Temporal properties of diagnosis code time series in aggregate

Area of Science:

• Health Informatics
• Biostatistics
• Data Mining

Background:

• Time series analysis is crucial for health data research.
• Administrative diagnoses, such as International Classification of Diseases, Ninth Revision (ICD9) codes, are widely available but often considered unreliable.
• Understanding the temporal dynamics of diseases is essential for effective healthcare management.

Purpose of the Study:

• To investigate the temporal information contained within ICD9 code time series.
• To develop and evaluate models for characterizing disease time courses using ICD9 data.
• To assess the reliability of ICD9 codes for representing disease progression.

Main Methods:

• Utilized differential entropy of ICD9 code time series as a measure of disease time course.
• Employed Gaussian kernel smoothing for a more fine-grained characterization of disease temporal profiles.
• Validated models against a gold standard established by a panel of clinicians.

Main Results:

• The first model achieved an area under the curve of 0.83 in classifying diseases as acute or chronic.
• Identified distinct temporal profiles including permanent, chronic, and acute disease patterns.
• Observed specific condition dynamics, such as the refractory period for childbirth.

Conclusions:

• ICD9 codes, despite known limitations, possess valid and valuable temporal information.
• Time series analysis of ICD9 codes can effectively model disease progression and characteristics.
• These findings support the use of administrative data for understanding disease dynamics.