Agreement between reported questionnaire data and medical records on diagnosis and COVID-19 symptoms at onset

Abstract

ABSTRACT The aim of this study was to assess whether there was agreement between self-reported data in a survey and medical records regarding diagnoses and symptoms at COVID-19 onset. The impact of sociodemographic factors on agreement between the two data sources was also assessed. Cross-sectional data were extracted from a Swedish longitudinal cohort study. In total, 401 non-hospitalized patients with a polymerase chain reaction-confirmed COVID-19 infection responded to a survey and agreed to a review of their electronic medical records. Agreement, estimated using the kappa statistic, sensitivity, and specificity were calculated for nine diagnoses and eleven symptoms. Differences between subgroups based on sociodemographic factors were assessed. The agreement between the self-reported data and medical records was at a substantial to moderate level for diagnoses such as diabetes mellitus (kappa 0.65, sensitivity 86%) and hypertension (kappa 0.59, sensitivity 56%) and at a fair level for more difficult-to-define conditions such as ongoing immunosuppressive treatment (kappa 0.27, sensitivity 25%). The agreement between the two data sources on symptoms was between fair and poor (kappa 0.36 for fever; kappa 0.05 for fatigue). Agreement for some diagnoses and symptoms varied across some sociodemographic subgroups, e.g. agreement in diabetes mellitus was significantly better in males (kappa 1.0) than females (kappa 0.52, homogeneity tests p = 0.02). In general, kappa values were lower for symptoms than diagnoses. The agreement between the two sources varied with diagnoses and symptoms and was also influenced by sociodemographic factors. This study illustrates that it is important to consider type of data used in the epidemiological studies as different information sources differ with quality and accuracy.