ziadatalabs/FreeSyntheticAppointmentScheduling50M

Free Synthetic Clinic Appointment Scheduling (50M)

A free, fully synthetic dataset of 50,000,000 clinic appointment records, built for developers and researchers working on healthcare scheduling software, no-show prediction, capacity planning, or clinic-operations analytics.

This dataset is operational only — it describes the logistics of scheduling appointments, not anything clinical. There are no symptoms, diagnoses, treatments, medications, or lab values anywhere in it. Every field is about when and how an appointment was booked and whether it happened — the kind of data a scheduling system or an operations dashboard runs on. Every value is artificially generated; no real patients, providers, or clinics are represented.

No-show behavior is deliberately correlated with booking lead time — appointments booked far in advance are missed more often, a well-documented operational pattern — so a no-show-prediction model trained or tested against this data has to actually learn the relationship rather than pattern-match on noise.

Schema

Column	Type	Description
appointment_id	string	Unique identifier for the appointment
patient_id	string	Synthetic patient reference (no personal attributes)
provider_id	string	Synthetic provider reference
department	string	Clinic department name (operational label only, no clinical detail)
appointment_type	string	One of: new_patient, follow_up, routine_checkup, consultation, procedure
scheduled_date	string	ISO date (YYYY-MM-DD) the appointment is booked for
scheduled_time	string	Time slot (HH:MM), clinic hours
duration_minutes	int	Scheduled slot length
booking_lead_time_days	int	Days between booking and the appointment
appointment_status	string	One of: completed, no_show, cancelled, rescheduled
wait_time_minutes	float	Actual wait past scheduled time — only populated for completed appointments, null otherwise
is_telehealth	bool	Whether the appointment is virtual

How the correlations work

• No-show rate rises with booking_lead_time_days — appointments booked further out are missed more often.
• wait_time_minutes is only recorded for completed appointments — null for no-shows, cancellations, and reschedules.
• Certain departments (e.g. radiology, cardiology) carry longer average wait times, reflecting real operational load differences.
• Telehealth is more common for follow-ups and consultations than for procedures.

Format

Single Parquet file, Snappy compression, ~0.9 GB, 50,000,000 rows.

Quick start

import pandas as pd
df = pd.read_parquet("appointments_50M.parquet")

from datasets import load_dataset
ds = load_dataset("ziadatalabs/FreeSyntheticAppointmentScheduling50M")

import duckdb
con = duckdb.connect()
con.sql("SELECT * FROM 'appointments_50M.parquet' LIMIT 10").show()

Notes

• Entirely operational/logistical — contains no clinical information of any kind (no symptoms, diagnoses, treatments, or lab data).
• wait_time_minutes is null for any appointment that wasn't completed — don't treat a missing wait as zero.
• Use appointment_status as your prediction target; booking_lead_time_days is the primary signal driving no-shows.
• patient_id and provider_id are bare synthetic references for grouping/joins — they carry no personal or demographic attributes.

License & Usage

Released under CC BY-NC 4.0 — free for personal, research, and educational use, with attribution. Not licensed for commercial use.

No real people, appointments, or healthcare facilities are represented in this data. It is entirely synthetic.

---

Created by Zia Data Labs. Questions or feedback: zia.data.team@protonmail.com