AI-Driven Patient Scheduling and No-Show Prediction
How machine learning predicts appointment no-shows and enables targeted interventions to reduce missed visits and idle clinic time.
Health system · Medical group · FQHC / community health centre
What this is
Machine learning models predict which scheduled appointments are most likely to result in a no-show, enabling targeted interventions and intelligent scheduling adjustments. The models typically use historical appointment data, patient demographics, scheduling patterns, and contextual features like weather and time of day.
What the evidence shows
A retrospective study at Boston Children’s Hospital (Liu et al., npj Digital Medicine, April 2022) analyzed 161,822 appointments made by 19,450 patients with a 20.3% baseline no-show rate. Their neural network model achieved an AUROC of 0.975 and correctly identified 83% of no-shows at the time of scheduling, with a false-alert rate below 17%.
A decision analysis framework study (Deina et al., BMC Health Services Research, January 2024) tested prediction models across two healthcare datasets and achieved sensitivity above 0.94 for both, with an AUC of 0.94 on the higher-prevalence dataset. The study estimated annual financial losses of $12,574–$21,149 from just 557 no-show cases in a single department.
What changes operationally
The scheduling workflow gains a risk-stratification layer. Each appointment receives a no-show probability score at the time of booking, and staff follow tiered protocols: low-risk appointments receive standard reminders; moderate- risk appointments receive additional outreach; high-risk appointments may trigger overbooking, waitlist backfill, or proactive patient contact.
The key organizational decision is what interventions to pair with the predictions. Overbooking alone shifts risk to patients who do attend. Evidence-based alternatives include targeted reminder calls, transportation assistance, and flexible rescheduling options.
Where this sits in the RUAIH framework
Classified as low risk. The model informs scheduling logistics and resource allocation. It does not make clinical decisions, interact directly with patients, or modify clinical workflows. However, organizations should monitor for disparate impact across demographic groups.
The operational record
| Accountable owner | VP of ambulatory operations or patient access director, with analytics team support and frontline scheduling staff involvement in workflow redesign |
| Baseline to measure first | Appointment no-show rate (typically 5–30% depending on specialty and setting), clinic utilization rate, patient wait time for next available appointment, and revenue lost to unfilled slots |
| Reported effect size | In a retrospective study of 161,822 appointments at Boston Children's Hospital, Liu et al. (2022) found that a neural network model achieved an AUROC of 0.975 for predicting patient no-shows, correctly identifying 83% of no-shows at the time of scheduling with a false-alert rate below 17%, published in npj Digital Medicine on 20 April 2022 |
| What changes in the workflow | Before: scheduling staff book appointments into fixed time slots without visibility into each patient's likelihood of attending, and overbooking decisions are based on gut feel or blanket policies. After: at the time of booking, the model scores each appointment with a no-show probability. High-risk appointments trigger targeted interventions — additional reminder calls, transportation assistance offers, or strategic overbooking of specific slots. Scheduling staff see risk scores in their booking interface and follow tiered intervention protocols rather than applying uniform reminder policies to every patient. |
| Risk tier for governance | Lower risk tier |
How it fails
Three failure modes, written before deployment rather than discovered after it. Each one belongs in the monitoring plan for this tool.
- The model encodes and amplifies socioeconomic bias — patients from lower-income zip codes or those with public insurance may receive systematically higher risk scores, and if overbooking is the primary intervention, those patients experience longer wait times or are deprioritized, deepening access inequity.
- Overbooking driven by prediction scores leads to clinic overflow on days when the model overestimates no-shows — resulting in extended wait times, rushed appointments, clinician frustration, and paradoxically worse patient experience that increases future no-shows.
- The model is trained on pre-intervention data, and as interventions (reminders, ride services) take effect, the underlying no-show patterns change — but the model is not retrained, leading to degrading prediction accuracy and increasingly misallocated resources over 6–12 months.
When not to do this
Do not deploy predictive scheduling where the primary driver of no-shows is a systemic access barrier that the health system has not addressed — such as lack of public transportation, inability to take time off work, or inadequate childcare. The model will accurately predict who will not show up but will not fix the reason. Also inappropriate in settings where overbooking is the only planned intervention without corresponding investment in patient outreach, because overbooking without support services merely shifts the burden of system inefficiency onto patients who do attend.
Published under the Healthcare AI Institute editorial standard.
Written and reviewed against the standard by a physician-executive whose career spans three national healthcare systems. Last reviewed on 2026-08-17.
Researched from peer-reviewed studies in npj Digital Medicine and BMC Health Services Research, then structured against the seven-field use-case framework. Every quantitative claim links to its source.
The Institute accepts no vendor sponsorship, holds no vendor equity and takes no referral fees.