Background Silicosis remains a global health problem. Chest radiography (CXR) is widely used for diagnosis and screening, despite concerns regarding its sensitivity. This systematic review and meta-analysis aimed to determine the pooled diagnostic accuracy of CXR compared to computed tomography (CT), high-resolution CT (HRCT) and autopsy, investigate heterogeneity according to severity of disease, and model the exposure-response relationship between silica exposure and CXR sensitivity.

Methods Medline, Embase, Scopus, and Web of Science databases were searched for relevant studies. Methodological quality was assessed using QUADAS-2. Meta-analyses were performed separately for different reference standards. We combined a linear meta-regression model to investigate differential heterogeneity according to severity of disease (determined by the ILO classification) and the exposure-response relationship with a previously unpublished logistic regression model from the Institute of Occupational Medicine. We estimated the number of cases missed using CXR.

Results Twenty studies met the inclusion criteria. CXR had moderate sensitivity (0.77; 95% CI: 0.64–0.87, I² = 84%, n=12) and high specificity (0.96, 95% CI: 0.82–0.99, I² = 27%, n=12) compared to HRCT (figure 1), and lower sensitivity (0.50, 95% CI: 0.45–0.55, I² = 0%, n=2) and high specificity (0.91, 95% CI: 0.87–0.93, I² = 20%, n=2) compared to autopsy. Compared to CT, CXR had high sensitivity (0.99, 95% CI: 0.86–1.00, I² = 0%, n=2) and moderate specificity (0.78, 95% CI: 0.15–0.99, I² = 0%, n=2). The linear meta-regression model (n=7 studies) showed that the proportion of severe cases (ILO category ≥2) explained a substantial proportion of the heterogeneity in sensitivity (R²= 88.56%). Modelling the exposure-response relationship revealed that CXR sensitivity increased with increasing silica exposure, reaching 1.0 at 7.3 mg/m³-years (95% CI: 3.1–18.6). However, the absolute missed cases increased with higher exposures due to the concomitant increase in prevalence; for example, at 4 mg/m³-years, 673/2058 (33%) cases were missed by CXR.

Discussion We report the sensitivity of CXRs in detection of silicosis and demonstrate that reliance on CXR alone in screening and health surveillance programmes will miss a significant burden of silicosis, especially in high-exposure settings, which may be mitigated through the wider use of HRCT.

• Download figure
• Open in new tab
• Download powerpoint

Abstract S17 Figure 1

Forest plots describing the sensitivity (left) and specificity (right) of CXR compared to HRCT