Family doctor

In several valleys, finding a family doctor still willing to take on new patients comes down to luck. AI will not treat patients in the doctor's place: it will hand back the hours paperwork had taken from them, at the very moment the canton needs those hours most.

The profession today

Primary care is becoming scarce exactly where it is most needed. Valley practices closing without a successor add up year after year, a significant share of practising general practitioners are nearing retirement age, and the next generation, trained in the city, settles in the city. Meanwhile demand keeps growing: one Valaisan in ten will be over 80 by 2035, and it is older, chronically ill, polymedicated patients who most need a doctor who knows them.

A settled general practitioner's week is split across very different registers:

• Consultations: from a sore throat to a complex case, twenty to thirty situations a day, each with its own story
• Clinical documentation: consultation notes, maintaining the patient record, often pushed to the evening
• Letters and reports: referral letters to specialists, insurance reports, certificates, forms
• Care coordination: specialists, the Hôpital du Valais, community health centres, pharmacies, families
• Follow-up of chronic patients: diabetes, hypertension, polymedication in the elderly, treatment plans
• Primary-care on-call and emergency duty: the regional rota, heavier as the ranks of practitioners thin
• Practice management: billing, staff, suppliers, premises

Part of the evening is spent at the screen. This administrative time is regularly cited by younger doctors among the reasons they prefer hospital salaried positions to setting up a practice, and by older doctors among the reasons for retiring earlier.

What AI is preparing

Dictated, validated clinical documentation. The consultation is transcribed, structured into a clinical note in the record's format, and submitted to the doctor, who reviews, corrects, and validates it before it is entered. Nothing goes into the record without the doctor's explicit approval: the note carries the doctor's accountability, whatever tool prepared it. The gain amounts to hours each week, and above all to the evenings.

Letters and reports. Referral letters, insurance reports, certificates: drafted from the record, in the register expected by each recipient, signed after review. The signature remains personal, and it is what gives the document its value for the colleague who receives it.

Decision support as a second opinion. Drug interactions, differential diagnoses not to be dismissed, updated guidelines: the tool flags, questions, reminds. The diagnosis itself remains an act of the doctor. This second opinion is worth exactly what the practitioner receiving it makes of it: one who keeps hold of their own reasoning, knows why they follow a suggestion, and documents why they depart from it.

Telemedicine woven in. Triaging the day's requests, preparing teleconsultations, following up between appointments: telemedicine works alongside in-person consultation rather than claiming to replace it. For an elderly patient living far from the practice, a well-prepared teleconsultation avoids a half-day trip; the clinical examination retains all its uses, and it is the doctor who draws the line case by case.

Health data: the prerequisite

Practice data is doubly protected: by the medical confidentiality rule under Article 321 of the Swiss Criminal Code and by the revised Federal Act on Data Protection, in force since 1 September 2023, which classifies health data among sensitive personal data. Three requirements precede any rollout: hosting in Switzerland, with subcontracting agreements documenting who accesses what; a strict ban on feeding any patient data into a free consumer-grade tool; and traceability of every generated document (who dictated it, who validated it, when). Medical confidentiality survives the introduction of tools on one condition: that the doctor knows, at every moment, where the data goes.

What rises in importance for judgment

The clinical examination. The patient's body remains the first source of information: a gait that has changed since last time, a paleness, a hand held. No transcription captures that. The time freed up by automated documentation flows back exactly there, into the eye and the hands.

Longitudinal knowledge. Following three generations of the same family, knowing what this patient usually withholds, what this particular pain means for them and not for someone else: this clinical and human memory is what gives meaning to the machine's second opinion. It builds over years of presence, and it disappears with every practice that closes.

Contextual therapeutic judgment. Between the current guideline and the real patient, their means, their isolation, their wishes, their family, the doctor calibrates. AI cites the literature; the general practitioner knows this patient will not take three more medications, and that an imperfect plan that gets followed beats a perfect plan that gets abandoned.

Running a practice as a team. The practice taking shape combines in-person consultation, telemedicine, and advanced practice nursing, whose legal framework is currently under discussion at federal and cantonal level. Deciding who sees whom, delegating without stepping back from responsibility, keeping overall clinical accountability: this is a doctor's work, in an expanded form that initial training barely prepared for.

The decisive conversations. Breaking a serious diagnosis, accompanying a patient through the end of life, a decision shared with an exhausted family. None of this can be delegated. And it is often exactly why one becomes a family doctor.

Who keeps the final word?

Composite illustration. A saturated two-doctor valley practice was about to close its books to new patients. It rolls out validated dictated documentation and hands phone triage, built on protocols the doctors wrote themselves, to its medical assistant. Within a few months, evenings spent on data entry nearly disappear, and consultation slots reopen. The practice hires an advanced practice nurse to follow up stable chronic patients, supervised by the doctors. The patient list stays open, and the regional on-call rota keeps one more practice in it. (Fictional, composite situation; to be replaced with a real case at the incarnation stage.)

Job profile 2030

The first new competency is validating generated documentation: reviewing a note or letter produced by AI with the eye of the person who signs it, catching the plausible error, the most dangerous kind in medicine, because it looks like the truth, and ensuring the whole chain complies with Article 321 of the Criminal Code and data protection law. This competency in accountable review is distinct from the competency of drafting, and medical schools do not yet teach it.

The second is the critical use of the second opinion: knowing when to query the decision-support tool, how to question it, when to depart from it, and how to document that departure. A doctor who follows the tool without understanding it is as dangerous as a doctor who ignores it on principle; the professionalism of 2030 sits between the two.

The third is running an expanded care team: orchestrating telemedicine, advanced practice nursing, and in-person consultation, writing delegation protocols, and remaining accountable for overall clinical care. The family doctor becomes the architect of a local care system, in addition to being its practitioner.

Territorial anchoring

Chapter 9's thesis applies here without qualification: in an ageing canton, AI's productivity gains serve first to offset a shortage that training will not close in time. A practice whose documentation moves four times faster (the factor of four to five the author observed on documentary tasks holds fully for notes and letters) can follow more patients without degrading quality, or simply keep existing where it would otherwise have closed.

The territorial stakes are stark in their simplicity: every valley practice that closes redirects hundreds of patients toward the emergency department of the Hôpital du Valais and toward consultations an hour's drive away, hitting the oldest and least mobile first. The augmented practice, combining consultation, telemedicine, advanced practice nursing, and automated documentation, is the form in which family medicine can survive in the side valleys. Technology supplies part of this; the tariff, legal, and training framework will supply the rest, or prevent it.

What the decision-maker must do now

For a doctor in practice

Start with documentation, the safest gain and the least clinically risky: choose a compliant tool (Swiss hosting, subcontracting agreement, traceability), measure the time regained over a quarter, and only then extend to triage and follow-up. Involve the medical assistant from the outset: she is the one who will see the workflows change, and her role grows richer if the redesign is thought through with her.

For the Société Médicale du Valais

Establish a branch-wide framework: a list of tools verified against Article 321 and data-protection criteria, shared experience between practices, a clear public position on the named validation of generated documents. Also push the question of tariff recognition for validation and supervision work, which matches no existing position in current fee schedules.

For the cantonal public health department

Integrate the AI equipment of practices into the cantonal strategy against the primary-care shortage, on the same footing as installation grants, and clarify hosting and traceability requirements for health data. Follow closely the ongoing legal-framework discussion on advanced practice nursing: the team-based practice model, the only one capable of holding the valleys, depends directly on it.

---

Jérôme Deshaie is the founder of MCVA Consulting SA, an agency specialising in the AI transformation of organisations in Valais, and the author of Bisse Cognitif.