We examine the benefits of leveraging big data information technology with a flexible program evaluation approach, coarsened exact matching (CEM), to estimate the effect of a fee-for-value initiative over time on medical allowed costs for a national population of continuously insured non-Medicare patients with at least one preventive care visit in the past 24 months. Within a Hadoop distributed file system, we use Hive SQL to match patients on coarsened bins of: age range, gender, geographic area, risk score range, and insurance product type. Bin definitions and sizes were established through literature review. Matching patients are weighted up or down accordingly to balance all covariates between groups. Pre- and post- covariate balance is measured with the L1 statistic, a measure of bias varying between 0 and 1 (0 indicating perfect covariate balance). More than five million patients with at least one preventative care visit to a primary care provider in the previous two years. An algorithm based on frequency and recency of preventive visits determines patient/provider alignment. Approximately 26% of patients are aligned to a fee-for-value primary care provider group. Patients’ ages range from 2 to 65 years. There are no other restrictions on patient eligibility (e.g. diagnosis, procedure). The initial fee-for-value sample had 665 unique geographic bins; the fee-for-service sample had 3,340. The final matched population retained 96% (1.3 million) of the fee-for-value sample across 646 geographic bins and 43% (1.6 million) of the control patients. The bias between fee-for-value and fee-for-service groups decreases from 67% to 42% before weighting; weighting reduces the bias to 0. Using commonly-described variables, accurate balance can be achieved in an efficient timeframe for observational studies on datasets incorporating millions of subjects. We recommend this algorithm using defined covariate bins to achieve matching on large patient datasets with minimal loss of case subjects.