Data integration of structured and unstructured sources for assigning clinical codes to patient stays

Abstract

Objective: Enormous amounts of healthcare data are becoming increasingly accessible through the large-scale adoption of electronic health records. In this work, structured and unstructured (textual) data are combined to assign clinical diagnostic and procedural codes (specifically ICD-9-CM) to patient stays. We investigate whether integrating these heterogeneous data types improves prediction strength compared to using the data types in isolation.

Methods: Two separate data integration approaches were evaluated. Early data integration combines features of several sources within a single model, and late data integration learns a separate model per data source and combines these predictions with a meta-learner. This is evaluated on data sources and clinical codes from a broad set of medical specialties.

Results: When compared with the best individual prediction source, late data integration leads to improvements in predictive power (eg, overall F-measure increased from 30.6% to 38.3% for International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnostic codes), while early data integration is less consistent. The predictive strength strongly differs between medical specialties, both for ICD-9-CM diagnostic and procedural codes.

Discussion: Structured data provides complementary information to unstructured data (and vice versa) for predicting ICD-9-CM codes. This can be captured most effectively by the proposed late data integration approach.

Conclusions: We demonstrated that models using multiple electronic health record data sources systematically outperform models using data sources in isolation in the task of predicting ICD-9-CM codes over a broad range of medical specialties.

Keywords: clinical coding; data integration; data mining; electronic health records; international classification of diseases.

Figure 1:

The top graph shows the number of patient records in the available datasets per medical specialty; the bottom graph shows the number of unique ICD-9-CM codes (procedural: left, diagnostic: right) per specialty.

Figure 2:

Visualization of an example set of structured and unstructured data types. (A) The disparate databases. (B) An example structure of a patient medical record, with full lines representing connections of data types and their level and dashed lines to their given date. (C) The mapping on the partial stay level.

Figure 3:

Presence (in %) of the different data types over all datasets. The main bar shows the average presence, the error bars show the standard deviation between the different medical specialties. Abbreviations are explained in the “dataset” section.

Figure 4:

Example of data integration for the ICD-9-CM code “430.” This figure illustrates the difference between early and late data integration. (A) A pipeline for early data integration. (B) A pipeline for late data integration.

Figure 5:

Micro-averaged F-measure for increasing number of data sources. From left to right on the X axis, additional data sources are added (using late data integration). (A) ICD-9-CM diagnostic codes, (B) ICD-9-CM procedural codes. Results for specific specialties are in gray, averaged results in black.