How to Suspect Long Covid in General Practice: Two Clinical Cases, Two Interviews, and AI Assistance in a Belgian Dodécagroupe
Synopsis
This pedagogical experience report uses two recent clinical situations of women around 30 years old to show how a general practitioner can, by chronologically rereading the electronic medical record and combining it with recorded interviews conducted with patient consent, reassemble scattered fatigue, neurocognitive and language difficulties, respiratory and dysautonomic manifestations, sensory changes, and pain into a clinical configuration compatible with Long Covid, while using a language model only to extract clinical concepts from interview verbatims for medical validation and later comparison with the Human Phenotype Ontology (HPO), and while insisting on preserving differential diagnoses, documenting functional impact, and adapting activity to the patient's limits rather than increa
Interpretation
It proposes and demonstrates an actionable clinical path: starting from the current consultation, going back through the record, then using a structured interview to rebuild a timeline of before infection, infection, after infection, exertion, recovery, and functional impact, so that symptoms previously scattered across visits appear as a coherent trajectory. Rather than locating Long Covid recognition in a single test or isolated symptom, it places the emphasis on temporal reconstruction and comparison with previous functioning, and states that no examination alone can confirm or exclude the diagnosis. Based on a narrative analysis of two anonymized clinical situations and two recorded interviews; it is a single physician's pedagogical experience report with no control group and no pre/post measurement.
It confines the language model to a reading tool rather than a diagnostic tool: first transcription, then extraction of clinical concepts without imposing a diagnosis or ontology, grouping by clinical domain, subsequent term-by-term checking of possible correspondence with HPO, keeping the verbatim as a trace of the original statement, and medical validation of any retained correspondence. Unlike uses of AI for automatic diagnosis, it places AI in a secondary step of organizing narrative material and explicitly notes that it can err or invent a terminological correspondence. It describes a four-step sequential process, but the text states that the AI-assisted analysis and HPO mapping did not undergo independent methodological validation and remain exploratory.
It offers five simple consultation questions (what happens after physical, cognitive, or emotional exertion; how long recovery takes; whether memory, words, reading, or daily organization have changed; whether there are palpitations, malaise, dizziness, sweating, nausea or other manifestations suggesting dysautonomia; whether smell, taste, pain, or tolerance to noise or light have changed), intended not to score symptoms but to recover their dynamics. Compared with symptom-counting checklists, these questions focus on triggering, worsening, fluctuation, recovery, and functional loss. Derived from the two situations; the author explicitly states they do not constitute a diagnostic score.
It stresses that in the absence of an etiological treatment demonstrated for all patients, listening, recognizing, treating symptoms, coordinating care, and advocating for functional impact remain essential medical acts; when post-exertional symptom exacerbation is present, activity should be adapted to the patient's limits rather than increased according to a fixed program. It separates 'no causal treatment' from 'no medical action,' and cites WHO guidance on pacing and energy conservation, noting that in the presence of post-exertional symptom exacerbation, fixed and imposed increases in activity are not a neutral strategy. Based on paraphrase of international guidance (NICE, CDC, WHO, French HAS) and the author's clinical experience, not on new trial data.
Perspective
The result is intended for general practitioners working in uncertainty and for their peer groups, at the office level: rereading the record chronologically, conducting a structured interview, documenting functional impact, keeping concurrent diagnostic hypotheses open, and using pacing when post-exertional symptom exacerbation is present. It also offers an exploratory way to organize narrative material into a shareable phenotypic vocabulary, on the premise that the physician retains final interpretive and validation responsibility. The author identifies next steps as replicating the format in several peer groups, especially Dodécagroupes, collecting participants' questions, assessing changes in how they question patients, and comparing this learning model with other forms of continuing education.
A careful reader would still watch how feasible the process is across more physicians and more patients; how stable the language-model concept extraction and HPO mapping are under independent evaluation; whether the five questions change the timing of recognition and care coordination in actual consultations; and how writing a functional history for advocacy works in different care and administrative systems. The text provides no figures or data tables, so readers needing concrete operational detail would still need to consult the international guidance and resources it lists.