OBJECTIVES:  Link person-level inpatient drug and service utilization data with pre-admission and post-discharge histories, using nationally-representative hospital discharge and managed care claims databases. METHODS:  Linkages were developed from two de-identified health care databases:  1) discharge summaries and detailed billing data for the complete census of discharges from 171 US hospitals; 2) claims for inpatient services paid to these hospitals by private and public health plans that contribute to the MarketScan Research Databases.  Hospital discharge records were sorted by hospital name, patient year of birth, sex, principal diagnosis, date of admission, and date of discharge, and cases were identified that were uniquely identified by these variables. Paid claims were then searched for matching records with the same combination of the six variables.  These were considered to be the same patient, given that each combination of matching variables was unique within the hospital census.  To understand how this convenience sample relates to the universe of discharges from US hospitals, linked discharges were compared to the 2006 National Inpatient Sample (NIS). RESULTS:  For 2006 there were 77,277 linked discharges.  Compared with NIS, more were in Medicaid (52% v. 20%) and fewer in Medicare (20% v. 37%) or commercial (29% v. 34%) health plans, reflecting the payer mix of the claims database.  They were younger (44 v. 48 years) and more female (67% v. 58%) than NIS.   Average length of stay was 4.6 days in both samples.  Of the top 10 most frequent DRGs in NIS, accounting for 31% of US discharges, 8 were also in the top 10 of the linked sample. CONCLUSIONS:  Patient-level hospital discharge data can be enhanced by linking it to longitudinal histories from health plan administrative data.  Judicious use of this resource for outcomes research requires understanding potential selection biases.