Aims: Older adults in diverse populations may interpret and respond to patient-reported outcome measures (PROMs) differently depending on where they live and age. Ignoring diversity when analyzing PROMs may lead to inaccurate measurements of their health. This study examined 1) heterogeneity in older adults’ responses to a widely used generic PROM, the Veterans Rand 12-item health survey (VR-12), and 2) measurement biases resulting from ignoring heterogeneity. Methods: Older adults (≥ 65 years) across Canada participated in an online survey, which included the VR-12, a PROM measuring physical and mental health (PH and MH), and a tool measuring various social determinants of health (SDOH). A 2-factor Item Response Theory (IRT) model was used to measure the PH and MH dimensions. Latent variable mixture models (LVMMs) were used to examine measurement non-invariance by allowing measurement model parameters to vary across latent classes. Measurement bias was calculated as the difference between IRT scores from a 1-class model (assuming homogeneity) and a k-class model (accommodating for heterogeneity). Multivariable linear regression models were used to examine associations between SDOH and positive and negative measurement bias. Results: Responses on the VR-12 (n=1649) were found to be heterogeneous and best represented by a 2-class model (Class proportions = 0.44, 0.56; Bayesian Information Criterion for 1- and 2-class models=36188 and 35990; Loglikelihood Ratio Test = p<0.001). Average positive and negative measurement bias was 0.40 (SD = 0.32) and –0.38 (SD = 0.23) for PH and 0.27 (SD = 0.20) and –0.26 (SD = 0.10) for MH. SDOH explained 4.9% of variance in positive and 11.6% in negative measurement bias for PH, and 14.9% and 7.2%, respectively, for MH. Conclusions: Older adults were found to respond differently to questions about their health, resulting in measurement biases associated with SDOH. LVMMs can be applied to improve equitable measurements of health for older adults.