Introduction Only one third of patients suspected of having heart failure (HF) see this diagnosis confirmed. Gender, age, education and obesity are major determinants of unconfirmed suspicions.

Objective To assess the impact of including major confounders in classification schemes based on clinical data solely (model 1–11 symptoms/signs) or considering objective evidence of cardiac dysfunction (model 2).

Methods Cross-sectional evaluation of 1115 community participants aged ≥45 years, 2006–2008. The individuals were classified by Latent Class Analysis with concomitant variables. The classification's prognostic value was assessed by the association with 6-year mortality in an independent sample of 753 subjects.

Results Bayesian Information criteria suggested the best solutions for model 1 and 2 was 2- and 3-class, respectively; the best solution for both models considering concomitant variables was 3-class.

Class 1 had high endorsement probabilities for all items (symptomatic HF); class 2 had high probability for volume overload and objective evidence of cardiac dysfunction and lower probability for subjective troubled breathing (asymptomatic cardiac dysfunction); class 3 had low endorsement probabilities for all items (non-cases).

The sex- and age-adjusted 6-year absolute risk of death was 13.5%, 4.3% and 2.7% for class 1, 2 and 3, respectively, in model 1; for model 2 it was 10.2%, 4.2% and 3.2%, respectively.

Conclusions When relying only on clinical data and not considering confounders, we were only able to distinguish symptomatic HF from the normal population. Considering confounders and evidence of cardiac dysfunction improved the discriminative power to distinguish a third group with asymptomatic cardiac abnormalities.