Résumé : Background: In March 2008, the D:A:D study published results demonstrating an increased risk of myocardial infarction (MI) for patients on abacavir (ABC). We describe changes to the use of ABC since this date, and investigate changes to the association between ABC and MI with subsequent follow-up. Methods: A total of 49,717 D:A:D participants were followed from study entry until the first of an MI, death, 1 February 2013 or 6 months after last visit. Associations between a person's 10-year cardiovascular disease (CVD) risk and the likelihood of initiating or discontinuing ABC were assessed using multivariable logistic/Poisson regression. Poisson regression was used to assess the association between current ABC use and MI risk, adjusting for potential confounders, and a test of interaction was performed to assess whether the association had changed in the post-March 2008 period. Results: Use of ABC increased from 10 % of the cohort in 2000 to 20 % in 2008, before stabilising at 18-19 %. Increases in use pre-March 2008, and subsequent decreases, were greatest in those at moderate and high CVD risk. Post-March 2008, those on ABC at moderate/high CVD risk were more likely to discontinue ABC than those at low/unknown CVD risk, regardless of viral load (≤1,000 copies/ml: relative rate 1.49 [95 % confidence interval 1.34-1.65]; >1,000 copies/ml: 1.23 [1.02-1.48]); no such associations were seen pre-March 2008. There was some evidence that antiretroviral therapy (ART)-naïve persons at moderate/high CVD risk post-March 2008 were less likely to initiate ABC than those at low/unknown CVD risk (odds ratio 0.74 [0.48-1.13]). By 1 February 2013, 941 MI events had occurred in 367,559 person-years. Current ABC use was associated with a 98 % increase in MI rate (RR 1.98 [1.72-2.29]) with no difference in the pre- (1.97 [1.68-2.33]) or post- (1.97 [1.43-2.72]) March 2008 periods (interaction P = 0.74). Conclusions: Despite a reduction in the channelling of ABC for patients at higher CVD risk since 2008, we continue to observe an association between ABC use and MI risk. Whilst confounding cannot be fully ruled out, this further diminishes channelling bias as an explanation for our findings.