Primary-care no-shows: a practical playbook

Primary-care no-shows are not a mystery personality trait. They are a scheduling and contact problem with published interventions that move rates when you measure the right denominator. A randomized trial in academic primary care cut no-shows among high-risk patients from 29.2% to 22.8% with a targeted staff call seven days out — on top of usual automated reminders. This playbook turns that evidence into an operating sequence a US practice can run without inventing a new EHR.

Define the metric before you buy a tactic

No-show rate is only useful if the denominator is explicit. The Shah et al. primary-care trial defined no-show as no-shows divided by no-shows plus arrivals — cancellations and reschedules were tracked separately. Many practices bury cancellations inside “non-arrival” and then wonder why interventions look weak.

Write your definitions:

• No-show — patient did not arrive and did not cancel/reschedule before the visit.
• Early cancel — cancel/reschedule with enough lead time to offer the slot.
• Late cancel — cancel too late to refill.
• Arrival — completed check-in / visit start per your EHR rule.

Without that glossary, every dashboard argument is politics.

What the evidence actually supports

Targeted calls for high-risk primary-care patients

Shah and colleagues randomized 2,247 adult primary-care patients with predicted no-show risk ≥15%. Everyone still received the practice’s usual automated call; the intervention arm also entered a queue for a staff phone call seven days before the visit, with training to push concrete planning (“What are you doing before your appointment?”). Result: no-show 22.8% vs 29.2% (absolute risk difference −6.4%, p < 0.01). Cancellations and reschedules happened slightly earlier in the intervention arm — which is how capacity is recovered.

Read that carefully: the win was targeted, not universal. Staff time went to the high-risk slice (~10.7% of screened patients in their setting).

Staff vs. automated vs. none

An earlier outpatient RCT (Parikh et al., academic practice) compared clinic-staff reminders, automated reminders, and no reminder. No-show rates were 13.6%, 17.3%, and 23.1% respectively. Staff calls won; automation still beat silence. For a primary-care playbook, that means: automation is table stakes; human or high-quality agent contact is the lever for the risky cohort — not a replacement for logging and eligibility.

Open access is a scheduling redesign, not a slogan

A 2024 systematic review of open-access (advanced-access) scheduling in outpatient clinics found that 10 of 16 included studies (62.5%) reported a significant decrease in no-show rates after OA; others showed non-significant reductions or no change. Family medicine and pediatrics dominated the sample. The authors stress needs assessment, stakeholder training, and design around real demand — not renaming the template “open access” while lead times stay long.

OA belongs in the playbook as a capacity design option alongside outreach — not as a substitute for confirmation.

Published signals

Figures you can cite without inventing a ‘national’ rate

22.8%

No-show · targeted arm (high-risk PC)

Shah 2016

29.2%

No-show · control (high-risk PC)

Shah 2016

62.5%

OA studies with significant drop

HSR 2024

The playbook: five moves that fit together

1. Score risk with boring, auditable features

Useful predictors in published models and ops practice: prior no-shows, long wait days (scheduling lead time), age band, insurance class, visit type. Start simple. A transparent rules score beats a black box you cannot explain to a patient or an auditor.

Threshold idea (adapt locally): flag patients above your historical no-show base — Shah used ≥15% predicted risk in a clinic whose overall average was much lower (~7%). Your threshold should reflect staff capacity, not a paper’s number.

2. Run a two-layer reminder stack

• Layer A (all visits): automated confirmation at T-72/T-48 with clear confirm / cancel / reschedule paths. HIPAA treats appointment reminders as treatment; keep content minimum necessary and respect channel rules (see HIPAA voice reminders).
• Layer B (high-risk): staff or governed agent outreach at T-7 with concrete planning, then stop-on-confirm.

Do not blast Layer B to everyone. That is how you recreate the cost problem the trial avoided.

3. Convert “I can’t make it” into a live slot

A confirmation that cannot cancel/reschedule in the same conversation is half a system. Train scripts and tools so the outcome codes include reschedule completed and slot released to waitlist.

Early cancel without refill is a vanity metric.

4. Own same-day and short-lead refill

Name a person (not “the front desk”) who owns the released slot list between 7 a.m. and noon. Connect that list to patients who wanted earlier access. This is where no-show work meets after-hours and overflow demand — the demand often already called; you failed to capture it.

5. Only then consider open-access redesign

If lead times are long and third-next-available is stuck, outreach alone will plateau. OA/advanced access needs demand measurement, protected same-week capacity, and clinician agreement. The review evidence says it often helps no-shows when those conditions exist.

Primary-care contact

Ungoverned reminders vs. playbook stack

Before

After

One automated blast to everyone; no risk flag; no outcome codes.

↓→

Layer A for all + Layer B for high-risk with logged outcomes.

Cancel means ‘thanks’ and an empty chair.

↓→

Cancel triggers waitlist offer the same morning.

No-show rate mixes late cancels and true misses.

↓→

Denominators split: no-show, early cancel, late cancel, arrival.

What not to do

Anti-patterns

Common US primary-care mistakes

Mistake

Why it fails

Playbook fix

Universal staff calls

Burns capacity; trial targeted ~11% of patients

Score risk; call the high-risk slice

Automation only for high-risk

Parikh: staff beat AUTO; AUTO still beats none

Keep AUTO for all; add human/agent for high-risk

Fees as first lever

Doesn’t free the slot early; trust cost

Prioritize early cancel + refill

‘Open access’ rename

Review: redesign + training required

Measure demand; protect same-week capacity

Staffing and AI without replacing the desk

The Shah intervention used patient service coordinators with a short training — not clinicians. That is the right labor class for confirmation and concrete planning. An AI voice layer can absorb Layer A overflow and some Layer B attempts if — and only if — handoff, scripts, and logs meet the same governance bar as front-desk overflow design. Clinical advice stays with humans.

Never promise the agent will “reduce no-shows by X%.” Promise a measured pilot on a defined cohort with the Shah-style endpoints: no-show rate, early cancel lead time, refill rate.

Metrics for a 30-day pilot

Track weekly:

1. No-show rate (your definition) overall and in the high-risk cohort.
2. Share of high-risk visits with a completed Layer B attempt.
3. Median days from cancel/reschedule to original appointment time.
4. Same-day / next-day refill rate of released slots.
5. Opt-outs and complaint codes (should stay flat or fall).

If (1) falls but (4) is zero, you improved attendance optics without recovering access for others.

Worked example: one pod, four weeks

Assume a primary-care pod with 400 visits/week and a true no-show rate of 12% (~48 missed arrivals). Roughly 11% of visits (~44) clear a high-risk flag. Layer B completes attempts on 80% of those (~35). If you reproduce even half of Shah’s absolute risk reduction on that slice, you recover a handful of arrivals per week — and, more importantly, you pull cancels earlier so the refill owner can place waitlist patients.

That math will not match your site until you measure. It is enough to justify a pod pilot without promising enterprise ROI in a sales deck.

Equity and fairness without freezing the program

Risk scores can encode structural inequities (insurance type, address proxies). Mitigations:

• Prefer features tied to behavior and lead time over zip-code stereotypes.
• Audit whether outreach completion rates differ by language or payer in ways that worsen access.
• Offer the same Layer A quality to everyone; use Layer B to add help, not to withhold it.
• Document the threshold rule so a patient can ask why they were called.

Fairness review belongs in the same weekly meeting as opt-outs.

Connecting outreach to phone capacity

No-show work fails when the inbound line cannot take a reschedule. Pair this playbook with overflow and after-hours coverage so “I need another day” becomes a completed booking, not an abandoned call. See front-desk overflow and after-hours cost.

What your practice can do this week

Primary-care no-show checklist

• Publish the denominator glossaryNo-show vs early/late cancel vs arrival.
• Build a transparent high-risk flagPrior miss + long lead time is enough for v1.
• Stand up Layer A + Layer BAUTO for all; T-7 outreach for high-risk only.
• Name the refill ownerOne accountable person for released slots before noon.
• Pilot 30 days on one podDo not enterprise-rollout before the metrics move.

1. 1
   Week 1 — Measure truthfully

   Pull 90 days of arrivals, no-shows, cancels; set baseline for overall and high-risk.

2. 2
   Week 2 — Wire Layer B

   Queue, script with concrete planning, outcome codes, stop-on-confirm.

3. 3
   Week 3 — Connect refill

   Every early cancel offers the slot to a short waitlist the same day.

4. 4
   Week 4 — Review

   Compare no-show ARD in high-risk to baseline; decide scale, tweak threshold, or stop.

Closing

US primary care does not need another motivational poster about “showing up.” It needs a playbook: define the metric, target the risky slice, stack automation with purposeful outreach, recover the slot early, and only then redesign access. The trials already told you the order of operations. Your job is to run it on your denominator — and to keep outreach inside the treatment lane HHS described for reminders.

Start with definitions. Add Layer B only where risk justifies staff time. Count refill, not vibes. Scale what moves the high-risk rate without breaking the desk.

For how Arbol thinks about US access operations, start at United States.

Script skeleton for Layer B (concrete planning)

Keep it short and non-clinical:

1. Identify the practice and the visit day/time without diagnosis detail.
2. Ask whether the patient can attend.
3. If yes: ask what they are doing beforehand (transport, childcare, work break) — the Shah-style prompt.
4. If no: offer reschedule options and release the slot.
5. Confirm how to reach the practice for changes; honor opt-out language as required by channel policy.

Train coordinators to avoid debating clinical urgency. Escalate clinical questions. Log the outcome before hanging up.

When fees enter the conversation

Some US practices ask about no-show fees. Fees do not free the chair early and can damage trust. If leadership insists, treat fees as a last lever after Layer A/B and refill are live — and measure whether cancels move earlier or simply become hostile. This playbook prioritizes recovered access over punitive optics.

Scaling beyond one pod

Only scale when high-risk no-show moves and refill is real. Then clone: same definitions, same outcome codes, same weekly sample. Change thresholds per specialty if lead times differ. Do not clone a black-box score you cannot explain.

Specialty twins inside the same building

Primary care often shares a phone tree with specialty pods. Do not assume one threshold fits cardiology follow-ups and same-day sick visits. Keep Layer A shared; tune Layer B eligibility and windows per template. Report no-show by department so a win in primary care is not masked by specialty lead-time effects.

Double-booking and predictive models — use with care

Some clinics double-book high-risk slots. That can protect clinician time and punish on-time patients with waits. If you experiment, cap it, measure lobby delay, and prefer Layer B outreach before aggressive overbooking. Predictive models help target calls; they should not silently decide who waits in a crowded waiting room without an ethics and ops review.

After the pilot: keep or kill

Scale when high-risk no-show falls and refill rises without complaint spikes. Kill or redesign when Layer B completion is low (bad numbers), when cancels do not move earlier, or when staff bypass the log. A killed pilot with learning is better than a zombie reminder program that burns trust.

Documentation for the next medical director

Leave behind: denominator glossary, risk-flag rule, Layer A/B windows, refill owner name, 30-day chart, and the three citations (Shah, Parikh, OA review). That packet survives leadership turnover better than a Slack thread about “the reminder thing.”