Why do DPC doctors still get patient texts at 2 a.m.?

The short answer
Patients choose direct primary care for fast access to their doctor, and in a small practice the overnight texts often go straight to the physician's phone. Studies tie after-hours inbox work to clinician exhaustion. A triage agent is designed to sort each message by urgency, answer booking and price questions, queue refills for signature and page the on-call clinician.
Direct primary care (DPC) gives patients more access to their doctor and more time with them, for a flat monthly, quarterly or annual fee 1. It did not fix the phone.
Fast access to the doctor is one reason patients sign up 1. In a one- or two-doctor practice, that access often means the physician's own phone. So the doctor reads a refill request at 2:14 a.m., decides it can wait, and lies awake anyway.
Why do DPC doctors get patient texts overnight?
Because access is the product. The American Academy of Family Physicians (AAFP) says patients choose DPC for fast access to their doctor, and it tells new practice owners they will need a plan for urgent requests while they are on vacation or away from the office 1. With one or two clinicians, there is often nobody else to forward the line to, so the plan is the physician.
Does after-hours message work wear doctors down?
The research points that way. A study of 122 primary care clinicians at one academic medical center, published in the Journal of the American Medical Informatics Association in 2020, compared burnout surveys with the time each clinician logged in the electronic health record (EHR) 2:
- Clinicians above the median of 68 minutes of after-hours EHR time a week on clinic days, per full-time clinician, had higher odds of high exhaustion.
- Clinicians in the top quarter for message volume, more than 307 messages a week per full-time clinician, had 6.17 times the odds of high exhaustion.
A second study, in the Journal of General Internal Medicine, followed 150 primary care clinicians at one health system 3. Their average quarterly count of patient messages asking for medical advice rose from 340 at the end of 2019 to 695 at the end of 2021. Each additional 10 messages went with about 12 more minutes of EHR work outside scheduled hours per quarter 3.
Both studies come from large health systems, not DPC practices, and both show associations rather than cause. Neither measured texts to a doctor's own phone at night, so read them as a signal, not a measurement of your practice.
What kinds of messages arrive overnight?
We designed our triage agent around four kinds:
- Routine refills for stable chronic medications.
- Scheduling: "Can I come Friday afternoon?"
- Questions with a known answer: a price, a lab result already released, directions.
- The few messages that need a clinician now: chest pain, a child with a high fever, a patient who says they are not safe.
We expect the first three to be most of the volume. The fourth is why the doctor keeps the phone on.
What would a triage agent do with them?
Aria is the triage agent we are building into ZunovaMD. In the pilot design, it verifies who is texting, reads the chart through tools that only expose that patient, and runs the intake. Each message lands in one of four tiers:
- Tier 0 is a crisis or an emergency. A fixed safety check runs before the model, so certain words and patterns never wait on the model's judgment. On a hit, Aria sends a scripted response, offers a connection to the 988 Suicide & Crisis Lifeline and pages the on-call clinician. The Lifeline takes calls, texts and chats around the clock 4.
- Tier 1 is urgent today. Aria compiles the history, possible causes and a suggested order set for the clinician to review, and pages the clinician.
- Tier 2 is routine clinical review. The summary waits in the morning queue for a signature. Protocol refills land here.
- Tier 3 is what Aria can close by itself, from a list the clinic writes: booking, prices, directions.
The physician is meant to wake up to a queue that says what was handled, what waits for a signature, and why.
Where does the agent stop?
By design, Aria does not prescribe. It drafts a refill under a protocol the clinic approved, and the physician signs it. It does not diagnose. It sorts and summarizes, and each statement it makes about a patient is checked against the record before anyone sees it. It never drafts a controlled substance prescription; those follow the rules for e-prescribing controlled substances and the prescriber's own judgment. A patient can say "talk to a person" at any time, and a human takes the thread.
The evidence backs a human in the loop. A 2023 systematic review of online symptom checkers found their triage accuracy suboptimal overall, found undertriage in several studies, and concluded they are "currently not a viable replacement for other triage and diagnostic options such as telephone triage or in-person consultations" 5. That is why every Tier 0 and Tier 1 message goes to a clinician.
What is ZunovaMD building?
ZunovaMD is in pilot and enrolling DPC and rural clinics. Aria is designed to do the overnight sorting while every clinical decision stays with the physician. The triage engine on our platform page shows the tiers, our security page covers how patient data is handled, and what HIPAA and the TCPA require for texting patients covers the consent side.
Questions
Does a triage agent prescribe medication?
Not in the ZunovaMD design. Aria drafts a refill under a protocol the clinic approved, and the physician reviews and signs it. It never drafts a controlled substance prescription.
What happens when a patient texts about an emergency at night?
In the ZunovaMD pilot design, a fixed safety check runs before the AI model reads the message. On a hit, the patient gets a scripted response and an offer to connect to the 988 Suicide & Crisis Lifeline, and the on-call clinician is paged.
Can a patient reach a person instead of the agent?
Yes. In the ZunovaMD design, a patient can say "talk to a person" at any time and a human takes over the thread. Urgent messages go to the on-call clinician either way.
Will patient messages be used to train AI models?
No. ZunovaMD does not train models on patient data, and that includes de-identified data. Patient messages are used to answer and route that patient's request.
Does after-hours inbox work cause physician burnout?
Studies show an association, not proof of cause. A 2020 study of 122 primary care clinicians linked more after-hours EHR time and higher message volume with higher odds of exhaustion. A 2024 study of 150 primary care clinicians found that each 10 extra patient messages went with about 12 more minutes of after-hours EHR work per quarter. Both studies come from large health systems, not DPC practices.
Sources
- American Academy of Family Physicians, Direct Primary Care Model for Family Physicians, checked 2026-10-26
- Adler-Milstein et al., Electronic health records and burnout: Time spent on the electronic health record after hours and message volume associated with exhaustion but not with cynicism among primary care clinicians, Journal of the American Medical Informatics Association 2020;27(4):531-538, via PMC, checked 2026-10-26
- Martinez et al., Patient Portal Message Volume and Time Spent on the EHR: an Observational Study of Primary Care Clinicians, Journal of General Internal Medicine 2024;39(4):566-572, via PMC, checked 2026-10-26
- Substance Abuse and Mental Health Services Administration, 988 Suicide & Crisis Lifeline (last updated 09/26/2025), checked 2026-10-26
- Riboli-Sasco et al., Triage and Diagnostic Accuracy of Online Symptom Checkers: Systematic Review, Journal of Medical Internet Research 2023;25:e43803, via PMC, checked 2026-10-26
Not medical advice. This post is general information about running a clinic and the software behind it. For a question about your health or a patient's care, ask a licensed clinician.
You've tried the workarounds.This one runs itself.
Thirty minutes on a clinic like yours: the overnight queue, a visit with the scribe on, a dispense from the note and a state report. Bring an export and we run the dry-run import while we talk.