The study enrolled 542 adolescents and young adults with mental disorders who underwent in-hospital treatment between October 2022 and December 2023. For each participant, the researchers gathered a broad set of information: clinical characteristics, scores on the Brøset Violence Checklist, a structured instrument for assessing short-term risk of aggression, and serum hormone levels measured in blood samples. The goal was to determine which of these many variables, alone or in combination, were most strongly associated with a baseline history of suicide attempt, meaning an attempt that had occurred before or at the time of admission.

To build a model that would generalize well rather than simply fit the data at hand, the team randomly divided the cohort into a training set of 379 patients and a validation set of 163 patients, in a seven-to-three ratio. All statistical model building was performed on the training set, while the validation set was reserved to test how well the resulting tool performed on patients it had never seen. This internal validation strategy is considered a cornerstone of responsible predictive modeling in clinical research, because a model that only performs well on the data used to build it is of little practical value at the bedside.

Before fitting the final model, the researchers applied least absolute shrinkage and selection operator regression, commonly known as LASSO, a technique that shrinks the coefficients of weak predictors toward zero and effectively filters out redundant or uninformative variables. LASSO regression is particularly useful in medical datasets where many candidate predictors are correlated with one another, as it reduces overfitting and produces a more parsimonious set of features. The variables that survived this screening step were then entered into a multivariate logistic regression model, the standard statistical framework for estimating the independent contribution of each factor to a binary outcome, in this case the presence or absence of a suicide-attempt history.

Five variables ultimately earned a place in the final nomogram: a history of non-suicidal self-injury, serum prolactin levels, plasma total cortisol levels, the number of previous hospitalizations, and educational level. Each of these factors carries its own clinical logic. Non-suicidal self-injury, the deliberate harming of one’s own body without suicidal intent, has long been recognized as a powerful signal of underlying emotional distress and is one of the most consistent correlates of later suicidal behavior in young people. Repeated hospitalizations suggest a chronic or unstable course of illness, while lower educational attainment may reflect the cumulative disruption that mental disorders impose on development and social functioning.

The inclusion of two hormones adds a biological dimension that sets this study apart from many purely psychosocial risk models. Prolactin, best known for its role in lactation, is also influenced by stress and by several psychiatric medications, and altered prolactin levels have been linked in previous research to self-directed violence and emotional dysregulation. Cortisol, the body’s primary stress hormone, reflects the activity of the hypothalamic-pituitary-adrenal axis, the neuroendocrine system that orchestrates the physiological response to stress. Dysregulation of this axis has repeatedly been implicated in mood disorders and suicidal behavior, and measuring plasma total cortisol offers an objective, laboratory-based window into a patient’s stress biology that cannot be captured by interview alone.

A nomogram translates the coefficients of a logistic regression model into a visual chart on which a clinician can plot the value of each predictor, read off assigned points, sum them, and convert the total into an estimated probability. In this case, the estimated probability refers to the likelihood that a given hospitalized adolescent or young adult has a baseline history of suicide attempt. The appeal of the format lies in its simplicity: no calculator or software is required, and the relative weight of each factor is immediately visible, which can help clinicians understand why a particular patient scores as high or low risk.

The performance metrics reported by the team suggest a moderately strong tool. The area under the receiver operating characteristic curve, or AUC, a measure of how well a model distinguishes between those with and without the outcome, reached 0.771 in the training set, with a 95 percent confidence interval of 0.715 to 0.827, and 0.788 in the validation set, with a confidence interval of 0.711 to 0.865. Values in this range indicate discrimination that is useful but not definitive, which the authors explicitly acknowledge. Calibration, meaning the agreement between predicted probabilities and observed outcomes, was strong: the Hosmer-Lemeshow goodness-of-fit test yielded a chi-square of 12.672 with a p-value of 0.123 in the training set and a chi-square of 8.919 with a p-value of 0.349 in the validation set, both indicating satisfactory model fit. Decision curve analysis, a method for quantifying the clinical net benefit of a model across a range of decision thresholds, identified effective thresholds from 19 percent to 73 percent in the training set and from 10 percent to 74 percent in the validation set, supporting the tool’s utility as a risk-stratification reference.

The authors are careful to position the nomogram as an auxiliary quantitative reference rather than a standalone screening instrument. Suicide risk assessment in young people is a complex clinical judgment that must integrate interview findings, collateral information, mental state examination and safety planning, and no statistical model can replace that process. What this tool offers is a structured way to visualize how multiple dimensions of a patient’s profile, from self-injury history to endocrine markers, converge to shape risk, potentially flagging high-risk patients who might otherwise receive insufficient attention during a busy admission. The researchers suggest it may help inform targeted suicide-prevention interventions within inpatient psychiatric settings.

Because the study is cross-sectional, it captures associations at a single point in time and cannot establish that the five predictors cause suicide attempts, nor can it predict future attempts prospectively. The cohort was drawn from a single hospital network in China, so external validation in other countries, cultures and healthcare systems will be needed before broader adoption. Nonetheless, the work represents a meaningful step toward integrating biological and psychosocial data into accessible bedside tools for one of medicine’s most urgent challenges. As the authors note, characterizing the risk profiles of hospitalized adolescents and young adults with mental disorders is essential for targeted suicide prevention, and a validated, visual, multidimensional model gives clinicians one more way to see that risk clearly.

Subject of Research: A nomogram integrating psychosocial and hormonal correlates of baseline suicide-attempt history in hospitalized adolescents and young adults with mental disorders

Article Title: Development and internal validation of a nomogram for visualising multidimensional correlates of baseline suicide‑attempt history among hospitalized adolescents and young adults with mental disorders

Article References: Ren, Y., Liu, Q., Zhu, Y., Wu, J., Chen, H., Pu, L., Dong, Z., Zhu, H., & Zhang, X. (2026). Development and internal validation of a nomogram for visualising multidimensional correlates of baseline suicide‑attempt history among hospitalized adolescents and young adults with mental disorders. BMC Psychiatry. https://doi.org/10.1186/s12888-026-08613-w

Image Credits: AI Generated

DOI: 10.1186/s12888-026-08613-w

Keywords: nomogram, suicide-attempt history, adolescents and young adults, mental disorders, non-suicidal self-injury, prolactin, cortisol, LASSO regression, logistic regression, risk prediction, BMC Psychiatry, inpatient psychiatry