高中生心理健康的异质性模式及其与焦虑抑郁的关联
Heterogeneous Patterns of Mental Health and Their Associations with Anxiety and Depression in High School Students
DOI: 10.12677/ap.2026.161040, PDF, HTML, XML,   
作者: 吴纪林*:西南医科大学人文与管理学院,四川 泸州;西南医科大学附属临床医院精神科,四川 泸州;陈 晶#:西南医科大学附属临床医院精神科,四川 泸州
关键词: 高中生心理健康异质性亚型困难与长处潜剖面分析症状网络分析随机森林模型High School Student Mental Health Heterogeneous Subtypes Strengths and Difficulties Latent Profile Analysis Symptom Network Analysis Random Forest Model
摘要: 目的:探究高中生内化/外化问题与亲社会行为的潜在剖面亚型及其与情绪症状的症状网络关联,并检验网络特征预测剖面归属的效能。方法:以2844名在读高中生为样本,采用长处与困难问卷、广泛性焦虑量表和9项患者健康问卷评估受试的心理状况,并运用潜剖面分析、症状网络分析及随机森林模型进行统计分析。结果:识别出“高亲社会–适应型”(47.12%)、“多动–情绪问题型”(34.28%)和“品行–同伴问题型”(18.60%)三类亚型,问题型剖面的情绪症状水平显著高于适应性剖面,良好亲子关系可降低归入问题型剖面的风险。各剖面在网络结构上存在显著差异,总体网络中情绪症状、品行问题、静坐困难和情绪低落为核心节点,亲社会行为在各网络中均呈保护性嵌入,在适应型剖面中作用最强。基于网络拓扑特征的随机森林模型预测剖面归属的准确率为86.2%,桥接特征与节点中心性贡献最大。结论:本研究揭示了高中生心理健康的质性差异及其症状网络的异质性结构,为发展精准、适龄的心理健康评估和干预提供实证依据。
Abstract: Objective: To explore the latent profile subtypes of internalizing/externalizing problems and prosocial behavior among high school students, their symptom network associations with emotional symptoms, and to examine the efficacy of network characteristics in predicting profile membership. Methods: A sample of 2844 high school students was assessed using the Strengths and Difficulties Questionnaire, the Generalized Anxiety Disorder Scale, and the 9-item Patient Health Questionnaire to evaluate psychological conditions. Statistical analyses included latent profile analysis, symptom network analysis, and random forest modeling. Results: Three subtypes were identified: “High Prosocial-Adaptive” (47.12%), “Hyperactivity-Emotional Problems” (34.28%), and “Conduct-Peer Problems” (18.60%). Problematic profiles exhibited significantly higher levels of emotional symptoms than the adaptive profile. Positive parent-child relationships were found to reduce the risk of belonging to a problematic profile. The network structures differed significantly across profiles. In the overall network, emotional symptoms, conduct problems, difficulty sitting still, and depressed mood served as central nodes, while prosocial behavior was protectively embedded in all networks, with the strongest effect observed in the adaptive profile. A random forest model based on network topology features achieved an accuracy of 86.2% in predicting profile membership, with bridging features and node centrality contributing the most. Conclusion: This study reveals qualitative differences in the mental health of high school students and heterogeneous structures in their symptom networks, providing empirical evidence for developing precise, age-appropriate mental health assessments and interventions.
文章引用:吴纪林, 陈晶 (2026). 高中生心理健康的异质性模式及其与焦虑抑郁的关联. 心理学进展, 16(1), 333-346. https://doi.org/10.12677/ap.2026.161040

1. 引言

在青少年群体中,高中生处于向成年初期过渡的关键时期,其心理发展兼具特殊性和敏感性(Tervo-Clemmens et al., 2023)。他们不仅承受着学业压力,还需应对身心成熟和同伴关系重组等多重发展挑战(Kwon & Telzer, 2022)。前人研究指出,高中生抑郁症状的检出率显著高于初中生,提示该群体在心理健康方面的特殊风险(Tang et al., 2019)。基于“长处和困难”视角的研究表明,在“困难”维度的高分者,常表现出更严重的负性情绪症状与学习功能受损(Yu et al., 2022)。高困难且低亲社会的学生更易陷入“情绪–行为–学业”相互加剧的恶性循环,而高亲社会行为则与更稳定的情绪状态和功能恢复相关联(Byrne et al., 2023)。同时,亲子关系、学业负担等情境因素,也与情绪、行为问题密切相关(Li et al., 2023)。由此可见,高中生的心理社会困境可能呈现多重累积与交互加剧的特征(Kwon & Telzer, 2022)。因此,对高中生开展涵盖心理困境与积极资源的多维评估,将有助于全面评估其心理适应水平,为分级预防和干预提供科学依据。

现有变量中心范式难以揭示高中生心理健康的系统性与群体异质性,依赖症状的均值和总分划界分析,可能忽视了症状组合的质异性和对症状互动模式的洞察(Cai et al., 2022; van Loon et al., 2022; Wolke et al., 2025)。因此,有必要区分具有不同心理困境与资源优势的典型亚群体,并系统探究其核心症状间的关联模式,以解答“如何干预最大化”的关键问题。潜剖面分析、症状网络理论及机器学习的方法为构建多维理解框架、系统解析心理健康的复杂本质提供了重要契机(Sun et al., 2025; Zavlis, 2024)。潜剖面分析能在多个指标的综合基础上识别出具有相似特征的潜在亚群,为揭示群体内部的异质性和后续的分层研究、干预奠定结构化基础(王雨薇等,2025)。症状网络理论将心理问题视为症状互动的动态系统,通过识别高中心性与桥接症状,以定位干预靶点和解析互动模式(Schumacher et al., 2023)。随机森林模型在处理大样本数据时优势明显:其不仅通过交叉验证提升稳健性、筛选关键预测特征,还可作为连接网络分析与潜剖面分析的工具,验证网络特征对风险亚型的预测效能(de Lacy et al., 2023)。

基于此,本研究通过整合潜剖面分析、症状网络估计与随机森林模型方法,构建一个“识别风险亚群–解析症状结构–特征验证”的递进式研究框架,为理解高中生心理问题的复杂机制及发展精准干预策略,提供一条从描述到机制,再到验证的闭环证据链。

2. 对象与方法

2.1. 研究对象

采用便利取样法,对四川省某地一所普通高中的在校学生进行在线问卷调查。共回收有效问卷2844份,有效率为98.59%。有效样本中,女生1538人(54.08%),男生1306人(45.92%);高一、高二、高三年级学生分别为911人、908人和1025人。所有参与者均在线阅读并自愿签署了电子版知情同意书,并报告了年龄、性别和亲子关系质量等信息。

2.2. 研究工具

2.2.1. 长处与困难问卷

长处与困难问卷(Strengths and Difficulties Questionnaire, SDQ)学生版评估受试的情绪与行为问题的特征,中文版具有良好的信效度 (许文兵等,2019)。本量表共25个条目,分为5个子量表。其中情绪困扰、品行问题、多动/注意、同伴问题等构成“困难”维度,高分表示适应问题突出;亲社会行为则为“长处”维度,高分反映积极行为良好。条目均为0~2分三级评分,第7、11、14、21、25题为反向计分。在社区样本中,品行问题与同伴问题子量表的心理测量学特性欠佳,采用“内化问题”、“外化问题”等二阶维度更具优势(Goodman, Lamping, & Ploubidis, 2010)。据此,本研究以亲社会行为、“内化问题”(情绪困扰 + 同伴问题)和“外化问题”(品行问题 + 多动/注意问题)的得分作为剖面指示变量,Cronbach’s α系数分别为0.804、0.693、0.714。

2.2.2. 广泛性焦虑症状量表

中文版广泛性焦虑量表(Generalized Anxiety Disorder-7, GAD-7)评估受试近两周的焦虑症状。该量表共7个条目,按“完全没有”至“几乎每天”4级评分(0~3分),总分范围为0~21分,得分越高表示焦虑症状越严重。既有研究表明,中文版GAD-7在青少年群体中具有良好的信效度,是评估焦虑水平的有效工具(Chen et al., 2021)。本研究中,该量表的Cronbach’s α系数为0.924。

2.2.3. 9项患者健康问卷

中文版9项患者健康问卷(Patient Health Questionnaire-9, PHQ-9)评估受试近两周的抑郁症状。该量表共9个条目,分别对应抑郁障碍的一种核心症状,采用0 (“完全没有”)至3 (“几乎每天”)的4级计分。总分范围为0~27分,得分越高表明抑郁症状越严重。该量表具有良好的信效度与敏感度,适用于抑郁症状的筛查与严重程度评估(Barry et al., 2023)。本研究中,该量表的Cronbach’s α系数为0.897。

2.3. 统计分析

常规统计分析在SPSS 27.0中完成,潜剖面分析在Mplus 8.3中进行,症状网络与随机森林分析在R 4.5.2中完成(主要使用bootnet、qgraph、networktools、NetworkComparisonTest、mgm、caret、pROC、randomForest等包)。采用Pearson相关检验SDQ各维度与GAD、PHQ的关联。综合AIC、BIC、aBIC、熵、LMR、BLRT及模型可解释性以确定最优剖面结构,并用R3STEP分析年龄、性别和亲子关系对剖面归属的预测,BCH法比较不同剖面的情绪症状差异(Rowe et al., 2025)。症状网络由扩展贝叶斯信息准则图模型(Extended Bayesian Information Criterion graphical lasso, EBICglasso)估计和正则化,并用qgraph包可视化。根据强度、预期影响力和桥接强度识别核心与桥接症状,利用bootnet和mgm评估中心性精确性、稳定性及节点可预测性(Kiakos et al., 2025)。采用网络比较检验(Network Comparison Test, NCT)和BH法校正,进行网络全局结构与强度的两两比较(Piazza et al., 2024)。基于网络拓扑特征(网络距离、节点中心性、桥接、网络聚类及相对位置特征)构建随机森林分类器,按7:3划分训练/测试集,设1000棵决策树并采用五折交叉验证评估泛化能力,模型性能以准确率、Kappa、AUC以及各类别的精确率、召回率和F1分数表征(Hill et al., 2025)。

3. 结果

3.1. 描述性统计及相关性分析

采取单因子法提取出6个特征值大于1的因子,最大因子解释率为30.82% (<40%),表明共同方法偏差不严重。相关分析(表1)表明:年龄与内化问题正相关,与焦虑、抑郁负相关;亲子关系与亲社会行为正相关,与其他变量均负相关;各心理症状间正相关,且均与亲社会行为负相关(ps < 0.05)。

Table 1. Descriptive statistics and correlation analysis of research variables (n = 2844)

1. 研究变量的描述性统计及相关分析(n = 2844)

变量

M ± SD

1

2

3

4

5

6

1) 年龄

16.280 ± 0.744

1

2) 亲子关系

1.830 ± 0.406

−0.004

1

3) 焦虑症状

6.170 ± 4.129

−0.047*

−0.295**

1

4) 抑郁症状

7.730 ± 5.013

−0.042*

−0.353**

0.772**

1

5) 外化问题

6.349 ± 2.950

−0.016

−0.258**

0.547**

0.556**

1

6) 内化问题

6.314 ± 3.254

0.038*

−0.267**

0.641**

0.623**

0.649**

1

7) 亲社会行为

6.548 ± 2.236

−0.034

0.133**

−0.073**

−0.127**

−0.359**

−0.249**

注:*p < 0.05,**p < 0.01。

3.2. 潜剖面分析

潜剖面分析支持将样本划分为3个类别(图1(A)):AIC、BIC与aBIC在3类后下降趋缓,3类模型的熵值最高(0.940),LMR与BLRT检验显著。该模型分类准确性高(平均归属概率 > 0.96,图1(B))。根据三类群体在内化、外化及亲社会行为上的特征(图1(C)),分别命名为“高亲社会–适应型”(47.12%)、“多动–情绪问题型”(34.28%)和“品行–同伴问题型”(18.60%)。

逻辑回归结果见表2。与“高亲社会–适应型”相比,良好亲子关系显著降低归属“多动–情绪问题型”(OR = 0.249)与“品行–同伴问题型”(OR = 0.219)的风险。男生归入“品行–同伴问题型”和“多动–情绪问题型”的风险分别为女性的2.390倍、0.739倍。年龄增加显著降低归入“多动–情绪问题型”(风险降低12%)。如图2所示,各剖面在焦虑和抑郁症状水平上均存在显著差异(p < 0.001):“高亲社会–适应型”水平均较低,“多动–情绪问题型”更为突出,而“品行–同伴问题型”则处于中间水平。

Figure 1. Fit indices of latent profile analysis for strengths and difficulties among senior high school students (n = 2844)

1. 高中生长处与困难的潜在剖面模型拟合结果(n = 2844)

Figure 2. Difference test results of anxiety and depression symptoms across latent profile groups

2. 不同剖面人群在焦虑和抑郁症状上的差异检验结果

Table 2. Multivariate logistic regression analysis of different subgroups

2. 不同剖面的多因素逻辑回归分析

预测变量

Class 2 (多动–情绪问题型)

Class 3 (品行–同伴问题型)

B (SE)

OR

B (SE)

OR

亲子关系

−1.391 (0.150)***

0.249

−1.521 (0.156)***

0.219

年龄

−0.128 (0.064)*

0.880

0.094 (0.071)

1.098

性别(女生 = 0)

−0.302 (0.095)**

0.739

0.871 (0.111)***

2.390

注:对照组 = Class 1 (高亲社会–适应型),OR = odd rations;*p < 0.05,**p < 0.01,***p < 0.001。

3.3. 心理症状网络分析

3.3.1. 网络估计及网络比较检验

症状网络估计如图3所示,总样本及各剖面的平均边权重分别为0.041、0.036、0.039、0.040。总样本网络中,GAD与PHQ条目形成密集连接簇,并与内外化问题症状正相关;亲社会行为与多数的问题行为负相关。风险剖面(Class 2/3)表现出更强的症状间连接与更弱的亲社会保护作用;而适应型剖面(Class 1)中亲社会行为的广泛负向关联更为突出。网络比较检验(表3)表明,三类网络在全局结构上均存在显著差异(p < 0.05);在全局强度上,仅Class 3显著强于Class 1 (S = 1.350, p = 0.030)。

Figure 3. Network models of anxiety, depression, and strengths and difficulties symptoms in senior high school students

3. 高中生焦虑、抑郁、长处和困难症状的网络模型图

Table 3. Results of network comparative test for profile subtypes

3. 剖面亚型的网络比较检验结果

网络比较

M

M-p

全局强度1

全局强度2

S

S-p

Class 1 vs Class 2

0.220

0.015

8.492

8.996

0.504

0.158

Class 1 vs Class 3

0.233

0.040

8.492

9.843

1.350

0.030

Class 2 vs Class 3

0.268

0.015

8.996

9.843

0.847

0.158

注:M和S统计量分别为两个网络间的最大边权、整体连通强度差异,以检验整体结构和强度差异;M-p值和S-p值为BH矫正后的结果。

3.3.2. 中心性、桥接性、节点可预测性及稳定性

表4所示,各网络强度中心性、期望中心性及桥接强度的CS均>0.50,估计较为稳定;除总样本外,中介中心性CS均<0.50,故不作解释。

Table 4. Table of CS coefficients for centrality metrics and bridging metrics

4. 中心性指标和桥接性指标的CS系数表

网络

强度中心性

期望中心性

桥接强度

中介中心性

Total Sample

0.750

0.750

0.750

0.750

Class 1

0.750

0.750

0.750

0.439

Class 2

0.750

0.750

0.672

0.283

Class 3

0.673

0.750

0.750

0.049

注:CS > 0.7,优秀;CS > 0.5,良好;CS > 0.25,可接受。

图4显示,总样本网络中EM (情绪困扰,Z = 1.35)与CO (品行问题,Z = 1.31)为强度最高节点,G5 (静坐困难)和P2 (情绪低落)亦较突出(Z > 1.00),其中EM也为最高正向期望中心性(Z = 1.20)。Class 1、Class 2、Class 3中最强的强度节点分别为G5 (Z = 1.33)、G1 (紧张,Z = 1.38)和G3 (过度担心,Z = 1.42),最大正向期望中心性节点依次为P4 (精力疲乏,Z = 1.20)、G1 (Z = 0.97)和G2 (控制不住担心,Z = 1.06)。跨剖面可见,G1、G2、P4在至少两个剖面中反复呈现较高中心性,可视为相对稳定的共性关键情绪症状;PS (亲社会行为)在各网络中期望中心性均为最强负值(Z约−2.45~−3.19),在网络中稳定表现为保护性节点。图5显示,CO为总样本(Z = 2.12)和Class 3 (Z = 2.51)桥接强度的最高节点,Class 1与Class 2中桥接强度最高节点分别为PS (Z = 2.29)和EM (Z = 2.46);各网络中EM为1-step与2-step桥接预期影响力最高的正向节点,PS则在两类指标上均为最强负向节点。节点可预测性见图6:总体网络中,GAD各条目及PHQ的P4、P2、P6 (自我评价低)、P7 (注意力问题),以及EM、HY(多动/注意)的可预测性均>50%;Class 1各节点可预测性整体偏低(<50%),Class 2中G2、G3、G4、G1可预测性较高,Class 3中GAD各条目及P7、P2、P6、P4均表现出较高可预测性,EM亦超过50%;G2的可预测性在所有网络中均最高。

3.4. 随机森林

随机森林分类结果见图6。ROC曲线(图7(A))显示,三类潜在剖面的判别效能较好,Class 1、Class 2和Class 3的一对多AUC分别为0.983 (Z = 133.13, p < 0.001)、0.944 (Z = 60.14, p < 0.001)和0.921 (Z = 32.70, p < 0.001)。混淆矩阵(图7(B))表明模型总体准确率为86.2%,Kappa系数为0.775;其中,真实为

Figure 4. Centrality metrics of the network model

4. 网络模型中心性指标图

Figure 5. Bridging metrics of the network model

5. 网络模型桥接性指标图

Figure 6. Node predictability heatmap of the network model

6. 网络模型节点可预测性热图

Table 5. Classification performance metrics of random forest model based on 5-fold cross-validation

5. 随机森林模型五折交叉验证的分类性能指标

折次

训练集 样本数

测试集 样本数

准确率

Kappa 系数

宏平均 精确率

宏平均 召回率

宏平均 F1

宏平均AUC

1

1593

399

0.872

0.792

0.867

0.825

0.840

0.963

2

1594

398

0.887

0.818

0.873

0.847

0.857

0.973

3

1594

398

0.844

0.747

0.840

0.796

0.810

0.945

4

1593

399

0.865

0.780

0.868

0.817

0.833

0.949

5

1594

398

0.854

0.765

0.855

0.822

0.833

0.950

注:五折交叉验证将样本随机划分为5个子集,轮流以其中1折为测试集、其余4折为训练集,表中为各折及其平均的分类性能指标。

Figure 7. Prediction results of random forest model for latent profile classification based on network features

7. 基于网络特征对潜在剖面分类的随机森林模型预测结果

Class 1的个体中约95.5%被正确识别,Class 2和Class 3则分别约为84.2%、65.8%,错误分类主要发生在Class 2与Class 3之间。各剖面的F1分数、精确率和召回率均在0.70以上,Class 1最高,Class 3相对较低(图7(D))。特征重要性分析(图7(E))显示,桥接特征的平均重要性最高,其次为节点中心性特征和网络聚类特征,相对位置特征贡献最小;PS的节点中心性、内化及外化问题的桥接特征的特征重要性相对较高(图7(C))。具体变量层面(图7(F)),PS、G5、P7等指标对潜剖面分类贡献最大。五折交叉验证结果(表5)表明基于网络特征的潜剖面分类具有较高判别效能和泛化能力。

4. 讨论

本研究系统探讨了高中生的内外化问题和亲社会行为的异质性组合及其与情绪症状的症状网络互动关联模式,并验证了症状网络特征预测剖面归属的较高准确性与有效性。我们识别出“高亲社会–适应型”、“多动–情绪问题型”和“品行–同伴问题型”三类剖面亚型,它们在焦虑和抑郁症状负担、症状网络方面存在显著差异,剖面归属受到亲子关系等因素显著影响,亲社会行为的中心性及内外化问题的桥接能力是区分不同剖面的关键结构信息。

心理健康双因素模型指出,心理困扰与积极功能共同决定了整体适应状况(Magalhães, 2024)。“高亲社会–适应型”以高亲社会和低内/外化问题为特征,符合“心理健康”的类型;而两类问题型剖面则呈现脆弱性和资源的不同组合(Magalhães, 2024)。“多动–情绪问题型”群体虽具备较正常的亲社会倾向,但难以缓解突出的情绪困扰或多动/注意问题。这与前人研究一致,即多动/注意问题可能与负面情绪加剧相关联,二者可能削弱了亲社会行为的保护作用(Olivier et al., 2020; Reimann et al., 2025)。“品行–同伴问题型”则以低亲社会行为伴随突出的品行–同伴问题为特征,攻击性或冲突性行为等品行问题易破坏同伴关系、引发同伴排斥,并进一步限制亲社会行为的展现(Girard et al., 2024; Reimann et al., 2025)。相较而言,男生归入“品行–同伴问题型”的风险较高,而女生归入“多动–情绪问题型”的风险较高,这与既有研究揭示的性别分布模式一致(Barbieri et al., 2024)。随年龄增加,个体归入“多动–情绪问题型”的风险显著降低,这可能反映了青少年时期执行功能的成熟,也可能与选择性流失有关(即困扰更严重者更易在主流学习路径上被边缘化) (Folker et al., 2025; Ringbom et al., 2022)。此外,良好亲子关系降低了归入问题型剖面的风险,表明支持性亲子互动在青少年中晚期仍是关键的保护因素。前人研究证实,亲子温暖不仅能缓解内化症状、行为问题及同伴问题,还可提升情绪调节能力和亲社会行为(Isdahl-Troye et al., 2025; Zhou et al., 2024)。最后,问题型剖面的焦虑、抑郁水平显著高于适应性剖面,尤以“多动–情绪问题型”最为突出,支持了群体内部异质性对应不同心理适应轨迹的观点(Bista et al., 2025)。

症状网络分析表明,在全样本网络中情绪困扰与品行问题分别呈现最高中心性或桥接性,静坐困难、无法控制的担心、情绪低落、精力匮乏和自我评价低亦处于关键位置,与负性情感、认知反刍和行为失调等跨诊断过程相符(Bähr et al., 2025),也与将“悲伤、担忧、自我批评”视为核心节点的研究一致(Ma et al., 2024)。亲社会行为在各网络中均与品行、多动/注意和同伴问题呈稳定负相关,表明它作为一种结构性优势资源,削弱了外化与人际问题向情绪困扰的“传导”,为心理韧性提供潜在路径(Hysaj et al., 2023; Memmott-Elison & Toseeb, 2023)。在适应型剖面,亲社会行为与其他症状的连接更强、辐射更广,而在两类问题型剖面(尤其是“品行–同伴问题型”)中则明显“脱钩”,提示心理韧性取决于其在症状网络中的嵌入深度(Reimann et al., 2025)。网络比较结果证实三类剖面的全局结构不同,其中“品行–同伴问题型”的连接性显著强于适应型剖面。剖面桥接症状也呈现出特异模式:适应性剖面中,亲社会行为作为负向桥接枢纽,抑制外化、同伴与情绪问题之间的病理耦合(Memmott-Elison & Toseeb, 2023);“品行–同伴问题型”中,品行问题是最强桥接节点,与纵向证据中“持续行为问题与同伴困难预示后续内化障碍”的发现相符(Gong & Zhou, 2025);“多动–情绪问题型”中,情绪困扰成为主要连接注意力/行为失调与焦虑–抑郁的通道(Antony et al., 2022)。紧张、无法控制担心和精力缺乏在多个网络中均呈现较高中心性,与既往研究中“担忧与疲劳是青少年跨诊断网络核心症状”的结论相呼应(Liu et al., 2025; Lu et al., 2025)。两类问题型剖面中节点更易被邻近节点预测,说明其内部耦合更紧密、症状动态且更具自我维持性,更难被打断,更易转变为临床综合征(Kelley & Gillan, 2022)。

重要的是,随机森林模型结果表明,症状间的连接模式可作为可解释的风险与韧性标记,并能够较为准确地识别个体所属的潜在剖面类型(Zhang et al., 2025)。亲社会行为的中心性特征和内外化问题的桥接性特征是最具信息量的预测因子,这既支持了“症状网络的连接模式本身蕴含个体潜在风险表型和预后信息”的观点(Borsboom et al., 2021),也提示亲社会行为与其他症状节点的关联强度以及内外化问题的跨社区桥接能力是有效区分不同特征剖面的有效因素(Eagle et al., 2023; Jones, Ma, & McNally, 2021)。

更为重要的是,本研究为高中生人群的分层评估与干预提供了关键的实证依据。对于亲社会–适应型学生,应维持并丰富支持性环境,以巩固现有的良好适应状态(Phan et al., 2022)。对于“多动–情绪问题型”,则宜侧重注意力调节、学业应对和焦虑/反刍思维等方面,例如通过认知行为疗法、正念训练等,以缓解学业困境与情绪困扰(Jeppesen et al., 2021)。对于“品行–同伴问题型”,则应侧重于重建亲社会规范、矫正冲突性行为和修复同伴关系,如社交技能训练、亲社会行为实践以及强化家校协作,以打破“攻击–排斥–孤立”的恶性循环(Jeppesen et al., 2021)。鉴于紧张、无法控制的担心和精力缺乏的跨剖面网络高中心性,以放松训练、压力管理与疲劳恢复为核心的干预也具有较高的跨剖面适用性(Juhász et al., 2024)。亲子关系所展现出的稳健保护效应进一步表明,将家庭纳入干预目标、发展以家庭为单位的支持性策略,应成为学校心理健康服务体系中的关键组成部分(Buehler, 2020)。

综上所述,高中生的心理健康可由三种潜剖面类型有效刻画,这些模式不仅在长处与困难的平衡上不同,也在症状网络结构上存在系统差异。亲社会行为作为结构性嵌入的保护性节点,而持续担忧、精力缺乏、情绪困扰和品行问题则是连接内化与外化领域的核心或桥接症状。基于网络特征构建的随机森林模型能够较准确地区分三类模式,提示这些网络特征有望成为机制可解释的风险分层标志物。通过整合潜在类别分析、症状网络与预测建模,构建了一个框架,用于识别异质性风险表型、探究亚剖面症状网络关联模式,并为更精准、更符合高中生发展阶段的心理健康评估和干预提供实证的证据。

本研究亦有若干局限。横断面设计无法推断因果,紧密连接结构也可能部分由未测量共同因素所致。数据来自同一地区的自陈问卷,存在共同方法偏差与情境局限,需在多信息源、多中心样本中加以验证。网络结果依赖EBICglasso等建模选择及症状纳入范围,若将睡眠、欺凌等重要维度纳入,网络拓扑与潜在模式可能发生变化。尽管采用交叉验证,随机森林模型仍在同一样本上开发与测试,尚需在独立队列中复现,以检验其泛化能力。

NOTES

*第一作者。

#通讯作者。

参考文献

[1] 王雨薇, 陈晓彤, 余红玉(2025). 青少年亲社会行为潜在类别及其与生命意义感的关系. 中国心理卫生杂志, 39(7), 624-630.[CrossRef
[2] 许文兵, 王孟成, 邓嘉欣, 刘惠华, 曾红, 杨文登(2019). 长处和困难问卷中文版的信度概化分析. 中国临床心理学杂志, 27(1), 67-72.[CrossRef
[3] Antony, E. M., Pihlajamäki, M., Speyer, L. G., & Murray, A. L. (2022). Does Emotion Dysregulation Mediate the Association between ADHD Symptoms and Internalizing Problems? A Longitudinal Within-Person Analysis in a Large Population‐Representative Study. Journal of Child Psychology and Psychiatry, 63, 1583-1590.[CrossRef] [PubMed]
[4] Bähr, C., Phan, D., Murawska, N., Gerber, J., Jordan, A., Hoffmann, K. et al. (2025). Together and Beyond: A Systematic Review on Characteristics and Efficacy of Transdiagnostic Psychotherapeutic Group-Based Interventions for Children and Adolescents. Clinical Psychology Review, 121, Article ID: 102642.[CrossRef
[5] Barbieri, V., Piccoliori, G., Engl, A., & Wiedermann, C. J. (2024). Parental Mental Health, Gender, and Lifestyle Effects on Post-Pandemic Child and Adolescent Psychosocial Problems: A Cross-Sectional Survey in Northern Italy. International Journal of Environmental Research and Public Health, 21, Article 933.[CrossRef] [PubMed]
[6] Barry, M. J., Nicholson, W. K., Silverstein, M., Chelmow, D., Coker, T. R., Davidson, K. W. et al. (2023). Screening for Depression and Suicide Risk in Adults: US Preventive Services Task Force Recommendation Statement. JAMA, 329, 2057-2067.[CrossRef] [PubMed]
[7] Bista, S., Tait, R. J., Straker, L. M., Lin, A., Steinbeck, K., Graham, P. L. et al. (2025). Joint Developmental Trajectories of Internalizing and Externalizing Problems from Mid-Childhood to Late Adolescence and Childhood Risk Factors: Findings from a Prospective Pre-Birth Cohort. Development and Psychopathology, 37, 176-191.[CrossRef] [PubMed]
[8] Borsboom, D., Deserno, M. K., Rhemtulla, M., Epskamp, S., Fried, E. I., McNally, R. J. et al. (2021). Network Analysis of Multivariate Data in Psychological Science. Nature Reviews Methods Primers, 1, Article No. 58.[CrossRef
[9] Buehler, C. (2020). Family Processes and Children’s and Adolescents’ Well‐Being. Journal of Marriage and Family, 82, 145-174.[CrossRef
[10] Byrne, M., Tan, R. K. J., Wu, D., Marley, G., Hlatshwako, T. G., Tao, Y. et al. (2023). Prosocial Interventions and Health Outcomes: A Systematic Review and Meta-Analysis. JAMA Network Open, 6, e2346789.[CrossRef] [PubMed]
[11] Cai, H., Bai, W., Liu, H., Chen, X., Qi, H., Liu, R. et al. (2022). Network Analysis of Depressive and Anxiety Symptoms in Adolescents during the Later Stage of the COVID-19 Pandemic. Translational Psychiatry, 12, Article No. 98.[CrossRef] [PubMed]
[12] Chen, X., Qi, H., Liu, R., Feng, Y., Li, W., Xiang, M. et al. (2021). Depression, Anxiety and Associated Factors among Chinese Adolescents during the COVID-19 Outbreak: A Comparison of Two Cross-Sectional Studies. Translational Psychiatry, 11, Article No. 148.[CrossRef] [PubMed]
[13] de Lacy, N., Ramshaw, M. J., McCauley, E., Kerr, K. F., Kaufman, J., & Nathan Kutz, J. (2023). Predicting Individual Cases of Major Adolescent Psychiatric Conditions with Artificial Intelligence. Translational Psychiatry, 13, Article No. 314.[CrossRef] [PubMed]
[14] Eagle, S. R., Jain, S., Sun, X., Preszler, J., McCrea, M. A., Giacino, J. T. et al. (2023). Network Analysis and Relationship of Symptom Factors to Functional Outcomes and Quality of Life Following Mild Traumatic Brain Injury: A TRACK-TBI Study. Frontiers in Neurology, 14, Article 13085040.[CrossRef] [PubMed]
[15] Folker, A., Bertrand, C., Hong, Y., Steinberg, L., Duell, N., Chang, L. et al. (2025). A Longitudinal Study of Adolescent‐to‐Young Adult Executive Function Development in Seven Countries. Developmental Science, 28, e70040.[CrossRef] [PubMed]
[16] Girard, L., Bøe, T., Nilsen, S. A., Askeland, K. G., & Hysing, M. (2024). Developmental Trajectories of Conduct Problems and Time-Varying Peer Problems: The Bergen Child Study. Social Psychiatry and Psychiatric Epidemiology, 59, 2237-2246.[CrossRef] [PubMed]
[17] Gong, X., & Zhou, J. (2025). Peer Victimization and Internalizing and Externalizing Problems among Early Adolescents: The Longitudinal between × within Moderation Role of the FKBP5 Gene. Development and Psychopathology, 37, 2672-2683.[CrossRef] [PubMed]
[18] Goodman, A., Lamping, D. L., & Ploubidis, G. B. (2010). When to Use Broader Internalising and Externalising Subscales Instead of the Hypothesised Five Subscales on the Strengths and Difficulties Questionnaire (SDQ): Data from British Parents, Teachers and Children. Journal of Abnormal Child Psychology, 38, 1179-1191.[CrossRef] [PubMed]
[19] Hill, E. D., Kashyap, P., Raffanello, E., Wang, Y., Moffitt, T. E., Caspi, A. et al. (2025). Prediction of Mental Health Risk in Adolescents. Nature Medicine, 31, 1840-1846.[CrossRef] [PubMed]
[20] Hysaj, M., Crone, M. R., Kiefte-de Jong, J. C., & Vermeiren, R. R. J. M. (2023). Do Parental Attachment and Prosocial Behavior Moderate the Impairment from Depression Symptoms in Adolescents Who Seek Mental Health Care? Child and Adolescent Psychiatry and Mental Health, 17, Article No. 133.[CrossRef] [PubMed]
[21] Isdahl-Troye, A., Villar, P., Álvarez-Voces, M., & Romero, E. (2025). Unraveling the Dynamics of Emotional Regulation and Parental Warmth across Early Childhood: Prediction of Later Behavioral Problems. Scientific Reports, 15, Article No. 23294.[CrossRef] [PubMed]
[22] Jeppesen, P., Wolf, R. T., Nielsen, S. M., Christensen, R., Plessen, K. J., Bilenberg, N. et al. (2021). Effectiveness of Transdiagnostic Cognitive-Behavioral Psychotherapy Compared with Management as Usual for Youth with Common Mental Health Problems: A Randomized Clinical Trial. JAMA Psychiatry, 78, 250-260.[CrossRef] [PubMed]
[23] Jones, P. J., Ma, R., & McNally, R. J. (2021). Bridge Centrality: A Network Approach to Understanding Comorbidity. Multivariate Behavioral Research, 56, 353-367.[CrossRef] [PubMed]
[24] Juhász, Á., Sebestyén, N., Árva, D., Barta, V., Pártos, K., Vokó, Z. et al. (2024). We Need Better Ways to Help Students Avoid the Harms of Stress: Results of a Meta-Analysis on the Effectiveness of School-Based Stress Management Interventions. Journal of School Psychology, 106, Article ID: 101352.[CrossRef] [PubMed]
[25] Kelley, S. W., & Gillan, C. M. (2022). Using Language in Social Media Posts to Study the Network Dynamics of Depression Longitudinally. Nature Communications, 13, Article No. 870.[CrossRef] [PubMed]
[26] Kiakos, D., Alameda, L., Lepreux, I., Conchon, C., Alerci, L., Gorgellino, M. et al. (2025). Pathways between Childhood Trauma, Clinical Symptoms, and Functioning in New-Onset Psychosis: Novel Insights from a Network Analysis Approach. Schizophrenia, 11, Article No. 75.[CrossRef] [PubMed]
[27] Kwon, S., & Telzer, E. H. (2022). Social Contextual Risk Taking in Adolescence. Nature Reviews Psychology, 1, 393-406.[CrossRef
[28] Li, M., Wang, J., Ma, P., Sun, W., Gong, H., & Gao, Y. (2023). The Relationship between Harsh Parenting and Adolescent Depression. Scientific Reports, 13, Article No. 20647.[CrossRef] [PubMed]
[29] Liu, C., Zhou, L., Pi, X., Liu, B., Zhang, X., Wei, W. et al. (2025). Network Characteristics of the Youth’s Insomnia and Emotional Symptoms and Their Gender Differences. Frontiers in Psychiatry, 16, Article 1597652.[CrossRef] [PubMed]
[30] Lu, G., Zhu, L., Huang, R., Lai, P., & Wang, C. (2025). The Association between Adaptability and the Symptoms of Depression and Anxiety in Early Adolescents: A Network Analysis in a Longitudinal Design. BMC Psychiatry, 25, Article No. 117.[CrossRef] [PubMed]
[31] Ma, H., Zhao, M., Liu, Y., & Wei, P. (2024). Network Analysis of Depression and Anxiety Symptoms and Their Associations with Life Satisfaction among Chinese Hypertensive Older Adults: A Cross-Sectional Study. Frontiers in Public Health, 12, Article 1370359.[CrossRef] [PubMed]
[32] Magalhães, E. (2024). Dual-Factor Models of Mental Health: A Systematic Review of Empirical Evidence. Psychosocial Intervention, 33, 89-102.[CrossRef] [PubMed]
[33] Memmott-Elison, M. K., & Toseeb, U. (2023). Prosocial Behavior and Psychopathology: An 11-Year Longitudinal Study of Inter-and Intraindividual Reciprocal Relations across Childhood and Adolescence. Development and Psychopathology, 35, 1982-1996.[CrossRef] [PubMed]
[34] Olivier, E., Morin, A. J. S., Langlois, J., Tardif-Grenier, K., & Archambault, I. (2020). Internalizing and Externalizing Behavior Problems and Student Engagement in Elementary and Secondary School Students. Journal of Youth and Adolescence, 49, 2327-2346.[CrossRef] [PubMed]
[35] Phan, M. L., Renshaw, T. L., Caramanico, J., Greeson, J. M., MacKenzie, E., Atkinson-Diaz, Z. et al. (2022). Mindfulness-Based School Interventions: A Systematic Review of Outcome Evidence Quality by Study Design. Mindfulness, 13, 1591-1613.[CrossRef] [PubMed]
[36] Piazza, G. G., Allegrini, A. G., Eley, T. C., Epskamp, S., Fried, E., Isvoranu, A. et al. (2024). Polygenic Scores and Networks of Psychopathology Symptoms. JAMA Psychiatry, 81, 902-910.[CrossRef] [PubMed]
[37] Reimann, G. E., Lahey, B. B., Jeong, H. J., Durham, E. L., Archer, C., Cardenas‐Iniguez, C. et al. (2025). Longitudinal Associations between Youth Prosocial Behavior and Dimensions of Psychopathology. JCPP Advances, 5, e12282.[CrossRef] [PubMed]
[38] Ringbom, I., Suvisaari, J., Kääriälä, A., Sourander, A., Gissler, M., Ristikari, T. et al. (2022). Psychiatric Disorders Diagnosed in Adolescence and Subsequent Long-Term Exclusion from Education, Employment or Training: Longitudinal National Birth Cohort Study. The British Journal of Psychiatry, 220, 148-153.[CrossRef] [PubMed]
[39] Rowe, J., Nogovitsyn, N., Mazurka, R., Squires, S. D., Hassel, S., Poppenk, J. et al. (2025). Latent Profile Analysis of Childhood Maltreatment and Neural Markers in Depression. JAMA Network Open, 8, e2525147.[CrossRef] [PubMed]
[40] Schumacher, L., Klein, J. P., Elsaesser, M., Härter, M., Hautzinger, M., Schramm, E. et al. (2023). Implications of the Network Theory for the Treatment of Mental Disorders: A Secondary Analysis of a Randomized Clinical Trial. JAMA Psychiatry, 80, 1160-1168.[CrossRef] [PubMed]
[41] Sun, J., Lu, T., Shao, X., Han, Y., Xia, Y., Zheng, Y. et al. (2025). Practical AI Application in Psychiatry: Historical Review and Future Directions. Molecular Psychiatry, 30, 4399-4408.[CrossRef] [PubMed]
[42] Tang, X., Tang, S., Ren, Z., & Wong, D. F. K. (2019). Prevalence of Depressive Symptoms among Adolescents in Secondary School in Mainland China: A Systematic Review and Meta-Analysis. Journal of Affective Disorders, 245, 498-507.[CrossRef] [PubMed]
[43] Tervo-Clemmens, B., Calabro, F. J., Parr, A. C., Fedor, J., Foran, W., & Luna, B. (2023). A Canonical Trajectory of Executive Function Maturation from Adolescence to Adulthood. Nature Communications, 14, Article No. 6922.[CrossRef] [PubMed]
[44] van Loon, A. W. G., Creemers, H. E., Vogelaar, S., Saab, N., Miers, A. C., Westenberg, P. M. et al. (2022). Trajectories of Adolescent Perceived Stress and Symptoms of Depression and Anxiety during the COVID-19 Pandemic. Scientific Reports, 12, Article No. 15957.[CrossRef] [PubMed]
[45] Wolke, D., Zhou, Y., Liu, Y., Eves, R., Mendonça, M., & Twilhaar, E. S. (2025). A Systematic Review of Conceptualizations and Statistical Methods in Longitudinal Studies of Resilience. Nature Mental Health, 3, 1088-1099.[CrossRef
[46] Yu, W., Abbey, C., Qian, Y., Wang, H., Rozelle, S., & Singh, M. K. (2022). Behavioral Strengths and Difficulties and Their Associations with Academic Performance in Math among Rural Youth in China. Healthcare, 10, Article 1642.[CrossRef] [PubMed]
[47] Zavlis, O. (2024). Computational Approaches to Mental Illnesses. Nature Reviews Psychology, 3, 650.[CrossRef
[48] Zhang, Z., Chen, H., Ye, Y., Chen, H., Guo, H., & Zhou, J. (2025). Entropy-Based Risk Network Identification in Adolescent Self-Injurious Behavior Using Machine Learning and Network Analysis. Translational Psychiatry, 15, Article No. 299.[CrossRef] [PubMed]
[49] Zhou, S., Leng, M., Zhang, J., Zhou, W., Lian, J., & Yang, H. (2024). Parental Emotional Warmth and Adolescent Internet Altruism Behavior: A Moderated Mediation Model. Humanities and Social Sciences Communications, 11, Article No. 446.[CrossRef