COVID-19 and chest CT: do not put the sensitivity value in the isolation room and look beyond the numbers

With interest we read the systematic review and meta-analysis by Kim et al. (1) on the value of chest CT in diagnosing COVID-19 infection. Kim et al. reported chest CT to have a high pooled sensitivity of 94% (95% confidence interval: 91%-96%), but a low specificity of 37% (95% confidence interval 26%-50%). However, we believe that there is no convincing evidence yet that chest CT achieves such a high sensitivity in diagnosing COVID-19 in clinical practice. Note that the far majority of studies that were included in the meta-analysis by Kim et al. (1) (58 of 63 studies) only enrolled patients with proven COVID-19 infection while cases without the disease were lacking. Strikingly, this is not in line with their exclusion criterion number 3: “lack of extractable data for a two-by-two contingency table”. As a result, these 58 studies only allowed for the calculation of sensitivity, and not specificity. However, the diagnostic value of a test depends on its ability to discriminate between diseased and non-diseased cases (2). Sensitivity and specificity are intertwined entities and are both dependent on the threshold value which is applied to discriminate between positive and negative cases (2). Generally, creating a high sensitivity by applying a low threshold is at the expense of specificity (2). Multiple studies in Kim et al.’s meta-analysis did not report which criteria were used as threshold value (1). The possibility that a low threshold was used, remains a realistic scenario. Applying a low threshold in cohorts of patients suspected of the disease (both with and without an actual COVID-19 infection) may result in virtually all cases being classified as having the disease. As a result, sensitivity values in these individual studies and the pooled estimate that was calculated by Kim et al. (1) may have been overestimated. It should also be noted that the 5 studies that did provide a 2×2 diagnostic contingency table, suffered from numerous methodological flaws. The lack of high-quality evidence, rather than the mathematical numbers, should have been the main conclusion in the otherwise excellent work by Kim et al. (1).