Autoverification of test results in the core clinical laboratory

Multivariate anomaly detection models enhance identification of errors in routine clinical chemistry testing

OBJECTIVES

Conventional autoverification rules evaluate analytes independently, potentially missing unusual patterns of results indicative of errors such as serum contamination by collection tube additives. This study assessed whether multivariate anomaly detection algorithms could enhance the detection of such errors.

METHODS

Multivariate Gaussian, k-nearest neighbours (KNN) distance, and one-class support vector machine (SVM) anomaly detection models, along with conventional limit checks, were developed using a training dataset of 127,451 electrolyte, urea, and creatinine (EUC) results, with a 5 % flagging rate targeted for all approaches. The models were compared with limit checks for their ability to detect atypical EUC results from samples spiked with additives from collection tubes: EDTA, fluoride, sodium citrate, or acid citrate dextrose (n=200 per contaminant). The study additionally assessed the ability of the models to identify 127,449 single-analyte errors, a potential weakness of multivariate models.

RESULTS

The KNN distance and SVM models outperformed limit checks for detecting all contaminants (p-values <0.05). The multivariate Gaussian model did not surpass limit checks for detecting EDTA contamination but was superior for detecting the other additives. All models surpassed limit checks for identifying single-analyte errors, with the KNN distance model demonstrating the highest overall sensitivity.

CONCLUSIONS

Multivariate anomaly detection models, particularly the KNN distance model, were superior to the conventional approach for detecting serum contamination and single-analyte errors. Developing multivariate approaches to autoverification is warranted to optimise error detection and improve patient safety.

Benefit analysis of the auto-verification system of intelligent inspection for microorganisms

In recent years, the automatic machine for microbial identification and antibiotic susceptibility tests has been introduced into the microbiology laboratory of our hospital, but there are still many steps that need manual operation. The purpose of this study was to establish an auto-verification system for bacterial naming to improve the turnaround time (TAT) and reduce the burden on clinical laboratory technologists. After the basic interpretation of the gram staining results of microorganisms, the appearance of strain growth, etc., the 9 rules were formulated by the laboratory technologists specialized in microbiology for auto-verification of bacterial naming. The results showed that among 70,044 reports, the average pass rate of auto-verification was 68.2%, and the reason for the failure of auto-verification was further evaluated. It was found that the main causes reason the inconsistency between identification results and strain appearance rationality, the normal flora in the respiratory tract and urine that was identified, the identification limitation of the mass spectrometer, and so on. The average TAT for the preliminary report of bacterial naming was 35.2 h before, which was reduced to 31.9 h after auto-verification. In summary, after auto-verification, the laboratory could replace nearly 2/3 of manual verification and issuance of reports, reducing the daily workload of medical laboratory technologists by about 2 h. Moreover, the TAT on the preliminary identification report was reduced by 3.3 h on average, which could provide treatment evidence for clinicians in advance.

Design of an algorithm for the detection of intravenous fluid contamination in clinical laboratory samples

OBJECTIVES

Contamination of blood samples from patients receiving intravenous fluids is a common error with potential risk to the patient. Algorithms based on the presence of aberrant results have been described but have the limitation that not all infusion fluids have the same composition. Our objective is to develop an algorithm based on the detection of the dilution observed on the analytes not usually included in infusion fluids.

METHODS

A group of 89 cases was selected from samples flagged as contaminated. Contamination was confirmed by reviewing the clinical history and comparing the results with previous and subsequent samples. A control group with similar characteristics was selected. Eleven common biochemical parameters not usually included in infusion fluids and with low intraindividual variability were selected. The dilution in relation to the immediate previous results was calculated for each analyte and a global indicator, defined as the percentage of analytes with significant dilution, was calculated. ROC curves were used to define the cut-off points.

RESULTS

A cut-off point of 20 % of dilutional effect requiring also a 60 % dilutional ratio achieved a high specificity (95 % CI 91-98 %) with an adequate sensitivity (64 % CI 54-74 %). The Area Under Curve obtained was 0.867 (95 % CI 0.819-0.915).

CONCLUSIONS

Our algorithm based on the global dilutional effect presents a similar sensitivity but greater specificity than the systems based on alarming results. The implementation of this algorithm in the laboratory information systems may facilitate the automated detection of contaminated samples.

Artificial Intelligence Applications in Clinical Chemistry

Artificial intelligence (AI) applications are an area of active investigation in clinical chemistry. Numerous publications have demonstrated the promise of AI across all phases of testing including preanalytic, analytic, and postanalytic phases; this includes novel methods for detecting common specimen collection errors, predicting laboratory results and diagnoses, and enhancing autoverification workflows. Although AI applications pose several ethical and operational challenges, these technologies are expected to transform the practice of the clinical chemistry laboratory in the near future.

Technical quality assurance and quality control for medical laboratories: a review and proposal of a new concept to obtain integrated and validated QA/QC plans

Technical quality assurance (QA) and quality control (QA/QC) are important activities within medical laboratories to ensure the adequate quality of obtained test results. QA/QC tools available at medical laboratories include external QC and internal QC, patient-based real-time quality control (PBRTQC) tools such as moving average quality control (MAQC), limit checks, delta checks, and multivariate checks, and finally, analyzer flagging. Recently, for PBRTQC tools, new optimization and validation methods based on error detection simulation have been developed to obtain laboratory-specific insights into PBRTQC error detection. These developments have enabled implementation and application of these individual tools in routine clinical practice. As a next step, they also enable performance comparison of the individual QA/QC tools and integration of all the individual QA/QC tools in order to obtain the most powerful and efficient QA/QC plans. In this review, a brief overview of the individual QA/QC tools and their characteristics is provided and the error detection simulation approaches are explained. Finally, a new concept entitled integrated quality assurance and control (IQAC) is presented. To enable IQAC, a conceptual framework is suggested and demonstrated for sodium, based on available published data. The proposed IQAC framework provides ways and tools by which the performance of different QA/QC tools can be compared in a so-called QA/QC error detection table to enable optimization and validation of the overall QA/QC plan in terms of alarm rate as well as pre-analytical, analytical, and post-analytical error detection performance.

Designing and validating an autoverification system of biochemical test results in Hatay Mustafa Kemal University, clinical laboratory

Introduction

Autoverification (AV) is a postanalytical tool that uses algorithms to validate test results according to specified criteria. The Clinical and Laboratory Standard Institute (CLSI) document for AV of clinical laboratory test result (AUTO-10A) includes recommendations for laboratories needing guidance on implementation of AV algorithms. The aim was to design and validate the AV algorithm for biochemical tests.

Materials and methods

Criteria were defined according to AUTO-10A. Three different approaches for algorithm were used as result limit checks, which are reference range, reference range ± total allowable error, and 2nd and 98th percentile values. To validate the algorithm, 720 cases in middleware were tested. For actual cases, 3,188,095 results and 194,520 reports in laboratory information system (LIS) were evaluated using the AV system. Cohen’s kappa (κ) was calculated to determine the degree of agreement between seven independent reviewers and the AV system.

Results

The AV passing rate was found between 77% and 85%. The highest rates of AV were in alanine transaminase (ALT), direct bilirubin (DBIL), and magnesium (Mg), which all had AV rates exceeding 85%. The most common reason for non-validated results was the result limit check (41%). A total of 328 reports evaluated by reviewers were compared to AV system. The statistical analysis resulted in a κ value between 0.39 and 0.63 (P < 0.001) and an agreement rate between 79% and 88%.

Conclusions

Our improved model can help laboratories design, build, and validate AV systems and be used as starting point for different test groups.

An Objective Approach to Deriving the Clinical Performance of Autoverification Limits

This study describes an objective approach to deriving the clinical performance of autoverification rules to inform laboratory practice when implementing them. Anonymized historical laboratory data for 12 biochemistry measurands were collected and Box-Cox-transformed to approximate a Gaussian distribution. The historical laboratory data were assumed to be error-free. Using the probability theory, the clinical specificity of a set of autoverification limits can be derived by calculating the percentile values of the overall distribution of a measurand. The 5th and 95th percentile values of the laboratory data were calculated to achieve a 90% clinical specificity. Next, a predefined tolerable total error adopted from the Royal College of Pathologists of Australasia Quality Assurance Program was applied to the extracted data before subjecting to Box-Cox transformation. Using a standard normal distribution, the clinical sensitivity can be derived from the probability of the Z-value to the right of the autoverification limit for a one-tailed probability and multiplied by two for a two-tailed probability. The clinical sensitivity showed an inverse relationship with between-subject biological variation. The laboratory can set and assess the clinical performance of its autoverification rules that conforms to its desired risk profile.

Optimization and Validation of Limit Check Error-Detection Performance Using a Laboratory-Specific Data-Simulation Approach: A Prerequisite for an Evidence-Based Practice

BACKGROUND

Autoverification procedures based on limit checks (LCs) provide important support to preanalytical, analytical, and postanalytical quality assurance in medical laboratories. A recently described method, based on laboratory-specific error-detection performances, was used to determine LCs for all chemistry analytes performed on random-access chemistry analyzers prior to application.

METHODS

Using data sets of historical test results, error-detection simulations of limit checks were performed using the online MA Generator system (www.huvaros.com). Errors were introduced at various positions in the data set, and the number of tests required for an LC alarm to occur was plotted in bias detection curves. Random error detection was defined as an LC alarm occurring in 1 test result, whereas systematic error detection was defined as an LC alarm occurring within an analytical run, both with ≥97.5% probability. To enable the lower limit check (LLC) and the upper limit check (ULC) to be optimized, the simulation results and the LC alarm rates for specific LLCs and ULCs were presented in LC performance tables.

RESULTS

Optimal LLCs and ULCs were obtained for 31 analytes based on their random and systematic error-detection performances and the alarm rate. Reliable detection of random errors greater than 60% was only possible for analytes known to have a rather small variation of results. Furthermore, differences for negative and positive errors were observed.

CONCLUSIONS

The used method brings objectivity to the error-detection performance of LCs, thereby enabling laboratory-specific LCs to be optimized and validated prior to application.

Use of the WISN method to assess the health workforce requirements for the high-volume clinical biochemical laboratories

BACKGROUND

The clinical laboratory services, as an essential part of health care, require appropriate staff capacity to assure satisfaction and improve outcomes for both patients and clinical staff. This study aimed to apply the Workload Indicators of Staffing Need (WISN) method for estimating required laboratory staff requirements for the high-volume clinical biochemical laboratories.

METHODS

In 2019, we applied the WISN method in all 13 laboratories within the Center for Medical Biochemistry of the University Clinical Centre of Serbia (CMB UCCS). A review of annual routinely collected statistics, laboratory processes observations, and structured interviews with lab staff helped identify their health service and additional activities and duration of these activities. The study outcomes were WISN-based staff requirements, WISN ratio and difference, and a recommendation on the new staffing standards for two priority laboratory workers (medical biochemists and medical laboratory technicians).

RESULTS

Medical biochemists' and laboratory technicians' annual available working time in 2019 was 1508 and 1347 working hours, respectively, for the workload of 1,848,889 samples. In general, the staff has four health service, eight support, and 15 additional individual activities. Health service activities per sample can take from 1.2 to 12.6 min. Medical biochemists and medical laboratory technicians spend almost 70% and more than 80% of their available working time, undertaking health service activities. The WISN method revealed laboratory workforce shortages in the CMB (i.e. current 40 medical biochemists and 180 medical laboratory technicians as opposed to required 48 medical biochemists and 206 medical laboratory technicians). Workforce maldistribution regarding the laboratory workload contributes to a moderate-high workload pressure of medical biochemists in five and medical laboratory technicians in nine organizational units.

CONCLUSIONS

The WISN method showed mainly a laboratory workforce shortages and workload pressure in the CMB UCCS. WISN is a simple, easy-to-use method that can help decision-makers and policymakers prioritize the recruitment and equitable allocation of laboratory workers, optimize their utilization, and develop normative guidelines in the field of clinical laboratory diagnostics. WISN estimates require periodic reviews.

Combined strategy of knowledge-based rule selection and historical data percentile-based range determination to improve an autoverification system for clinical chemistry test results

Background

Current autoverification, which is only knowledge‐based, has low efficiency. Regular historical data analysis may improve autoverification range determination. We attempted to enhance autoverification by selecting autoverification rules by knowledge and ranges from historical data. This new system was compared with the original knowledge‐based system.

Methods

New types of rules, extreme values, and consistency checks were added and the autoverification workflow was rearranged to construct a framework. Criteria for creating rules for extreme value ranges, limit checks, consistency checks, and delta checks were determined by analyzing historical Zhongshan laboratory data. The new system's effectiveness was evaluated using pooled data from 20 centers. Efficiency improvement was assessed by a multicenter process.

Results

Effectiveness was evaluated by the true positive rate, true negative rate, and overall consistency rate, as compared to manual verification, which were 77.55%, 78.53%, and 78.3%, respectively for the new system. The original overall consistency rate was 56.2%. The new pass rates, indicating efficiency, were increased by 19%‒51% among hospitals. Further customization using individualized data increased this rate.

Conclusions

The improved system showed a comparable effectiveness and markedly increased efficiency. This transferable system could be further improved and popularized by utilizing historical data from each hospital.

Customized middleware experience in a tertiary care hospital hematology laboratory

Background

In the clinical laboratory, middleware is a software application that sits between the analyzer and the laboratory information system (LIS). One of the more common uses of middleware is to perform more efficient result autoverification than can be achieved by the LIS or analyzer alone. In addition to autoverification, middleware can support highly customized rules to handle samples and results from specific patient locations. The objective of this study was to review the impact of customized middleware rules that were designed and implemented in the hematology laboratory of a 1000-bed tertiary care adult academic center hospital.

Methods

Three novel initiatives using middleware rules to achieve workflow efficiencies were retrospectively reviewed over different audit periods: preliminary neutrophil resulting for oncology patients, microcytosis interpretive comments, and 1 white blood cell differential (WBCD) reported per day. In addition, autoverification rates for complete blood count and differential (CBCD) and coagulation tests were calculated.

Results

A preliminary neutrophil count was released from middleware on average 64 min before the final CBCD for Leukemia/Bone Marrow Transplant (L/BMT) outpatients, and on average 59 min earlier for oncology patients. Reflexing interpretive comments for select instances of microcytosis removed on average 500 slides per month from technologist review with an estimated cost savings of approximately $3383.33 CAD per month. The 1 WBCD per day rule resulted in a 5.1% cancelation rate, resulting in an estimated monthly cost savings of $943.46 CAD in reagents and technologist time. Finally, middleware rules achieved very high autoverification rates of 97.2% and 88.3% for CBC and CBCD results, respectively.

Conclusions

Implementation of customized middleware hematology rules in our institution resulted in multiple positive impacts on workflow, achieving high autoverification rates, reduced slide reviews, cost savings, and improved standardization.

The development of autoverification system of lymphocyte subset assays on the flow cytometry platform

OBJECTIVES

Peripheral blood lymphocyte subsets are important parameters for monitoring immune status; however, lymphocyte subset detection is time-consuming and error-prone. This study aimed to explore a highly efficient and clinically useful autoverification system for lymphocyte subset assays performed on the flow cytometry platform.

METHODS

A total of 94,402 lymphocyte subset test results were collected. To establish the limited-range rules, 80,427 results were first used (69,135 T lymphocyte subset tests and 11,292 NK, B, T lymphocyte tests), of which 15,000 T lymphocyte subset tests from human immunodeficiency virus (HIV) infected patients were used to set customized limited-range rules for HIV infected patients. Subsequently, 13,975 results were used for historical data validation and online test validation.

RESULTS

Three key autoverification rules were established, including limited-range, delta-check, and logical rules. Guidelines for addressing the issues that trigger these rules were summarized. The historical data during the validation phase showed that the total autoverification passing rate of lymphocyte subset assays was 69.65% (6,941/9,966), with a 67.93% (5,268/7,755) passing rate for T lymphocyte subset tests and 75.67% (1,673/2,211) for NK, B, T lymphocyte tests. For online test validation, the total autoverification passing rate was 75.26% (3,017/4,009), with 73.23% (2,191/2,992) for the T lymphocyte subset test and 81.22% (826/1,017) for the NK, B, T lymphocyte test. The turnaround time (TAT) was reduced from 228 to 167 min using the autoverification system.

CONCLUSIONS

The autoverification system based on the laboratory information system for lymphocyte subset assays reduced TAT and the number of error reports and helped in the identification of abnormal cell populations that may offer clues for clinical interventions.

Improving the laboratory result release process in the light of ISO 15189:2012 standard

The ISO 15189:2012 standard section 5.9.1 requires laboratories to review results before release, considering quality control, previous results, and clinical information, if any, and to issue documented procedures about it. While laboratory result reporting is generally regarded as part of the post-analytical phase, the result release process requires a general view of the total examination process. Reviewing test results may follow with troubleshooting and test repetition, including reanalyzing an individual sample or resampling. A systematic understanding of the result release may help laboratory professionals carry out appropriate test repetition and ensure the plausibility of laboratory results. In this paper, we addressed the crucial steps in the result release process, including evaluation of sample quality, critical result notification, result reporting, and recommendations for the management of the result release, considering quality control alerts, instrument flags, warning messages, and interference indexes. Error detection tools and plausibility checks mentioned in the present paper can support the daily practice of results release.

Serum index rules prevent risk of analysing uncentrifuged tubes on automated biochemistry analysers

Around 1.5% of the total clinical biochemistry tests performed in laboratories are affected by preanalytical errors. Large, automated chemistry analysers prevent errors and interference by using control systems such as spectrophotometric measurements to evaluate serum indices, i.e. haemolysis (H), icterus (I), and lipemia/turbidity (L). However, still preanalytical errors can remain undetected. Our laboratory experienced an incident caused by laboratory-induced preanalytical errors, where approximately 100 sedimented lithium heparin samples bypassed centrifugation and entered our automated analyser. Based on index results, we investigated the possibility of using turbidimetry measurement, as a mean to prevent analysis on uncentrifuged sedimented whole blood. 14078 L-indices from 8 days in August 2019 were extracted from the middleware and used to develop and evaluate stop rules. Similarly, a one-day validation dataset was identified in December 2020 and used for an independent validation. Three different types of stop rules were evaluated: (1) A single L-index result above a cut-off; (2) A sequence of an L-index results above a cut-off; (3) A simple moving average of n results above a cut-off. A stop rule using 3 consecutive L-indices of 40-60 was found to be superior. However, practical implementation in the instrument middleware on a Roche Cobas 8000 only allowed a simple moving average of 110 (n = 5). This rule was found to be able to identify and stop sedimented whole blood analysis. Additionally, the rule has minimal impact on daily routine production in the laboratory.

Use of Middleware Data to Dissect and Optimize Hematology Autoverification

Background:

Hematology analysis comprises some of the highest volume tests run in clinical laboratories. Autoverification of hematology results using computer-based rules reduces turnaround time for many specimens, while strategically targeting specimen review by technologist or pathologist.

Methods:

Autoverification rules had been developed over a decade at an 800-bed tertiary/quarternary care academic medical central laboratory serving both adult and pediatric populations. In the process of migrating to newer hematology instruments, we analyzed the rates of the autoverification rules/flags most commonly associated with triggering manual review. We were particularly interested in rules that on their own often led to manual review in the absence of other flags. Prior to the study, autoverification rates were 87.8% (out of 16,073 orders) for complete blood count (CBC) if ordered as a panel and 85.8% (out of 1,940 orders) for CBC components ordered individually (not as the panel).

Results:

Detailed analysis of rules/flags that frequently triggered indicated that the immature granulocyte (IG) flag (an instrument parameter) and rules that reflexed platelet by impedance method (PLT-I) to platelet by fluorescent method (PLT-F) represented the two biggest opportunities to increase autoverification. The IG flag threshold had previously been validated at 2%, a setting that resulted in this flag alone preventing autoverification in 6.0% of all samples. The IG flag threshold was raised to 5% after detailed chart review; this was also the instrument vendor's default recommendation for the newer hematology analyzers. Analysis also supported switching to PLT-F for all platelet analysis. Autoverification rates increased to 93.5% (out of 91,692 orders) for CBC as a panel and 89.8% (out of 11,982 orders) for individual components after changes in rules and laboratory practice.

Conclusions:

Detailed analysis of autoverification of hematology testing at an academic medical center clinical laboratory that had been using a set of autoverification rules for over a decade revealed opportunities to optimize the parameters. The data analysis was challenging and time-consuming, highlighting opportunities for improvement in software tools that allow for more rapid and routine evaluation of autoverification parameters.

A model to establish autoverification in the clinical laboratory

OBJECTIVES

Autoverification is the process of evaluating and validating laboratory results using predefined computer-based algorithms without human interaction. By using autoverification, all reports are validated according to the standard evaluation criteria with predefined rules, and the number of reports per laboratory specialist is reduced. However, creating and validating these rules are the most demanding steps for setting up an autoverification system. In this study, we aimed to develop a model for helping users establish autoverification rules and evaluate their validity and performance.

DESIGN & METHODS

The proposed model was established by analyzing white papers, previous study results, and national/international guidelines. An autoverification software (myODS) was developed to create rules according to the model and to evaluate the rules and autoverification rates. The simulation results that were produced by the software were used to demonstrate that the determined framework works as expected. Both autoverification rates and step-based evaluations were performed using actual patient results. Two algorithms defined according to delta check usage (Algorithm A and B) and three review limits were used for the evaluation.

RESULTS

Six hundred seventeen rules were created according to the proposed model. 1,976 simulation results were created for validation. Our results showed that manual review limits are the most critical step in determining the autoverification rate, and delta check evaluation is especially important for evaluating inpatients. Algorithm B, which includes consecutive delta check evaluation, had higher AV rates.

CONCLUSIONS

Systemic rule formation is a critical factor for successful AV. Our proposed model can help laboratories establish and evaluate autoverification systems. Rules created according to this model could be used as a starting point for different test groups.

Chronometric vs. Structural Hypercoagulability

Prolonged tourniquet stasis induced by venepuncture can lead to the release of the plasma of cell lysis products, as well as tissue factor (TF), impairing the quality of coagulation test results. The accidental presence of TF in vitro can trigger the coagulation mechanism, generating a false decrease in prothrombin time (PT). Background and Objectives: Identification of short PT tests below the normal reference value that could suggest a situation of hypercoagulability. The study aimed to compare the results of the shortened PT tests at their first determination with the eventual correction following duplication of the analysis from the same sample. Materials and methods: Identification of the shortened PT tests has been carried out for a period of 4 months, upon 544 coagulation samples referred to the Hematology department of Sf. Spiridon County Clinical Emergency Hospital from Iasi, Romania. Results: Out of the 544 samples of which the results indicated a state of hypercoagulability, by repeating the determination from the same sample, for 200 (36.76%) PT tests (p = 0.001) the value was corrected, falling within the normal reference range. For 344 (63.24%) tests, the results suggested a situation of hypercoagulability. Conclusions: In order to guarantee the highest quality of the laboratory services, a proper interpretation and report of the patients' results must be congruent and harmoniously associated to the actual clinical condition of the patient. Duplication of the PT determination from the same sample would exclude situations of false hypercoagulability and would provide significant improvement for the patient's safety.

Using machine learning to develop an autoverification system in a clinical biochemistry laboratory

OBJECTIVES

Autoverification systems have greatly improved laboratory efficiency. However, the long-developed rule-based autoverfication models have limitations. The machine learning (ML) algorithm possesses unique advantages in the evaluation of large datasets. We investigated the utility of ML algorithms for developing an artificial intelligence (AI) autoverification system to support laboratory testing. The accuracy and efficiency of the algorithm model were also validated.

METHODS

Testing data, including 52 testing items with demographic information, were extracted from the laboratory information system and Roche Cobas^® IT 3000 from June 1, 2018 to August 30, 2019. Two rounds of modeling were conducted to train different ML algorithms and test their abilities to distinguish invalid reports. Algorithms with the top three best performances were selected to form the finalized ensemble model. Double-blind testing between experienced laboratory personnel and the AI autoverification system was conducted, and the passing rate and false-negative rate (FNR) were documented. The working efficiency and workload reduction were also analyzed.

RESULTS

The final AI system showed a 89.60% passing rate and 0.95 per mille FNR, in double-blind testing. The AI system lowered the number of invalid reports by approximately 80% compared to those evaluated by a rule-based engine, and therefore enhanced the working efficiency and reduced the workload in the biochemistry laboratory.

CONCLUSIONS

We confirmed the feasibility of the ML algorithm for autoverification with high accuracy and efficiency.

Highly accurate and explainable detection of specimen mix-up using a machine learning model

Background

Delta check is widely used for detecting specimen mix-ups. Owing to the inadequate specificity and sparseness of the absolute incidence of mix-ups, the positive predictive value (PPV) of delta check is considerably low as it is labor consuming to identify true mix-up errors among a large number of false alerts. To overcome this problem, we developed a new accurate detection model through machine learning.

Methods

Inspired by delta check, we decided to conduct comparisons with the past examinations and broaden the time range. Fifteen common items were selected from complete blood cell counts and biochemical tests. We considered examinations in which ≥11 among the 15 items were measured simultaneously in our hospital; we created individual partial time-series data of the consecutive examinations with a sliding window size of 4. The last examinations of the partial time-series data were shuffled to generate artificial mix-up cases. After splitting the dataset into development and validation sets, we allowed a gradient-boosting-decision-tree (GBDT) model to learn using the development set to detect whether the last examination results of the partial time-series data were artificial mixed-up results. The model’s performance was evaluated on the validation set.

Results

The area under the receiver operating characteristic curve (ROC AUC) of our model was 0.9983 (bootstrap confidence interval [bsCI]: 0.9983–0.9985).

Conclusions

The GBDT model was more effective in detecting specimen mix-up. The improved accuracy will enable more facilities to perform more efficient and centralized mix-up detection, leading to improved patient safety.