A meta-analysis of three decades of validating thoracic impedance cardiography

Objective: To provide a meta-analysis of current literature concerning the validation of thoracic impedance cardiography (TIC) and to explain the vari ations in the reported results from the differences in the studies. Data Sources: A computer-assisted search of English-language, German, and D utch literature was performed for the period January 1966 to April 1997. Mo reover, references from review articles were obtained. Study Selection: A total of 154 studies comparing measurements of cardiac o utput or related variables obtained from TIC and a reference method were an alyzed. Data Extraction: Articles were classified by differences in Tie met hodology, reference method, and subject characteristics. Fisher's Z(f) tran sformed correlation coefficients were used to compare results. Data were po oled using the random-effects method. Data Synthesis: An overall pooled r(2) value of .67 (95% confidence interva l, 0.64-0.71) was found. However, the correlation was higher in repeated-me asurement designs than in single-measurement designs (r(2) = .53; 95% confi dence interval, 0.43-0.62). Further research using analysis of variance rev ealed a significant influence of the reference method and the subject chara cteristics on the correlation coefficient. The correlation was significantl y better in animals than in cardiac patients. Subgroup analysis revealed th at TIC correlated significantly better to the indirect Fick method than to echocardiography in healthy subjects. No significant influence of the appli ed TIC methodology was found. Discussion: The overall r(2) value of .67 indicates that TIC might be usefu l for trend analysis of different groups of patients. However, for diagnost ic interpretation, a r(2) value of .53 might not meet the required accuracy of the study. Great care should be taken when TIC is applied to the cardia c patient. However, because the applied reference method was of significant influence, differences between TIC and the reference method are incorrectl y attributed to errors in TIC alone.