Determination of Age- and Sex-specific Reference Intervals for Cholinesterase and Gamma-Glutamyl Transferase Using Real-World Data of the Marienhospital in Stuttgart, Germany: Comparison of Three Different Indirect Methods
A. Meyer (Wiesbaden, DE)
Accurate reference intervals are crucial for the correct interpretation of laboratory results. A fast and cost-effective alternative to direct reference interval estimation is provided by indirect methods using routine laboratory data. The objective of this study was to compare reference intervals provided by the manufacturer to estimates calculated with three indirect methods. For this comparison, we used the reference intervals stated in Abbott’s instruction for use and real-world data of cholinesterase (CHE) and gamma-glutamyl transferase (GGT) provided by the Marienhospital in Stuttgart.
Routine laboratory data of CHE and GGT that had been analyzed on the Alinity c platform were used to determine refence intervals. The datasets were cleaned, and data pre-processing steps were applied to decrease the number of presumably diseased subjects in the data. This helps the indirect methods to identify and mathematically model the result distribution of the healthy population. Reference intervals were estimated with a modified Hoffmann method (RefLim), based on a quantile-quantile plot, the Truncated Maximum Likelihood method (TML), and an inverse modeling approach (refineR) for both sexes and for different age groups.
For CHE, all three indirect methods showed a good agreement with comparable results. However, for women, the lower reference limits were consistently higher than the expected values provided by the manufacturer, and for men the same was observed for the upper reference limits.
For GGT, the situation was more complex due to a poor agreement between the three methods and an age-dependent increase of the upper reference limits for both sexes. In general, RefLim tended to estimate wider and refineR narrower reference intervals, while TML was in between. All three methods confirmed the lower reference limits reported by the manufacturer and obtained different age-dependent upper reference limits.
Our study reinforces the call of numerous guidelines and recommendations that reference intervals reported by a manufacturer should be compared with results from the laboratory’s own subject population. Using CHE as an example, we showed that three different indirect methods based on completely different statistical procedures yield comparable results. The derived statistical models of the healthy population were similar, which indicates that they reflect the true reference population. The example of GGT, on the other hand, showed that these methods can result in different and possibly false reference limits if the presumably pathological results in a mixed population cannot be correctly identified and separated with the statistical techniques of the indirect methods.
A tool for plausibility checks of reference intervals
S. Klawitter (Braunschweig, DE)
Background: Laboratory information systems typically contain hundreds or even thousands of reference limits stratified by sex and age. Since under these conditions a manual plausibility check is hardly feasible, we have developed a simple algorithm that facilitates this check. A user-friendly open-source R tool is available as a Shiny application at github.com/SandraKla/Zlog_AdRI.
Methods: Based on the zlog standardization, we can possibly detect critical jumps at the transitions between age groups, regardless of the analytical method or the measuring unit. Its advantage compared to the standard z-value is that means and standard deviations are calculated from the reference limits rather than from the underlying data itself. The purpose of the tool is illustrated by the example of reference intervals of children and adolescents from the Canadian Laboratory Initiative on Pediatric Reference Intervals (CALIPER).
Results: The Shiny application identifies the zlog values, lists them in a colored table format and plots them additionally with the specified reference intervals. The algorithm detected several strong and rapid changes in reference intervals from the neonatal period to puberty. Remarkable jumps with absolute zlog values of more than 5 were seen for 29 out of 192 reference limits (15.1%). This might be attenuated by introducing shorter time periods or mathematical functions of reference limits over age.
Discussion: Age-partitioned reference intervals will remain the standard in laboratory routine for the foreseeable future, and as such, algorithmic approaches like our zlog approach in the presented Shiny application will remain valuable tools for testing their plausibility on a wide scale.
Refining Diagnosis Paths for Medical Diagnosis based on an Augmented Knowledge Graph
J. Kirchhoff (Karlsruhe, DE)
F. Stumpe (Karlsruhe, DE)
Medical diagnosis is the process of making a prediction of the disease a patient is likely to have, given a set of symptoms and observations. This requires extensive expert knowledge, in particular when covering a large variety of diseases. Such knowledge can be coded in a knowledge graph -- encompassing diseases, symptoms, and diagnosis paths. Since both the knowledge itself and its encoding can be incomplete, refining the knowledge graph with additional information helps physicians making better predictions. At the same time, for deployment in a hospital, the diagnosis must be explainable and transparent. In this paper, we present an approach using diagnosis paths in a medical knowledge graph. We show that those graphs can be refined using latent representations with RDF2vec, while the final diagnosis is still made in an explainable way. Using both an intrinsic as well as an expert-based evaluation, we show that the embedding-based prediction approach is beneficial for refining the graph with additional valid conditions.
In this paper, we have introduced the medicalvalues knowledge graph, which is used for medical diagnosis using so-called diagnosis paths. Those paths allow for a transparent prediction of a patient’s disease. Since the paths are developed
manually, they are notoriously incomplete.To tackle this incompleteness, we have introduced an approach which first
enriches the medicalvalues knowledge graph into a augmented graph, connecting it to a large dataset of patient records. On that augmented graph, we have trained vector embeddings with RDF2vec, which are used to predict completions.
Both in an internal validation as well as in an expert evaluation, we have shown that the prediction of such extensions is possible with high precision. This methodology of enriching the graph and producing predictions therewith is independent of the task and domain at hand. One key limitation of the approach is the external data used, which is data gathered from intensive care units. Therefore, diseases which do rarely lead to treatments in intensive care are not well covered. In order to augment diagnosis paths for as versatile diseases as possible, other external datasets should be considered as well. Here, the connectors to clinic information systems (CIS) and laboratory information systems (LIS) may also add large-scale instance data in the future, which can also be exploited with the same methodology.
So far, drugs are not represented in the medicalvalues knowledge graph. In the future, we would like to include them, both as a part of a patient’s medical history (i.e., existing medication), as well as possible treatments once a diagnosis
is made. To that end, we plan to augment the graph with existing datasets on drugs and drug interactions.
Full paper: https://arxiv.org/pdf/2204.13329.pdf
RefLim: A Graphical and Numerical Approach to Reference Interval Verification
G. Hoffmann (Grafrath, DE)
There is a wide range of direct and indirect methods, by which reference limits can be determined . Comparably little attention has been paid to the verification of specified reference intervals from test package inserts and other external sources. The respective guideline procedure  is simple, but due to the low number of just 20 values, it suffers from unacceptably wide confidence limits , and the criterion that ≥ 90% of the measured values must fall within the specified limits cannot detect excessively wide reference intervals .
We aimed to develop an alternative method that is as easy to perform, but works with a moderate number of routine laboratory data and detects all kinds of deviations between expectation and observation.
We suggest a modification of a previously published iterative boxplot method  to create substantially stronger truncation. We derive the quantiles q of the expected central 95% of reference values from the quartiles Q1, Q2, and Q3 according to the following algorithm:
var = min(Q2 - Q1, Q3 - Q2)
q(0.025, 0.975) = Q2 ± f ∙ var
where f is 2.91 for the first truncation step and 3.08 for all subsequent iterations.
This algorithm is applied to log-transformed values assuming that unknown distributions of reference values can be modelled as lognormal . Outliers beyond the estimated quantiles are removed and the algorithm is iterated, until no more outliers are detected. From the truncated data, zlog values are calculated  with reference to the specified limits, and a normal quantile-quantile plot with 39 equidistant quantiles is generated . A linear regression line y = a + b ∙ x is calculated from the central 33 data points, and the deviations of a and b from the expected values 0 and 1, resp., are used as measures of discordance between expectation and observation.
Based on public data (archive.ics.uci.edu/ml/datasets/HCV+data) for routine liver biomarkers (e. g. transaminases, bilirubin) from 540 blood donors and 75 hepatitis C patients, we demonstrate four types of deviations, which can be quantified by our method: the intercept a signifies positive or negative shifts of the given reference interval, whereas the slope b indicates whether that interval is too narrow or too wide. We use empirical thresholds for both constants to visually represent the magnitude of the deviations by traffic light colors. Comparable results are obtained for blood donors and mixed data. Out of 16 manufacturer-derived reference intervals, nine should be rejected (red), and four should at least be given more careful consideration (yellow).
Our method provides an easy and robust way to verify given reference intervals using routine data and intuitive graphics. Our little study makes it likely that a substantial proportion of manufacturer-derived limits needs to be reevaluated using more sophisticated methods for reference interval determination .
Artificial Intelligence in laboratory medicine and the responsibility of laboratory information system providers
J. Gebauer (Meißen, DE)
Artificial intelligence and machine learning have made substantial advances during the last decade in a broad field of applications ranging from fraud detection over speech recognition to self-driving cars. In healthcare machine learning approaches are state of the art in medical imaging. In laboratory science however applications are uncommon despite the fact that laboratory medicine with its numerical and structured data would be a very suitable field for machine learning. Meaningful but yet unknown correlations are expected to be discovered. First models e.g. for estimation of iron deficiency anemia, liver function parameters or low-yield repetitive laboratory tests are already described in basic research. However, their validation in prospective studies or their application in clinical routine seems to be inhibited by rigid laboratory and clinical information systems. Therefor we evaluated technical possibilities for establishing data science approaches in routine diagnostics.
The authors evaluated possible work flows for the application of data science and machine learning methods in laboratory routine or prospective study settings.
Interfaces to a plethora of devices and applications are a key feature of every LIS. However most LIS providers tend to distribute these as so-called highly customized software solutions at a high price, even if the connected device is used in virtually every laboratory. One possible and cost-effective workaround could be intercepting interface communication. Another more desirable and efficient approach would be the usage of standardized resources and application programming interfaces like HL7 FIHR.
While granular and inconsistent interfaces resulted from the first steps of digitalization in laboratory medicine, nowadays well-structured and consistent standards for exchanging health care data like HL7 FIHR are available as open source standards. Therefor we demand that laboratory and hospital information systems provider anticipate in the evaluation and validation of innovative data science techniques by catalyzing their development through deployment of open-source interfaces or free application programming interfaces for research use or development and validation of novel data science techniques in laboratory diagnostic routine.
Identification of Putative Non-Substrate-Based XT-I Inhibitors by Small Molecule Library Screening
T. Ly (Bad Oeynhausen, DE)
Instruction: Fibroproliferative disorders are characterized by excessive accumulation of extracellular matrix (ECM) components, including collagens and proteoglycans (PGs), which can cause organ dysfunction. The transition of fibroblasts to ECM-synthesizing myofibroblasts in the presence of fibrotic mediators such as transforming growth factor-β1 (TGF-β1) is a key event of this process. The rise in myofibroblast content is marked by an increase in intracellular and extracellular activity of xylosyltransferase-I (XT-I), which is the initial enzyme of PG biosynthesis. Therefore, human XT-I resembles not only a myofibroblast marker but also a serum biomarker for accessing the proteoglycan biosynthesis rate under fibrotic conditions. Accordingly, the inhibition of XT-I would be a promising treatment option for fibrosis reducing ECM accumulation.
Methods: We used a natural product-derived molecular library to identify non-substrate-based inhibitors of the human XT-I by ultra-performance liquid chromatography/electrospray ionization tandem mass spectrometry. We combined this cell-free approach with virtual and molecular biological analyses to confirm and prioritize the inhibitory potential of the identified compounds. The characterization of the compound’s efficacy in TGF-β1-mediated XYLT1 transcriptional regulation in primary human dermal fibroblasts, the key cells of ECM remodeling, was investigated by gene expression analyses.
Results: Through this approach, we identified amphotericin B and celastrol as novel non-substrate-based XT-I protein inhibitors. The XT-I inhibitory effect of amphotericin B was mediated by a non-competitive inhibition mode, while that of celastrol was based on a competitive mode of inhibition. Both compounds reduced the XYLT1 mRNA-expression levels and cellular XT-I activity of the primary cells. Furthermore, we demonstrated that the cellular effects mediated by amphotericin B and celastrol were due to inhibitor-induced changes in the TGF-β and microRNA-21 signaling pathway.
Conclusion: The results of this study provide a promising basis for the optimization and future use of the XT-I inhibitors amphotericin B and celastrol as therapeutic agents for the treatment of fibroproliferative diseases.
Erythropoietin determinations in proficiency tests - variation in laboratory results and method precision
L. Toll (Villingen-Schwenningen, DE)
Erythropoietin determinations in proficiency tests - variation in laboratory results and method precision
Luisa Toll1, 3, *, Folker Wenzel1, 2, Nathalie Wojtalewicz2, Laura Vierbaum2, Ingo Schellenberg2, Mario Thevis3
1 Faculty of Medical and Life Sciences, Furtwangen University, Villingen Schwenningen, Germany
2 INSTAND e.V., Society for Promoting Quality Assurance in Medical Laboratories, Düsseldorf, Germany
3 Institute of Biochemistry/ Center for Preventive Doping Research, German Sport University Cologne, Cologne, Germany
* Corresponding author. E-mail address: firstname.lastname@example.org
The aim of the study was to summarize the results of the proficiency tests for Erythropoietin (EPO) determination managed by INSTAND e.V. and to evaluate the measurement quality of the specific methods used by the participating laboratories.
Materials and Methods
In each of a total of nine test series, two samples (sample a and b) with different EPO concentrations were send to the participating laboratories. The median of each round was calculated for sample a and b, respectively. Deviation from the median and the relative error-values was calculated. The criteria of acceptance were set to 20 % around the median. In consideration of the used methodology, the relative deviation of one method around the total mean as well as the mean without the respective method was calculated and the outcomes were contrasted.
The first proficiency test was performed with a number of 10 participants and in the following test series it has increased up to a maximum of 85 participating laboratories in series 8. The average fraction of measurements not meeting the defined criteria of acceptance was 14.3 ± 17.4 % (sample a) and 8.2 ± 6.1 % (sample b). The relative deviation tended to be higher in a lower concentration range. Enzyme linked immunosorbent assay (ELISA) showed significantly higher values compared to chemiluminescent immunoassay (CLIA) and luminescence-enhanced enzyme immunoassay (LEIA) in all cases.
The results appeared to be well harmonized in most cases. However, individual proficiency tests showed higher scatterings of the measured values. This should be observed in future test series.
SARS-CoV-2-PCR pool testing: high efficiency without sensitivity loss
K. Stiehler (Dresden, DE)
Aims: The COVID-19 pandemic brought enormous challenges for clinical laboratories: demand for timely and precise SARS-CoV-2-PCR results increased rapidly while consumables for sampling and testing were in ever shorter supply. Laboratories had to build up adequate testing capacities in terms of machines and personnel under great time pressure. We implemented a multiple-swab pool testing method without any volume dilution  in a hospital lab setting, which increased efficiency dramatically, providing economic SARS-CoV-2-PCR screening without loss of sensitivity.
Materials and Methods: Up to 10 nasopharyngeal swabs were pooled directly into one 3,0 ml UTM (Universal Transport Medium) tube. Additionally, dry nasopharyngeal swabs were collected from each participant of the pool, serving as secondary samples for eventual re-testing. After RNA extraction and RT-qPCR, negative pool test results were reported. For positive pool tests, the corresponding secondary samples were tested to identify the infected individual(s) causing the positive pool test result.
Results: From April 2020 to March 2022 a total of 25.243 pool tests were analyzed, resulting in 770 positive pool samples with 840 participants found SARS-CoV-2-PCR positive with a mean difference between pool and secondary sample of 0,1 Ct values (standard deviation: 3,6 Ct values). In approx. 4 % of positive pool tests no positive participant could be identified. Overall, 7 results per one PCR pool test were produced using this protocol, which reduces the price per result accordingly. Time to result is very short for SARS-CoV-2-PCR negative pools (approx. 3 h) and still acceptable for individual SARS-CoV-2-PCR results for the participants of positive pools (approx. 7 h).
Discussion: Our pool testing method shows good sensitivity for time efficient SARS-CoV-2-PCR screening. The disadvantage of our pool testing method is the necessity to collect secondary samples, which in consequence may be of different sampling quality, as seen in cases where in spite of positive pool tests no positive participant could be identified, probably due to low viral load in early or late infection – SARS-CoV-2-PCR can be (weakly) positive for several weeks to months after infection. The widely used alternative method by Dorfman et al.  is based on pooling individual samples in the lab. No secondary samples are required, but sensitivity is reduced markedly due to the dilution effect . Also, sample preparation takes longer because of the additional pooling step, and instrumentation complexity and lab consumable use are higher, e.g. for some PCR clean filter pipette tips in short supply up to today.
SARS-CoV-2 nucleocapsid protein mutations disrupt N gene amplification in frequently employed multiplex RT-PCR assays
D. Hilti (Buchs SG, CH)
Absence of SARS-CoV-2 PCR target signals may indicate the presence of new variants. We analysed the increase of N gene dropouts in the TaqPath COVID-19 CE-IVD RT-PCR Kit from Thermo Fisher Scientific (TaqPath) covering the viral ORF1ab, N and S genes. Whole genome sequencing was used to identify potential issues with the N gene PCR efficacy.
Materials and Methods
We analysed nasopharyngeal swabs and saliva samples from symptomatic patients as well as asymptomatic carriers from mass testing programs by RT-PCR with the TaqPath Kit. Complete N gene target failures (NGTF) were defined as missing N gene amplification in the presence of intact amplification of ORF1ab and S genes. Ultimately, whole genome sequencing was performed on matching samples with CT-values < 30 using a GridIon Nanopore sequencer (Oxford Nanopore Technologies - ONT, Oxford, UK).
From 2nd of August 2021 to 1st of May 2022, we observed a total of 218’815 SARS-CoV-2 positive samples, with 168’101 samples matching the defined criteria. Out of these, 194 specimen with NGTF were identified (0.12 %). The samples originated from all regions of Switzerland and were mostly clonally unrelated. In large part, samples with NGTF were collected from October to December 2021 correlating with the infection wave attributable to the Delta variant (B.1.617.2) and its sub-lineages. During this time, a proportional increase of NGTF among all positive samples with a peak frequency of 1 % was observed. Frequency of NGTF swiftly fell once the Omicron lineages BA.1 and BA.2 became prevalent.
Sequencing revealed the nucleotide substitution G28922T (A217S) in 148 cases (88.6 %). 10 samples (6 %) carried the deletion 28913 – 28918 (del214/215), 8 samples (4.8 %) the deletion 28913 – 28915 (del214) and 1 sample (0.6 %) the deletion 28892 – 28930 (del207 – 219). Samples with intact N gene amplification lacked the specified mutations. Lineages included the Delta variant parental lineage B.1.617.2 (n=4), sub-lineages thereof (AY.4 (n=99), AY.4.3 (n=4), AY.33 (n=2), AY.36 (n=9), AY.39 (n=1), AY.43 (n=7), AY.43.3 (n=2), AY.46.6 (n=1), AY.98.1 (n=4), AY.122 (n=1) and AY.125 (n=11)) as well as the Omicron lineage BA.2 (n=1). The lineage of 24 specimen could not be determined.
The substitution A217S, as well as the deletions G214-, G214/215- and del207-219 in SARS-CoV-2 appear to be associated with the NGTF in the TaqPath Kit. These mutations were identified in several sub-lineages of the Delta variant as well as the Omicron BA.2 lineage and were apparently not linked to a particular variant. Importantly, a selection advantage associated with the NGTF could not be identified. The N gene is a common target in RT-PCRs e.g., Cepheid’s Xpert Xpress SARS CoV 2 or the VIASURE SARS-CoV-2 (N1 + N2) Real Time PCR Detection Kit for BD MAX. Continuous monitoring and timely re-assessment using whole genome sequencing can thus improve the development of diagnostic tests.
Liquid chromatography tandem mass spectrometry for quantitative analysis of 11 oxygenated androgens in human serum.
R. Zeidler (Leipzig, DE)
MULTIMODAL SPATIALLY RESOLVED INVESTIGATION OF LIPIDS SIGNATURES IN NEEDLE BIOPSIES OF LIVER NEOPLASMS WITHIN THE MANNHEIM MOLECULAR INTERVENTION ENVIRONMENT (M2OLIE)
M. Rittel (Mannheim, DE)
Introduction and aims
Mass spectrometry imaging (MSI) is an indispensable tool for label-free spatially-resolved investigations of biological processes and of the molecular composition of various samples in fundamental research. However, the translation of MSI-based techniques into clinical routine for in-depth medical diagnosis, e.g. molecular pathology, remains a challenging task. Requirements in sample preparation for molecular analysis and time restrictions, as well as feasibility in a clinical context are often on opposing sides of the spectrum. Especially, targeting cancer and cancer sub-types in a fast and reliable way by MSI with diagnostic value for further treatment decisions is of particular interest. Here, we present a multimodal approach aligned with clinical routine practice for the assessment of molecular composition of different tumor tissue samples by fusing modern analytical tools and sophisticated clinical routine.
Within the framework of the M2OLIE research campus, we combined various spatially resolved techniques like infrared spectroscopy, immunohistochemistry and hematoxylin and eosin stainings with MSI-based molecular analysis to investigate the lipid and metabolite composition of primary hepatocellular carcinoma and metastases residing in the liver. Fresh-frozen liver tissue section were measured by means of high-precision untargeted matrix-assisted laser desorption/ionization (MALDI)-MSI in order to reveal the molecular fingerprint of cancer types for biomarker discovery. This includes resected tissue samples as well as biopsies. To this end, we have developed a device for biopsy embedding and MSI sampling, as well as a workflow for data fusion of modern molecular analysis with sophisticated clinical routine. By mapping this sample-specific information onto the MSI data a higher precision in analysis of neoplastic lesions can be accomplished.
We analyzed liver cancer tissue sections with a multimodal approach. To this end, we combined the spatial information from infrared spectroscopy and histopathological evaluation to map tissue morphology-specific features onto the MSI data for in-depth molecular analysis, especially for tumorous regions. By precise co-registration this guided approach enabled transfer of information between adjacent tissue sections and of data from different modalities. Preliminary data revealed a differentiation of the tissue morphology and thus differences in the lipid and metabolite composition of specific histological features.
These results promote a MSI-based routine that is feasible in clinical practice. Fusion of molecular information and clinical routine assessment for cancer classification could potentially result in benefits for treatment decisions based on higher level sample analysis.
Estimation of Continuous Reference Intervals using Real-World Data and refineR
T. Ammer (Erlangen, DE)