Development of an Algorithm to Identify Patients with Physician-Documented Insomnia

Area of Science:

• Medical Informatics
• Clinical Data Analysis
• Sleep Medicine Research

Background:

• Insomnia is a prevalent sleep disorder impacting public health.
• Accurate identification of insomnia patients is crucial for effective treatment and research.
• Traditional methods relying on billing codes may underestimate insomnia prevalence.

Purpose of the Study:

• To develop and validate an advanced algorithm for classifying insomnia patients using electronic medical records (EMR).
• To compare the performance of the developed algorithm against traditional billing code-based methods.
• To assess the algorithm's ability to identify a larger and more comprehensive cohort of insomnia patients.

Main Methods:

• Developed an insomnia classification algorithm by analyzing a large electronic medical records (EMR) database (314,292 patients).
• Combined structured data (ICD-9 codes, prescriptions, lab results) and unstructured data (clinical notes mentioning sleep and psychiatric disorders).
• Evaluated algorithm performance using the area under the receiver operating characteristic curve (AUROC) and compared it to billing codes alone.

Main Results:

• The algorithm achieved a high classification performance (AUROC = 0.83) when combining structured and unstructured variables.
• This performance was significantly superior to using billing codes alone (AUROC = 0.55).
• The algorithm identified 36,810 insomnia patients, with less than 17% having an insomnia billing code, highlighting underdiagnosis.

Conclusions:

• An insomnia classification algorithm incorporating clinical notes significantly outperforms methods based solely on billing codes.
• This novel algorithm offers a more accurate, comprehensive, and efficient approach to identifying large insomnia patient cohorts.
• The findings suggest a potential for improved insomnia diagnosis and management through advanced data analytics in EMR.