Attempts to use artificial intelligence (AI) in psychiatric disorders show moderate success, highlighting the potential of incorporating information from clinical assessments to improve the models. This study focuses on using large language models (LLMs) to detect suicide risk from medical text in psychiatric care.

To extract information about suicidality status from the admission notes in electronic health records (EHRs) using privacy-sensitive, locally hosted LLMs, specifically evaluating the efficacy of Llama-2 models.

We compared the performance of several variants of the open source LLM Llama-2 in extracting suicidality status from 100 psychiatric reports against a ground truth defined by human experts, assessing accuracy, sensitivity, specificity and F1 score across different prompting strategies.

A German fine-tuned Llama-2 model showed the highest accuracy (87.5%), sensitivity (83.0%) and specificity (91.8%) in identifying suicidality, with significant improvements in sensitivity and specificity across various prompt designs.

The study demonstrates the capability of LLMs, particularly Llama-2, in accurately extracting information on suicidality from psychiatric records while preserving data privacy. This suggests their application in surveillance systems for psychiatric emergencies and improving the clinical management of suicidality by improving systematic quality control and research.

Individuals at risk for bipolar disorder (BD) have a wide range of genetic and non-genetic risk factors, like a positive family history of BD or (sub)threshold affective symptoms. Yet, it is unclear whether these individuals at risk and those diagnosed with BD share similar gray matter brain alterations.

In 410 male and female participants aged 17–35 years, we compared gray matter volume (3T MRI) between individuals at risk for BD (as assessed using the EPIbipolar scale; n = 208), patients with a DSM-IV-TR diagnosis of BD (n = 87), and healthy controls (n = 115) using voxel-based morphometry in SPM12/CAT12. We applied conjunction analyses to identify similarities in gray matter volume alterations in individuals at risk and BD patients, relative to healthy controls. We also performed exploratory whole-brain analyses to identify differences in gray matter volume among groups. ComBat was used to harmonize imaging data from seven sites.

Both individuals at risk and BD patients showed larger volumes in the right putamen than healthy controls. Furthermore, individuals at risk had smaller volumes in the right inferior occipital gyrus, and BD patients had larger volumes in the left precuneus, compared to healthy controls. These findings were independent of course of illness (number of lifetime manic and depressive episodes, number of hospitalizations), comorbid diagnoses (major depressive disorder, attention-deficit hyperactivity disorder, anxiety disorder, eating disorder), familial risk, current disease severity (global functioning, remission status), and current medication intake.

Our findings indicate that alterations in the right putamen might constitute a vulnerability marker for BD.

Individuals with bipolar disorder are commonly correctly diagnosed a decade after symptom onset. Machine learning techniques may aid in early recognition and reduce the disease burden. As both individuals at risk and those with a manifest disease display structural brain markers, structural magnetic resonance imaging may provide relevant classification features.

Following a pre-registered protocol, we trained linear support vector machine (SVM) to classify individuals according to their estimated risk for bipolar disorder using regional cortical thickness of help-seeking individuals from seven study sites (N = 276). We estimated the risk using three state-of-the-art assessment instruments (BPSS-P, BARS, EPIbipolar).

For BPSS-P, SVM achieved a fair performance of Cohen's κ of 0.235 (95% CI 0.11–0.361) and a balanced accuracy of 63.1% (95% CI 55.9–70.3) in the 10-fold cross-validation. In the leave-one-site-out cross-validation, the model performed with a Cohen's κ of 0.128 (95% CI −0.069 to 0.325) and a balanced accuracy of 56.2% (95% CI 44.6–67.8). BARS and EPIbipolar could not be predicted. In post hoc analyses, regional surface area, subcortical volumes as well as hyperparameter optimization did not improve the performance.

Individuals at risk for bipolar disorder, as assessed by BPSS-P, display brain structural alterations that can be detected using machine learning. The achieved performance is comparable to previous studies which attempted to classify patients with manifest disease and healthy controls. Unlike previous studies of bipolar risk, our multicenter design permitted a leave-one-site-out cross-validation. Whole-brain cortical thickness seems to be superior to other structural brain features.

Mental illness is known to come along with a large mortality gap compared to thegeneral population and it is a risk for COVID-19 related morbidity andmortality. Achieving high vaccination rates in people with mental illness is therefore important. Reports are conflicting on whether vaccination rates comparable to those of the general population can be achieved and which variables represent risk factors for nonvaccination in people with mental illness.

The COVID Ψ Vac study collected routine data on vaccination status, diagnostic groups, sociodemographics, and setting characteristics from in- and day-clinic patients of 10 psychiatric hospitals in Germany in August 2021. Logistic regression modeling was used to determine risk factors for nonvaccination.

Complete vaccination rates were 59% (n = 776) for the hospitalized patients with mental illness versus 64% for the regionally and age-matched general population. Partial vaccination rates were 68% (n = 893) for the hospitalised patients with mental illness versus 67% for the respective general population and six percentage (n = 74) of this hospitalized population were vaccinated during the hospital stay. Rates showed a large variation between hospital sites. An ICD-10 group F1, F2, or F4 main diagnosis, younger age, and coercive accommodation were further risk factors for nonvaccination in the model.

Vaccination rates were lower in hospitalized people with mental illness than in the general population. By targeting at-risk groups with low-threshold vaccination programs in all health institutions they get in contact with, vaccination rates comparable to those in the general population can be achieved.

Response to lithium in patients with bipolar disorder is associated with clinical and transdiagnostic genetic factors. The predictive combination of these variables might help clinicians better predict which patients will respond to lithium treatment.

To use a combination of transdiagnostic genetic and clinical factors to predict lithium response in patients with bipolar disorder.

This study utilised genetic and clinical data (n = 1034) collected as part of the International Consortium on Lithium Genetics (ConLi+Gen) project. Polygenic risk scores (PRS) were computed for schizophrenia and major depressive disorder, and then combined with clinical variables using a cross-validated machine-learning regression approach. Unimodal, multimodal and genetically stratified models were trained and validated using ridge, elastic net and random forest regression on 692 patients with bipolar disorder from ten study sites using leave-site-out cross-validation. All models were then tested on an independent test set of 342 patients. The best performing models were then tested in a classification framework.

The best performing linear model explained 5.1% (P = 0.0001) of variance in lithium response and was composed of clinical variables, PRS variables and interaction terms between them. The best performing non-linear model used only clinical variables and explained 8.1% (P = 0.0001) of variance in lithium response. A priori genomic stratification improved non-linear model performance to 13.7% (P = 0.0001) and improved the binary classification of lithium response. This model stratified patients based on their meta-polygenic loadings for major depressive disorder and schizophrenia and was then trained using clinical data.

Using PRS to first stratify patients genetically and then train machine-learning models with clinical predictors led to large improvements in lithium response prediction. When used with other PRS and biological markers in the future this approach may help inform which patients are most likely to respond to lithium treatment.

Previous studies in individual countries have identified inconsistent predictors of length of stay (LoS) in psychiatric inpatient units. This may reflect methodological inconsistencies across studies or true differences of predictors. In this study we assessed predictors of LoS in five European countries and explored whether their effect varies across countries.

Prospective cohort study. All patients admitted over 14 months to 57 psychiatric inpatient units in Belgium, Germany, Italy, Poland and United Kingdom were screened. Putative predictors were collected from medical records and in face-to-face interviews and tested for their association with LoS.

Average LoS varied from 17.9 days in Italy to 55.1 days in Belgium. In the overall sample being homeless, receiving benefits, social isolation, diagnosis of psychosis, greater symptom severity, substance use, history of previous admission and being involuntarily admitted predicted longer LoS. Several predictors showed significant interaction effects with countries in predicting LoS. One variable, homelessness, predicted a different LoS even in opposite directions, whilst for other predictors the direction of the association was the same, but the strength of the association with LoS varied across countries.

The same patient characteristics have a different impact on LoS in different contexts. Thus, although some predictor variables related to clinical severity and social dysfunction appear of generalisable relevance, national studies on LoS are required to understand the complex influence of different patient characteristics on clinical practice in the given contexts.

Patient satisfaction is a key indicator of inpatient care quality and is associated with clinical outcomes following admission. Different patient characteristics have been inconsistently linked with satisfaction. This study aims to overcome previous limitations by assessing which patient characteristics are associated with satisfaction within a large study of psychiatric inpatients conducted across five European countries.

All patients with a diagnosis of psychotic (F2), affective (F3) or anxiety/somataform (F4) disorder admitted to 57 psychiatric inpatient units in Belgium, Germany, Italy, Poland and the UK were included. Data were collected from medical records and face-to-face interviews, with patients approached within 2 days of admission. Satisfaction with inpatient care was measured on the Client Assessment of Treatment Scale.

Higher satisfaction scores were associated with being older, employed, living with others, having a close friend, less severe illness and a first admission. In contrast, higher education levels, comorbid personality disorder and involuntary admission were associated with lower levels of satisfaction. Although the same patient characteristics predicted satisfaction within the five countries, there were significant differences in overall satisfaction scores across countries. Compared to other countries, patients in the UK were significantly less satisfied with their inpatient care.

Having a better understanding of patient satisfaction may enable services to improve the quality of care provided as well as clinical outcomes for all patients. Across countries, the same patient characteristics predict satisfaction, suggesting that similar analytical frameworks can and should be used when assessing satisfaction both nationally and internationally.

In Europe, at discharge from a psychiatric hospital, patients with severe mental illness may be exposed to one of two main care approaches: personal continuity, where one clinician is responsible for in- and outpatient care, and specialisation, where various clinicians are. Such exposure is decided through patient-clinician agreement or at the organisational level, depending on the country’s health system. Since personal continuity would be more suitable for patients with complex psychosocial needs, the aim of this study was to identify predictors of patients’ exposure to care approaches in different European countries.

Data were collected on 7302 psychiatric hospitalised patients in 2015 in Germany, Poland, and Belgium (patient-level exposure); and in the UK and Italy (organisational-level exposure). At discharge, patients were exposed to one of the care approaches according to usual practice. Putative predictors of exposure at patients’ discharge were assessed in both groups of countries.

Socially disadvantaged patients were significantly more exposed to personal continuity. In all countries, the main predictor of exposure was the admission hospital, except in Germany, where having a diagnosis of psychosis and a higher education status were predictors of exposure to personal continuity. In the UK, hospitals practising personal continuity had a more socially disadvantaged patient population.

Even in countries where exposure is decided through patient-clinician agreement, it was the admission hospital, not patient characteristics, that predicted exposure to care approaches. Nevertheless, organisational decisions in hospitals tend to expose socially disadvantaged patients to personal continuity.