Algorithm-based validation of sleep apnea treatment : lessons from primary care after one year

Context
Algorithm-based validation of medical prescriptions is increasingly used to standardise care. Since June 2024 in France, prescriptions for continuous positive airway pressure (CPAP) therapy for sleep apnea must be validated through a national algorithm based on French National Authority for Health (HAS) criteria. While this system aims to improve efficiency and equity, its impact on complex, real- life primary care situations—where clinical judgement is essential—remains poorly documented.
Methods
We conducted a practice-based retrospective review of electronically submitted CPAP prescriptions between June 2024 and September 2025. Data from 127 submissions were analysed to document algorithm-based decisions, administrative outcomes at the Commission de Recours Amiable (CRA), and the role of the general practitioner (GP) in the appeal process.
Originality
This contribution provides a rare real-world primary care perspective on the practical limits of algorithm-based medical validation. It explores how digital decision-support systems may outweigh general practitioners’ clinical assessment of treatment benefit. Rather than a theoretical evaluation of artificial intelligence, this reflection highlights how administrative frameworks can shape the balance between algorithms and professional judgement in everyday practice.
Lessons learned and future implications
The analysis showed that 30% of prescriptions were rejected by the validation system. Rejections mainly concerned patients whose clinical profiles were consistent with international recommendations but did not meet national threshold-based criteria. Several patients had an apnea–hypopnea index below 15/h but significant respiratory events during REM sleep or in the supine position. The appeal process relied exclusively on the initial algorithm-based opinion, without medical reassessment, and did not integrate the GP’s documentation of clinical improvement. Older patients with multimorbidity or cognitive impairment were particularly affected.
Discussion
This reflection highlights tensions between algorithm-driven decision-making and the realities of primary care. Reliance on global numerical thresholds may overlook clinically meaningful situations, while the absence of a medical reassessment pathway during appeals undermines continuity of care and professional responsibility.
Conclusion
Algorithm-based validation should support, not replace, clinical judgement. Digital decision-support systems need to evolve to integrate contextual clinical information and observed treatment benefit, ensuring that standardisation does not limit patient- centred care.