Front office · Scheduling & predictive analytics
Smart Scheduling
A logistic-regression no-show model (trained in Python, scored live here) ranks every appointment, explains why, and recommends the cheapest intervention that works. Open slots are backfilled from the waitlist automatically.
model: logistic_regressionAUC 0.7296explainable featureswaitlist matchingtwo-way SMS
13.4→8.6%
no-show rate
+38
visits recovered / wk
Tuesday, Nov 17 · all locations
10 appointments · expected no-shows: 2.7
Maria Gonzalez · 69% no-show risk
8:15 · Follow-up · Medicaid
Why: contribution to log-odds
Prior no-shows (12 mo)
Booking lead time
Medicaid coverage
Travel distance
Before 9 AM
← lowers riskraises risk →
Recommended: Call personally, offer Medicaid transportation (MAS) or a closer location, and backfill from the waitlist if unconfirmed by 24 h.
What-if simulator
Change the inputs and the model re-scores instantly
80%
risk
Waitlist backfill · Thursday
Open slots matched to waitlisted patients by visit type, provider, insurance, language and distance
- OpenThu 2:00 PM · Garden City · Dr. OkaforAna Pereira · Waitlisted 9 d · same visit type · 3.1 mi · auth approved
- OpenThu 2:30 PM · Garden City · Dr. OkaforLuis Moreno · Asked for earlier date · ES speaker · Medicare
- OpenThu 3:15 PM · Garden City · D. Park, PA-CGrace Kim · New patient referral 2 d old · 1.8 mi
- OpenThu 3:45 PM · Garden City · D. Park, PA-CDaniel Shah · Post-op check pulled forward · confirmed by SMS
Model card
Transparent by design
- Algorithm
- Logistic regression (L2)
- Training data
- 12000 synthetic appointments (seed 7)
- Test AUC
- 0.7296
- Base rate
- 13.4%
- Source
- ml/no_show_model.py
- Production plan
- Retrain monthly on de-identified EMR history; monitor drift and fairness by payer and language