The use of sequential pattern mining to predict next prescribed medications.

Area of Science:

• Data mining
• Pharmacology
• Health informatics

Background:

• Many medical treatments follow a stepwise approach with sequential medication prescriptions.
• Sequential pattern mining (SPM) is a data mining technique for identifying ordered event patterns.

Purpose of the Study:

• To assess the efficacy of SPM in identifying temporal medication relationships.
• To evaluate SPM's accuracy in predicting the next prescribed medication for patients.

Main Methods:

• Utilized Blue Cross Blue Shield of Texas claims data (2008-2011) for patients with diabetes medication prescriptions.
• Applied the CSPADE algorithm to mine sequential prescription patterns at drug class and generic drug levels.
• Evaluated prediction accuracy using a 90% training set and 10% test set, with 10-fold cross-validation.

Main Results:

• Identified 161,497 patients and mined stepwise therapy patterns aligning with clinical guidelines.
• Achieved 90.0% prediction accuracy at the drug class level and 64.1% at the generic drug level.
• Predictions were stable under cross-validation; using 1-2 historical prescriptions improved accuracy compared to no history.

Conclusions:

• Sequential pattern mining is effective for uncovering temporal medication relationships.
• SPM can accurately predict subsequent medications in a patient's regimen.
• Optimal prediction accuracy can be achieved without utilizing the patient's complete medication history.