Rates of childhood mental health problems are increasing in the UK. Early identification of childhood mental health problems is challenging but critical to children’s future psychosocial development. This is particularly important for children with social care contact because earlier identification can facilitate earlier intervention. Clinical prediction tools could improve these early intervention efforts.

Characterise a novel cohort consisting of children in social care and develop effective machine learning models for prediction of childhood mental health problems.

We used linked, de-identified data from the Secure Anonymised Information Linkage Databank to create a cohort of 26 820 children in Wales, UK, receiving social care services. Integrating health, social care and education data, we developed several machine learning models aimed at predicting childhood mental health problems. We assessed the performance, interpretability and fairness of these models.

Risk factors strongly associated with childhood mental health problems included age, substance misuse and being a looked after child. The best-performing model, a gradient boosting classifier, achieved an area under the receiver operating characteristic curve of 0.75 (95% CI 0.73–0.78). Assessments of algorithmic fairness showed potential biases within these models.

Machine learning performance on this prediction task was promising. Predictive performance in social care settings can be bolstered by linking diverse routinely collected data-sets, making available a range of heterogenous risk factors relating to clinical, social and environmental exposures.

We sought to describe perspectives among Black nursing professionals and community leaders regarding the return of genetic test results, and place perspectives into context with aggregated findings in the All of Us Research Program’s Data Browser.

Semi-structured, virtual interviews were held with adults (≥18 years of age) self-identifying as Black. A 2-step thematic analysis process was used to assess interviewee perspectives with (sub)themes identified in the literature across two topics: drug/medication response and hereditary disease risk. Themes were placed into context with Data Browser content, focusing on genes and their respective alleles with frequencies ≥0.10 in African ancestry populations in All of Us.

Interviewee perspectives aligned with previously identified major themes in the literature (motivations to engage or disengage; integrating research and care), with five (5) subthemes emerging across major themes. Seven (7) alleles were observed with frequencies ≥0.10 for three (3) pharmacogenomic (PGx) biomarkers in the Data Browser for African ancestry populations: CYP2C19 (SNV, 10-94761900-C-T; SNV,10-94775367-A-G; SNV 10-94781859-G-A), DPYD (SNV, 1-97883329-A-G; SNV, 1-97515839-T-C), UGT1A1 (insertion, 2-233760233-C-CAT; SNV, 2-233757136-G-A). Four (4) alleles were observed with frequencies ≥0.10 for three (3) genes implicated in hereditary disease risk, two of which contemporaneously hold PGx implications for African ancestry populations: CACNA1S (PGx, SNV, 1-201112815-C-T; SNV, 1-201110107-C-T), SCN5A (no PGx, SNV, 3-38603929-T-C), TP53 (PGx, SNV, 17-7676154-G-C).

Our findings convey important clinical and translational science considerations for individuals and community leaders of African ancestry and researchers seeking reputable, publicly available information to understand, communicate, and act on genomic findings.

Scalable assessment tools for precision psychiatry are of increasing clinical interest. One clinical risk assessment that might be improved by such approaches is assessment of violence perpetration risk. This is an important adverse outcome to reduce for some people presenting to services for first-episode psychosis. A prediction tool (Oxford Mental Illness and Violence (OxMIV)) has been externally validated in these services, but clinical acceptability and role need to be examined and developed.

This study aimed to understand clinical use of the OxMIV tool to support violence risk management in early intervention in psychosis services in terms of acceptability to clinicians, patients and carers, practical feasibility, perceived utility, impact and role.

A mixed methods approach integrated quantitative data on utility and patterns of use of the OxMIV tool over 12 months in two services with qualitative data from interviews of 20 clinicians and 12 patients and carers.

The OxMIV tool was used 141 times, mostly in new assessments. Required information was available, with only family history items scored unknown to any notable degree. The OxMIV tool was deemed helpful by clinicians in most cases, especially if there were previous risk concerns. It was acceptable practically, and broadly for the service, for which its concordance with clinical judgement was important. Patients and carers thought it could improve openness. There was some limited impact on plans for clinical support.

The OxMIV tool met an identified clinical need to support clinical assessment for violence risk. Linkage to intervention pathways is a research priority.

Physical activities are widely implemented for non-pharmacological intervention to alleviate depressive symptoms. However, there is little evidence supporting their genotype-specific effectiveness in reducing the risk of self-harm in patients with depression.

To assess the associations between physical activity and self-harm behaviour and determine the recommended level of physical activity across the genotypes.

We developed the bidirectional analytical model to investigate the genotype-specific effectiveness on UK Biobank. After the genetic stratification of the depression phenotype cohort using hierarchical clustering, multivariable logistic regression models and Cox proportional hazards models were built to investigate the associations between physical activity and the risk of self-harm behaviour.

A total of 28 923 subjects with depression phenotypes were included in the study. In retrospective cohort analysis, the moderate and highly active groups were at lower risk of self-harm behaviour. In the followed prospective cohort analysis, light-intensity physical activity was associated with a lower risk of hospitalisations due to self-harm behaviour in one genetic cluster (adjusted hazard ratio, 0.28 [95% CI, 0.08–0.96]), which was distinguished by three genetic variants: rs1432639, rs4543289 and rs11209948. Compliance with the guideline-level moderate-to-vigorous physical activities was not significantly related to the risk of self-harm behaviour.

A genotype-specific dose of light-intensity physical activity reduces the risk of self-harm by around a fourth in depressive patients.

Cardiovascular disease (CVD) is twice as prevalent among individuals with mental illness compared to the general population. Prevention strategies exist but require accurate risk prediction. This study aimed to develop and validate a machine learning model for predicting incident CVD among patients with mental illness using routine clinical data from electronic health records.

A cohort study was conducted using data from 74,880 patients with 1.6 million psychiatric service contacts in the Central Denmark Region from 2013 to 2021. Two machine learning models (XGBoost and regularised logistic regression) were trained on 85% of the data from six hospitals using 234 potential predictors. The best-performing model was externally validated on the remaining 15% of patients from another three hospitals. CVD was defined as myocardial infarction, stroke, or peripheral arterial disease.

The best-performing model (hyperparameter-tuned XGBoost) demonstrated acceptable discrimination, with an area under the receiver operating characteristic curve of 0.84 on the training set and 0.74 on the validation set. It identified high-risk individuals 2.5 years before CVD events. For the psychiatric service contacts in the top 5% of predicted risk, the positive predictive value was 5%, and the negative predictive value was 99%. The model issued at least one positive prediction for 39% of patients who developed CVD.

A machine learning model can accurately predict CVD risk among patients with mental illness using routinely collected electronic health record data. A decision support system building on this approach may aid primary CVD prevention in this high-risk population.

The Personalized Advantage Index (PAI) shows promise as a method for identifying the most effective treatment for individual patients. Previous studies have demonstrated its utility in retrospective evaluations across various settings. In this study, we explored the effect of different methodological choices in predictive modelling underlying the PAI.

Our approach involved a two-step procedure. First, we conducted a review of prior studies utilizing the PAI, evaluating each study using the Prediction model study Risk Of Bias Assessment Tool (PROBAST). We specifically assessed whether the studies adhered to two standards of predictive modeling: refraining from using leave-one-out cross-validation (LOO CV) and preventing data leakage. Second, we examined the impact of deviating from these methodological standards in real data. We employed both a traditional approach violating these standards and an advanced approach implementing them in two large-scale datasets, PANIC-net (n = 261) and Protect-AD (n = 614).

The PROBAST-rating revealed a substantial risk of bias across studies, primarily due to inappropriate methodological choices. Most studies did not adhere to the examined prediction modeling standards, employing LOO CV and allowing data leakage. The comparison between the traditional and advanced approach revealed that ignoring these standards could systematically overestimate the utility of the PAI.

Our study cautions that violating standards in predictive modeling may strongly influence the evaluation of the PAI's utility, possibly leading to false positive results. To support an unbiased evaluation, crucial for potential clinical application, we provide a low-bias, openly accessible, and meticulously annotated script implementing the PAI.

Psychiatric research applies statistical methods that can be divided in two frameworks: causal inference and prediction. Recent proposals suggest a down-prioritisation of causal inference and argue that prediction paves the road to ‘precision psychiatry’ (i.e., individualised treatment). In this perspective, we critically appraise these proposals.

We outline strengths and weaknesses of causal inference and prediction frameworks and describe the link between clinical decision-making and counterfactual predictions (i.e., causality). We describe three key causal structures that, if not handled correctly, may cause erroneous interpretations, and three pitfalls in prediction research.

Prediction and causal inference are both needed in psychiatric research and their relative importance is context-dependent. When individualised treatment decisions are needed, causal inference is necessary.

This perspective defends the importance of causal inference for precision psychiatry.

Patients with cystic fibrosis (CF) experience frequent episodes of acute decline in lung function called pulmonary exacerbations (PEx). An existing clinical and place-based precision medicine algorithm that accurately predicts PEx could include racial and ethnic biases in clinical and geospatial training data, leading to unintentional exacerbation of health inequities.

We estimated receiver operating characteristic curves based on predictions from a nonstationary Gaussian stochastic process model for PEx within 3, 6, and 12 months among 26,392 individuals aged 6 years and above (2003–2017) from the US CF Foundation Patient Registry. We screened predictors to identify reasons for discriminatory model performance.

The precision medicine algorithm performed worse predicting a PEx among Black patients when compared with White patients or to patients of another race for all three prediction horizons. There was little to no difference in prediction accuracies among Hispanic and non-Hispanic patients for the same prediction horizons. Differences in F508del, smoking households, secondhand smoke exposure, primary and secondary road densities, distance and drive time to the CF center, and average number of clinical evaluations were key factors associated with race.

Racial differences in prediction accuracies from our PEx precision medicine algorithm exist. Misclassification of future PEx was attributable to several underlying factors that correspond to race: CF mutation, location where the patient lives, and clinical awareness. Associations of our proxies with race for CF-related health outcomes can lead to systemic racism in data collection and in prediction accuracies from precision medicine algorithms constructed from it.

Less than a third of patients with depression achieve successful remission with standard first-step antidepressant monotherapy. The process for determining appropriate second-step care is often based on clinical intuition and involves a protracted course of trial and error, resulting in substantial patient burden and unnecessary delay in the provision of optimal treatment. To address this problem, we adopt an ensemble machine learning approach to improve prediction accuracy of remission in response to second-step treatments.

Data were derived from the Level 2 stage of the STAR*D dataset, which included 1439 patients who were randomized into one of seven different second-step treatment strategies after failing to achieve remission during first-step antidepressant treatment. Ensemble machine learning models, comprising several individual algorithms, were evaluated using nested cross-validation on 155 predictor variables including clinical and demographic measures.

The ensemble machine learning algorithms exhibited differential classification performance in predicting remission status across the seven second-step treatments. For the full set of predictors, AUC values ranged from 0.51 to 0.82 depending on the second-step treatment type. Predicting remission was most successful for cognitive therapy (AUC = 0.82) and least successful for other medication and combined treatment options (AUCs = 0.51–0.66).

Ensemble machine learning has potential to predict second-step treatment. In this study, predictive performance varied by type of treatment, with greater accuracy in predicting remission in response to behavioral treatments than to pharmacotherapy interventions. Future directions include considering more informative predictor modalities to enhance prediction of second-step treatment response.

The purpose of this study is to evaluate the validity of the standard approach in expert judgment for evaluating precision medicines, in which experts are required to estimate outcomes as if they did not have access to diagnostic information, whereas in fact, they do.

Fourteen clinicians participated in an expert judgment task to estimate the cost and medical outcomes of the use of exome sequencing in pediatric patients with intractable epilepsy in Thailand. Experts were randomly assigned to either an “unblind” or “blind” group; the former was provided with the exome sequencing results for each patient case prior to the judgment task, whereas the latter was not provided with the exome sequencing results. Both groups were asked to estimate the outcomes for the counterfactual scenario, in which patients had not been tested by exome sequencing.

Our study did not show significant results, possibly due to the small sample size of both participants and case studies.

A comparison of the unblind and blind approach did not show conclusive evidence that there is a difference in outcomes. However, until further evidence suggests otherwise, we recommend the blind approach as preferable when using expert judgment to evaluate precision medicines because this approach is more representative of the counterfactual scenario than the unblind approach.