Prognostic and discriminatory abilities of imaging scoring systems in predicting COVID-19 adverse outcomes

Background

To evaluate the discriminatory ability of imaging modalities' scoring systems in the prediction of COVID-19 adverse outcomes like ICU admission, ventilatory support, or mortality.

Methods

We searched PUBMED, EBSCO, WEB OF SCIENCE, and SCOPUS. Two authors independently screened the resulting papers for fulfillment criteria. Meta-DiSc version 1.4, RevMan version 5.4, and MedCalc version 19.1 were used for test accuracy analysis, sensitivity and specificity analysis, and pooling Area under the curve for discriminatory assessment, respectively.

Results

Regarding mortality prediction, the computed tomography (CT) showed significantly higher sensitivity [80%; 95% CI 0.74–0.85] and positive likelihood ratio (PLR) [4.41 95% CI 2.94–6.61] relative to the Lung Ultrasound Score (LUS) approach, while the LUS approached the CT scan with specificity of 81% [95% CI 0.78–0.83] and negative likelihood ratio (NLR) of [0.32; 95% CI 0.16–0.64]. The pooled area under ROC for LUS was [AUC = 0.777, 95% CI 0.701–0.852; p < 0.001, I² = 74.86%, p = 0.019] while the pooled area under ROC for CT severity score was [AUC = 0.855, 95% CI 0.78–0.93; p < 0.001, I² = 93.73%, p < 0.001]. Regarding adverse outcomes prediction, the LUS had a slightly higher specificity of [78%; 95% CI 0.75–0.80] and PLR of [3.60; 95% CI 2.28–5.68] compared to CT score. The pooled AUC using LUS was (0.77, 95% CI 0.719–0.832; p < 0.001), while using CT severity score was (0.843, 95% CI 0.787–0.898; p < 0.001), and using X-ray scores was (0.814, 95% CI 0.751–0.878; p < 0.001).

Conclusion

CT severity score showed a better discriminatory ability in predicting COVID-19 adverse outcomes, as in-hospital mortality, ICU admission, and need for ventilatory support compared to LUS and X-RAY scores, while the LUS, being more specific, had a slightly better prognostic value.