Related Experiment Video
Updated: Sep 21, 2026
Establishing a Competing Risk Regression Nomogram Model for Survival Data
Published on: October 23, 2020
A Competing-Risk Nomogram for in-Hospital Non-Suicidal Self-Injury Recurrence Using Inflammatory and Clinical
1Department of Affective Disorders, The Third Honorable Military Hospital of Hebei Province, Baoding, Hebei, 071000, People's Republic of China.
Background:
Non-suicidal self-injury (NSSI) recurrence is an important safety concern in psychiatric inpatient care. This study aimed to identify predictors of subsequent in-hospital NSSI recurrence and develop an internally validated prediction model accounting for discharge as a competing event.
Methods:
This single-center retrospective cohort included 842 psychiatric inpatients aged 12-40 years. Admission clinical characteristics and inflammatory biomarkers were collected within the first 24 hours, while NSSI occurring during this period was recorded separately as an early in-hospital behavioral predictor. Subsequent NSSI recurrence was analyzed using Fine-Gray proportional subdistribution hazards models, with discharge without recurrence treated as a competing event. Non-linear associations were examined using restricted cubic splines. Internal validation was performed using 1000 bootstrap resamples with optimism correction.
Results:
The 14-day cumulative incidence of NSSI recurrence was 13.8%. Baseline NLR showed a significant non-linear association with recurrence risk. For exploratory model simplification, a maximally selected Gray statistic identified 2.45 as a data-derived NLR cutpoint. The multivariable Fine-Gray model identified six predictors: NSSI during the first 24 hours (sHR, 2.15), primary borderline personality disorder diagnosis (sHR, 1.88), childhood trauma or abuse (sHR, 1.65), prior suicide attempts (sHR, 1.42), NLR >2.45 (sHR, 1.58), and HAMD score (sHR per point, 1.05). The apparent C-index was 0.765, with an optimism-corrected value of 0.751. Calibration was acceptable at 7 and 14 days, and decision curve analysis suggested potential net benefit across selected threshold ranges. In sensitivity analysis excluding early NSSI, the C-index was 0.741 and the remaining predictor associations were generally maintained.
Conclusion:
Clinical history, depressive symptom burden, early in-hospital self-injury, and baseline NLR were associated with subsequent NSSI recurrence. The non-linear NLR association warrants further investigation, whereas early NSSI primarily reflects an already active crisis. The nomogram remains a research-stage risk-support model and requires prospective external validation before clinical implementation.
