Randomized clinical trials (RCTs) are considered the gold standard to demonstrate treatment efficacy. However, if patients are heterogeneous with respect to treatment outcomes, RCT estimates may overestimate the effectiveness of treatments in practice. Instrumental Variable (IV) estimation techniques applied to retrospective health care data can yield unbiased treatment estimates for patients at the “extensive margin” of practice. As a result, it can be shown that IV estimates are better suited to evaluate the cost-effectiveness of policies that modify existing treatment rates. OBJECTIVE: To estimate cost-effectiveness ratios using treatment estimates from IV techniques and contrast these results to cost-effectiveness ratios obtained using RCT estimates. METHODS: Data from 18,795 Iowa Medicaid eligible children with an index event of acute otitis media (AOM) from 1989-1995 were collected. Instrumental variables were constructed for each patient based on their relative access to various provider types. IV estimates were used to estimate cost-effectiveness ratios for AOM patients on the extensive margin. RESULTS: For these patients, IV estimates of the difference in cure rates between patients that were treated with antibiotics and those not treated ranged from 11% to 13% and were statistically significant from zero at the 1% level. These estimates are less than 15% difference in cure rates from RCTs (Rosenfeld et al. Journal of Pediatrics. 1994). Using the average antibiotic prescription charge observed from our sample, IV estimates translate into cost-effectiveness ratios between $123 and $145 per additional cured patient. In contrast, the RCT estimate yields a cost-effectiveness ratio of $107 per additional cured patient. CONCLUSION: If patients are heterogenous with respect to treatment outcomes, cost-effectiveness ratios estimated using RCT treatment estimates will understate the costs per outcome from changing existing treatment rates. IV estimation with retrospective data provides a useful alternative for policy-makers to evaluate costs per outcome at the margins of practice.