OBJECTIVETo study the interobserver variation and its sources in classifying chest radiographs of pneumoconiosis for improving diagnostic quality.

METHODSA set of 240 high-voltage chest radiographs including different stages of pneumoconiosis was developed and 5 sets of copies were made in the same condition. It was read by 7 experienced readers independently in eligible situation. Results of reading were recorded in unified reading sheets. The profusion of small opacities and the stage of pneumoconiosis were recorded as corresponding numerical value for analysis.

RESULTSThe coefficients of variation of 7 readers in total profusion of small opacities, average profusion of small opacities and stage of pneumoconiosis were 15.17%, 15.91%, 12.06%, respectively. The judgement of the presence or absence of small opacities, large opacities and coalescence of small opacities were closer, and the rates of consistency were 91.09%, 91.93% and 76.41% respectively. Technical quality of the radiographs in this study could not make a considerable impact on the reading results. If the shape of small opacities were recorded consistently by most readers as small rounded opacities or small irregular opacities, the judgement of total profusion of small opacities and stage of pneumoconiosis were more consistent; if the shape of small opacities were recorded inconsistently, the difference of judgement was higher.

CONCLUSIONInterobserver variation exists actually among different readers and the main sources of the variation come from uncertain judgement of the shape of small opacities.