CoChat
Sign Up
Cited by 17Clinical Laboratory Practices and Quality ControlHealthcare Operations and Scheduling OptimizationArtificial Intelligence in Healthcare

Turnaround time prediction for clinical chemistry samples using machine learning

Open access
Tsai, Eline R., Demirtas, Derya, Hoogendijk, Nick, Tintu, Andrei N., Boucherie, Richard J.·2022·Clinical Chemistry and Laboratory Medicine (CCLM)
DOI 10.1515/cclm-2022-0668
Open this paper in CoChat — free accountOpen PDF

Abstract

Objectives: Turnaround time (TAT) is an essential performance indicator of a medical diagnostic laboratory. Accurate TAT prediction is crucial for taking timely action in case of prolonged TAT and is important for efficient organization of healthcare. The objective was to develop a model to accurately predict TAT, focusing on the automated pre-analytical and analytical phase. Methods: A total of 90,543 clinical chemistry samples from Erasmus MC were included and 39 features were analyzed, including priority level and workload in the different stages upon sample arrival. PyCaret was used to evaluate and compare multiple regression models, including the Extra Trees (ET) Regressor, Ridge Regression and K Neighbors Regressor, to determine the best model for TAT prediction. The relative residual and SHAP (SHapley Additive exPlanations) values were plotted for model evaluation. Results: The regression-tree-based method ET Regressor performed best with an R 2 of 0.63, a mean absolute error of 2.42 min and a mean absolute percentage error of 7.35%, where the average TAT was 30.09 min. Of the test set samples, 77% had a relative residual error of at most 10%. SHAP value analysis indicated that TAT was mainly influenced by the workload in pre-analysis upon sample arrival and the number of modules visited. Conclusions: Accurate TAT predictions were attained with the ET Regressor and features with the biggest impact on TAT were identified, enabling the laboratory to take timely action in case of prolonged TAT and helping healthcare providers to improve planning of scarce resources to increase healthcare efficiency.

AI snapshot

Written from the paper’s full text.

TL;DR

The study developed a machine learning model to predict turnaround time (TAT) for clinical chemistry samples in the automated pre-analytical and analytical phase at Erasmus MC. The Extra Trees Regressor performed best, achieving R²=0.63, MAE=2.42 min, and MAPE=7.35% for TAT, with workload in pre-analysis and number of modules as key predictors.

Key findings

  • Among the PyCaret regression models compared, the Extra Trees Regressor performed best for predicting TAT_pipet and final TAT; for TAT it achieved R²=0.63, MAE=2.42 min, and MAPE=7.35%, where the historical average TAT was 30.09 min.
  • For TAT_pipet, the tuned ET Regressor on the test set achieved R²=0.52, MAE=2.42 min, and MAPE=11.71%, where the historical average TAT_pipet was 19.41 min.
  • For TAT prediction, 76.6% of test samples had a relative residual between −0.1 and 0.1, and 95.7% between −0.25 and 0.25; a myopic mean-TAT policy achieved 42.6% and 84.8%, respectively.
  • SHAP analysis identified the most important features as the number of modules a sample is allocated to, the workload in pre-analysis (c8100) upon sample arrival, and whether the sample requires the CO₂-L test.

Methods

  • Retrospective data from January to March 2019 at Erasmus MC’s automated clinical chemistry laboratory; 96,126 initial samples were reduced to 90,543 after removing samples with >40 min duration in c8100 or c8000 as outliers (5.8%).
  • A total of 39 features were used, including priority level, workload in different stages upon sample arrival, number of modules, number of tests, and test-mix indicators.
  • Python 3.9 with PyCaret 2.2.0 compared multiple regression models; a 95% training / 5% test split performed best, with 10-fold cross-validation on the training set, parameter tuning of the ET Regressor, and evaluation via MAE, MAPE, R², relative residual plots, and SHAP values.

Limitations

  • The paper states that the data are right-skewed; this is generally not an issue for non-parametric methods such as the ET Regressor, but parametric methods could have been improved by transforming the data to be Gaussian.
  • The paper states that samples with >40 min duration in c8100 or c8000 were removed as outliers because delays were not explainable under regular circumstances or with the feature set, and that broadening TAT prediction to include manual transport between p612 and c8100 and result reporting remains challenging.
  • Evident from the text: the study uses single-center retrospective data from a three-month period and reports no external validation; SHAP computation also took approximately 3.5 days on a 16 GB RAM computer cluster.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1515/cclm-2022-0668

Cites

19 of 19 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

12 of 17 cited sources checked · 10 supported · 0 flagged · accuracy not yet measured

  • Findingown data

    The ET Regressor performed best among the compared regression models for predicting TAT_pipet in terms of MAE, R2, and MAPE.

    “Out of the regression models in PyCaret, the ET Regressor [16] performed best in terms of MAE, R 2 and MAPE (Table 3).”
    • Also stated in: § Abstract · plain
      “The regression-tree-based method ET Regressor performed best with an R 2 of 0.63, a mean absolute error of 2.42 min and a mean absolute percentage error of 7.35%, where the average TAT was 30.09 min.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [16] Geurts, P, Ernst, D, Wehenkel, L. Extremely randomized trees. Mach Learn 2006;63:3–42. https://doi.org/10.1007/s10994-006-6226-1. source
  • Premisecitation

    The ET Regressor is computationally efficient.

    “The ET Regressor is also computationally efficient [16].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [16] Geurts, P, Ernst, D, Wehenkel, L. Extremely randomized trees. Mach Learn 2006;63:3–42. https://doi.org/10.1007/s10994-006-6226-1. source
  • Premisecitation

    To minimize bias, the ET Regressor uses the full original training set instead of bootstrap replicas.

    “At the same time, to minimize bias, the ET Regressor uses the full original training set instead of bootstrap replicas [16].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [16] Geurts, P, Ernst, D, Wehenkel, L. Extremely randomized trees. Mach Learn 2006;63:3–42. https://doi.org/10.1007/s10994-006-6226-1. source
  • Premisecitation

    Only one study has predicted TAT for a specific sample, namely the time in post-analysis between result generation and result reporting.

    “However, we found only one study on predicting TAT for a specific sample, namely the time in post-analysis between result generation and result reporting [6].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [6] Eminağaoğlu, M, Vahaplar, A. Turnaround time prediction for a medical laboratory using artificial neural networks. Bilişim Teknol Derg 2018;11:357–68. source
  • Findingnone given

    The top 7 performing models in terms of MAE were also the top 7 performing models in terms of R2 and MAPE.

    “The top 7 performing models in terms of MAE were also the top 7 performing models in terms of R 2 and MAPE.”
    Quote in the paperNot a substantial claim
  • May combine two claims. 26
  • Classified as probably not a substantial claim. 27

  • May combine two claims.

    Premisecitation

    The high level of randomization of the ET Regressor drastically reduces variance and thus the probability of overfitting.

    “Furthermore, the high level of randomization of the ET Regressor drastically reduces the variance [16], and thus the probability of overfitting.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [16] Geurts, P, Ernst, D, Wehenkel, L. Extremely randomized trees. Mach Learn 2006;63:3–42. https://doi.org/10.1007/s10994-006-6226-1. source

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

    Premisecitation

    MAE and MAPE are commonly used measures in machine learning and provide a more natural and intuitive measure of model performance.

    “MAE and MAPE are commonly used measures in ML [9] and provide a more natural and intuitive measure of model performance [12,13].”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites [9] Shalev-Shwartz, S, Ben-David, S. Understanding machine learning: from theory to algorithms. New York: Cambridge University Press; 2014. source
    • cites [12] Hyndman, RJ, Koehler, AB. Another look at measures of forecast accuracy. Int J Forecast 2006;22:679–88. https://doi.org/10.1016/j.ijforecast.2006.03.001. source
    • cites [13] Willmott, CJ, Matsuura, K. Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance. Research 2005;30:79–82. https://doi.org/10.3354/cr030079. source

      The Concluding Remarks (page not supplied) call MAE the most natural measure of average error magnitude, supporting that part of the claim. The paper compares MAE with RMSE, not MAPE, and does not establish that MAPE is natural or intuitive. The separate assertion about common use in machine learning is attached to citation [9], not this source. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    There are very few studies on machine learning to predict TAT or investigate factors affecting TAT in medical diagnostic laboratories.

    “However, there are very few studies on ML to predict TAT or investigate factors affecting TAT in medical diagnostic laboratories [5,6].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [5] Thiha, S, Shewade, HD, Philip, S, Aung, TK, Kyaw, NTT, Oo, MM, et al.. Factors associated with long turnaround time for early infant diagnosis of HIV in Myanmar. Glob Health Action 2017;10:1–7. https://doi.org/10.1080/16549716.2017.1395657. source

      The Discussion (Key findings) says few studies had systematically analyzed factors associated with long turnaround time or nonreceipt of results for early infant HIV diagnosis. The Methods (Analysis and statistics) describe Poisson regression, not machine learning. The source supports the scarcity of studies investigating TAT factors in this specific setting, but not the claim about ML prediction or medical diagnostic laboratories generally. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [6] Eminağaoğlu, M, Vahaplar, A. Turnaround time prediction for a medical laboratory using artificial neural networks. Bilişim Teknol Derg 2018;11:357–68. source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The best choice for T (number of generated trees) is a tradeoff between computation time and accuracy.

    “The best choice for T (number of generated trees) is a tradeoff between computation time and accuracy, i.e., the more trees generated, the better the accuracy but the higher the computation time [16].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [16] Geurts, P, Ernst, D, Wehenkel, L. Extremely randomized trees. Mach Learn 2006;63:3–42. https://doi.org/10.1007/s10994-006-6226-1. source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    R2 is the most informative performance measure in terms of how often the model correctly predicted TAT.

    “Of these performance measures, R 2 is the most informative in terms of how often did the model correctly predict TAT [11].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [11] Chicco, D, Warrens, MJ, Jurman, G. The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE and RMSE in regression analysis evaluation. PeerJ Comput Sci 2021;7:e623. https://doi.org/10.7717/peerj-cs.623. source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The noisier the output, the higher the optimal value of n_min (minimum number of samples for splitting a node) to minimize overfitting by creating smaller trees.

    “The noisier the output, the higher the optimal value of n min (minimum number of samples for splitting a node) to minimize overfitting by creating smaller trees [16].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [16] Geurts, P, Ernst, D, Wehenkel, L. Extremely randomized trees. Mach Learn 2006;63:3–42. https://doi.org/10.1007/s10994-006-6226-1. source

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    The smaller F (number of features evaluated at a node), the stronger the randomization of the trees and the weaker the dependence of the structure of the trees on TAT values of the training set samples.

    “The smaller F (number of features evaluated at a node), the stronger the randomization of the trees and the weaker the dependence of the structure of the trees on TAT values of the training set samples [16].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [16] Geurts, P, Ernst, D, Wehenkel, L. Extremely randomized trees. Mach Learn 2006;63:3–42. https://doi.org/10.1007/s10994-006-6226-1. source

  • May combine two claims.

    Findingown data

    The optimal value of n_min for this study is 2, which is lower than the default value of 5.

    “Our study shows that for our case the optimal value of n min is 2, which is lower than the default value 5.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The mean TAT_pipet in the training set was 19.29 min.

    “The mean TAT pipet in the training set was 19.29 min.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    The distributions of the duration in the c8100, the duration in the c8000, TAT_pipet and TAT were right-skewed.

    “The distributions of the duration in the c8100, the duration in the c8000, TAT pipet and TAT were right-skewed (Supplementary Figure 1).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    The historical average TAT and TAT_pipet were 29.96 and 19.30 min, respectively.

    “The historical average TAT and TAT pipet were 29.96 and 19.30 min, respectively.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    For predicting TAT, 76.6% of test set samples had a relative residual between -0.1 and 0.1, and 95.7% had a relative residual between -0.25 and 0.25.

    “For predicting TAT, 76.6% of samples had a relative residual between -0.1 and 0.1, and 95.7% of samples had a relative residual between -0.25 and 0.25.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The model takes 0.38 s to predict TAT for all the test set samples.

    “The model takes 0.38 s to predict TAT for all the test set samples.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Model performance does not greatly improve after T=350, while training time increases linearly.

    “Furthermore, the model performance does not greatly improve after T=350, while the training time increases linearly.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Using a myopic prediction policy in which each TAT is predicted to be the mean TAT, 42.6% of samples had a relative residual between -0.1 and 0.1, and 84.8% had a relative residual between -0.25 and 0.25.

    “Using a myopic prediction policy in which each TAT is predicted to be the mean TAT, 42.6% of the samples had a relative residual between -0.1 and 0.1, and 84.8% had a relative residual between -0.25 and 0.25.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For predicting TAT_pipet, 59.5% of test set samples had a relative residual between -0.1 and 0.1, and 88.2% had a relative residual between -0.25 and 0.25.

    “The relative residual plot for TAT pipet (Figure 3) shows that 59.5% of samples had a relative residual between -0.1 and 0.1, and 88.2% of samples had a relative residual between -0.25 and 0.25.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For predicting TAT_pipet, the ET Regressor achieved an R2 of 0.52, MAE of 2.42 min, and MAPE of 11.71%, with a historical average TAT_pipet over the test set of 19.41 min.

    “This resulted in R 2 of 0.52, MAE of 2.42 min and MAPE of 11.71%, where the historical average TAT pipet over the test set was 19.41 min.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The test set is representative of the population of clinical chemistry samples at Erasmus MC.

  • May combine two claims.

    Findingown data

    The initial dataset contained 96,126 samples; after data cleaning, 90,543 samples remained, and 5.8% of the initial dataset was considered outliers and discarded.

    “The initial dataset contained 96,126 samples.After data cleaning, 90,543 samples remained, while 5.8% of the initial dataset was considered as outliers and discarded.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    For predicting TAT, the ET Regressor achieved an R2 of 0.63, MAE of 2.42 min, and MAPE of 7.35%, with a historical average TAT over the test set of 30.09 min.

    “Accordingly, for predicted TAT (TAT i = TAT pipet,i + max{incubation time sample i}), we obtained R 2 of 0.63, MAE of 2.42 min and MAPE of 7.35%, where the historical average TAT over the test set was 30.09 min.”
    • Also stated in: § Discussion · plain
      “In this paper, we presented a method to accurately predict TAT of clinical chemistry samples (R 2 =0.63,MAE=2.42,MAPE=7.35%,historical average TAT=30.09min)”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The test set is representative of the population of clinical chemistry samples at Erasmus MC.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Findingnone given

    The time spent between arrival at the p471 and arrival at the p612 typically takes less than 2 min, therefore not having a big contribution to TAT.

    “The time spent between arrival at the p471 and arrival at the p612 typically takes less than 2 min, therefore not having a big contribution to TAT.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Findingnone given

    The vast majority of Erasmus MC test results are automatically confirmed and reported, so the result generation timestamp differs typically only a few seconds from the result reporting timestamp.

    “As the vast majority of Erasmus MC test results are automatically confirmed and reported, the result generation timestamp differs typically only a few seconds from the result reporting timestamp.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Samples requiring the CO2-L test are prone to a higher TAT because they are not allowed to be uncapped for too long before analysis and wait in the c8100 for manual uncapping before insertion into a c8000.

    “Samples requiring the CO 2 -L test are prone to a higher TAT as they are not allowed to be uncapped for too long before analysis.After centrifugation, these samples wait in the c8100 for a laboratory technician to pick them up and manually uncap them before being inserted into a c8000.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The ET Regressor model had sufficiently small errors for predicting TAT and had 1.8 times more samples with a relative residual between -0.1 and 0.1 compared to using the myopic prediction policy.

    “Thus, the model had sufficiently small errors for predicting TAT and had 1.8 times (76.6 vs. 42.6%)more samples with a relative residual between -0.1 and 0.1 as compared to using the myopic policy.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given

    The proposed methodology is widely applicable for TAT prediction in medical diagnostic laboratories and potentially beyond.

    “The proposed methodology is widely applicable for TAT prediction in medical diagnostic laboratories and potentially beyond.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Average performance measures obtained from 10-fold cross validation on the training set were considered to get a more accurate view of model performance.

    “When comparing the models and tuning the parameters, the average performance measures obtained from 10-fold cross validation on the training set were considered to get a more accurate view of model performance.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    A machine learning model may effectively learn the relation between various features and TAT.

    “A machine learning (ML) model may effectively learn the relation between various features and TAT.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given

    This is the first TAT prediction study for predicting the time from the start of the automated pre-analytical phase until the end of the analytical phase for medical diagnostic laboratories.

    “This is the first TAT prediction study for predicting the time from the start of the automated pre-analytical phase until the end of the analytical phase for medical diagnostic laboratories.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    TAT prediction requires advanced algorithms because the diagnostic process is complex with different test-mix and workload dependent processing times, varying set-up, wait-to-batch and cycle times of the instruments, and sample prioritization.

    “The diagnostic process is complex with different test-mix and workload dependent processing times, varying set-up, wait-to-batch and cycle times of the instruments, and sample prioritization.Therefore, TAT prediction requires advanced algorithms.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductnone given

    The ET Regressor was fit on the entire training set and its performance was evaluated on the test set.

    “The ET Regressor was fit on the entire training set and its performance was evaluated on the test set.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    Accurate TAT prediction enables laboratories to take timely action in case of prolonged TAT and helps healthcare providers to improve planning of scarce resources to increase healthcare efficiency.

    “Accurate TAT prediction enables laboratories to take timely action in case of prolonged TAT and helps healthcare providers to improve planning of scarce resources to increase healthcare efficiency.”
    • Also stated in: § Conclusions · plain
      “ML techniques allow for accurate laboratory TAT predictions and the identification of features with the biggest impact on TAT, enabling the laboratory to take timely action in case of prolonged TAT and helping healthcare providers to improve planning of scarce resources to increase healthcare efficiency.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    The dataset was divided into training and test sets by random splitting, and several ratios were tested to determine the best one.

    “The dataset was divided into two sets, a training set, and a test set, by random splitting.Several ratios for splitting the data were tested, namely 95-5%, 90-10%, 80-20% and 70-30%, respectively, to determine the best one for our experiment.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The test set is representative of the population of clinical chemistry samples at Erasmus MC.

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    Samples arriving within a short time before a sample have a bigger influence on the TAT of this sample compared to samples that arrived much earlier and may almost be finished.

    “Furthermore, samples arriving within a short time before a sample have a bigger influence on TAT of this sample as compared to samples that arrived much earlier and may almost be finished.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    Sample characteristics and test-mix related features are expected to be more important for TAT prediction in pre-analysis and analysis than for post-analysis.

    “We also included sample characteristics and test-mix related features, which are expected to be more important for TAT prediction in pre-analysis and analysis than for post-analysis.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Using the workload for TAT prediction is especially important if the arrival pattern of the samples is not monotone.

    “Using the workload for TAT prediction is especially important if the arrival pattern of the samples is not monotone.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationreasoning

    Applying a priori feature selection is not necessary for the ET Regressor because it is a tree-based algorithm and tree-based algorithms naturally filter out less significant features.

    “As the ET Regressor is a tree-based algorithm and as tree-based algorithms naturally filter out less significant features, applying a priori feature selection is not necessary.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    The behavior of TAT lags behind the behavior of the workload.

    “In general, the behavior of TAT lags behind the behavior of the workload.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

    Interpretationnone given

    The described parameter tuning approach is widely applicable, while the best parameter settings depend on the study.

    “The described parameter tuning approach is widely applicable, while the best parameter settings depend on the study.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Understanding which features prolong TAT can help laboratories to reduce TAT.

    “Understanding which features prolong TAT can help laboratories to reduce TAT.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    Performance of the parametric methods could have been improved by first transforming the data to have a Gaussian distribution.

    “Performance of the parametric methods could have been improved by first transforming the data to have a Gaussian distribution.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Machine learning techniques allow for accurate laboratory TAT predictions and the identification of features with the biggest impact on TAT.

    “ML techniques allow for accurate laboratory TAT predictions and the identification of features with the biggest impact on TAT, enabling the laboratory to take timely action in case of prolonged TAT and helping healthcare providers to improve planning of scarce resources to increase healthcare efficiency.”
    Quote in the paperSubstantial claim
40 of 40

  • The paper relies on this without support.

    The relationship between the features and TAT observed in the January–March 2019 data will hold for future samples.

    “Therefore, we expect to have a good estimate of the performance of the model on an unseen dataset.”

  • The paper relies on this without support.

    The test set is representative of the population of clinical chemistry samples at Erasmus MC.

    “The dataset was divided into two sets, a training set, and a test set, by random splitting.”

  • The paper relies on this without support.

    The 39 features used are sufficient to capture the main determinants of TAT.

    “These high delays were not explainable under regular circumstances and with our feature set.”
3 of 3

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Tsai, E. R., Demirtas, D., Hoogendijk, N., Tintu, A. N., & Boucherie, R. J.
Affiliations
  • Erasmus MC (NL)
  • University of Twente (NL)
Corresponding author
Tintu, Andrei N.
ORCID
Venue
Clinical Chemistry and Laboratory Medicine (CCLM)
Publisher
Walter de Gruyter GmbH
Published
2022-10-12
Volume
60
Issue
12
Pages
1902-1910
ISSN
  • 1434-6621 (ISSN-L)
  • 1437-4331
DOI
10.1515/cclm-2022-0668
OpenAlex
W4304588850
Semantic Scholar
30ae2140826aed0cc8519cd955ffab5144d7c80e
Cited by
17
References
16
Citations by year
2026: 1 · 2025: 5 · 2024: 7 · 2023: 4
FWCI
2.06
Type
journal article
Language
en
Open access
Licence
Locations
  • Clinical Chemistry and Laboratory Medicine (CCLM) (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • University of Twente Research Information (publishedVersion, cc-by)
  • EUR Research Repository (Erasmus University Rotterdam) (acceptedVersion, cc-by)
  • EUR Research Repository (Erasmus University Rotterdam) (submittedVersion, cc-by)
  • University of Twente Research Information (submittedVersion, cc-by)
Topics
  • Health Sciences › Medicine › Physiology › Clinical Laboratory Practices and Quality Control
  • Health Sciences › Health Professions › Emergency Medical Services › Healthcare Operations and Scheduling Optimization
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
Keywords
pre-analytical phase, laboratory workload, SHAP feature importance, extra trees regressor, ridge regression, k-nearest neighbors regression
MeSH terms
Machine Learning, Chemistry, Clinical, Humans, Laboratories, Time Factors

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-23.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-23.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review