Scholarlink

Background <p>Cluster headache (CH) is a rare and disabling primary headache disorder characterized by severe unilateral pain and cranial autonomic symptoms. Although its diagnostic criteria are well defined, the conditional relationships among symptoms, reported triggers, and pain locations are not well characterized. Network analysis may provide an exploratory framework for examining these co-occurrence patterns.</p> Objective <p>To characterize conditional associations among symptoms, reported triggers, and headache regions in patients with CH using Ising model-based network analysis and to evaluate the stability of the estimated edge and centrality patterns.</p> Methods <p>A cross-sectional analysis was conducted using data from the Iranian National Cluster Headache Registry. Binary variables indicating the presence or absence of symptoms, reported triggers, and headache regions were analyzed. Separate symptom, trigger, and headache-region networks, together with an integrated network, were estimated using EBIC-regularized Ising models (γ = 0.25) in R (v4.4.3). Strength and bridge-strength point estimates were calculated. Nonparametric and case-dropping bootstrap procedures (1,000–3,000 iterations) were used to assess edge-weight accuracy and correlation stability (CS); CS coefficients below 0.25 were considered insufficient for robust interpretation of centrality rankings.</p> Results <p>Data from 321 patients were analyzed. In the symptom network, the largest estimated positive edges included epiphora–runny nose (1.80) and miosis–runny nose (1.05); epiphora and miosis had the highest strength point estimates, but strength centrality did not meet the prespecified stability threshold (CS = 0.206). The trigger network was unstable (edge CS = 0.05; strength CS = 0.00), and trigger edge and centrality rankings were therefore not interpreted. The headache-region network showed stronger stability (edge CS = 0.748; strength CS = 0.361), with a positive neck–occipital association (2.16) and a negative lower-jaw–cheek association (−2.31). In the integrated network, edge stability was moderate (CS = 0.439), whereas strength centrality remained below the interpretation threshold (CS = 0.206); bridge-strength estimates, including runny nose (2.17), were treated as exploratory.</p> Conclusion <p>Ising network analysis identified conditional co-occurrence patterns among CH features, with the most robust findings arising from the headache-region network. Symptom centrality, trigger-network estimates, and integrated centrality and bridge rankings require cautious interpretation because of limited stability. These cross-sectional findings are hypothesis-generating and support validation in larger and longitudinal datasets; they do not establish causal mechanisms or clinically actionable treatment targets.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Exploratory Ising model network analysis of cluster headache: mapping conditional associations across symptoms, triggers, and pain localization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月29日

查看全文
  • Mansoureh Togha,
  • Omid Kohandel Gargari,
  • Fariborz Khorvash,
  • Mohsen Hoseinnezhad,
  • Elham Jafari,
  • Arya Derakhshesh,
  • Shiva Rahimi,
  • Vahid Shaygannejad,
  • Zhale Salami,
  • Mahdi Badiee-Gavarti,
  • Reza Shokrani Foroushani,
  • Sheida Shaafi,
  • Faraidoon Haghdoost

摘要

Background

Cluster headache (CH) is a rare and disabling primary headache disorder characterized by severe unilateral pain and cranial autonomic symptoms. Although its diagnostic criteria are well defined, the conditional relationships among symptoms, reported triggers, and pain locations are not well characterized. Network analysis may provide an exploratory framework for examining these co-occurrence patterns.

Objective

To characterize conditional associations among symptoms, reported triggers, and headache regions in patients with CH using Ising model-based network analysis and to evaluate the stability of the estimated edge and centrality patterns.

Methods

A cross-sectional analysis was conducted using data from the Iranian National Cluster Headache Registry. Binary variables indicating the presence or absence of symptoms, reported triggers, and headache regions were analyzed. Separate symptom, trigger, and headache-region networks, together with an integrated network, were estimated using EBIC-regularized Ising models (γ = 0.25) in R (v4.4.3). Strength and bridge-strength point estimates were calculated. Nonparametric and case-dropping bootstrap procedures (1,000–3,000 iterations) were used to assess edge-weight accuracy and correlation stability (CS); CS coefficients below 0.25 were considered insufficient for robust interpretation of centrality rankings.

Results

Data from 321 patients were analyzed. In the symptom network, the largest estimated positive edges included epiphora–runny nose (1.80) and miosis–runny nose (1.05); epiphora and miosis had the highest strength point estimates, but strength centrality did not meet the prespecified stability threshold (CS = 0.206). The trigger network was unstable (edge CS = 0.05; strength CS = 0.00), and trigger edge and centrality rankings were therefore not interpreted. The headache-region network showed stronger stability (edge CS = 0.748; strength CS = 0.361), with a positive neck–occipital association (2.16) and a negative lower-jaw–cheek association (−2.31). In the integrated network, edge stability was moderate (CS = 0.439), whereas strength centrality remained below the interpretation threshold (CS = 0.206); bridge-strength estimates, including runny nose (2.17), were treated as exploratory.

Conclusion

Ising network analysis identified conditional co-occurrence patterns among CH features, with the most robust findings arising from the headache-region network. Symptom centrality, trigger-network estimates, and integrated centrality and bridge rankings require cautious interpretation because of limited stability. These cross-sectional findings are hypothesis-generating and support validation in larger and longitudinal datasets; they do not establish causal mechanisms or clinically actionable treatment targets.