Background Systematic loss to follow-up (LFU) creates selection bias and hinders generalisability in longitudinal cohort studies. Little is known about LFU risks in underserved populations including immigrants, those with depressive symptoms and language minorities. We used the Canadian Longitudinal Study on Aging (baseline 2012–2015 and 3-year follow-up 2015–2018) comprehensive and tracking cohorts to examine the association of language with LFU and its effect modification by immigrant status and depressive symptoms among participants from Quebec and those from outside Quebec.

Methods Language was English-speaking, French-speaking and Bilingual according to the language participants’ reported being able to converse in. Language minorities were French-speakers outside Quebec and English-speakers inside Quebec. LFU was withdrawal or not providing follow-up data. Logistic regression models assessed the associations of interest.

Results Our cohort included 49 179 individuals (mean age 63.0, SD 10.4 years; 51.4% female). Overall, 7808 (15.9%) were immigrants and 7902 (16.1%) had depressive symptoms. Language was 4672 (9.5%) French-speaking, 33 532 (68.2%) English-speaking and 10 976 (22.3%) Bilingual. Immigration ≤20 years (OR 1.84, 95% CI 1.34 to 2.53) or arrival at age >22 years (1.32, 95% CI 1.10 to 1.58) and depressive symptoms (1.23, 95% CI 1.13 to 1.46) had higher LFU risks. Bilingual (vs French-speaking) had lower LFU risk outside (0.45, 95% CI 0.24 to 0.86) and inside Quebec (0.78, 95% CI 0.63 to 0.98). LFU risk was higher in French-speakers (vs English-speakers) outside (2.33, 95% CI 1.19 to 4.55), but not inside Quebec. Female, higher income, higher education and low nutritional risk had lower LFU risks.

Conclusion Speaking only French (vs Bilingual), having depressive symptoms and immigrant status increased LFU risks, with the latter not modifying the language effect.