OBJECTIVES: To determine whether a new cancer diagnosis affects drug therapy for diabetes among Medicare beneficiaries. METHODS: We identified diabetes patients from a 5% random sample of Medicare beneficiaries enrolled in Medicare Part D in 2007 and 2008.  We then identified the subset with newly diagnosed cancer between January and December 2007 (N=4,457).  The comparison group represented beneficiaries with diabetes who had no evidence of cancer during the study period (N=29,187).  Use and adherence with oral hypoglycemic agents, RAAS-inhibitors, and statins were tracked during 6-month periods prior to and after cancer diagnosis and a matching index date for non-cancer controls. Monthly adherence was measured by the percentage of days covered (PDC).  We estimated difference-in-difference-in-difference (DDD) equations to assess the independent impact of cancer diagnosis and poor prognosis (died in month 7 to 12 after index date) on PDC controlling for individual characteristics.  RESULTS: We found that adherence declined between the pre-index and post-index periods for all drug classes with the sharpest drop evident among cancer patients who subsequently died in month 7 to 12.  Depending on drug class, a cancer diagnosis reduced PDC by 4.7% to 6.1% among survivors and by 7.6% to 16.4% for decedents.  Among non-cancer controls, the declines ranged between 1.5% and 3.2% (survivors) and between 5.7% and 9.1% (decedents). Similarly, the DDD results indicated small declines in medication adherence for cancer survivors relative to beneficiaries without a cancer diagnosis in all three drug classes (3 to 5 percentage points), whereas longer term cancer survivors had much better adherence to all three drug classes (10 to 12 percentage points higher) relative to beneficiaries with cancer who had a poor prognosis. CONCLUSIONS: A diagnosis of cancer among Medicare beneficiaries with diabetes significantly reduces adherence with evidence-based medications recommended in diabetes treatment guidelines.