Objective Data errors are a well-documented part of clinical datasets as is their potential to confound downstream analysis. In this study, we explore the reliability of manually transcribed data across different pathology fields in a prostate cancer database and also measure error rates attributable to the source data.

Design Descriptive study.

Setting Specialist urology service at a single centre in metropolitan Victoria in Australia.

Participants Between 2004 and 2011, 1471 patients underwent radical prostatectomy at our institution. In a large proportion of these cases, clinicopathological variables were recorded by manual data-entry. In 2011, we obtained electronic versions of the same printed pathology reports for our cohort. The data were electronically imported in parallel to any existing manual entry record enabling direct comparison between them.

Outcome measures Error rates of manually entered data compared with electronically imported data across clinicopathological fields.

Results 421 patients had at least 10 comparable pathology fields between the electronic import and manual records and were selected for study. 320 patients had concordant data between manually entered and electronically populated fields in a median of 12 pathology fields (range 10–13), indicating an outright accuracy in manually entered pathology data in 76% of patients. Across all fields, the error rate was 2.8%, while individual field error ranges from 0.5% to 6.4%. Fields in text formats were significantly more error-prone than those with direct measurements or involving numerical figures (p<0.001). 971 cases were available for review of error within the source data, with figures of 0.1–0.9%.

Conclusions While the overall rate of error was low in manually entered data, individual pathology fields were variably prone to error. High-quality pathology data can be obtained for both prospective and retrospective parts of our data repository and the electronic checking of source pathology data for error is feasible.