Sleep Disturbance Subgroups in TCM Nurses: Who Falls Into the Severe Category?

Nursing combines high physical and emotional demands. Night shifts, rotating schedules, emergency response, and emotional labor intertwine to make sleep problems a long-recognized health concern in this population. Historically, however, most studies have looked at average scores, such as a single mean sleep disturbance score for an entire group of nurses. That approach is like averaging the speed of a trained sprinter and a toddler learning to walk: it flattens out real differences between individuals.

This study focused specifically on TCM nurses, a workforce that, in addition to conventional nursing training, must understand TCM theory and techniques, creating particular role and workload characteristics. The research team noted that while these nurses face significant occupational stressors, few studies have applied person-centered approaches to categorize their sleep patterns into distinct subgroups. This is precisely why the study's value lies not in confirming that nurses sleep poorly, but in attempting to answer a different question: among those who sleep poorly, might there actually be several different patterns worth distinguishing?

Latent profile analysis showed that the 7721 surveyed nurses could be divided into three sleep disturbance subgroups: mild-stable (29.8%), moderate-fluctuating (60%), and severe-persistent (10.2%). The differences across groups extended beyond severity to the pattern of symptom fluctuation; fluctuation means disturbance levels rise and fall over time, whereas stable and persistent indicate disturbance remains relatively fixed at a given level.

On the machine learning classification side, the five models achieved test AUCs ranging from 0.71 to 0.84, with XGBoost performing best at AUC 0.84 (95%CI: 0.83-0.85). This indicates the model has considerable accuracy in distinguishing which sleep disturbance subgroup an individual belongs to. SHAP interpretability analysis identified the top five features contributing most to classification, in order: monthly income, organizational support, hospital level, self-compassion, and resilience. These features collectively span the economic dimension (income), the workplace environment (organizational support, hospital level), and personal psychological resources (self-compassion, resilience).

Overall, most surveyed nurses fell into the moderate-fluctuating group, meaning their sleep disturbance is not all-or-nothing but rather a fluctuating intermediate state. This carries practical implications for workplace health management: resources should not focus exclusively on the roughly one in ten with the most severe symptoms, but should also consider how to stabilize sleep quality for the six in ten whose symptoms fluctuate.

For the general reader, the most important takeaway from this study is that when we discuss nurses' sleep problems, we should not assume everyone is in the same situation. The mild-stable subgroup may consist of workers who only occasionally sleep poorly; the moderate-fluctuating subgroup may experience ups and downs tied to shift work and on-call pressure; and the severe-persistent subgroup represents long-term, marked sleep disturbance. The three patterns may call for different support strategies.

Several key terms also warrant brief explanation. Psychological resilience refers to a person's ability to recover and adapt when facing stress or adversity. Self-compassion refers to the tendency to treat oneself with warmth and without harsh self-criticism during difficulty. Both features ranked in the SHAP top five, suggesting that personal internal psychological resources may be meaningfully associated with whether someone falls into the severe sleep disturbance group. Organizational support captures the assistance and recognition nurses perceive from their unit, supervisors, and colleagues; it is an important indicator of the workplace psychosocial environment.

Readers must understand, however, that the study shows statistical associations, not causal relationships. For example, the link between lower monthly income and falling into the severe sleep disturbance group does not directly mean low income causes worse sleep. Likewise, resilience and self-compassion could be outcomes rather than causes; only future longitudinal research can clarify the direction. Furthermore, the sample was limited to nurses at 130 TCM healthcare institutions in Liaoning Province, so findings should not be directly extrapolated to nurses in other specialties, other provinces, or other countries.

This cross-sectional study has clear limitations. The most critical point is that the design is a single time-point survey; it does not track the same nurses over time. Therefore, whether the three identified subgroups are truly stable, whether someone remains in the severe-persistent category long-term or is simply passing through a difficult period, has not been verified. The research team explicitly noted that future longitudinal studies are required to confirm subgroup stability and establish causal relationships.

In addition, the study population was limited to TCM nurses at healthcare institutions in Liaoning Province. Local work culture, salary structure, hospital-level distribution, and TCM-specific workflow practices may all shape how the findings apply elsewhere. For nurses in other specialties, other provinces, or other countries, whether these three subgroups appear in the same proportions with the same associated factors remains unknown from this study. When citing these numbers, readers should treat them as observations from one specific population rather than as universal rules.

這篇發表於 2026-02-24 的研究，標題為《Characterizing Sleep Disturbance Subgroups and Identifying Associated Factors in Traditional Chinese Medicine Nurses: A Latent Profile Analysis and Explainable Machine Learning Approach.》，探討中醫護理人員的睡眠困擾。文章指出，中醫科護理人員因職業壓力，面臨顯著的睡眠挑戰，但以「以人為中心」的方式——也就是把人依特徵分群而非只看整體平均——來分類這些睡眠模式的研究仍然相當有限。為了回答「哪些人屬於哪種睡眠困擾類型」，研究採用了一種稱為「潛在剖面分析」（Latent Profile Analysis，簡稱 LPA）的統計方法。這種方法的概念是：即使大家填同一份量表，每個人的困擾程度與波動模式其實不盡相同，因此可以依作答反應把人分成幾個不同的亞群。研究工具包括 PROMIS 睡眠障礙短量表 8a，以及涵蓋人口統計、職業與心理變項的自填問卷。資料分析採取兩階段：先以 LPA 區分亞群類型，再以遞迴特徵消去法結合隨機森林（RFE-RF）篩選變項，餵入五種機器學習模型，其中 70% 資料用於訓練，30% 作為獨立測試集；最終以 XGBoost 模型結合 Shapley 加性解釋（SHAP）方法，揭示每個變項對分類結果的貢獻度。

結果顯示，LPA 辨識出三組睡眠困擾亞群：輕度穩定型佔 29.8%、中度波動型佔 60%、重度持續型佔 10.2%。換言之，每十位中醫護理人員中，約有一位落入最嚴重的類別，而大多數落在中等程度。這種「以人為中心」的分群視角，比單純計算平均分更能呈現差異。透過 SHAP 解釋，最能區分亞群成員身分的五個特徵依序為：月收入、組織支持、醫院層級、自我關懷與心理韌性。

文章的結論提醒讀者，這些僅是相關性而非因果關係，且樣本限於遼寧省 130 所中醫醫療機構的 7721 位中醫護理人員，無法直接推論到其他族群。作者建議未來以縱貫性研究驗證亞群的穩定性，並嘗試建立因果鏈結。整體而言，研究展示了一條由「分群描述」到「解釋變項」的工作流程，讓讀者看見統計方法如何把一群人的睡眠困擾，切出三種可辨識的樣貌。