Validating epilepsy diagnoses in routinely collected data

Area of Science:

• Medical Informatics
• Epidemiology
• Health Services Research

Background:

• Routinely collected, anonymised healthcare data offers a valuable resource for epilepsy research.
• Validating data sources is crucial for ensuring the reliability of research findings.

Purpose of the Study:

• To validate algorithms for identifying individuals with epilepsy using primary care records.
• To assess the accuracy of different code combinations (diagnosis and anti-epileptic drug prescriptions) within the Secure Anonymised Information Linkage (SAIL) databank.

Main Methods:

• A reference population of 150 individuals with epilepsy and 150 without was established from hospital records.
• Three algorithms were tested using combinations of General Practitioner (GP) diagnosis codes and anti-epileptic drug (AED) prescription codes.
• These algorithms were applied to anonymised GP records within the SAIL databank, covering 2.4 million individuals.

Main Results:

• Combining diagnosis and AED prescription codes achieved 84% sensitivity and 98% specificity.
• Diagnosis codes alone showed 86% sensitivity and 97% specificity.
• AED prescription codes alone had 92% sensitivity and 73% specificity, but were highly accurate in children (88% sensitivity, 98% specificity).

Conclusions:

• General practitioner diagnosis and anti-epileptic drug prescription codes can be reliably used to identify individuals with epilepsy in anonymised Welsh healthcare records.
• These validated algorithms enhance the utility of routinely collected data for epilepsy research in the UK.
• The findings support the confident use of anonymised primary care data for epidemiological studies on epilepsy.