Objective: High rates of dementia are evident in First Nations populations, and modifiable risk factors may be contributing to this increased risk. This study aimed to use a longitudinal dataset to gain insights into the long-term risk and protective factors for dementia and cognitive impairment not dementia (CIND) in a Torres Strait Islander and Aboriginal population in Far North Queensland, Australia. Study Design and Setting: Probabilistic data linkage was used to combine baseline health check data obtained in 1998/2000 and 2006/2007 for 64 residents in remote communities with their results on a single dementia assessment 10–20 years later (2015–2018). The relationship between earlier measures and later CIND/dementia status was examined using generalized linear modeling with risk ratios (RRs). Due to the small sample size, bootstrapping was used to inform variable selection during multivariable modeling. Results: One third of participants (n = 21, 32.8%) were diagnosed with dementia (n = 6) or CIND (n = 15) at follow-up. Secondary school or further education (RR = 0.38, 95% CI 0.19–0.76, p = 0.006) and adequate levels of self-reported physical activity (RR = 0.26, 95% CI 0.13–0.52, p < 0.001) were repeatedly selected in bootstrapping and showed some evidence of protection against later CIND/dementia in final multivariate models, although these had moderate collinearity. Vascular risk measures showed inconclusive or unexpected associations with later CIND/dementia risk. Conclusions: The preliminary findings from this small study highlighted two potential protective factors for dementia that may be present in this population. A tentative risk profile for later CIND/dementia risk is suggested, although the small sample size limits the applicability of these findings.