The goal of this clinical trial is to learn whether right dorsolateral prefrontal cortex (right DLPFC)-targeted fNIRS-BCI online closed-loop neurofeedback, delivered with slow-wave acoustic cueing, can reduce anxiety symptoms and improve cardiac autonomic regulation in women with recurrent pregnancy loss (RPL) and comorbid anxiety (women aged 18-45 years, right-handed, currently not pregnant or in a missed miscarriage state).

The main questions it aims to answer are:

Does real neurofeedback increase the proportion of participants who achieve an anxiety treatment response (defined as ≥50% reduction in Hamilton Anxiety Rating Scale [HAMA] total score from baseline) compared with sham feedback, at end of treatment and at 3-month follow-up? Is the intervention safe and well tolerated, as reflected by between-group differences in adverse events during the training period? Do brain and autonomic measures show between-group differences during the first formal session, including right DLPFC HbO downregulation, interhemispheric DLPFC synchronisation, heart rate (HR), and heart rate variability (HRV) indices? Researchers will compare real right DLPFC neurofeedback to sham feedback (identical procedures and displays but weakened coupling to real-time neural activity) to see if real neurofeedback improves anxiety outcomes and brain-heart autonomic regulation.

Participants will:

Complete screening, baseline clinical assessments, and physical examination Be randomly assigned (1:1) to real neurofeedback or sham feedback Complete 3 days of adaptation training followed by 3 weeks of training (15 sessions; one weekday session per day; ~20 minutes each) using a block design with slow-wave acoustic cueing (1 Hz amplitude-modulated tone; 20 s rest + 40 s cueing per block; 20 blocks/session) Undergo fNIRS recording in all sessions, with ECG recorded in session 1 only (for HR/HRV analyses) Receive matched, guideline-informed cognitive-behavioural therapy (CBT) during the intervention period Complete anxiety-related assessments at baseline, ~1 hour after the final session, and 3 months after treatment, with adverse events monitored throughout the intervention period

This study is a mechanistically informed, prospective, randomized, sham-controlled, parallel-group clinical trial designed to evaluate whether right dorsolateral prefrontal cortex (right DLPFC)-targeted fNIRS-BCI online closed-loop neurofeedback, delivered with slow-wave acoustic cueing, improves anxiety symptoms and cardiac autonomic regulation in women with recurrent pregnancy loss (RPL) and comorbid anxiety. Participants who meet eligibility criteria will be randomized in a 1:1 ratio to either real neurofeedback or sham feedback, with identical visit structure and procedures across groups.

The intervention will use an online neurofeedback pipeline to estimate right DLPFC activation from fNIRS signals in real time and present a visual activation bar to guide volitional downregulation of the target region during slow-wave acoustic cueing. The sham condition will use the same interface, cueing, and workflow but with feedback parameters configured to substantially weaken effective coupling between the displayed feedback and the participant's instantaneous neural state. fNIRS will be recorded across all training sessions, and ECG will be recorded during the first formal training session only to quantify heart rate and heart rate variability and to support mechanistic analyses of short-term brain-heart coupling during the task.

Clinical outcomes will focus on changes in anxiety severity over time, assessed at baseline, at the end of treatment, and at 3-month follow-up. Neurophysiological endpoints will be derived from the first formal session to characterize task-related right DLPFC downregulation, interhemispheric DLPFC synchrony, and cardiac autonomic responses during the neurofeedback task. Safety and tolerability will be evaluated by comparing the incidence and profile of adverse events between groups throughout the intervention period. In addition, both groups will receive matched, guideline-informed cognitive-behavioural therapy (CBT) during the intervention period as background standard care, and any clinically indicated psychotropic medication use during follow-up will be documented for analytic control.

Inclusion Criteria:

• (i) Women aged 18-45 years, right-handed;
• (ii) Diagnosis of recurrent pregnancy loss (RPL), defined as ≥2 consecutive spontaneous pregnancy losses occurring before 28 weeks of gestation;
• (iii) Not currently pregnant, or currently in a missed miscarriage state;
• (iv) Meet DSM-5 diagnostic criteria for an anxiety disorder, with at least moderate severity, defined as Clinical Global Impression-Severity (CGI-S) ≥4;
• (v) Hamilton Anxiety Rating Scale (HAMA) ≥16, with 17-item Hamilton Depression Rating Scale (HAMD-17) <17, to ensure anxiety is the predominant affective disturbance.

Exclusion Criteria:

• (i) Markedly unstable blood pressure (systolic BP >180 mmHg or <90 mmHg);
• (ii) Clinically important comorbid organic diseases, including but not limited to hyperthyroidism, history of atrial fibrillation, sinus bradycardia, major neurological disorders, cerebrovascular disease, or severe pulmonary disease;
• (iii) Significant suicide risk, judged by psychiatric assessment to be unsuitable for study participation;
• (iv) Other severe psychiatric disorders, including substance use disorder, schizophrenia, delusional disorder, unspecified psychotic disorder, bipolar disorder, or delirium;
• (v) Use of any oral antidepressant, anxiolytic, or antipsychotic medication within the past 4 weeks, or fluoxetine within the past 6 weeks, or any long-acting injectable antipsychotic within the past 3 months.

Masking (blinding) will be implemented at multiple levels. Participants will be blinded to group assignment, and both groups will receive identical slow-wave acoustic cueing, session structure, instructions, and an indistinguishable on-screen feedback interface with the numerical threshold concealed. Outcome assessors (clinical raters) and data analysts/statisticians will remain blinded throughout data collection and primary analyses. Only the session operator responsible for loading group-specific neurofeedback parameters will be unblinded, and will not participate in outcome assessments or statistical analyses. Emergency unblinding will be permitted only when clinically necessary for participant safety and will be documented according to prespecified procedures.