OBJECTIVES

A database technical guidance in China summarized the strengths and limitations of using EMR and claims databases to estimate HCRU indicators. A major limitation of EMR is not being able to capture patient’s information in other hospitals, which may cause misclassification of new patients (new presenters to one hospital were misclassified as newly-diagnosed) and underestimate of HCRU. This study aims to quantify the issues using two case examples and propose real-world solutions.

METHODS

A city-level administrative claims database and a single-tertiary-hospital EMR database during the same study period were used. Sensitivity analyses using three other tertiary hospital EMR databases were performed to test the robustness of findings. Case 1 was to assess the misclassification of new osteoporosis patients. In DM patients, all osteoporosis were firstly identified, then a new osteoporosis patient identification algorithm (having any non-osteoporosis record in a one-year baseline) was applied in each database. Case 2: HCRU indicators in T2DM patients in each database.

RESULTS

Case 1: using the claims database, 89% of DM patients with osteoporosis had medical records in the 1-year baseline and 39% had no evidence of osteoporosis (39% new). The corresponding percentages was 67% and 51% in a single-tertiary hospital EMR (51% new). Misclassification rate was 24%. Extending the baseline to 2-year can reduce the misclassification rate to 15%. Case 2: Number of visits using claims was 8.0 times of that in a single-center EMR, sensitivity analysis results ranged 6.3-9.4; number of hospitalization (4.1 times (3.5-5.5)); length of stay per hospitalization (1.3 times (1.2–1.3)) and the average direct medical cost per-person-per-year (5.6 times (3.9–5.6)).

CONCLUSIONS

Compared with claims data, EMR data overestimates percentage of DM patients with a newly-diagnosed comorbid condition. Extending baseline may reduce misclassification; however cannot eliminate bias. EMR data also markedly underestimates HCRU per-person-per-year; and may be feasible for HCRU assessment in each hospitalization.