Comparing two approaches for studying symptom clusters: factor analysis and structural equation modeling.

PURPOSE: We investigated alternative ways of understanding the relationships among co-occurring symptoms in individuals with advanced cancer. While factor analysis has been increasingly used to identify symptom clusters, we argue that structural equation modeling is more appropriate because it permits investigating and testing of a greater variety of potential causal interconnections among symptoms.
METHODS: The sample included 82 palliative patients whose symptom scores were obtained from a database of the Capital Health Regional Palliative Care Program in Alberta, Canada, from 1995 to 2000. Data were analyzed using exploratory factor analysis (SPSS PASW 18.0.0, 2009) and compared to previous results obtained using structural equation modeling (LISREL 8.8, 2009).
RESULTS: Factor models failed to fit the covariance data, even though a single factor "explained" nearly half the variance. Structural equation models fit the data and explained an average of 66 % of the variance in the dependent latent variables. The factor analytic estimates were not clinically useful because they failed to correspond to the reasonable underlying common causes of the symptoms. Structural equation models, on the other hand, incorporated and tested specific clinically anticipated causal relationships among the symptoms and changes in those symptoms over time.
CONCLUSION: We used factor analysis to reanalyze data previously investigated with structural equation modeling and found that the structural equation models fit the data better and were more interpretable from a clinical perspective. We caution that factor models should be tested for consistency with the data and critically examined for inconsistencies with clinical understandings of the causal foundations of coordinated symptoms.