OBJECTIVES: Dynamic Panel Models (DPMs) capture time-varying adherence effect as well as control for state-dependence and unobserved biases. We applied DPMs to estimate the direct impact of adherence to diabetes medications on HbA1C and adherence to hypertension medications on systolic blood pressures (SBP) and diastolic blood pressures (DBP) for patients who have both type II diabetes and comorbid hypertension.  METHODS: A retrospective cohort study for patients newly starting oral diabetes or hypertension medication therapy between July 2006 and June 2007 with the pre-existing comorbid hypertension or diabetes prescription history was conducted using administrative claims from a large physician group in Southern California (N=4633). Time-varying medication adherence, covariates (prior outcomes, experience in hypoglycemia, experience in hospitalization, occurrence of micro- or macro-complications) and outcomes were measured every 6 months over 33 months of follow-up period. Available lag dependent and independent variables were used as instrumental variables. Parameter estimates were compared using fixed effects (FE), random effects (RE), and DPMs. The validity of the model specification was tested through 2nd order autocorrelation test and Sargan test of overidentifying restrictions. RESULTS: Applying DPMs, medication adherence, prior outcomes, and age showed significant impact on clinical outcomes. We found a 10 percentage point increase in adherence was associated with a 0.05 percentage point decrease (p=0.04) in HbA1C, a 0.62 mmHg decrease (p=0.02) in SBP, and a 0.47 mmHg decrease in DBP (p=0.002) applying DPMs. Using the FE or RE models, the adherence impact became smaller. A 10 percentage point increase in adherence was associated with a 0.04 percentage point decrease (p<0.001) in HbA1C, a 0.22 mmHg decrease (p<0.001) in SBP, and a 0.13 mmHg decrease (p<0.001) in DBP applying the FE model.  CONCLUSIONS: DPMs estimated higher adherence to diabetes or hypertension medications improved clinical outcomes, and the estimated impact was greater than the results from conventional models.