modeling represents a “model-based” approach to survey data analysis, in which dependencies in the data introduced by complex sampling features are generally accounted for by sound specification of the underlying probability model.5,6 Advocates of this approach argue that any information contained in the sample design features should be accounted for in the model specification, making the sampling uninformative.5 However, analysts may not have access to covariates capturing all of this information. In this case, the use of weighted estimation when fitting multilevel models provides some protection against potential biases introduced by informative sampling.6 Informed by recent methodological and computational developments in this area,1–3,6,7 we show that changes in inferences are possible when fitting multilevel models to complex sample survey data and ignoring the sampling weights. We analyzed data from the 2013 Medical Monitoring Project HIV Provider Survey, sponsored by the Centers for Disease Control and Prevention, for which a probability sample of HIV care providers was selected from outpatient HIV care facilities in 16 states and Puerto Rico.8,9 Briefly, the provider survey followed a 2-stage probability-proportionate-to-size sample design, first sampling states and territories and then HIV facilities and selecting all providers within a facility. Unbiased estimation of multilevel model parameters requires the use of weights at all levels of a given data hierarchy,7 so we used previously calculated sampling weights adjusted for nonresponse at the facility level and inverses of estimated response probabilities at the provider level. We focus on only facilities with multiple responding providers and include covariates that are both theoretically relevant for the dependent variables described later in this article and related to the sampling weights (e.g., an indicator of the provider serving more than 200 patients). Details about computation of the Medical Monitoring Project sampling weights for both providers and facilities are available on request.10 We scaled the final provider-level weights to sum to the sample sizes within each facility. A failure to do this would overstate actual sample sizes within each higher-level unit (facility), possibly resulting in biased estimates of model parameters.2,3,7 We fit multilevel logistic regression models to 2 binary dependent variables, indicating whether the responding provider delivered adequate drug use risk reduction and sexual risk reduction services to patients (defined as delivering approximately 70% of recommended risk reduction services to most or all of the patients). The models included random intercepts to capture between-facility variation in each proportion, in addition to fixed effects of several provider- and facility-level covariates of interest. We fit these models with the new GLIMMIX command11 in SAS/STAT version 13.1 (SAS Institute, Cary, NC), which can fit multilevel models to complex sample survey data. Identical results can be obtained with the new svy: melogit command in Stata version 14 (StataCorp LP, College Station, TX). We did not test whether the parameter differences in the weighted and unweighted models were significant,12 but we did observe several shifts in inference when using weighted estimation (Table A; available as a supplement to the online version of this article at http://www.ajph.org). In both models, the intercept became more negative and significant, suggesting that the probability of using adequate risk reduction was being overstated for the type of provider represented by zeroes on all of the covariates (which may not be entirely meaningful in all models). For drug risk reduction, the coefficient for delivering care in a language other than English became nonsignificant. For the sexual risk reduction outcome, the male provider coefficient became significant, and the Black provider, nurse practitioner, and integrated team effects became even stronger. Finally, the estimated variability of the random facility intercepts was clearly being overstated when ignoring the weights, and the weighted models explained more of the variance in the outcomes at each level. The weights at each level were clearly informative about the parameters defining these models, and ignoring them in analysis would have led to erroneous inferences with respect to the sample design used. Notably, these results held despite the inclusion of available covariates related to the sampling weights in the models. In practice, covariates used to compute the weights or the weights at each level of the data hierarchy may not be available to the public, making appropriate design-adjusted estimation of multilevel models difficult or impossible. We encourage analysts fitting multilevel models to survey data to carefully examine the variables available for weighted estimation in these data sets, make use of the powerful software1–3,11 that has been developed in this area, and (when possible) examine whether weighted estimation or adjustment for covariates related to the weights affects their inferences.

BackgroundA miscarriage or a spontaneous loss of a pregnancy that occurs before the 20th week is an abortion. Even though numerous recommendations state that pregnant women should abstain from alcohol at all stages of pregnancy, alcohol intake among pregnant women is common. However, there are few papers addressing the effect of alcohol use on miscarriage using nationally representative data. Moreover, the association of alcohol use with abortion and its mechanisms is not well studied in the Ethiopian region. Therefore, the objective of the current study was to estimate the association of alcohol use with abortion rates among reproductive age (15–49) women in Ethiopia.MethodsUsing the most recent findings of the Ethiopian Demographic and Health Survey (EDHS), secondary data analysis was performed among pregnant women in Ethiopia. A total of 11,396 women between the ages of 15 and 49 years who were of reproductive age were included in the research. To characterize the study population, descriptive statistics were used. The variability was considered using the multilevel binary logistic regression model. A multilevel binary logistic model was used to determine the effect of alcohol intake on abortion while controlling for potential confounders. In the multivariable analysis, variables with a P-value of less than 0.05 were considered statistically significant for the response variable.ResultsThe proportion of women who had an abortion was 10.46% with a 95% CI of 9.92–11.03. In the final model of the multilevel analysis, age group [adjusted odds ratio (AOR) = 6.13; 95% CI: 3.86–9.73], education level (AOR = 1.29; 95 and CI: 1.10–1.51), alcohol consumption (AOR = 1.38; 95% CI: 1.18–1.61), age at first sex (AOR = 1.20; 95% CI: 1.03–1.39), media exposure (AOR = 1.28, CI: 1.10–1.48), contraceptive use (AOR = 1.34, CI: 1.16–1.56), and occupation of respondent (AOR = 1.21, CI: 1.06–1.38) were identified to be significant determinants of abortion in Ethiopia.ConclusionSexual and reproductive health education and family planning programs should target older women in the reproductive age group, women with primary educational status, working women, and those who initiated sexual intercourse at a younger age considering it could reduce abortion and unintended pregnancy. Furthermore, as part of sexual and reproductive health education, the adverse effect of alcohol consumption on abortion should be emphasized.

Multilevel logistic regression models are increasingly being used to analyze clustered data in medical, public health, epidemiological, and educational research. Procedures for estimating the parameters of such models are available in many statistical software packages. There is currently little evidence on the minimum number of clusters necessary to reliably fit multilevel regression models. We conducted a Monte Carlo study to compare the performance of different statistical software procedures for estimating multilevel logistic regression models when the number of clusters was low. We examined procedures available in BUGS, HLM, R, SAS, and Stata. We found that there were qualitative differences in the performance of different software procedures for estimating multilevel logistic models when the number of clusters was low. Among the likelihood-based procedures, estimation methods based on adaptive Gauss-Hermite approximations to the likelihood (glmer in R and xtlogit in Stata) or adaptive Gaussian quadrature (Proc NLMIXED in SAS) tended to have superior performance for estimating variance components when the number of clusters was small, compared to software procedures based on penalized quasi-likelihood. However, only Bayesian estimation with BUGS allowed for accurate estimation of variance components when there were fewer than 10 clusters. For all statistical software procedures, estimation of variance components tended to be poor when there were only five subjects per cluster, regardless of the number of clusters.

A multilevel logistic regression model for identifying the relevance of environmental risk factors on Gestational Diabetes Mellitus

Traditional statistical analyses can be compromised when data are collected from groups or multiple observations are collected from individuals. We present an introduction to multilevel models designed to address dependency in data. We review current use of multilevel modeling in 3 personality journals showing use concentrated in the 2 areas of experience sampling and longitudinal growth. Using an empirical example, we illustrate specification and interpretation of the results of series of models as predictor variables are introduced at Levels 1 and 2. Attention is given to possible trends and cycles in longitudinal data and to different forms of centering. We consider issues that may arise in estimation, model comparison, model evaluation, and data evaluation (outliers), highlighting similarities to and differences from standard regression approaches. Finally, we consider newer developments, including 3-level models, cross-classified models, nonstandard (limited) dependent variables, multilevel structural equation modeling, and nonlinear growth. Multilevel approaches both address traditional problems of dependency in data and provide personality researchers with the opportunity to ask new questions of their data.

Trade of animals and animal products imposes an uncertain and variable risk for exotic animal diseases introduction into importing countries. Risk analysis provides importing countries with an objective, transparent, and internationally accepted method for assessing that risk. Over the last decades, European Union countries have conducted probabilistic risk assessments quite frequently to quantify the risk for rare animal diseases introduction into their territories. Most probabilistic animal health risk assessments have been typically classified into one-level and multilevel binomial models. One-level models are more simple than multilevel models because they assume that animals or products originate from one single population. However, it is unknown whether such simplification may result in substantially different results compared to those obtained through the use of multilevel models. Here, data used on a probabilistic multilevel binomial model formulated to assess the risk for highly pathogenic avian influenza introduction into Spain were reanalyzed using a one-level binomial model and their outcomes were compared. An alternative ordinal model is also proposed here, which makes use of simpler assumptions and less information compared to those required by traditional one-level and multilevel approaches. Results suggest that, at least under certain circumstances, results of the one-level and ordinal approaches are similar to those obtained using multilevel models. Consequently, we argue that, when data are insufficient to run traditional probabilistic models, the ordinal approach presented here may be a suitable alternative to rank exporting countries in terms of the risk that they impose for the spread of rare animal diseases into disease-free countries.

tional stress researchers may investigate individual factors related to stress and well-being (e.g. personality, work-life balance, emotions, coping skills, etc.); job and organizational factors (e.g., work hours and schedules, role overload/underload/ambiguity, emotional labour, job insecurity, organizational climate); and even societal and national factors (e.g. cultural values; social, economic and environmental indicators). Yet, all too often, these variables are all measured at a single level, usually at the individual level. Thus, while we acknowledge that individual, group, organizational and even cultural variables are theoretically necessary to understand and predict stress outcomes, we rarely measure these variables at their appropriate level. The purpose of this editorial is to encourage researchers to take a multilevel approach to considering the relationships between occupational stress and wellbeing. This entails intentionally refl ecting on the individual, job, organizational and macroeconomic antecedents of employee stress. Such a multilevel framework can be fruitful not only for understanding the stress-strain process, but also crucial in developing and conceptualizing stress interventions. Thus, a multilevel approach can enhance both our theoretical and empirical understanding of occupational stress as well as the practical implications stemming from our research. Using a multi-level modelling (MLM) approach means explicitly taking into account the often hierarchical nature of our data structure (e.g. individuals working in groups, students in classrooms, children in families). Such a hierarchical dataset has traditionally posed three problems within an ordinary least squares (OLS) regression approach: (1) the units of analysis problem; (2) the violation of independent observations assumption problem; and (3) the heterogeneity of slopes problem. Let us consider each in turn. Most behavioural science models of stress acknowledge that a complete understanding of the structure and process of stress invariably requires a consideration of individual, group and environmental/situational characteristics (Sulsky & Smith, 2005). Such models originally arose from an awareness that purely biological models of stress fail to reliably predict which potential stressors will be perceived as stressful and how this process is infl uenced by personal and contextual factors. For example, the cognitive-transactional model of stress (Lazarus & Launier, 1978) posits that there are few, if any, universal stressors. Rather, the perception of and reactions to a potential stressor will vary within and across individuals and may differ over occasions and time. Whether the potential stressor is perceived to be stressful, therefore, intimately depends on characteristics of the individual (e.g. needs, abilities, personality) and the situation (e.g. demands, resources). As a result, most contemporary models of stress include some variation of a person-environment (P-E) fi t perspective [e.g. Karasek’s (1979) demands-control model; Hobfall’s (1989) Conservation of Resources theory; French and colleagues’ P-E fi t model, (French, Caplan, & Van Harrison, 1982, Van Harrison, 1978)]. Despite the fact that our models of stress explicitly consider individual and contextual factors, researchers often do not appropriately model these environmental and individual factors either in their research design, data collection or subsequent analyses. Thus, organiza-

Anemia is a major global public health problem that had tremendous impacts on human health, social and economic development. African countries contribute to the highest-burden of anemia among women, particularly in adolescent females and young women. Anemia among young women remains a public health problem in most parts of Africa, including Ethiopia. Therefore, this study aimed to investigate the prevalence and determinants of anemia among young women in Ethiopia. A secondary data analysis was conducted based on the 2016 Ethiopian Demographic and Health Survey (EDHS) data. A total weighted sample of 5796 young women (15-24 years) was included in this study. We employed a multilevel analysis to assess factors associated with anemia since the EDHS has hierarchical nature. Deviance, Intraclass Correlation Coefficient (ICC) and Median Odds Ratio (MOR) were used for model comparison as well as for assessing model fitness. Variables with a p-value of less than 0.20 at bivariable multilevel analysis were considered for the multivariable multilevel analysis. In the multivariable multilevel analysis variables with p-value ≤ 0.05 were declared to be a significant factor associated with anemia, and the Adjusted Odds Ratio (AOR) with the 95% Confidence Interval (CI) were reported to assess the strength and direction of the association. The overall prevalence of anemia among young women was 21.7% (95%CI = 20.7%, 22.8%). In the multivariable multilevel binary logistic regression analysis; being Muslim religion follower [adjusted odds ratio (AOR) = 1.31, 95%CI = 1.07, 1.70] and being protestant religion follower [AOR = 1.31; 95%CI = 1.01, 1.71], being rural dweller [AOR = 1.34; 95%CI = 1.02, 1.78], and being married [AOR = 1.46; 95%CI = 1.22, 1.74] were significantly associated with higher odds of anemia among young women. While, modern contraceptive use (AOR = 0.66; 95%CI = 0.53, 0.83) were significantly associated with lower odds of anemia among young women. In this study, the prevalence of anemia among young women was high. Being a follower of Muslim and protestant religions, being married women, modern contraceptive use and being from the rural area were found to be significant determinants of anemia among young women. Therefore, giving special attention to these high-risk groups and distributing modern contraceptives for those in need of it could decrease this devastating public health problem in young women.

BackgroundSelf-harm and suicidality are major public health concerns and are notably higher among the Indigenous population in Australia. We aimed to identify factors protecting Indigenous adolescents from self-harm and/or suicidality...