The impact of fine-tuning of optical recognition system on database reliability

Although optical reading systems (ORS) are useful tools to transfer data from paper forms (PF) to electronic database, clinical researchers may be unaware that the system may represent a direct source of errors in final data base, and the impact of system fine-tuning on the final error rate is not usually considered. Present investigation was performed within the frame of HYDY study to measure error rate introduced in the final data base by an ORS. The impact of fine-tuning, the relationship between error rate and the number of areas to be blackened for each field, shadow errors, and test retest variability, were investigated. Ink threshold changes from 10% (default setting) to 4% (final setting) reduced the error rate from 0.65% to 0.05% (p<0.001), comparable to the rate achieved by visual checking (0.07%; OR 1.24; 0.79 to 1.97; p=0.526). Error rate was affected by the number of areas to be blackened in the form to fill a single field (0.02%, 0.07% and 0.09% for single, double and triple area respectively, p=0.008). The final error rate did not differ between overlapped and non-overlapped questions on paper forms (0.05% for both). In repeated readings no inter - form or inter- mark variability were found when ink threshold values was < 7%. Fine-tuning should be introduced as a standard procedure while using optical reading systems to reach the acceptable cut-point (0.1%) of error rate.