Most busy outpatient departments overbook already. A receptionist double-books the 11:00 slot because "that patient never comes", and sometimes she is right. The trouble is that the rule lives in one person's head, applies to the wrong patients and produces a waiting room that backs up on the days when everyone does turn up.
This article is for clinic owners, OPD managers and hospital operations leads who already send reminders and want a written scheduling rule that uses a predicted no-show probability. It is a method, not a result. We do not report outcomes from any clinic, and the published evidence on overbooking is thinner than vendors suggest. Treat everything below as a pilot design that your clinical leadership approves before it touches a live OPD.
If you have not yet tackled the basics, start with reducing no-shows through reminders and automation and the broader view of predictive analytics for no-shows in hospitals. This post begins where those stop: you have a risk score, and you must decide what the schedule does with it.
What the evidence does and does not say
A 2023 rapid systematic review in the Journal of the American Medical Informatics Association looked at interventions that use predictive models to target outpatients likely to miss appointments. It found good evidence for targeted text reminders and moderate evidence for targeted phone calls and patient navigator calls, but the abstract states that the effect of predictive model-based overbooking was uncertain. The authors also found little on cost-effectiveness, acceptability or equity.
Newer work is looking at the same question from the scheduling side. A 2025 BMJ Open paper, Development and evaluation of prediction models to improve the hospital appointments overbooking strategy at a large tertiary care hospital in the Sultanate of Oman, is a retrospective analysis of prediction models built for exactly this use. A retrospective model tells you the score is usable; it does not tell you that your waiting room will cope. Only your own pilot can do that.
So the order of operations is: reminders first, targeted outreach second, overbooking last and only for the sessions where the arithmetic and the clinical team both support it.
Where overbooking does not belong
Write the exclusions before the rules. A sensible default list:
- First visits that need pre-visit preparation, such as fasting investigations or bringing records the doctor must review.
- Sessions with a hard clinical constraint: procedures, chemotherapy chairs, dialysis, anything where a "late" patient cannot simply be seen later.
- Paediatric and elderly-heavy sessions where long waits carry a higher cost, unless the clinical lead decides otherwise.
- Any slot already booked as an urgent or follow-up-within-days appointment.
- Doctors who have not agreed to it. Overbooking is a workload decision as much as a revenue one.
What remains is typically routine review visits, repeat consultations and low-acuity first visits in high-volume general OPDs.
Inputs for the no-show score
The model does not need to be sophisticated. It needs honest features that exist at the time of booking and that you may lawfully use. Common candidates:
- Lead time: days between booking and appointment.
- History: previous attended, cancelled and missed visits for this patient.
- Confirmation status: whether the patient replied to the reminder. This is often the strongest late signal and it changes the score the day before.
- Booking channel: call centre, walk-in, website form, WhatsApp.
- Slot features: weekday, time of day, doctor, visit type.
- Distance or travel band, if captured.
Keep clinical details out of the model. A marketing or scheduling model has no need for diagnoses, and the less sensitive data it touches, the smaller your compliance exposure under the Digital Personal Data Protection Act. The DPDP Rules require a plain-language notice that itemises the data and the specified purposes, so "scheduling and reminders" should appear in that notice if you use attendance history this way.
If confirmation status and follow-ups sit in one place, for example in Pioneer CRM, the score can be refreshed when a patient confirms or goes silent without anyone exporting spreadsheets.
Calibrate before you act
A score that ranks patients well is not the same as a score that gives true probabilities, and overbooking depends on the second. Before use:
- Hold out the most recent two or three months of appointments as a test set. Do not test on the data the model learned from.
- Group test appointments by predicted probability (for example, tenths) and compare the average predicted show rate in each group with the actual show rate.
- If the groups drift apart, adjust or recalibrate. A model that says 0.9 for a group that actually attends 0.75 of the time will overbook the wrong sessions.
- Repeat monthly. Reminder changes, holidays and fee changes shift the baseline.
Do this separately for each department. A single hospital-wide model hides large differences between a physiotherapy clinic and an oncology review clinic.
The rule: cap the chance of overflow
The simplest defensible rule treats overbooking as a risk budget. For a session with C consultation slots, you add an extra booking only if the probability that more than C patients arrive in that period stays below a tolerance T.
You can compute this by multiplying probabilities. For a small block, list each booked patient's probability of showing and work out the chance that all of them appear.
Worked example (hypothetical numbers). An hour of three 20-minute slots has three booked patients with show probabilities of 0.90, 0.85 and 0.60. Expected arrivals are 2.35, so on average the hour is under-used. The team considers a fourth booking and the tolerance is set at 0.15.
- A fourth patient with show probability 0.70: the chance all four arrive is 0.90 x 0.85 x 0.60 x 0.70, about 0.32. That is above 0.15, so the booking is not added.
- A fourth patient with show probability 0.30: the chance all four arrive is about 0.14. That is inside the tolerance, so the booking is allowed.
Notice what the rule does. A booking is added only where the existing patients make an overflow unlikely, and the less likely the extra patient is to arrive, the safer the addition. Overbooking therefore pays off in blocks where attendance is already uncertain, not in blocks that merely look full.
Overflow also is not failure by itself. If the doctor can absorb one extra patient with a 10-minute delay, a tolerance of 0.15 may be fine. If the session runs on a tight chair or procedure schedule, set T near zero or switch overbooking off.




