A Multimodal Data-Based Diagnostic Model for Predicting Vestibular Migraine: A Retrospective Study

Ping Gu^1*Sai Zhang¹ShuaiJie Yin²Shuo Qin²Yilin Lang¹Wenting Wang¹Shaona Liu¹Ting Zhang¹Shuangmei Yan¹Dong Li¹Yongci Hao¹

• ¹The First hospital of Hebei Medical University, Shijiazhuang, China
• ²Hebei Medical University, Shijiazhuang, China

Objective: Vestibular migraine (VM) is a common neurological disorder characterized by recurrent vertigo and migraine symptoms. Due to its heterogeneous clinical presentation and lack of objective biomarkers, VM is often misdiagnosed. This study aimed to develop a diagnostic prediction model for VM based on multimodal data to improve diagnostic accuracy. Methods: A total of 288 patients who visited the Vertigo Clinic of our Hospital between January 2023 and December 2024 were enrolled, including 141 VM patients and 147 non-VM controls. Multimodal data were collected, including clinical features, vestibular function tests, hematological indicators, contrast transthoracic echocardiography, and psychological assessments. Logistic regression was used to construct the prediction model, and its performance was evaluated using receiver operating characteristic (ROC) curve analysis. Results: VM patients were more likely to be female, younger, and had lower body mass index (BMI) compared to controls. They also exhibited higher rates of photophobia, phonophobia, tinnitus, emotional triggers, insomnia, and family history of migraine or vertigo. Vestibular function tests showed fewer peripheral abnormalities and more central pathway dysfunction in VM patients. Hematological analysis revealed lower levels of vitamin D and D-dimer, and higher platelet counts and calcium levels in VM patients. Right-to-left shunt (RLS) was more prevalent in VM patients. The final model included six variables: BMI, emotional triggers, insomnia triggers, history of motion sickness, and abnormal otoacoustic emissions at 8000 Hz (left ear) and 6000 Hz (right ear). The model achieved an Area Under the ROC curve of 0.8788 (95% CI: 0.8374–0.9202), indicating strong diagnostic performance. Conclusion: The multimodal diagnostic prediction model developed in this study demonstrates high preliminary accuracy. It shows potential as a clinical tool for improving the diagnosis of VM, but its generalizability requires validation in larger, prospective cohorts.