Using machine learning to link electronic health records in cancer registries - On the tradeoff between linkage quality and manual effort

Aug 3, 2026

Cancer registries connect large volumes of electronic health record (EHR) data reported by medical institutions to already registered records for the same individual and tumor. While deterministic and probabilistic approaches dominate automated linkage, uncertain records often require manual processing. The study evaluates how five machine learning models balance linkage accuracy with coder effort when linking reported tumor records to registry targets, using a real dataset.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages11

Reading Time28 min

AccessFree to download

Introduction

Record linkage in cancer registries

Matching rules and approaches

Methods and evaluation setup

Machine learning models vs deterministic baseline

Two-class and three-class linkage settings

Results

Accuracy and F1 score comparisons

Impact of manual processing volume

Conclusion

International Journal of Medical Informatics 185 (2024) 105387 Contents lists available at ScienceDirect International Journal of Medical Informatics journal [homepage: www.elsevier.com/locate/ijmedinf](homepage: www.elsevier.com/locate/ijmedinf) | Using machine learning to link electronic health records in cancer registries: On the tradeoﬀ between linkage quality and manual eﬀort Philipp Röchner a,b,∗ , Franz Rothlaufb | | | | | --- | --- | --- | --- | | a Cancer Registry, Institute for Digital Health Data Rhineland-Palatinate, Große Bleiche 46, Mainz, 55116, Germany
b Information Systems and Business Administration, Johannes Gutenberg University, Jakob-Welder-Weg 9, Mainz, 55128, Germany | | | | | A R T I C L E I N F O | | A B S T R A C T | | | Keywords:
Record linkage
Data matching
Cancer registry Electronic health records Machine learning
Data quality | | Background: Cancer registries link a large number of electronic health records reported by medical institutions to already registered records of the matching individual and tumor. Records are automatically linked using deterministic and probabilistic approaches; machine learning is rarely used. Records that cannot be matched automatically with suﬃcient accuracy are typically processed manually. For application, it is important to knowhow well record linkage approaches match real-world records and how much manual eﬀort is required to achieve the desired linkage quality. We study the task of linking reported records to the matching registered tumor in cancer registries.
Methods: We compare the tradeoﬀ between linkage quality and manual eﬀort of ﬁve machine learning methods (logistic regression, random forest, gradient boosting, neural network, and a stacked method) to a deterministic baseline. The record linkage methods are compared in a two-class setting (no-match/ match) and a three-class setting (no-match/ undecided/ match). A cancer registry collected and linked the dataset consisting of categorical variables matching 145,755 reported records with 33,289 registered tumors.
Results: In the two-class setting, the gradient boosting, neural network, and stacked models have higher accuracy and 􀀂1 score (accuracy: 0.968 −0 .978, 􀀂1 score: 0.983 −0 .988) than the deterministic baseline (accuracy: 0.964,􀀂1 score: 0.980) when the same records are manually processed (0 . 89% of all records). In the three-class setting, these three machine learning methods can automatically process all reported records and still have higher accuracy and 􀀂1 score than the deterministic baseline. The linkage quality of the machine learning methods studied, except for the neural network, increase as the number of manually processed records increases. Conclusion: Machine learning methods can signiﬁcantly improve linkage quality and reduce the manual eﬀort required by medical coders to match tumor records in cancer registries compared to a deterministic baseline. Our results help cancer registries estimate how linkage quality increases as more records are manually processed. | | 1. Introduction Identifying records that describe the same real-world entity is called record linkage, data deduplication, data matching, or entity resolution. Cancer registries collect and link information about cancer patients ina particular population. This information is used to: “1) deﬁne and monitor cancer incidence at the local, state, and national levels; 2) investigate patterns of cancer treatment; and 3) evaluate the eﬀectiveness of public health eﬀorts to prevent cancer cases and improve cancer survival” [1]. After matching reported records to individuals, cancer registries link incoming records to the corresponding registered tumor for that individual. Reported records match registered tumors either when all values describing the tumor are identical or when the data meet certain criteria. The latter case is of medical interest because tumors can change over time, diﬀerent

FAQs

Why is record linkage important for cancer registries?

Cancer registries use record linkage to identify matching records for individuals and to connect incoming reported data to the corresponding registered tumor, enabling incidence monitoring, treatment pattern analysis, and public health evaluation.

How does the study compare machine learning to deterministic record linkage?

It compares five machine learning methods (logistic regression, random forest, gradient boosting, neural network, and a stacked method) against a deterministic baseline in two-class (no-match/match) and three-class (no-match/undecided/match) settings.

What tradeoff does the study find between linkage quality and manual effort?

Machine learning models can improve linkage accuracy and F1 score while reducing manual work by medical coders; additionally, for most studied models, linkage quality increases as more records are processed manually.

Aug 3, 2026

Upload by Valentina