Invention and Validation of an Automated Camera System That Uses Optical Character Recognition to Identify Patient Name Mislabeled Samples

Abstract

BACKGROUND Mislabeled samples are a serious problem in most clinical laboratories. Published error rates range from 0.39/1000 to as high as 1.12%. Standardization of bar codes and label formats has not yet achieved the needed improvement. The mislabel rate in our laboratory, although low compared with published rates, prompted us to seek a solution to achieve zero errors. METHODS To reduce or eliminate our mislabeled samples, we invented an automated device using 4 cameras to photograph the outside of a sample tube. The system uses optical character recognition (OCR) to look for discrepancies between the patient name in our laboratory information system (LIS) vs the patient name on the customer label. All discrepancies detected by the system's software then require human inspection. The system was installed on our automated track and validated with production samples. RESULTS We obtained 1 009 830 images during the validation period, and every image was reviewed. OCR passed approximately 75% of the samples, and no mislabeled samples were passed. The 25% failed by the system included 121 samples actually mislabeled by patient name and 148 samples with spelling discrepancies between the patient name on the customer label and the patient name in our LIS. Only 71 of the 121 mislabeled samples detected by OCR were found through our normal quality assurance process. CONCLUSIONS We have invented an automated camera system that uses OCR technology to identify potential mislabeled samples. We have validated this system using samples transported on our automated track. Full implementation of this technology offers the possibility of zero mislabeled samples in the preanalytic stage.

Reduction in Hospital-Wide Clinical Laboratory Specimen Identification Errors following Process Interventions: A 10-Year Retrospective Observational Study

“…An electronic ID system is an additional successful measure to minimize specimen ID errors [ 6 , 8 , 20 , 28 – 33 ]. Reducing patient ID errors in the demanding ED environment is critical and should therefore be the first target for incorporation of an electronic ID system.…”

Integration of artificial intelligence in clinical laboratory medicine: Advancements and challenges

“…69,70 An automated camera system using optical character recognition was utilized to detect samples with patient name labeling errors. 71 An ML model was applied to identify instances of incorrect blood samples in the complete blood count results. 72 Recently, Yang et al developed a DL-based system for evaluating serum quality by analyzing sample images.…”

“…Studies employing ML models to identify mislabeled samples have shown that even relatively straightforward ML models can outperform humans 69,70 . An automated camera system using optical character recognition was utilized to detect samples with patient name labeling errors 71 . An ML model was applied to identify instances of incorrect blood samples in the complete blood count results 72 .…”

Patient and Sample Identification. Out of the Maze?

“…Innovative approaches are also emerging for univocal patient identification, such as human hand back skin texture detection (32), probabilistic matching (33) or near field communication (NFC). Interestingly, Hawker et al recently developed and validated an automated device based on four cameras which photographs the outside of a sample tube and then recognizes discrepancies between patient identity in the LIS versus that on the blood tube label by means of optical character recognition (OCR) (34). The system was found to have a high sensitivity for labeling errors (i.e., out of 742,977 total images that were passed, zero were mislabeled by patient name or exhibited patient name spelling discrepancies, yielding to 1.00 sensitivity; 95% CI 0.97–1.00), whereas the specificity was still modest (only 121 true patient mislabels were identified in the 266,853 images classified as fails by the system, yielding to 0.74 specificity; 95% CI 0.73–0.74).…”