Machine learning models outperform manual result review for the identification of wrong blood in tube errors in complete blood count results

INTRODUCTION

Wrong blood in tube (WBIT) errors are a significant patient-safety issue encountered by clinical laboratories. This study assessed the performance of machine learning models for the identification of WBIT errors affecting complete blood count (CBC) results against the benchmark of manual review of results by laboratory staff.

METHODS

De-identified current and previous (within seven days) CBC results were used in the computer simulation of WBIT errors. 101 015 sets of samples were used to develop machine learning models using artificial neural network, extreme gradient boosting, support vector machine, random forest, logistic regression, decision trees (one complex and one simple) and k-nearest neighbours algorithms. The performance of these models, and of manual review by laboratory staff, was assessed on a separate data set of 1940 samples.

RESULTS

Volunteers manually reviewing results identified WBIT errors with an accuracy of 85.7%, sensitivity of 80.1% and specificity of 92.1%. All machine learning models exceeded human-level performance (p-values for all metrics were <.001). The artificial neural network model was the most accurate (99.1%), and the simple decision tree was the least accurate (96.8%). Sensitivity for the machine learning models varied from 95.7% to 99.3%, and specificity varied from 96.3% to 98.9%.

CONCLUSION

This study provides preliminary evidence supporting the value of machine learning for detecting WBIT errors affecting CBC results. Although further work addressing practical issues is required, substantial patient-safety benefits await the successful deployment of machine learning models for WBIT error detection.

Proficiency-based progression intern training to reduce critical blood sampling errors including ‘wrong blood in tube’

Background: Blood sampling errors including ‘wrong blood in tube’ (WBIT) may have adverse effects on clinical outcomes. WBIT errors occur when the blood sample in the tube is not that of the patient identified on the label. This study aims to determine the effect of proficiency-based progression (PBP) training in phlebotomy on the rate of blood sampling errors (including WBIT). Methods: A non-randomised controlled trial compared the blood sampling error rate of 43 historical controls who had not undergone PBP training in 2016 to 44 PBP trained interventional groups in 2017. In 2018, the PBP training programme was implemented and the blood sampling error rate of 46 interns was compared to the 43 historical controls in 2016. Data analysis was performed using logistic regression analysis adjusting for sample timing. Results: In 2016, 43 interns had a total blood sample error rate of 2.4%, compared to 44 interns in 2017, who had error rate of 1.2% (adjusted OR=0.50, 95% CI 0.36-0.70; <0.01). In 2018, 46 interns had an error rate of 1.9% (adjusted OR=0.89, 95% CI 0.65-1.21; p=0.46) when compared to the 2016 historical controls. There were three WBITs in 2016, three WBITs in 2017 and five WBITs in 2018.  Conclusions: The study demonstrates that PBP training in phlebotomy has the potential to reduce blood sampling errors. Trial registration number: NCT03577561

Factors associated with wrong blood in tube errors: An international case series - The BEST collaborative study

BACKGROUND

A wrong blood in tube (WBIT) error signifies a blood sample that does not match the patient identified on the sample label. WBIT errors can result in ABO mistransfusions.

STUDY DESIGN AND METHODS

In this international, multicenter, descriptive study, healthcare facilities provided detailed information on WBIT errors occurring from 1/1/2019 to 12/31/2020. Factors contributing to WBIT errors were classified as protocol violations, knowledge gaps, and slips/lapses.

RESULTS

331 WBIT errors were compiled from 36 centers in 11 countries. WBIT errors were most frequently detected through pretransfusion sample testing (191, 58%), with 38 (20%) detected by a second ("check") sample. WBIT errors were divided almost evenly between intended patient drawn/wrong label applied (166, 50%) and wrong patient drawn/intended label applied (158, 48%). Information on contributing factors was available for 260 WBIT errors; most involved a combination of protocol violations and slips/lapses (139, 53%). The most frequent contributing factor was another patient's sample labels or tubes being available during phlebotomy (61%). Protocol violations were more likely to result in wrong patient being drawn (p = .0007). In 43 WBIT errors, electronic positive patient identification (ePPID) was not used when available or was used incorrectly.

CONCLUSIONS

Protocol violations and slips/lapses frequently contribute to WBIT errors. Sample collection processes should be designed to minimize error opportunities; staff should be educated on why protocol compliance is critical for patient safety. Using ePPID does not eliminate all WBIT errors. Institutions using ePPID may elect to require check sample verification as an added safety measure.

Emergency departments are higher-risk locations for wrong blood in tube errors

BACKGROUND

Wrong blood in tube (WBIT) errors can lead to ABO mistransfusions. It is unknown if WBIT errors are more likely in specific healthcare locations or if specific collection practices influence the commission of WBIT errors.

STUDY DESIGN AND METHODS

Data on pretransfusion samples from calendar year 2019 were collected retrospectively by 39 transfusion services in nine countries. We compared the proportion of WBIT errors made in emergency departments (EDs), inpatient wards, and outpatient clinics.

RESULTS

In total, 143 WBIT errors were detected among 1,394,862 samples for an unadjusted aggregate WBIT proportion of 1.03/10,000 samples. Using a pooled random effects model, the WBIT proportion was estimated to be significantly higher in EDs (1.23/10,000 samples, 95% CI 0.62-2.43) than inpatient wards (0.71/10,000, 95% CI 0.44-1.14; p < .001) or outpatient clinics (0.24/10,000, 95% CI 0.08-0.65; p < .001) and significantly higher in inpatient wards than outpatient clinics (p = .043). The use of electronic positive patient identification (ePPID) systems was associated with a significantly lower WBIT proportion in the ED (odds ratio, OR: 0.32, 95% CI: 0.11-0.96, p = .041), but not in inpatient wards (OR: 0.45, 95% CI: 0.20-1.01, p = .054) or outpatient clinics (OR: 1.95, 95% CI: 0.39-9.74, p = .415).

DISCUSSION

Normalized for the number of samples drawn per location, the WBIT proportion in EDs was 1.7 times higher than inpatient wards and 5.1 times higher than outpatient clinics. EDs represent higher-risk clinical locations for WBIT errors, and electronic positive patient identification (ePPID) may provide a greater impact on safety in EDs relative to other clinical areas.

Prevalence of Near-miss Events of Transfusion Practice and Its Associated Factors amongst House Officers in a Teaching Hospital

Objectives

A near miss in transfusion practice is defined as a deviation from standard procedures discovered before transfusion and can lead to a transfusion error. Information on near-miss events provides pivotal data on areas of improvement to prevent actual errors in the future. Our study sought to determine the prevalence and rate of near-miss events and their associated factors amongst house officers (HO) in Hospital Universiti Sains Malaysia.

Methods

The initial part of this study is a descriptive cross-sectional study involving data collection from all requests sent for group, screen, and hold (GSH) and group and cross match (GXM) tests from 2011 to 2017. The association between sociodemographic, workplace, and experience factors with near-miss events amongst HO was analyzed with a case-control study using logistic regression.

Results

We reported 83 near-miss events with a prevalence of 0.034% (95% confidence interval 0.027–0.042). The rate of near-miss events was one in every 2916 requests. The mean reporting rate was 11.9 events per year. Clinical near miss predominated at 89.2% compared to 10.8% laboratory near miss. Mislabeled events (33.7%) were more than miscollected events (10.8%). HO were implicated with most events (83.1%). Most events were predominantly in the medical and obstetrics and gynecology wards amounting to 31.3% each. We found a significant association between the ages of HO with near-miss events.

Conclusions

The prevalence of near-miss events in our hospital was relatively low. Our study has shown areas for improvement include improving sampling practices in clinical areas, adequate training of laboratory technicians, and providing proper transfusion education. Interventions such as encouraging compliance to guidelines and training in clinical and laboratory areas to minimize the risk of mistransfusion should be considered.

Reducing sample rejection in Durban, South Africa

Objectives

Rejections of clinical chemistry specimens delays the availability of results, which may impact patient management. The study aims to measure sample rejection rate, identify reasons for sample rejection, evaluate the effect of a campaign to reduce rejection rates and discover which clinical units produced the most insufficient specimen.

Methods

The study measured specimen rejection rates and the contributions of different rejection reasons in calendar 2016 and April 2018-March 2019. The study undertook a 7-intervention campaign to reduce specimen rejection during the 2018-2019 intervention period. It compared rejections rates, number of months with rejection rates ≤1.2%, and distribution of rejection reasons between the two year-long intervals. The study also determined the origin for specimens rejected for the most common rejection reason during one month in the second period.

Results

The overall rejection rate fell significantly from 1.4% in pre-intervention period to 1.2% in the intervention period. The number of months with rejection rates within the target range increased significantly from 2 in the post-intervention period to 6 in the intervention period. Insufficient, hemolysed, and 'too-old' specimen decreased significantly, however, insufficient specimen remained the most frequent rejection reason. In February 2019, one-third of all insufficient specimen came from neonatal units and 24% from the pediatric units.

Conclusions

Interventions decreased significantly both overall and monthly rejection rates above target levels. Insufficient, hemolysed, 'too-old' specimen, became significantly less frequent, however, insufficient specimen remained the most frequent rejection reason. Over a month, most insufficient specimen came from neonatal and pediatric sites.

Registration errors among patients receiving blood transfusions: a national analysis from 2008 to 2017

BACKGROUND AND OBJECTIVES

The key first step for a safe blood transfusion is patient registration for identification and linking to past medical and transfusion history. In Canada, any deviation from standard operating procedures in transfusion is an error voluntarily reportable to a national database (Transfusion Error Surveillance System [TESS]). We used this database to characterize the subset of registration-related errors impacting transfusion care, including where, when and why the errors occurred, and to identify frequent high-risk errors.

MATERIALS AND METHODS

A retrospective analysis was conducted on transfusion errors reported to TESS by sentinel reporting sites relating to patient registration and patient armbands, between 2008 and 2017. Free-text comments describing the error were coded to further categorize into common error types. The number of specimens received in the transfusion laboratory was used as the denominator for rates to allow for comparison between hospital sites.

RESULTS

Five hundred and fifty-four registration errors were reported from 10 hospitals, for a global error rate of 5·4/10 000 samples (median 5·0 [interquartile range 3·7-7·0]). The potential severity was high in 85·7% of errors (n = 475). The patient experienced a consequence in 10·8% of errors (n = 60), but none resulted in patient harm. Rates varied widely and differed by nature across sites. Errors most commonly occurred in outpatient clinics or procedure units (n = 160, 28·8%) and in emergency departments (n = 130, 23·5%).

CONCLUSION

Registration errors affect transfusion at every step and location in the hospital and are commonly high risk. Further research into common root causes is warranted to identify preventative strategies.

Sample collection and sample handling errors submitted to the transfusion error surveillance system, 2006 to 2015

BACKGROUND

In Canada, transfusion-related errors are voluntarily reported to a tracking system with the goal to systematically improve transfusion safety. This report provides an analysis of sample collection (SC) and sample handling (SH) errors from this national error-tracking system.

STUDY DESIGN AND METHODS

Errors from 2006 to 2015 from 23 participating sites were extracted. A survey was conducted to obtain information regarding institutional policies. Samples received in the blood bank were used to calculate rates. "Wrong blood in tube" (WBIT) errors are blood taken from wrong patient and labeled with intended patient's information, or blood taken from intended patient but labeled with another patient's information.

RESULTS

A total of 42,363 SC and 14,666 SH errors were reported. Predefined low-severity (low potential for harm) and high-severity errors (potential for fatal outcomes) increased from 2006 to 2015 (low SC, SH: 13-27, 3-12 per 1000; high SC, SH: 1.9-3.7, 0.5-2.0 per 1000). The WBIT rate decreased from 12 to 5.8 per 10,000 between 2006 and 2015 (p < 0.0001). The overall WBIT rate was 6.2 per 10,000, with variability by site (median, 0.3 per 10,000; range, 0-17 per 10,000). Sites with error detection mechanisms, such as regrouping second sample requirements, had lower error rates than sites that did not (SC, SH: 12, 1 per 1000 samples vs. 17, 3 per 1000 samples; p < 0.0001).

CONCLUSION

WBIT rates decreased significantly. Low-severity error rates are climbing likely due to increased ascertainment and reporting. Prevention studies are necessary to inform changes to blood transfusion standards to eliminate these errors.

Effectiveness of Laboratory Practices to Reducing Patient Misidentification Due to Specimen Labeling Errors at the Time of Specimen Collection in Healthcare Settings: LMBP™ Systematic Review

Background

Specimen labeling errors have long plagued the laboratory industry putting patients at risk of transfusion-related death, medication errors, misdiagnosis, and patient mismanagement. Many interventions have been implemented and deemed to be effective in reducing sample error rates. The objective of this review was to identify and evaluate the effectiveness of laboratory practices/ interventions to develop evidence based recommendations for the best laboratory practices to reduce labeling errors.

Content

The standardized LMBP™ A-6 methods were used to conduct this systematic review. Total evidence included 12 studies published during the time periods of 1980 to September 2015. Combined data from seven studies found that the interventions developed as a result of improved communication and collaboration between the laboratory and clinical staff resulted in substantial decrease in specimen labeling errors (Median relative percent change in labeling errors: -75.86; IQI: -84.77, -58.00). Further data from subset of four studies showed a significant decrease in specimen labeling errors after the institution of the standardized specimen labeling protocols (Median relative percent decrease in specimen labeling errors: -72.45; IQI: -83.25, -46.50).

Summary

Based on the evidence included in this review, the interventions that enhance the communication and collaboration between laboratory and healthcare professionals can decrease the specimen identification errors in healthcare settings. However, more research is needed to make the conclusion on the effectiveness of other evaluated practices in this review including training and education of the specimen collection staff, audit and feedback of labeling errors, and implementation of new technology (other than barcoding).

The role of comprehensive check at the blood bank reception on blood requisitions in detecting potential transfusion errors

Pre-transfusion testing includes proper requisitions, compatibility testing and pre-release checks. Proper labelling of samples and blood units and accurate patient details check helps to minimize the risk of errors in transfusion. This study was aimed to identify requisition errors before compatibility testing. The study was conducted in the blood bank of a tertiary care hospital in north India over a period of 3 months. The requisitions were screened at the reception counter and inside the pre-transfusion testing laboratory for errors. This included checking the Central Registration number (C.R. No.) and name of patient on the requisition form and the sample label; appropriateness of sample container and sample label; incomplete requisitions; blood group discrepancy. Out of the 17,148 blood requisitions, 474 (2.76 %) requisition errors were detected before the compatibility testing. There were 192 (1.11 %) requisitions where the C.R. No. on the form and the sample were not tallying and in 70 (0.40 %) requisitions patient's name on the requisition form and the sample were different. Highest number of requisitions errors were observed in those received from the Emergency and Trauma services (27.38 %) followed by Medical wards (15.82 %) and the lowest number (3.16 %) of requisition errors were observed from Hematology and Oncology wards. C.R. No. error was the most common error observed in our study. Thus a careful check of the blood requisitions at the blood bank reception counter helps in identifying the potential transfusion errors.

Pathology consultation on electronic crossmatch

OBJECTIVES

The full crossmatch is traditionally the final step in compatibility testing, acting as a serologic double check for ABO compatibility and unexpected RBC antibodies. In this review, we discuss the development of electronic crossmatch (EXM), an approach for determining when EXM can be used, and its strengths and weaknesses.

METHODS

Because EXM relies on highly sensitive screening assays, antibodies are frequently encountered whose clinical significance must be investigated and interpreted. Our approach is to obtain further history, perform enhanced tube testing, and consider tests of immune reactivity or RBC survival.

RESULTS

For those without clinically significant antibodies, we found two alternatives: immediate-spin crossmatch (IS XM) and EXM. IS XM is prone to error related to serologic interference, whereas EXM depends on the accuracy of the sample label, accurate data entry, and informatics to avoid errors.

CONCLUSION

EXM is an alternative to the serologic test in patients who have no clinically significant antibodies.