Validating the accuracy of asthma outcomes in routine UK primary care data

Problem

Routine clinical data offers the opportunity to observe healthcare outcomes in populations but the measures used are rarely validated against the full medical record which includes free text and correspondence.

Approach

We compared data on unscheduled asthma care and action plan provision over 12-months extracted manually by inspection of the electronic healthcare records (EHR) (the reference standard) with electronically-extracted coded data from the same 500 patients. Combinations of Read codes and prescribing data were tested to derive the most accurate algorithm compared to the reference standard.

Findings

Ten practices each provided data on 50 people with asthma of whom 34% had an unscheduled asthma care event in the manually-extracted data. The best performing algorithm gave a sensitivity/specificity of 71% (95%CI 63% to 78%) and 82% (95%CI 77% to 86%) respectively. The intra-cluster correlation was 0.12 (95%CI 0.050 to 0.33). For action plan provision, the best performing algorithm only achieved a sensitivity of 34% (95%CI 18% to 54%).

Consequences

Unscheduled care, but not provision of action plans, can be detected with acceptable accuracy in routine data, though the intra-cluster correlation was high. Validating coded data against a reference standard is an important step in designing, analysing and interpreting the findings of clinical initiatives, implementation and real-life studies.