OBJECTIVES: Differences in the frequency and severity of hypoglycemic episodes associated with alternative treatments often contribute significantly to differences in QALYs (the denominator of the cost-effectiveness ratio).  The 2013 T2DM CADTH report examining third-line therapy used disutility values from studies conducted before 2009 based on small samples.  A new study with over 1,600 individuals with T2DM (including Canadian respondents) found disutility estimates that were meaningfully greater (e.g. 0.005 versus 0.000004767 per non-severe hypoglycemic event). This analysis examined the impact of these alternative estimates on QALYs in an evaluation of canagliflozin 300mg versus sitagliptin 100mg in patients inadequately controlled on MET+SU. METHODS: Two 40-year simulations were performed using a validated economic model (ECHO-T2DM), differing only in the hypoglycemia disutility weights: (A) CADTH and (B) new estimates. Disutilities for T2DM-related complications and other adverse events were sourced from the literature. Data for patient characteristics, treatment effects (i.e. A1C, SBP, BMI, and cholesterol), and rates of adverse events were obtained from a previously reported 52-week trial, where canagliflozin 300mg demonstrated statistically superior A1C-lowering, as well as reductions in blood pressure and weight versus sitagliptin 100mg. Hypoglycemic event rates were similar.  In the simulations, insulin (which has an excess risk of hypoglycemia) was added and titrated as necessary to maintain A1C < 7.0%. RESULTS: Discounted QALYs for canagliflozin and sitagliptin were (A) 8.4 and 8.32; (B) 7.91 and 7.81. The higher QALYs associated with canagliflozin versus sitagliptin are largely attributable to the delay in the need for insulin for canagliflozin-treated patients. The contribution of hypoglycemic events to these total QALY differences was greater in scenario B (20.8%) than scenario A (2.3%). CONCLUSIONS: These results illustrate the need to carefully consider downstream assumptions in economic evaluations. Using the older values yielded 20% fewer net QALYs, implying that the corresponding ICER would be approximately 25% greater.