Multivariate models of self-reported health often neglected essential candidate determinants and methodological issues

Self-reported health is an important indicator of overall well-being that may be influenced by diverse parameters. We intended to evaluate the variety of candidate determinants used in models of self-reported health (SRH) and to examine the methodological problems encountered in multivariate models used in recent studies in this field. Medline searches identified articles published in 2002 in which SRH was included as an outcome, at least one other variable was used as a determinant of SRH, and the study population was not defined by the presence of specific diseases. Of 1,991 initially identified reports, 56 were eligible. In 91% of the eligible articles, multivariate models were used. In total, 133 different determinants of SRH were considered (median 7 determinants considered per study with multivariate models). The proportions of studies with problems in multivariate modeling were: overfitting, 10%; nonconformity to a linear gradient, 29%; no report of tests for interactions, 63%; unspecified coding of variables, 49%; and unspecified selection of variables, 29%. Models that try to identify what influences SRH should consider appropriate lists of candidate determinants, with proper attention to methodological aspects of multivariate modeling.