Search strategy design
Based on previous literature18,19 and our clinical expertise, we identified 5 unique EMR information fields that could be used to determine asthma disease status, and search
parameters to identify patients with asthma within each field (Table 1). We defined additional parameters expected to exclude patients with chronic obstructive pulmonary disease (COPD) (Table 1).
To evaluate the accuracy of search algorithms in distinguishing asthma from other respiratory conditions, we identified patients who had a higher likelihood of having respiratory disease than the general population.20-22 We created approximately equal cohorts of patients likely
to have each of the following conditions: asthma, COPD, other respiratory conditions (ie, not asthma or COPD), andnonrespiratory conditions. Patients with COPD and with other respiratory conditions were included to ensure that algorithms could differentiate asthma from these clinically similar conditions, whereas patients with nonrespiratory conditions acted as healthy controls. For simplicity, these patients were limited to those with hypertension or musculoskeletal disorders.21 We first identified patients with a high likelihood of carrying 1 of these 4 diagnoses. For possible COPD, other respiratory conditions, and nonrespiratory conditions, we identified patients with relevant diagnoses in the electronic disease registry section of the EMR or a relevant corresponding billing diagnostic code billed within the past 3 years. Diagnoses listed in the electronic disease registry and the Ontario Health Insurance Plan billing codes used to identify patients are available from CFPlus.* For possible asthma, in addition to the above strategies, we identified patients who had been prescribed an inhaled asthma medication within the past 12 months (available from CFPlus*), while excluding patients who had been prescribed tiotropium bromide or ipratropium bromide (medications used predominantly for COPD). Any patients who fulfilled criteria to be included in more than 1 category were placed in the category identified by the most recent relevant billing
code or prescription.After identifying all potential patients within each of these 4 diagnostic categories, we used a random number generator to choose 150 patients in each (600 total) for
review, stratified by clinic site and by physician (Figure 1).