Editors: Liron Pantanowitz, MD, director of anatomical pathology, Department of Pathology, University of Michigan, Ann Arbor, and David McClintock, MD, associate chief medical information officer for pathology, Department of Pathology, University of Michigan.
Validation of a digital pathology system during the COVID-19 pandemic
September 2020—Whole slide imaging has been available for clinical, research, and educational use for decades, with several digital pathology systems cleared by the FDA for primary diagnosis. However, widespread adoption of this technology for routine practice has been slow. Likely reasons for the slow uptick in employing whole slide imaging (WSI) for sign-out include the cost of these systems, their lack of interoperability with laboratory information systems, pathologist resistance to using this digital modality, and regulatory restrictions on remote sign-out imposed by the Clinical Laboratory Improvement Amendments (CLIA). However, the COVID-19 pandemic led the Trump administration, on March 26, to temporarily waive these CLIA regulations, giving pathologists the flexibility to sign out cases digitally from their homes. The authors shared their experience validating a digital pathology system, at Memorial Sloan Kettering Cancer Center (MSKCC), for remote review during the COVID-19 pandemic. They used an Aperio GT450 scanner to scan surgical pathology slides at 40× magnification. The validation study included 108 cases composed of 1,196 slides with a mean of 11 slides per case. The study involved digitizing glass slides of formalin-fixed, paraffin-embedded and frozen tissue with H&E, IHC, and special stains. The glass slides were generated in-house or from consultation cases from referring institutions. Twelve pathologists with surgical pathology subspecialties were recruited to remotely review and report complete pathology cases using the digital pathology system from home. They were able to easily access scanned slides from within the laboratory information system by connecting to their institution’s workstations via a secure virtual private network (VPN). The pathologists used a custom vendor-agnostic viewer and consumer-grade computers and monitors (monitor size, 13.3–42 inches; resolution, 1,280 × 800–3,840 × 2,160 pixels). Diagnoses made from home using the digital setup were compared with those rendered when examining the corresponding glass slides using a light microscope in MSKCC’s CLIA-certified laboratory. The pathologists recorded several metrics apart from a top-line diagnosis, such as margin status, lymphovascular invasion, perineural invasion, pathology stage, and ancillary testing to measure intraobserver concordance. The authors reported a major diagnostic equivalency of 100 percent between digital and glass slide diagnoses. The validation study design and outcome satisfied the CAP guideline recommendations for validating WSI for diagnostic purposes, with the exception of the washout period, which was only two days. After completing the validation study, not only could MSKCC use its digital pathology system for primary diagnosis, but it could conduct remote review and reporting of pathology specimens, including from pathologists’ homes during the COVID-19 public health emergency.
Hanna MG, Reuter VE, Ardon O, et al. Validation of a digital pathology system including remote review during the COVID-19 pandemic. Mod Pathol. 2020;33(6). doi:10.1038/s41379-020-0601-5.
Correspondence: Dr. Matthew G. Hanna at hannam@mskcc.org
Using predictive analytics to detect falsely elevated POC potassium results
Providers primarily use their clinical judgment to detect preanalytical errors that can affect the accuracy and reliability of point-of-care testing results. Pseudohyperkalemia is a common preanalytical testing error that, while screened for regularly in central laboratories, is not detected using whole blood POC tests. Given that potassium POC testing typically is performed as part of a basic metabolic panel, which in turn generates specific testing analyte patterns, the authors posited using machine learning to predict POC analyte outliers indicative of preanalytical error—in this case, falsely elevated potassium results. They generated a logistic regression model to compare POC basic metabolic profiles with associated central laboratory data using the R statistical programming language. They used 3,489 unique emergency department patient encounters between August 2019 and December 2019 to produce POC and central laboratory panel results to train, test, and prospectively validate the model. The patients had both POC and central lab collections performed within a 10-minute timespan. In the test data set, overall model performance demonstrated high predictive value (area under the receiver operating characteristic curve, 0.995). The authors illustrated multiple ways to maximize the clinical applicability of their model using different methods to determine the optimal cutoff value, including relevant test performance characteristics, for various use cases. For example, they showed that by using a rule-in or rule-out cutoff approach for pseudohyperkalemia, it is possible to maximize the positive predictive value (90 percent PPV with a cutoff of 0.76) or negative predictive value (99.9 percent NPV with a cutoff of 0.16), respectively. Alternatively, the Youden method could be used if the intent of the use case was to maximize sensitivity and specificity for detecting pseudohyperkalemia (optimal cutoff value, 0.076; sensitivity, 100 percent; specificity, 96.6 percent). The authors concluded that patterns observed in POC testing analyte panels could be used to generate clinically applicable machine-learning models, which could improve the quality and accuracy of POC testing compared to similar central laboratory testing. This would ultimately lead to increased clinician confidence in POC testing methods. Finally, they believe their approach could be applied to other sources of preanalytical error at the point of care and in core laboratories.
Benirschke RC, Gniadek TJ. Detection of falsely elevated point-of-care potassium results due to hemolysis using predictive analytics. Am J Clin Pathol. 2020;154(2):242–247.
Correspondence: Dr. Robert C. Benirschke at rbenirschke@northshore.org