Predicting risk of undiagnosed COPD in primary care: Development and validation of the TargetCOPD model

Abstract

<b>Background:</b> Previous prediction models for assessing risk of undiagnosed COPD used data from routine diagnoses, which may be inaccurate because of under- and misdiagnosis. We developed and externally validated a primary care-based model using data from a unique case finding trial. <b>Methods:</b> Patients aged 40-79 years with no prior diagnosis of COPD received a screening questionnaire either by post or opportunistically at primary care attendances through a large case finding trial based in primary care in the West Midlands, UK. Those reporting chronic respiratory symptoms were assessed with spirometry. COPD was defined as presence of respiratory symptoms and post-bronchodilator FEV₁/FVC&lt;lower limit of normal. A prediction model was developed using logistic regression with predictor variables available from electronic health records from subjects who returned a postal questionnaire (n=2398, mean age 59.9 years, 52% male). The model was externally validated among subjects who returned an opportunistic questionnaire (n=1097). <b>Results:</b> A model containing age, smoking status, dyspnoea, and prescriptions of salbutamol and antibiotics discriminated between patients with and without undiagnosed COPD (validation c-statistic 0.74 [95% CI 0.68 to 0.80]). A cut-point of ≥7.5% predicted risk to prompt referral for diagnostic assessment had a sensitivity of 68.8% (95% CI 57.3 to 78.9%) and specificity of 68.8% (95% CI 65.8.1 to 71.6%), requiring 7 diagnostic assessments (95% CI 6 to 10) to identify 1 patient with undiagnosed COPD. <b>Conclusion:</b> We have developed and externally validated a readily applicable risk model for undiagnosed COPD using routine data from electronic health records in primary care.