Introduction: We examined the relationship between previous fluctuations in Mini-Mental State Examination (MMSE) scores, future changes in MMSE scores, and attrition from follow-up surveys, which helps in a more comprehensive interpretation of repeatedly collected MMSE scores. Methods: This 4-year longitudinal study included 2,073 community-dwelling older adults aged ≥65 years in Japan. The MMSE was administered at baseline (T0), 2 years (T1), and 4 years (T2) follow-up. We performed multinomial logistic regression analysis with the dependent variable, indicating the change in MMSE score from T1 to T2 (categorized as increase, no change [reference category], and decrease) and attrition at T2. The independent variables included the change in MMSE scores from T0 to T1 and MMSE scores at T0 and T1. Results: The mean MMSE score was 29 across the three time points. A one-point decrease in MMSE score from T0 to T1 was associated with 79% (95% confidence interval: 1.62, 1.97) higher odds of an increase in MMSE score from T1 to T2 and 28% (1.17, 1.40) higher odds of attrition at T2. A one-point decrement in the MMSE score at T0 and T1 was also associated with an increase in the MMSE score from T1 to T2 and attrition at T2. Conclusion: Focusing on cognitive fluctuation for 2 years, rather than cognitive function at a point in time, would have no remarkable advantage when focusing on future cognitive function and attrition. Our results emphasize the need for further studies to identify factors that distinguish between those who continue to attend follow-up surveys and show improvements in cognitive test scores and those who drop out.