A deep learning approach for predicting severity of COVID-19 patients using a parsimonious set of laboratory markers.

The SARS-CoV-2 virus has caused tremendous healthcare burden worldwide. Our focus was to develop a practical and easy-to-deploy system to predict the severe manifestation of disease in patients with COVID-19 with an aim to assist clinicians in triage and treatment decisions. Our proposed predictive algorithm is a trained artificial intelligence-based network using 8,427 COVID-19 patient records from four healthcare systems. The model provides a severity risk score along with likelihoods of various clinical outcomes, namely ventilator use and mortality. The trained model using patient age and nine laboratory markers has the prediction accuracy with an area under the curve (AUC) of 0.78, 95% CI: 0.77-0.82, and the negative predictive value NPV of 0.86, 95% CI: 0.84-0.88 for the need to use a ventilator and has an accuracy with AUC of 0.85, 95% CI: 0.84-0.86, and the NPV of 0.94, 95% CI: 0.92-0.96 for predicting in-hospital 30-day mortality.

Use of a model for A1c formation to estimate average glucose operative between short-interval measurements of %A1c.

BACKGROUND: Estimated average glucose (AG) is generally reported along with hemoglobin A1c measurements according to a standard calculation. Given a normal red blood cell lifetime of 120 days, serial A1c measurements at intervals <120 days are not completely independent. For short interval measurements, a change in AG (ΔAG) necessarily underestimates the change in average glucose operative during the interval (ΔG). We use a model for kinetics of HbA1c to evaluate the theoretical relationship between ΔAG and ΔG for HbA1c measurements made at intervals between 0 and 120 days. METHODS: From any given starting point for A1c, step changes in G were simulated using model calculations to determine the extent to which A1c could change as a function of the interval of exposure. Values for ΔAG were compared to the operative ΔG as a function of the interval between A1c measurements. RESULTS: Results of model simulations are a single graph for relationship of ΔAG to ΔG as a function of the interval between A1c measurements. ΔAG for (15, 30, 45, 60, 76, and 90) day intervals underestimated operative ΔG by (73, 51, 34, 21, 11, and 5)%, respectively. CONCLUSIONS: Model calculations predict the relationship between changes in estimated average glucose to changes in operative glucose for serial A1c measurements made at intervals <120 days. Given that serial measurements of A1c made at short intervals are not uncommon in practice, physicians may find this information to be useful.

Average glucose from hemoglobin A1c for altered red blood cell lifetimes: Predictions based on a model for hemoglobin A1c formation.

BACKGROUND: A model for hemoglobin A1c (HbA1c) formation was used to predict the relationship between average glucose (AG) and %HbA1c under conditions of altered red blood cell lifetime (RCL). METHODS: Using a kinetic mass balance model for formation of HbA1c in red blood cells as a function of age (time in circulation), whole blood %HbA1c vs. glucose was calculated based on the nonlinear distribution of red blood cells as a function of age across RCL. RESULTS: Model calculations provided a close fit to the standard relationship of estimated average glucose to %HbA1c for normal RCL (r>0.999). Results for altered RCL were calculated assuming simple time-scale compression or expansion of the distribution of red blood cells as a function of RCL. For a given %HbA1c, the operative average glucose needed to have achieved a given %HbA1c was predicted to be altered by RCL according to average glucose×RCL=constant. CONCLUSIONS: Model calculations estimate the extent to which standard reporting of AG vs. HbA1c underestimates or overestimates AG under conditions of altered RCL. Conditions of altered RCL may often be operative in patients with certain hemoglobin variants.

Effects of hemoglobin C, D, E and S traits on measurements of hemoglobin A1c by twelve methods.

BACKGROUND: Hemoglobin C, D Punjab, E or S trait can interfere with hemoglobin A1c (HbA1c) results. We assessed whether they affect results obtained with 12 current assay methods. METHODS: Hemoglobin AA (HbAA), HbAC, HbAD Punjab, HbAE and HbAS samples were analyzed on one enzymatic, nine ion-exchange HPLC and two Capillary Electrophoresis methods. Trinity ultra(2) boronate affinity HPLC was the comparative method. An overall test of coincidence of least-squared linear regression lines was performed to determine if HbA1c results were statistically significantly different from those of HbAA samples. Clinically significant interference was defined as >7% difference from HbAA at 6 or 9% HbA1c compared to ultra(2) using Deming regression. RESULTS: All methods showed statistically significant effects for one or more variants. Clinically significant effects were observed for the Tosoh G8 variant mode and GX (all variants), GX V1.22 (all but HbAE) and G11 variant mode (HbAC). All other methods (Abbott Architect c Enzymatic, Bio-Rad D-100, Variant II NU and Variant II Turbo 2.0, Menarini HA-8180T thalassemia mode and HA-8180V variant mode, Sebia Capillarys 2 and Capillarys 3) showed no clinically significant differences. CONCLUSIONS: Several methods showed clinically significant interference with HbA1c results from one or more variants which could adversely affect patient care.

Utilization of assay performance characteristics to estimate hemoglobin A1c result reliability.

BACKGROUND: Allowable total error (TE(a)) goals for hemoglobin (Hb) A(1c) require minimal assay imprecision and bias and implementation of a robust QC monitoring program. Here, we compare the combined influence on the risk of reporting unreliable results of TE(a) goals, a routine QC practice, and assay performance characteristics of 6 Hb A(1c) instruments across 4 academic medical centers. METHODS: The CLSI protocols EP-5 and EP-9 were applied to investigate Hb A(1c) result imprecision and bias on the Variant II Turbo and Variant II (Bio-Rad), G8 (Tosoh), Capillarys 2 Flex Piercing (Sebia), COBAS Integra 800 (Roche), and DCA Vantage (Siemens). Patient-weighted σ values and the risk of reporting unreliable Hb A(1c) results were determined for each assay at TE(a) specifications of 5%, 6%, and 7%. RESULTS: A large range of patient-weighted σ values spanning 0.5 orders of magnitude at a 6% TE(a) was observed. Although imprecision for all instruments was <3%, bias impacted the majority of the σ changes observed. Estimates for reporting unreliable results varied almost 500-fold based on analytical performance alone. CONCLUSIONS: Considerable differences in the probability of reporting unreliable Hb A(1c) results between different NGSP (formerly the National Glycohemoglobin Standardization Program)-certified platforms were observed. At a 6% TE(a), our study indicates all but the Capillarys 2 Flex Piercing requires that the maximum affordable QC be run. Risk estimates for individual laboratories' Hb A(1c) methods can be used to assess QC practices and residual risk of an unreliable Hb A(1c) result.

Pathology consultation on HbA(1c) methods and interferences.

OBJECTIVES: To review the various hemoglobin (Hb) A1c methods, with a focus on interferences resulting from Hb variants. METHODS: HbA1c is a marker used for the diagnosis and management of diabetes. Each available HbA1c method has advantages and challenges, such as its susceptibility to interferences. RESULTS: The presence of Hb variants and/or abnormalities in RBC turnover cannot only interfere analytically with HbA1c measurements but also may affect the clinical interpretation of HbA1c values. CONCLUSIONS: Familiarity with the advantages and challenges of the various methods used for HbA1c testing is essential when establishing therapeutic management and goals based on HbA1c results, especially in populations with a high prevalence of Hb variants.

Unintended reporting of misleading Hb A(1c) values when using assays incapable of detecting hemoglobin variants.

CONTEXT: It is recommended that hemoglobin (Hb) A1c (Hb A1c) not be used to assess average glycemia in patients who have altered red blood cell life span. OBJECTIVE: To investigate the frequency of reporting an Hb A1c value for Hb variant samples that do not include Hb A. DESIGN: Hb A1c samples (n = 500) were procured and screened for Hb variants that may affect Hb A1c interpretation (Hb SS, Hb SC, and Hb S-ß-thalassemia). Five of each of these samples were tested by ion-exchange high-performance liquid chromatography, immunoturbidimetric assay, second-generation immunoturbidimetric assay, and affinity chromatography. RESULTS: Eleven (2.2%) homozygous Hb SS, 6 (1.2%) Hb SC, and 5 (1.0%) Hb S-ß-thalassemia samples were identified out of the 500 samples tested. Three of 4 instruments investigated in this study are known to not be plagued by analytic interference from these Hb variants but disturbingly reported Hb A1c values in the absence of Hb A. CONCLUSIONS: The improved analytic specificity of Hb A1c platforms has by and large eliminated interferences from the most common heterozygous Hb variants. A consequence, however, is the potential for unintended reporting of Hb A1c results in the presence of homozygous and compound heterozygous Hb variants that lack Hb A and the inability to distinguish those samples not recommended to be used for patient care. The ability to identify samples harboring Hb variants that preclude the utility of Hb A1c may be beneficial in high prevalence populations.

Challenges in implementing clinical liquid chromatography-tandem mass spectrometry methods--the light at the end of the tunnel.

The use of liquid chromatography-tandem mass spectrometry (LC-MS/MS) in the clinical setting is a relatively new application. One of the significant barriers hampering the transition of LC-MS/MS from the research lab into a clinical setting is the uncertainty of how to successfully develop and validate a method that meets guidelines for clinical applications. Here, we have taken this seemingly overwhelming process and broken it into five general stages for consideration: assessing the clinical validity of a new LC-MS/MS assay, determination of feasibility, assay development, assay validation and post-implementation monitoring. Although various publications are available and serve as resources for determining development processes and acceptability criteria for specific LC-MS/MS assays, many of them are general recommendations or are specific to research applications that may not translate either practically or clinically. In this perspective special feature article, a resource is compiled that describes key differences between LC-MS/MS methods for research use versus clinical use. In addition, the challenges facing the expanding role of this technique in the clinical setting are discussed, including instrumentation/automation challenges, potential regulation of laboratory developed tests by the US Food and Drug Administration and standardization and harmonization of MS methods through the use of traceable materials and availability of guidance documents.

Direct total and free testosterone measurement by liquid chromatography tandem mass spectrometry across two different platforms.

OBJECTIVES: To develop and validate liquid chromatography tandem mass spectrometry (LC-MS/MS) methods for the direct measurement of total and free testosterone in patient samples on two different analytical systems. DESIGN AND METHODS: An API 4000 and 5000 triple quadropoles were used and compared; the former is reported to be 3-5 times less sensitive, as was used to set the quantitation limits. Free testosterone was separated from the protein-bound fraction by equilibrium dialysis followed by derivatization. Either free or total testosterone, and a deuterated internal standard (d3-testosterone) were extracted by liquid-liquid extraction. The validation results were compared to two different clinical laboratories. RESULTS: The use of d2-testosterone was found to be unacceptable for our method. The total testosterone LC-MS/MS methods on both systems were linear over a wide concentration range of 1.5-2000ng/dL. Free testosterone was measured directly using equilibrium dialysis coupled LC-MS/MS and linear over the concentration range of 2.5-2500pg/mL. Good correlation (total testosterone, R(2)=0.96; free testosterone, R(2)=0.98) was observed between our LC-MS/MS systems and comparator laboratory. However, differences in absolute values for both free and total testosterone measurements were observed while a comparison to a second published LC-MS/MS method showed excellent correlation. Free and total testosterone measurements correlated well with clinical observations. CONCLUSIONS: To our knowledge, this is the first published validation of free and total testosterone methods across two analytical systems of different analytical sensitivities. A less sensitive system does not sacrifice analytical or clinical sensitivity to directly measure free and total testosterone in patient samples.

Development of a liquid chromatography tandem mass spectrometry method for iothalamate measurement to assess renal function for potential kidney donation.

BACKGROUND: Chronic kidney disease often goes undetected due to the insensitivity of current methods to accurately assess glomerular filtration rate (GFR) in early stages of renal dysfunction. The clearance of exogenously introduced iothalamate, a commonly used radiopaque agent, is an alternative to inulin clearance for the assessment of renal function and its use in calculating GFR can serve as a screening tool for kidney transplant donors. METHODS: A method was developed to measure iothalamate in plasma and urine samples by HPLC combined with electrospray positive ionization tandem mass spectrometry (MS/MS). Iothalamate is isolated from plasma by methanol extraction and urine using a quick-spin filtration approach, then monitored by multiple reaction monitoring using the hydrogen adduct mass transitions. Iohexol was used as an internal standard. RESULTS: Iothalamate was measured within an analytical run time of 5 min, with a lower limit of quantification of 18.75 ng/ml. The intraassay and interassay variations of the plasma and urine iothalamate assays were both <9%. Recovery from plasma and urine samples ranged from 93.6% to 104.1%. GFR was calculated using the patient's urine flow rate and plasma and urine iothalamate values. Linear correlations tested by LC-MS/MS and an accepted capillary electrophoresis (CE) assay showed similar results (GFR, r=0.92, Sy/x=10.3). CONCLUSIONS: We developed and validated an LC-MS/MS method for quantitating iothalamate in plasma and urine to calculate GFR used for screening potential kidney donors in our hospital system. A less sensitive mass spectrometry system does not sacrifice analytical or clinical sensitivity for measuring GFR.

H-gemcitabine: a new gemcitabine prodrug for treating cancer.

In this report, we present a new strategy for targeting chemotherapeutics to tumors, based on targeting extracellular DNA. A gemcitabine prodrug was synthesized, termed H-gemcitabine, which is composed of Hoechst conjugated to gemcitabine. H-gemcitabine has low toxicity because it is membrane-impermeable; however, it still has high tumor efficacy because of its ability to target gemcitabine to E-DNA in tumors. We demonstrate here that H-gemcitabine has a wider therapeutic window than free gemcitabine.

Teaching laboratory medicine to medical students: implementation and evaluation.

CONTEXT: Laboratory medicine is an integral component of patient care. Approximately 60% to 70% of medical decisions are based on laboratory results. Physicians in specialties that order the tests are teaching medical students laboratory medicine and test use with minimal input from laboratory scientists who implement and maintain the quality control for those tests. OBJECTIVE: To develop, implement, and evaluate a 1.5-day medical student clinical laboratory experience for fourth-year medical students in their last month of training. DESIGN: The experience was devised and directed by laboratory scientists and included a panel discussion, laboratory tours, case studies that focused on the goals and objectives recently published by the Academy of Clinical Laboratory Physicians and Scientists, and medical-student presentations highlighting salient points of the experience. The same knowledge quiz was administered at the beginning and end of the experience and 84 students took both quizzes. RESULTS: A score of 7 or more was obtained by 16 students (19%) on the initial quiz, whereas 34 (40%) obtained the same score on the final quiz; the improvement was found to be statistically significant (P  =  .002; t  =  3.215), particularly in 3 out of the 10 questions administered. CONCLUSIONS: Although the assessment can only measure a small amount of knowledge recently acquired, the improvement observed by fourth-year medical students devoting a short period to learning laboratory medicine principles was encouraging. This medical student clinical laboratory experience format allowed teaching of a select group of laboratory medicine principles in 1.5 days to an entire medical school class.