Objectives: To review the diagnostic utility of the chest radiograph for left ventricular dysfunction.
Data sources: Structured MEDLINE searches, citation reviews of relevant primary research, review articles, and textbooks, personal files, and data from experts.
Study selection: Studies of patients without valvular disease that allowed calculation of the sensitivity and specificity of selected radiographic signs compared with a criterion standard of increased left ventricular preload or reduced ejection fraction.
Data extraction: Two independent readers reviewed 29 studies. Studies were pooled after stratification by radiographic finding, criterion standard, and clinical setting.
Main results: Redistribution best diagnosed increased preload with a sensitivity of 65% (95% confidence interval [CI] 55%, 75%) and specificity 67% (95% CI 53%, 79%). Cardiomegaly best diagnosed decreased ejection fraction with a sensitivity of of 51% (95% CI 43%, 60%) and specificity of 79% (95% CI 71%, 85%). Interrater reliability was fair to moderate for redistribution and moderate for cardiomegaly. The clinical setting affected results by decreasing the specificity of cardiomegaly to 8% in detecting increased preload in patients with severe systolic dysfunction. The absence of redistribution could only exclude increased preload in situations in which the suspicion (pretest probability) of disease was less than 9%, whereas redistribution could confirm increased preload when the pretest probability was greater than 91%. The absence of cardiomegaly could only exclude a reduced ejection fraction if the pretest probability was less than 8%, whereas cardiomegaly could confirm a reduced ejection fraction if the pretest probability was greater than 87%.
Conclusions: Redistribution and cardiomegaly are the best chest radiographic findings for diagnosing increased preload and reduced ejection fraction, respectively. Unfortunately, neither finding alone can adequately exclude or confirm left ventricular dysfunction in usual clinical settings. Redistribution is not always reliably interpreted.