OBJECTIVES: The use of patient level data (PLD) within cost-effectiveness models offers the potential to analyse the relationship between individual input profiles and predicted output.  The objective of this study was to ascertain if particular PLD input profiles were predictive of cost effectiveness sub-groups in Type 2 diabetes mellitus (T2DM) subjects. METHODS: This study used the IMS Core Diabetes Model (CDM), a validated and established diabetes model to evaluate the cost effectiveness of a new 2ndline oral therapy (Treatment) compared to metformin+ sulphonylurea (Control).  Delta treatment effects (favouring Treatment) were a 0.5% HbA1c reduction, 2kg weight change and a difference in symptomatic hypoglycaemia of 0.9/100 patient years. Annual diabetes specific therapy cost was £455 (Treatment) versus £70 (Control).  A PLD extract was obtained from NHANES over the period of 1999 to 2008 of T2DM subjects treated with oral therapy only. Costs (2010 UK£) and benefits were discounted at 3.5%.  Analysis of input/output data was undertaken using R. RESULTS: PLD for 1,858 T2DM subjects from NHANES were obtained with mean age 63.6 years of which 53% were male. Mean estimated cost per QALY of Treatment versus Control was £6,111.  Multivariate logistic regression identified age (p<0.05), SBP (p<0.001) and HbA1c (p<0.001) as model input variables significantly associated with cost effectiveness at a willingness to pay (WTP) threshold of £20,000. HbA1c was linearly and negatively correlated with incremental cost (-£569 per 1% increase (p<0.001)).  Subjects with baseline HbA1c>7.4% had significantly lower incremental costs compared to those £ 7.4% (£ 1,205 versus £3,462 respectively) and higher incremental QALY benefits (0.18 versus 0.15 respectively. CONCLUSIONS: The identification of patient characteristics associated with greater potential for health gain and reduced cost is an important goal.  The analysis of PLD alongside simulation model output provides an additional mechanism for informing healthcare decision-making.