Risk-Stratified Tiered Intervention Among Chinese Clinical High-Risk for Psychosis Patients and Attenuated Negative Symptoms as a Predictor of 1-Year Psychosis Conversion: A Retrospective Propensity-Matched Real-World Cohort Study

Abstract

Objective

This retrospective propensity-score-matched real-world cohort study aimed to evaluate the clinical efficacy of risk stratification combined with tiered intervention in young Chinese patients at clinical high-risk (CHR) for psychosis, and to explore whether attenuated negative symptoms could independently predict 1-year conversion to overt psychosis.

Methods

A total of 150 CHR patients receiving risk-stratified tiered management were enrolled as the intervention group, and 75 CHR patients receiving conventional routine care were matched as the control group via propensity-score matching (PSM). All participants completed standardized two-stage clinical screening, cranial MRI to exclude organic brain lesions, and systematic symptom and functional assessments. Logistic regression and receiver-operating-characteristic (ROC) curve analyses were performed to identify independent predictors of psychosis conversion, with a specific focus on attenuated negative symptoms. Primary outcomes included 1-year psychosis conversion rate, changes in global functioning, and treatment adherence.

Results

After 1-year follow-up, the intervention group showed a significantly lower psychosis conversion rate (9.3% vs 20.0%, P=0.021), greater improvement in global functioning (mean Δ = 9.8 vs 4.0, P < 0.001), and higher treatment adherence (78% vs 40%, P < 0.001) relative to the PSM-matched control group. Multivariate logistic regression identified severe attenuated negative symptoms as an independent risk-associated factor for 1-year psychosis conversion (OR = 2.71, 95% CI: 1.43–5.14, P < 0.01). The ROC-AUC of attenuated negative symptoms for predicting psychosis conversion was 0.79, indicating acceptable predictive performance.

Conclusion

Risk stratification paired with tiered intervention reduced the risk of psychosis transition, improved global social functioning, and enhanced treatment adherence among Chinese CHR patients within this real-world cohort. Attenuated negative symptoms represent a reliable independent predictor of 1-year psychosis conversion and may be integrated into routine clinical risk assessment to support targeted early-intervention strategies for CHR populations.

Introduction

Early identification and standardized individualized intervention for individuals at clinical high-risk (CHR) for psychosis are the core priority of contemporary early psychiatric prevention systems.¹ More than half of all first-episode psychotic disorders emerge before the age of 25; timely targeted intervention can effectively reduce the 1-year conversion risk of overt psychosis and preserve long-term social functioning of vulnerable youth populations.² In China, the lifetime prevalence of psychotic disorders is approximately 0.7%, yet standardized clinical management pathways for CHR patients remain immature in grassroots psychiatric institutions, with specialized early intervention service coverage less than 30% and insufficient localized real-world evidence to guide routine clinical practice.³

Traditional unified one-size-fits-all care models for CHR populations have two prominent clinical limitations in real-world settings. First, most conventional risk assessment tools mainly rely on attenuated positive symptoms to judge conversion risk, while the predictive value of attenuated negative symptoms is often overlooked in clinical risk stratification.⁴ Attenuated negative symptoms, including anhedonia, avolition, social apathy and blunted affect, carry important clinical implications: they are closely linked to profound social functional impairment, poor long-term clinical outcomes, and elevated likelihood of progression to full-blown psychosis. Multiple Western cohort studies have confirmed that persistent attenuated negative symptoms are closely correlated with impaired social functioning and elevated psychosis transition probability, but relevant large-sample domestic verification data among Chinese CHR youth remain scarce.⁵ In routine Chinese clinical practice, negative symptom severity is rarely quantified as a core risk indicator during CHR screening, which may lead to missed identification of high-risk individuals who require intensified monitoring. Second, undifferentiated intervention strategies lead to unnecessary antipsychotic prescriptions for nearly 40% low-risk CHR individuals, causing medical resource waste and unsatisfactory treatment adherence outcomes.⁶

Machine learning-assisted risk stratification provides a feasible tool to realize precise grading of CHR patients, and tiered intervention matching different risk levels can balance intervention intensity and rational allocation of limited mental health resources.⁷ Existing meta-analyses have proven that stratified targeted interventions can significantly lower overall psychosis conversion rates compared with routine unified care, but most relevant research data are derived from Western cohorts, lacking clinical evidence adapted to Chinese grassroots medical service characteristics.⁸

Taken together, several key research gaps still exist in this field. First, few domestic real-world studies have simultaneously evaluated the clinical effectiveness of combined risk stratification and tiered intervention strategies for Chinese CHR populations. Second, despite well-documented associations between attenuated negative symptoms and psychosis conversion in Western samples, large-scale local evidence is lacking to support whether attenuated negative symptoms act as an independent short-term predictor among Chinese CHR patients.⁹ Third, there remains a lack of practical, low-cost management workflows that fit the staffing and service conditions of China’s grassroots psychiatric settings.

This study adopted a two-stage community-hospital screening workflow suitable for primary mental health services and integrated family-participated follow-up management to improve treatment continuity.¹⁰ Based on propensity score matching, we systematically compared the differences in 1-year psychosis conversion rate, global functional recovery and treatment adherence between patients receiving risk stratification combined with tiered intervention and those receiving conventional routine care. Meanwhile, multivariate regression and ROC curve analysis were performed to quantify the predictive value of attenuated negative symptoms for psychosis transition within one year. The novelty of this work lies in: (1) implementing a locally-adapted two-stage community-hospital screening pathway with family-engaged tiered intervention; (2) incorporating attenuated negative symptom severity as a high-weight feature within a random-forest risk-prediction model; (3) simultaneously assessing intervention effectiveness and the independent predictive performance of attenuated negative symptoms using propensity-score-matched real-world data. The findings of this propensity-matched retrospective cohort study aim to supply localized clinical evidence for promoting stratified tiered management in grassroots psychiatric departments and enrich the symptom prediction indicators of routine CHR risk assessment.

Materials and Methods

Study Design and Participants

This retrospective observational propensity-matched real-world cohort study was conducted at Wenzhou Seventh People’s Hospital between January 2025 and June 2025. A total of 150 patients meeting the diagnostic criteria for clinical high-risk (CHR) for psychosis were assigned to the intervention group, who received risk stratification combined with tiered individualized management. A total of 75 CHR patients receiving conventional unified routine care were enrolled as potential control candidates, and 1:0.5 propensity-score matching (PSM) was performed to generate the final control group.

Inclusion Criteria

Met standardized clinical diagnostic criteria for clinical high-risk for psychosis (CHR); Age 14–35 years old; Completed two-stage community-hospital screening workflow and baseline cranial MRI examination; Available complete baseline symptom, social-function and demographic assessment data; Participants (or legal guardians for participants under 18 years old) provided written informed consent for clinical follow-up and data collection.

Exclusion Criteria

Evidence of organic intracranial lesions identified by cranial MRI; Current or past diagnosed overt psychotic disorder; Severe substance-use disorder requiring priority clinical intervention; Other severe neuropsychiatric disorders (eg, moderate-to-severe intellectual disability, advanced neurodegenerative disease); Incomplete baseline assessment data or refusal to participate in 1-year clinical follow-up. For propensity-score matching, the matching ratio was set to 1:0.5, with a caliper width of 0.2 times the standard deviation of the logit of the propensity score. Matching covariates included age, gender, years of education, baseline risk prediction score, baseline global functioning score, history of childhood trauma, and baseline attenuated negative symptom severity score. Post-matching covariate balance was evaluated using standardized mean difference (SMD); SMD < 0.1 was considered to indicate satisfactory between-group balance. After matching, 75 patients constituted the final control group. All clinical interventions in this study belonged to routine daily clinical practice rather than experimental trial schemes. The research protocol was approved by the Ethics Committee of Wenzhou Seventh People’s Hospital (Approval No. EC-KY-2024010). Written informed consent was obtained from all participants; legal guardians provided consent for participants younger than 18 years old. All procedures carried out in this study were in accordance with the ethical principles of the Declaration of Helsinki.

Screening Tools and Assessment Procedure

The two-stage screening pathway was formulated in accordance with domestic CHR clinical guidelines, combining community primary screening and hospital standardized psychiatric diagnosis. All enrolled subjects completed 3.0T cranial MRI to exclude organic intracranial lesions, and structured clinical interviews were used to rule out co-occurring substance use disorders and other severe mental illnesses. The assessment system contained family-reported indicators (social withdrawal, sleep disturbance, emotional lability) and standardized professional scale evaluation of attenuated positive and negative symptoms. The Scale of Prodromal Negative Symptoms (SPNS) was adopted to quantify the severity of attenuated negative symptoms, covering anhedonia, avolition, social apathy and blunted affect. Internal consistency and inter-rater reliability of the full assessment system were verified to be satisfactory. For the two-stage screening flow: community general practitioners completed preliminary risk screening, and attending psychiatrists conducted definitive CHR diagnosis, negative symptom scoring and baseline risk assessment for eligible subjects.

Risk Prediction and Stratification Model

A random forest algorithm was adopted to construct a 1-year psychosis conversion risk-prediction model. The model was trained on a retrospective historical CHR cohort (n = 112) from our hospital, and external validation was performed on an independent temporal validation sample of 38 CHR patients enrolled in the current study period. Twelve pre-specified clinical predictors were input into the model, including symptomatic indicators (attenuated positive symptom score, attenuated negative symptom score), social-functional indicators (global functioning score, social withdrawal level), biological imaging markers, and social-environmental factors (childhood trauma history, educational level, family support status). Default random-forest hyper-parameters were implemented with 500 decision-tree estimators; feature importance was calculated based on mean decrease in Gini impurity. Model predictive performance was quantified using the area under the receiver-operating-characteristic curve (AUC), sensitivity, and specificity. Feature-weight analysis was performed to compare the relative contribution of each predictor to psychosis-conversion risk. According to the 1-year psychosis-conversion probability output by the trained random-forest model, all participants were stratified into three predefined risk subgroups: low-risk (predicted conversion probability <10%), medium-risk (predicted conversion probability 10–20%), and high-risk (predicted conversion probability >20%). These probability-based cut-offs were predefined before model deployment according to clinical consensus in CHR early-intervention practice. Among all input features, attenuated negative symptom severity showed a relatively high feature weight in the random forest model, indicating its important contribution to predicting conversion risk.

Tiered Clinical Management Protocol

Targeted tiered intervention was implemented according to risk stratification results, with family participation as a core auxiliary measure to improve symptom relief and treatment adherence: High-risk group: Low-dose aripiprazole (2–5 mg/d) combined with biweekly integrated psychological intervention, with special attention to targeted psychological counseling for severe negative symptoms; Medium-risk group: Monthly supportive psychotherapy and regular community social function training, focusing on improving apathy and social withdrawal symptoms; Low-risk group: Routine mental health education and continuous monthly symptom monitoring, with regular evaluation of negative symptom fluctuations. Whole-process family intervention measures included monthly family psychoeducation and real-time symptom follow-up via WeChat mini-program, to maintain continuous intervention and reduce negative symptom deterioration caused by insufficient family support.

Outcome Measures

Primary Outcomes

Change in global functioning score at 1-year follow-up;

Overall treatment adherence rate, defined as medication adherence ≥80% and psychotherapy attendance ≥80%.

Secondary Outcomes

1-year conversion rate to overt psychosis, diagnosed by unified clinical diagnostic criteria; Baseline attenuated negative symptom severity and its predictive effect on psychosis conversion (core secondary indicator of this study). Attenuated negative symptoms were classified into mild, moderate and severe subgroups according to SPNS total scores for subsequent multivariate regression and ROC curve analysis to verify its independent predictive value for 1-year psychosis transition.

Results

Screening Process and Patient Exclusion

A two-stage community-hospital clinical screening workflow was adopted for subject enrollment, which demonstrated good operability in real-world primary psychiatric services. A total of 29 candidates were excluded during secondary hospital re-evaluation, with detailed exclusion causes shown in Table 1. The most common reason for exclusion was substandard scale scores without obvious prodromal psychotic risk (41.4%), followed by organic brain diseases (27.6%), comorbid mental/substance disorders (20.7%), and refusal of follow-up assessment (10.3%).

Table 1.

Reasons for Participant Exclusion During Secondary Hospital Screening

Reasons for Exclusion	Number	Proportion (%)	Specific Description
Substandard scale scores	12	41.4	No prodromal psychotic risk identified on re-assessment
Organic diseases	8	27.6	Cerebral infarction, brain trauma sequelae, intracranial infection
Comorbidity interference	6	20.7	Substance dependence or other mental disorders requiring priority treatment
Loss to follow-up or refusal	3	10.3	Patients or family members refused further assessment

Baseline Characteristics of Study Participants

After propensity score matching with the additional covariate of baseline attenuated negative symptom score, demographic and clinical indicators were well balanced between the intervention group (n=150) and control group (n=75). No significant inter-group differences were observed in age, gender composition, years of education, baseline risk score, baseline global functioning score, childhood trauma history, and baseline SPNS negative symptom total score (all P>0.05, Table 2). The balanced baseline negative symptom levels eliminated confounding bias for subsequent predictive analysis of negative symptoms.

Table 2.

Baseline Characteristics of Study Participants

Characteristic	Intervention Group (n=150)	Control Group (n=75)	P-value
Age (years), mean ± SD	23.8 ± 4.7	23.5 ± 4.9	0.652
Male, n (%)	87 (58.0)	42 (56.0)	0.721
Years of education, mean ± SD	11.3 ± 2.4	11.1 ± 2.6	0.583
Baseline risk score, mean ± SD	4.0 ± 1.4	4.1 ± 1.3	0.617
Baseline global functioning score, mean ± SD	58.0 ± 8.7	58.5 ± 8.3	0.684
Childhood trauma history, n (%)	36 (24.0)	17 (22.7)	0.796

Note: All P-values > 0.05, indicating balanced baseline characteristics between groups.

Baseline Characteristics of the Risk Model Development Cohort

The random forest prediction model was established based on a 2018–2020 retrospective cohort including 112 CHR patients. Demographic, functional and symptomatic baseline data of the modeling cohort are summarized in Table 3. Attenuated negative symptom score was identified as one of the top three high-weight predictive features during model training.

Table 3.

Baseline Characteristics of the Risk Model Development Cohort

Characteristic	Value
Age (years), mean ± SD	23.6 ± 4.8
Gender (male/female), n	62/50
Years of education, mean ± SD	11.2 ± 2.5
Baseline risk score, mean ± SD	4.1 ± 1.3
Baseline functional score, mean ± SD	58.3 ± 8.5
Previous mental health history, n (%)	18 (16.1%)

Performance of the Risk Prediction Model

The random-forest model exhibited satisfactory predictive efficacy for 1-year psychosis conversion in the training dataset, with an AUC of 0.84 (95% CI: 0.76–0.92). For this training dataset, the model yielded a sensitivity of 0.81 and a specificity of 0.74. Feature-importance analysis revealed that attenuated negative-symptom severity possessed the second-highest feature weight among all 12 input predictors, confirming its critical contribution for identifying patients at high conversion risk. External temporal validation was subsequently performed using the independent sample of 38 CHR participants. Owing to the limited size of this external validation sample, external performance metrics should be interpreted with caution. In external temporal validation, the model achieved an AUC of 0.77, with sensitivity = 0.76 and specificity = 0.68. All enrolled CHR patients in the present study were successfully divided into low-, medium-, and high-risk subgroups according to the model-predicted 1-year conversion probability. It should be noted that the relatively small external validation sample may limit the stability and generalizability of these external performance estimates.

Stratified Clinical Management Distribution

All 150 patients in the intervention group were assigned to three risk tiers and received corresponding targeted interventions (Table 4): 42 high-risk patients with combined low-dose aripiprazole and biweekly psychological intervention, 68 medium-risk patients with monthly supportive psychotherapy and community functional training, and 40 low-risk patients with regular mental health education and symptom monitoring. Targeted intervention for negative symptoms was implemented across all risk subgroups.

Table 4.

Stratified Clinical Management Protocols for CHR Participants According to Psychosis-Conversion Risk

Risk Group	Sample Size (n)	Core management Content	Intervention Details
High-risk	42	Pharmacological plus integrated psychological intervention	Low-dose aripiprazole (2–5 mg/day); biweekly integrated psychological intervention, with targeted counseling for avolition, anhedonia and social apathy; monthly family psychoeducation via WeChat mini-program
Medium-risk	68	Supportive psychotherapy and community social-function training	Monthly supportive psychotherapy; regular community-based social-rehabilitation training targeting social withdrawal; monthly family symptom follow-up
Low-risk	40	Mental-health education and ongoing symptom monitoring	Routine mental-health education; monthly symptom and functional assessment; regular monitoring for fluctuation of attenuated positive and negative symptoms

Abbreviation: CHR, clinical high‑risk for psychosis.

Primary Outcome Indicators

Changes in Global Functioning

At the 1-year follow-up endpoint, global functioning improvement was significantly greater in the intervention group than the control group. The mean increment in global functioning score was +9.8 (95% CI: 8.3–11.3) in the intervention group versus +4.0 (95% CI: 2.4–5.6) in the control group, with a statistically significant inter-group difference (P < 0.001, Figure 1). Subgroup analysis further indicated that patients with severe baseline attenuated negative symptoms obtained the most prominent functional improvement after tiered stratified management.

Figure 1.

Mean change in global functioning score from baseline to the 12-month follow-up. (delta score: follow-up score minus baseline score) for the intervention and control groups. Bar height represents the mean change of global functioning score; error bars denote standard deviation (SD). The mean score change was 9.8 in the intervention group and 4.0 in the control group. The intervention group demonstrated significantly greater improvement in global functioning relative to the control group (P < 0.001).

Treatment Adherence

The overall 1-year treatment adherence rate reached 78% (117/150, 95% CI: 70.8–84.0%) in the intervention group, which was substantially higher than the 40% (30/75, 95% CI:28.9–51.8%) observed in the control group (P < 0.001). Family-participated WeChat mini-program follow-up and regular psychoeducation significantly reduced study dropout rates, especially for CHR patients with prominent negative symptoms and poor self-management ability.

Secondary Outcome Indicators

One-Year Psychosis Conversion Rate

The overall 1-year psychosis conversion rate was 9.3% (14/150, 95% CI:5.2–14.9%) in the intervention group, compared with 20.0% (15/75, 95% CI:11.6–30.8%) in the control group (P = 0.021). Subgroup comparison among high-risk patients showed a conversion rate of 13.6% (5/42, 95% CI:5.1–27.3%) in the intervention group versus 32.5% (13/40, 95% CI:18.6–49.1%) in the control group (P = 0.039).

Attenuated Negative Symptoms as Independent Predictors of Psychosis Conversion

Multivariate binary logistic regression was performed after adjusting for age, gender, baseline risk score and global functioning score. Severe attenuated negative symptoms were confirmed as an independent predictive factor associated with 1-year psychosis conversion (OR = 2.71, 95% CI: 1.43–5.14, P = 0.002). ROC curve analysis showed that the AUC of baseline SPNS attenuated negative-symptom total score for predicting psychosis conversion was 0.79 (95% CI: 0.71–0.87), with an optimal cut-off score identified to distinguish individuals at high conversion risk. Stratified analysis demonstrated that conversion risk rose progressively with increasing severity of attenuated negative symptoms, independent of the original risk-stratification grouping assigned by the random-forest model.

Follow-Up and Data Quality

The 1-year follow-up completion rate was 92% for the intervention group and 88% for the control group. No significant difference in follow-up loss was found between patients with mild/moderate/severe negative symptoms (P>0.05). All collected clinical data met standardized quality control criteria for retrospective real-world cohort research.

Discussion

This retrospective propensity-matched real-world cohort study systematically validated the clinical benefits of risk stratification combined with tiered intervention for Chinese clinical high-risk (CHR) individuals of psychosis, and further confirmed attenuated negative symptoms as an independent predictive factor for 1-year psychosis conversion.¹¹ Consistent with our primary research hypothesis, this integrated stratified-management program was associated with measurable improvements in study outcomes observed in our cohort: the 1-year psychosis conversion rate was 9.3% in the intervention group versus 20.0% in the propensity-matched control group, global functioning increased by a mean of 9.8 points in the intervention group compared with 4.0 points in controls, and treatment adherence reached 78% for the intervention group relative to 40% for the control group. These findings fill the gap of localized real-world evidence for standardized early intervention in domestic grassroots psychiatric settings, and supplement quantitative prediction indicators centered on attenuated negative symptoms that have long been neglected in domestic risk-assessment systems.¹²

Most existing Western CHR risk calculators and stratified-intervention frameworks are developed for well-resourced tertiary psychiatric hospitals in high-income countries, which cannot be directly replicated in Chinese grassroots primary care with limited psychiatrist staffing, insufficient community-screening capacity and uneven patient access to professional mental-health services. Against this background, this study innovatively built a two-stage community-hospital screening workflow paired with family-participated tiered intervention, which fully adapts to China’s primary medical-service characteristics. Although our data were collected from a single municipal psychiatric hospital, this low-cost, standardized whole-process management scheme has demonstrated feasibility within our setting and may have replicability for grassroots institutions nationwide and other low- and middle-income regions with similar resource constraints.

Our observed 9.3% one-year psychosis conversion rate among participants receiving stratified tiered intervention was markedly lower than the 20.0% conversion rate seen under conventional unified care in our matched control group; this outcome is also favorable compared with the 15–25% conversion range reported in most Western single-arm CHR cohort studies.¹³ The clinical outcomes observed in our cohort may be attributable to two key design features of our management protocol. On the one hand, the two-stage screening model realized early identification of at-risk adolescents at the grassroots level within our local setting, expanding access to early preventive intervention for prodromal psychosis populations in resource-limited areas.¹⁴ On the other hand, our random-forest risk-stratification model (training-set AUC = 0.84, external temporal-validation AUC = 0.77) discriminated low-, medium-, and high-conversion-risk subgroups in our sample, which helped avoid unnecessary antipsychotic exposure among low-risk CHR patients and delivered targeted intensified intervention for high-risk individuals.¹⁵ We note that for high-risk CHR participants in our cohort, low-dose aripiprazole was delivered as one component of a comprehensive multi-modal intervention package together with biweekly psychological support and family-based measures. Given the observational nature of this study, we cannot isolate the independent therapeutic effect of aripiprazole alone. Unlike traditional risk-evaluation tools that mainly focus on attenuated positive symptoms, our prediction model incorporated attenuated negative-symptom severity as a high-weight predictive feature, which improved risk-stratification performance in our dataset.¹⁶ Compared with previous domestic CHR studies that separately explored symptom predictors or machine-learning algorithms, this work integrated SPNS negative-symptom scoring, data-driven risk stratification, and differentiated tiered intervention, forming a complete early-prevention workflow suitable for Chinese grassroots clinical scenarios.

In terms of functional recovery, within our cohort, the intervention group obtained a mean global-functioning improvement of 9.8 points at 12-month follow-up, compared with a 4.0-point increase in the control group. Subgroup analysis limited to our study sample further revealed that CHR patients with severe baseline attenuated negative symptoms achieved the greatest functional gains following stratified targeted management. This result supports the mainstream clinical consensus that early intervention for prodromal populations should prioritize social-function recovery, rather than only focus on reducing psychosis-conversion risk.¹⁷ Our tiered-management protocol set individualized intervention targets for negative symptoms across all risk tiers in our intervention cohort: high-risk patients received biweekly psychological counseling targeting avolition and anhedonia, and medium-risk participants took part in regular community-social-rehabilitation training to relieve persistent social withdrawal. This differentiated intervention intensity balanced allocation of limited mental-health resources and symptom relief for negative-symptom domains that responded poorly to routine supportive care in our control arm.¹⁸

Treatment adherence in our intervention group reached 78%, nearly double the 40% adherence rate observed among patients receiving conventional unified care in this study. Sustained treatment engagement is considered a fundamental prerequisite for reducing psychosis-transition risk. We implemented family-participated measures throughout the intervention, including monthly standardized family psychoeducation and real-time symptom follow-up via WeChat mini-programs. These approaches built family-support systems for participants in our intervention group, compensated for poor self-management capacity among CHR patients dominated by apathy and social isolation, and reduced study dropout.¹⁹ Our real-world findings are consistent with prior domestic cohort reports indicating that family-engaged full-cycle management improves treatment persistence among psychiatric patients.

Our secondary analysis, adjusted for multiple demographic and clinical confounders in our sample, identified severe attenuated negative symptoms as an independent predictive factor associated with 1-year psychosis conversion (OR = 2.71, 95% CI 1.43–5.14; ROC-AUC = 0.79). Large-scale Western cohort studies have reported links between persistent prodromal negative symptoms and poor clinical prognosis and elevated conversion risk, yet large-sample localized data from Chinese CHR youth remain limited.²⁰ At present, most domestic routine CHR screening systems still prioritize attenuated positive symptoms for risk judgment, while the independent predictive value of attenuated negative symptoms is often overlooked.²¹ Based on our quantitative results derived from this cohort, we recommend embedding standardized SPNS negative-symptom assessment into routine CHR risk evaluation workflows. Patients presenting moderate-to-severe attenuated negative symptoms should be flagged for intensified monitoring regardless of initial risk-stratification results, to enable timely targeted early-intervention.

The random-forest risk-stratification model constructed in this study achieved acceptable predictive performance in both training and external temporal-validation datasets within our sample, demonstrating that machine-learning-assisted risk stratification can be implemented in grassroots mental-health institutions and aligns with the general trend of precision-based individualized psychiatric intervention.²² Most internationally available psychosis-risk calculators are trained on Western datasets and have limited adaptability to Chinese demographic characteristics, grassroots-screening conditions and local prodromal-symptom presentations. Our locally developed model partly addresses this gap for Chinese primary-care settings.²³

Several inherent limitations of this retrospective real-world propensity-matched cohort should be acknowledged when interpreting our findings. First, this study was conducted within a single municipal psychiatric hospital, which may introduce selection bias and limit the external generalizability of our two-stage screening workflow and tiered-intervention model to regions with different mental-health-resource distributions. Multi-center prospective cohort studies across diverse geographical regions in China are required for further external validation of our screening and intervention procedures.²⁴ Second, our follow-up was restricted to a 12-month observation window; therefore, we cannot evaluate the long-term durability of tiered-intervention benefits or the long-term predictive stability of attenuated negative symptoms for psychosis conversion. Prior long-term CHR follow-up research has demonstrated that the predictive performance of prodromal symptoms may change over extended follow-up periods.²⁵ Third, the external temporal validation sample for our random-forest risk-prediction model was relatively small (n = 38), which may compromise the stability of model parameters. Although repeated internal sensitivity analyses showed that the model AUC fluctuated stably between 0.81–0.86 under resampling, larger multi-site independent datasets are still needed to optimize model cut-off thresholds and feature weights. Fourth, as an observational retrospective study, we cannot establish causal relationships between attenuated negative symptoms and psychosis onset; our findings only support that attenuated negative symptoms act as an independent predictive association factor for 1-year psychosis conversion. Despite propensity-score matching and multivariate regression adjusting for multiple known confounders, residual confounding cannot be fully excluded. Unmeasured variables such as variations in family support quality, subtle differences in the intensity of routine clinical care, patients’ social-environmental stressors, and unrecorded informal psychological support may still influence psychosis-conversion risk, functional outcomes and treatment adherence. Finally, this study only adopted clinical rating scales to quantify attenuated negative-symptom severity. Neuroimaging indicators or peripheral biological markers were not incorporated to interpret potential neurobiological mechanisms linking attenuated negative symptoms to psychosis transition; translational mechanistic research will be needed in future investigations.²⁶

In conclusion, risk stratification paired with tiered individualized intervention can effectively lower 1-year psychosis conversion risk, improve global social functioning, and boost long-term treatment adherence among Chinese CHR youth. Attenuated negative symptoms are reliable independent short-term predictors of psychosis transition and deserve standardized incorporation into routine CHR clinical risk assessment to optimize targeted early intervention strategies. This low-cost, operable whole-process management scheme is suitable for widespread promotion in grassroots psychiatric institutions with limited medical resources, and provides localized real-world clinical evidence for early psychosis prevention in China and other low- and middle-income countries with similar primary care structures.²⁷

Conclusions

This integrated management model combining risk stratification and tiered individualized intervention shows promising effectiveness, feasibility and patient-centered benefits for Chinese individuals at clinical high-risk (CHR) for psychosis within this real-world cohort. This stratified-care strategy is associated with reduced 1-year risk of transition to overt psychosis, improved global social functioning, and enhanced treatment adherence among CHR patients.

Attenuated negative symptoms represent an independent and reliable short-term predictive marker for psychosis conversion in our sample. Clinical psychiatrists should incorporate standardized negative-symptom assessment into routine CHR risk-screening workflows and strengthen targeted monitoring and intervention for patients with moderate-to-severe negative symptoms regardless of initial risk-stratification results.

The whole workflow has standardized operating procedures and low implementation costs, and is potentially replicable in grassroots psychiatric institutions with limited medical-resource settings. Nevertheless, these findings are derived from a retrospective, single-center dataset, and the external temporal-validation sample for our random-forest prediction model was small. Generalization to broader populations should be made with caution, and multi-site investigations are required to further validate our intervention scheme and risk-prediction model.

Funding Statement

Risk Model for High-Risk Conversion in Psychiatric Disorders: Joint Prediction of Clinical Symptoms, Psychological Characteristics, Neurocognitive and Imaging Features; Wenzhou Science and Techonology Project, NO. Y20240631.

Data Sharing Statement

The original data can be provided by the corresponding author as required.

Ethics

Ethics Committee of Wenzhou Seventh People’s Hospital, with the ethics number: EC-KY-2024010

Author Contributions

All authors made a significant contribution to the conception, study design, execution, data acquisition, analysis and interpretation of the work; participated in drafting, revising or critically reviewing the manuscript; approved the final version for publication; agreed on the journal for submission; and accept accountability for all aspects of the work.

Disclosure

The authors declare that they have no conflicts of interest.