BACKGROUND: Discontinuation of tyrosine kinase inhibitor (TKI) therapy is feasible for some patients with chronic myeloid leukaemia with deep molecular responses; however, patients with stable major molecular response (MMR), but not MR4, have not been studied, nor has the effect of treatment de-escalation rather than outright cessation. We aimed to examine the effects of treatment de-escalation as a prelude to complete cessation, not only in patients with MR4 or greater, but also in those with MMR but not MR4. METHODS: We did this interim analysis of a non-randomised, phase 2 trial at 20 hospitals in the UK. We recruited patients (aged ≥18 years) with chronic myeloid leukaemia in first chronic phase who had received TKI for 3 years or more and were either in stable MR4 (BCR-ABL1:ABL1 ratio <0·01%; MR4 cohort) or in stable MMR (BCR-ABL1:ABL1 ratio consistently <0·1%) but not MR4 (MMR cohort) for 12 months or longer. Participants received half their standard TKI dose (imatinib 200 mg daily, dasatinib 50 mg daily, or nilotinib 200 mg twice daily) for 12 months. Molecular recurrence was defined as loss of MMR (BCR-ABL1:ABL1 ratio >0·1%) on two consecutive samples. The primary endpoint of this interim analysis was the proportion of patients who lost MMR on de-escalation and regained MMR on TKI resumption. Analyses were by intention to treat. This study is registered with ClinicalTrials.gov, number NCT01804985. FINDINGS: Between Dec 16, 2013 and April 10, 2015, we enrolled 174 patients into the MMR cohort (n=49) or the MR4 cohort (n=125). During the 12 months of half-dose therapy, 12 patients (7%) had molecular recurrence, all of whom regained MMR within 4 months of full-dose TKI resumption (median time to recovery 77 days). Recurrence was significantly lower in the MR4 cohort (three [2%; 90% CI 0·2-4·8] of 121 evaluable patients) than in the MMR cohort (nine [19%; 90% CI 9·5-28·0] of 48 evaluable patients; hazard ratio 0·12, 90% CI 0·04-0·37; p=0·0007), but was unrelated to previous TKI or TKI therapy duration. Adverse events (eg, lethargy, diarrhoea, rash, and nausea) improved during the first 3 months of de-escalation, though not thereafter. 16 serious adverse events were reported, including one fatality due to worsening pre-existing peripheral arterial occlusive disease in a patient who had received only imatinib. INTERPRETATION: TKI de-escalation is safe for most patients with excellent responses to TKI therapy, and is associated with improvement in symptoms. These findings show that lower TKI doses might maintain responses in these patients, implying that such patients could be unnecessarily overtreated. Studies of more ambitious de-escalation are warranted. FUNDING: Newcastle University and Bloodwise.