Complete Guide to Patient Matching Models

This guide shows how to configure and tune a probabilistic matching model for MDMbox on your own: from the first definition of what a “correct match” means to expert-level model tuning. We walk through the full path of preparing a matching model for real-world healthcare data, with patient matching as the main example. Inside, you’ll learn how to run Goal Analysis and define match criteria, design the Model Structure through feature selection, grouping, and comparison order, optimize match time with blocking without losing quality, train the model on real data, evaluate matching quality, and perform Expert Tuning based on model behavior.

The guide also includes step-by-step examples of tuning a trained model using pattern distribution analysis, ways to reduce false positives by more than 98% while preserving recall, and common edge cases from large healthcare datasets — including family members, shared facility addresses, and noisy registration data. You’ll also find practical Splink-based configuration examples, weight adjustment workflows, and real tuning scenarios based on production-like datasets.