A primer on the use of machine learning to distil knowledge from data in biological psychiatry.

Applications of machine learning in the biomedical sciences are growing rapidly. This growth has been spurred by diverse cross-institutional and interdisciplinary collaborations, public availability of large datasets, an increase in the accessibility of analytic routines, and the availability of powerful computing resources. With this increased access and exposure to machine learning comes a responsibility for education and a deeper understanding of its bases and bounds, borne equally by data scientists seeking to ply their analytic wares in medical research and by biomedical scientists seeking to harness such methods to glean knowledge from data. This article provides an accessible and critical review of machine learning for a biomedically informed audience, as well as its applications in psychiatry. The review covers definitions and expositions of commonly used machine learning methods, and historical trends of their use in psychiatry. We also provide a set of standards, namely Guidelines for REporting Machine Learning Investigations in Neuropsychiatry (GREMLIN), for designing and reporting studies that use machine learning as a primary data-analysis approach. Lastly, we propose the establishment of the Machine Learning in Psychiatry (MLPsych) Consortium, enumerate its objectives, and identify areas of opportunity for future applications of machine learning in biological psychiatry. This review serves as a cautiously optimistic primer on machine learning for those on the precipice as they prepare to dive into the field, either as methodological practitioners or well-informed consumers.

Attention-deficit hyperactivity disorder in children is related to maternal screen time during early childhood in Taiwan: a national prospective cohort study.

BACKGROUND: It is unclear to familial screen time in early childhood is associated with the subsequent diagnosis of attention-deficit and hyperactivity disorder (ADHD). Our study is to evaluate the association between screen time during early childhood in families and the incidence of ADHD. METHODS: We conducted a population-based birth cohort study by using the Taiwan Birth Cohort Study, which recruited 24 200 mother-child pairs when children were 6 months old. Screen time exposure for children and parents were collected at the age of 18 and 36 months. Whether the child has ever been diagnosed with ADHD was determined at a follow-up interview at age 8. Factors including socioeconomic factors and screen time were analyzed using logistic regression to determine their association with the rate of ADHD. RESULTS: A total of 16 651 term singletons were included in the final analysis. Of them, 382 (2.3%) were diagnosed as having ADHD before the age of 8 years. No significant relationship between children's or fathers' screen time and ADHD was noted. When compared to children whose mothers spent less time on screens, those whose mothers spent more than 3 h a day on screens when the child was 3 years old exhibited a higher incidence of ADHD (adjusted OR [aOR]: 1.31, 95% CI: 1.03-1.66). CONCLUSION: Higher maternal screen time when the child was 3 years old was associated with an increased incidence of ADHD in this population-based study. However, children's screen time did not find related to ADHD. We found that it was the mother's screen time, who typically serves as the primary caregiver in our study participants, not the child's, that mattered. In addition to superficial screen use time, future research is needed to replicate the findings and clarify mechanisms underlying this association.

Number of words at age 2.5 years is associated with intellectual functioning at age 7 years in the SELMA study.

AIM: We examined the association between the number of words used at age 2.5 years and deficits in intellectual functioning at age 7 years, in 549 children, and whether such association is confirmed by parental concern about the child's development. METHODS: Parental reports of how many words their children used at age 2.5 years were analysed for the association to intellectual functioning (assessed with Wechsler Intelligence Scale for Children - Fourth Edition,WISC-IV) at age 7 years using linear regression, adjusting for sex, maternal education level, parental IQ and smoking during pregnancy. Parental concern at age 7 years was examined with the Early Symptomatic Syndromes Eliciting Neurodevelopmental Clinical Examinations-Questionnaire (ESSENCE-Q). RESULTS: Adjusted linear regression showed that use of 50 words or fewer at age 2.5 years, relative to use of more than 50 words, was associated with lower scores of Full-scale IQ (B = 7.27, p = 0.001), verbal comprehension (B = 8.53, p < 0.001), working memory (B = 9.04, p < 0.001) and perceptual reasoning (B = 4.21, p = 0.045), in the WISC-IV, at age 7 years. Parental concern was more common in the group that used 50 words or fewer (Mann-Whitney U test, p = 0.011). CONCLUSION: This easily accessible measure of number of words seems to be a valuable marker for intellectual functioning later in life.

Maternal serum levels of perfluoroalkyl substances in early pregnancy and offspring birth weight.

BACKGROUND: Perfluoroalkyl substances (PFASs) are widespread, bioaccumulating, and persistent and show placental transfer. Emerging research indicates associations between prenatal exposure and low birth weight. The aim of this study was to assess the associations between first trimester exposure to PFASs and birth weight (BW) in the Swedish Environmental, Longitudinal, Mother and child, Asthma and allergy (SELMA) study and examine whether associations differ between girls and boys. METHODS: Eight PFASs were analyzed in maternal serum (median: 10 weeks of pregnancy). Associations between prenatal PFAS exposure and birth outcomes with BW, BW for gestational age, and birth small for gestational age (SGA) were assessed in 1533 infants, adjusted for potential confounders and stratified by sex. RESULTS: Increased maternal perfluorooctane sulfonate (PFOS), perfluorooctanoic acid (PFOA), perfluorononanoic acid (PFNA), perfluorodecanoic acid (PFDA), and perfluoroundecanoic acid (PFUnDA) were associated with lower BW, lower BW for gestational age, and SGA birth. Associations were significant only in girls, where prenatal exposure in the upper quartile was associated with a 93-142-g lower BW when compared with that of the lowest quartile exposure. The associations were not mediated by effects on gestational age. CONCLUSIONS: We found associations between prenatal exposure for five different PFASs and birth weight, with more pronounced associations in girls than in boys.

Persistent organic pollutants, pre-pregnancy use of combined oral contraceptives, age, and time-to-pregnancy in the SELMA cohort.

BACKGROUND: We are exposed to several chemicals such as persistent organic pollutants (POPs) in our everyday lives. Prior evidence has suggested that POPs may have adverse effects on reproductive function by disrupting hormone synthesis and metabolism. While there is age-related decline of fertility, the use of hormonal combined oral contraceptives (COCs) and its association to return of fertility remains controversial. The goal of this study is to investigate the association between exposure to POPs, both individually and as a mixture, and fecundability measured as time-to-pregnancy (TTP) according to pre-pregnancy use of COCs and age. METHODS: Using the SELMA (Swedish Environmental Longitudinal Mother and Child, Allergy and Asthma) study, we have identified 818 pregnant women aged 18-43 years (mean 29 years) with data on how long they tried to get pregnant and what was their most recently used contraceptive method. These data were collected at enrollment to the study (median week 10 of pregnancy). Concentrations of 22 POPs and cotinine were analyzed in the blood samples collected at the same time as the questions on TTP and pre-pregnancy use of contraceptive. Analyses were done on the association between POPs exposure and TTP measured as continuous (months) and binary (infertile for those with TTP > 12 months). To study the chemicals individually, Cox regression and logistic regression were used to estimate fecundability ratios (FRs) and odds ratios (ORs), respectively. Weighted quantile sum (WQS) regression was used to investigate the chemicals as a mixture where chemicals of concern were identified above the 7.6% threshold of equal weights. To perform the subgroup analysis, we stratified the sample according to use of COCs as the most recent pre-pregnancy contraception method and age (< 29 years, and ≥ 29 years). The models were adjusted for parity, regularity of menses, maternal body mass index (BMI) and smoking status, and stratified as described above. RESULTS: Prior to stratification, none of the POPs were associated with fecundability while increased exposure to HCB, PCB 74 and 118 had higher odds of infertility. Upon stratification, POP exposure was significantly associated with longer TTP in women aged ≥29 years who did not use COC. Specifically, PCBs 156, 180, 183, and 187 were associated with reduced fecundability while PCBs 99, 153, 156, 180, 183, and 187 had higher odds of infertility. As a mixture, we identified the chemicals of concern for a longer TTP include PCBs 118, 156, 183, and 187. Moreover, chemicals of concern identified with increased odds of infertility were PCB 74, 156, 183, 187, and transnonachlor. CONCLUSION: Serum concentrations of selected POPs, both as individual chemicals and as a mixture, were significantly associated with lower fecundability and increased odds of infertility in women aged 29 years and above not using COC as their most recent pre-pregnancy contraceptive. Our findings suggest that pre-pregnancy use of oral contraceptive and age may modify the link between POPs and fecundability. The differences of specific chemicals in the individual analysis and as a mixture support the need to study combination effects of chemicals when evaluating reproductive outcomes.

Eye Gaze Patterns Associated with Aggressive Tendencies in Adolescence.

Social information processing theory hypothesizes that aggressive children pay more attention to cues of hostility and threat in others' behavior, consequently leading to over-interpretation of others' behavior as hostile. While there is abundant evidence of aggressive children demonstrating hostile attribution biases, less well documented is whether such biases stem from over-attendance and hypersensitivity to hostile cues in social situations. Over-attendance to hostile cues would be typified by deviations at any stage of the multi-stage process of social information processing models. While deviations at later stages in social information processing models are associated with aggressive behavior in children, the initial step of encoding has historically been difficult to empirically measure, being a low level automatic process unsuitable for self-report. We employed eye-tracking methodologies to better understand the visual encoding of such social information. Eye movements of ten 13-18 year-old children referred from clinical and non-clinical populations were recorded in real time while the children viewed scenarios varying between hostile, non-hostile and ambiguous social provocation. In addition, the children completed a brief measure of risk of aggression. Aggressive children did attend more to the social scenarios with hostile cues, in particular attending longest to those hostile scenarios where the actor in the scenario had a congruent emotional response. These findings corroborate social information processing theory and the traditional bottom-up processing hypotheses that aggressive behavior relates to increased attention to hostile cues.

Nestin depletion induces melanoma matrix metalloproteinases and invasion.

Matrix metalloproteinases (MMPs) are key biological mediators of processes as diverse as wound healing, embryogenesis, and cancer progression. Although MMPs may be induced through multiple signaling pathways, the precise mechanisms for their regulation in cancer are incompletely understood. Because cytoskeletal changes are known to accompany MMP expression, we sought to examine the potential role of the poorly understood cytoskeletal protein, nestin, in modulating melanoma MMPs. Nestin knockdown (KD) upregulated the expression of specific MMPs and MMP-dependent invasion both through extracellular matrix barriers in vitro and in peritumoral connective tissue of xenografts in vivo. The development of three-dimensional melanospheres that in vitro partially recapitulate noninvasive tumorigenic melanoma growth was inhibited by nestin KD, although ECM invasion by aberrant melanospheres that did form was enhanced. Mechanistically, nestin KD-dependent melanoma invasion was associated with intracellular redistribution of phosphorylated focal adhesion kinase and increased melanoma cell responsiveness to transforming growth factor-beta, both implicated in pathways of melanoma invasion. The results suggest that the heretofore poorly understood intermediate filament, nestin, may serve as a novel mediator of MMPs critical to melanoma virulence.

The relationship between mental health reforms and general population suicide rates in Australia over the past three and a half decades: 1987-2021.

Nearly 3,000 Australians tragically end their lives by suicide each year, underscoring a major national public health challenge with substantial socio-economic ramifications. Australia's National Mental Health Plans (NMHPs) aim to improve mental health and reduce suicide rates. This study investigates their effectiveness by analyzing how age-standardized suicide rates across Australian jurisdictions have fluctuated alongside the implementation of five NMHPs from 1987 to 2021. Findings reveal mixed impacts, with some plans linked to decreases and others associated with increases in suicide rates across different periods and regions. Notably, the recent decline in 2020 requires careful consideration amidst COVID-19 pandemic influences. These insights not only provide valuable evidence for shaping future mental health policies and initiatives but also for future health services research.

Pathway from attention-deficit/hyperactivity disorder to suicide/self-harm.

To evaluate the effect of ADHD diagnosis by the age of 10 on the suicide/self-harm risk at age 14 and examine factors that may modify/mediate the association, a longitudinal study based on the nationwide survey consisting of 5,107 children in Australia was used. Self-harm and suicidal risks in children with ADHD at age 14 were the main outcomes; ADHD medication, history of bullied or depression, and other sociodemographic disadvantages, were treated as covariates. The diagnosis of ADHD at age 10 is associated with elevated risks of self-harm, suicidal thoughts, planning, or attempts by age 14, with greater susceptibility in males. The pathway from ADHD symptoms to suicide/self-harm could also be mediated by depression and exposure to bullying assessed at age 12. Note that depression and exposure to bullying might at best contribute to less than 10 % of the total effect of ADHD diagnosis on either the risk of suicide or self-harm. Early gender-tailored intervention and prevention strategies are crucial in clinical practice and health policy.

The relationship between hair cortisol concentration and autism diagnosis.

BACKGROUND: Autistic children are prone to experience heightened levels of distress and physiological reactivity to a range of sensory, social, and emotional stimuli. In line with this, multiple studies have demonstrated that autistic children have higher acute cortisol stress responses to adverse or threatening stimuli and altered cortisol awakening responses. However, few studies have examined whether this sensitivity may relate to heightened levels of chronic stress and persistently elevated hypothalamic-pituitary-adrenal (HPA) axis activity. The measurement of cortisol accumulation in hair is considered a non-invasive biomarker of chronic stress and has been associated with several childhood diseases. Here, we investigated whether hair cortisol concentration in a large sample of autistic children differed from non-autistic children, and after accounting for a range of child, parental and family-level characteristics. METHODS: Hair cortisol concentration was measured in 307 autistic children and 282 non-autistic controls aged between 2 and 17 years recruited from four Australian states who participated in providing hair samples and demographic data to the Australian Autism Biobank. Independent samples t-test or one-way analysis of variance (ANOVA) were conducted to determine significant differences in the mean hair cortisol concentration (pg/mg) between potential covariates. Primary analysis included multivariable regression modelling of the collapsed sample to identify variables that were significantly associated with hair cortisol concentration after controlling for covariates. We also accounted for the potential interaction of multiple biological (e.g., age, sex, BMI) and psychosocial characteristics at the level of the child, the mother and the father, and the family unit. RESULTS: Our findings suggest that the diagnosis of autism was not a significant predictor of chronic stress, as measured by hair cortisol concentration. However, findings of the multivariable regression analysis showed that key factors such as area of residence (Queensland vs Victorian state of residence) and decrease in child's age were significantly associated with higher hair cortisol concentration whereas lower family income was significantly associated with higher hair cortisol concentration. CONCLUSION: To our knowledge, this is the first study to show that socioeconomic factors such as family annual income affect hair cortisol status in autistic children, indicating that the psychosocial environment may be a potential mediator for chronic stress in autistic children just as it has been demonstrated in non-autistic children.

Machine learning-based prediction for self-harm and suicide attempts in adolescents.

This study aimed to use machine learning (ML) models to predict the risk of self-harm and suicide attempts in adolescents. We conducted secondary analysis of cross-sectional data from the Longitudinal Study of Australian Children dataset. Several key variables at the age of 14-15 years were used to predict self-harm or suicide attempt at 16-17 years. Random forest classification models were used to select the optimal subset of predictors and subsequently make predictions. Among 2809 participants, 296 (10.54%) reported an act of self-harm and 145 (5.16%) reported attempting suicide at least once in the past 12 months. The area under the receiver operating curve was fair for self-harm (0.7397) and suicide attempt (0.7220), which outperformed the prediction strategy solely based on prior suicide or self-harm attempt (AUC: 0.6). The most important factors identified were similar, and included depressed feelings, strengths and difficulties questionnaire scores, perceptions of self, and school- and parent-related factors. The random forest classification algorithm, an ML technique, can effectively select the optimal subset of predictors from hundreds of variables to forecast the risks of suicide and self-harm among adolescents. Further research is needed to validate the utility and scalability of ML techniques in mental health research.

Appraising risk factors of aggression in children and adolescents at psychiatric inpatient units.

This study examined predictors of aggression and assessed whether different subgroups of children and young people (CYP) display varying risks of aggressive incidents during hospitalization. Data from 10,090 children admitted to the psychiatric inpatient units of Cincinnati Children's Hospital between April 2010 and June 2021 were analysed. Multivariable logistic regression models were used to determine significant predictors associated with aggression, followed by average marginal effects and cluster analyses to rank and establish clusters by the order of predictor importance. About 32.5% reported positive history of an aggressive incident. The mean BRACHA score was doubled compared to those without a prior history. The primary analysis showed that both younger and male CYPs had higher odds of aggressive incidents. We also found that CYP with an African descent, not being able to live with both biological parents, those who reported positive history of psychiatric hospitalisation, and prior externalising behaviours had higher odds of aggressive incidents. These findings have important clinical and public health implications, as they provide valuable knowledge for healthcare professionals to improve prevention strategies for aggression amongst this vulnerable population.

A nationwide study of COVID-19 impact on mental health-related presentations among children and adolescents to primary care practices in Australia.

This study explored the impacts of COVID-19 on the mental health (MH)-related visits to general practices (GPs) among children and young people (CYP) up to 18 years of age in Australia. This study analysed national-level data captured by the NPS MedicineWise program on monthly CYP MH-related visits per 10,000 visits to GPs from January 2014 to September 2021. We considered the pre-COVID-19 period (January 2014-February 2020) and the COVID-19 period (March 2020-September 2021). We used a Bayesian structural time series (BSTS) model to estimate the impact of COVID-19 on MH-related GP visits per 10,000 visits. A total of 103,813 out of 7,690,874 visits to GP (i.e., about 135 per 10,000 visits) were related to MH during study period. The BSTS model showed a significant increase in the overall MH-related visits during COVID-19 period (33%, 95% Credible Interval (Crl) 8.5%-56%), particularly, visits related to depressive disorders (61%, 95% Crl 29%-91%). The greatest increase was observed among females (39%, 95% Crl 12%-64%) and those living in socioeconomically least disadvantaged areas (36%, 95% Crl 1.2-71%). Our findings highlight the need for resources to be directed towards at-risk CYP to improve MH outcomes and reduce health system burden.

Screening for Media Use in the Emergency Department Among Young Australians: Cross-sectional Study.

BACKGROUND: Research on problematic internet use has largely adhered to addiction paradigms, possibly impeding the identification of specific internet behaviors related to psychopathology. This study presents a novel approach to screening for specific problematic internet behaviors by using a new measure, the emergency department media use screener (EDMUS). OBJECTIVE: The purpose of this study was to identify patterns of internet use in young people presenting with mental health concerns to the emergency department (ED), ascertain associations with their mental health, and evaluate whether the EDMUS can be used to predict subsequent ED presentations within 3 months. METHODS: This cross-sectional retrospective study of Australian young people (N=149, aged 11-25 years; female: n=92, 61.7%) sought to use the EDMUS, a 24-item questionnaire, to identify problematic internet behaviors, including accessing or posting prosuicidal or proeating disorder content, cyberbullying, and inappropriate digital content. Data on each person's mental health were extracted from electronic medical records to look for associations with EDMUS responses and ED re-presentation over 3 months. EDMUS items were grouped into clusters for analysis using chi-square tests, binary logistic regression, and path analyses. RESULTS: Sharing suicidal digital content was the most common problematic internet use pattern identified by the EDMUS. However, this did not correlate with having a prior mental health diagnosis or predict readmission. Most participants had families with a concern for their internet use; however, this was less likely in participants with a diagnosis of personality disorder. Diagnoses of personality disorder or posttraumatic stress disorder were independent predictors of readmission (P=.003; P=.048). CONCLUSIONS: Although a history of complex psychopathology increases the likelihood of subsequent ED presentations, its links to internet use-related behaviors are still unclear. The EDMUS has potential for identifying young people who are most vulnerable to problematic internet behaviors and offers the opportunity for early intervention and potential prevention of more entrenched difficulties.

COVID-19 and Pediatric Mental Health Hospitalizations.

OBJECTIVES: To analyze Australian national data to examine the impact of the coronavirus disease 2019 (COVID-19) pandemic on mental health-related hospital presentations among children and adolescents during the pandemic period with restrictions, and the period after the restrictions eased. METHODS: We analyzed the monthly mental health-related inpatient admissions and emergency department (ED) attendances data from 6 large pediatric hospitals across Australia, using the Bayesian structural time series models. The COVID-19 restriction period was from March 2020 to December 2021 and the COVID-19 restriction-eased period from January to June 2022. RESULTS: A total of 130 801 mental health-related hospital admissions (54 907) and ED attendances (75 894) were analyzed. During the COVID-19 restriction period, there was a significant increase in inpatient admissions related to deliberate self-harm behaviors (82%, 95% credible interval [CrI], 7%-160%) and ED attendances related to overall mental health disorders (15%, 95% CrI, 1.1%-30%) and eating disorders (76%, 95% CrI, 36%-115%). The increase was higher among females and those living in the least socioeconomically disadvantaged areas, suggesting a widening gap between mental health-related presentations by sex and socioeconomic status. After the restrictions eased, there were slight declines in mental health-related hospital presentations; however, the numbers remained higher than the pre-COVID-19 levels. CONCLUSIONS: The increase in mental health-related hospital presentations during the COVID-19 period calls for additional support for pediatric mental health care, particularly for eating disorders and deliberate self-harm among female adolescents. It is important to monitor pediatric mental health service use as we enter "COVID-19 normal" period.