Graham Ross Dallas Jones

A widely accepted approach to setting Analytical Performance Specifications (APS) in laboratory medicine is the Milan models. Milan model 1b is an approach where APS are developed based on patient classification, or clinical decision-making, noting that these factors link to the probability of patient outcomes. This paper describes limitations that should be considered for any APS derived using model 1b, based on any available information from model 2 (biological variation), and model 3 (state of the art). It is also proposed that ”state of the art”, when defined as being tests which are routinely available, influences studies based on clinical surveys as this the background used by doctors to develop their clinical experience, as well as the ability for the APS to be put into routine use. Additionally, when ”state of the art” is defined as the method used in an outcome study, this becomes a reference where such studies cannot support a smaller assay precision, as the assay performance is only one aspect of the variation in the links between laboratory results and clinical findings. Model 1b studies remain a key approach to assigning clinical meaning to assay performance, however outcomes of these studies need to be taken with awareness of relevant limitations.

Background and objectives: CYP2C19 phenotype is a known contributor to the interpatient variability in voriconazole response. Alternative therapy is recommended for ultrarapid/rapid and poor metabolizers due to increased probability of subtherapeutic exposure and side effects, respectively. We aimed to evaluate whether CYP2C19 phenotype is associated with switching from voriconazole to alternative antifungal therapy, in settings where genetic results were not available at the time prescribing decisions were made. Methods: A multicentre, retrospective observational study was conducted in three Australian hospitals. Patients who had previously taken voriconazole (from 1 May 2019 to 31 May 2024) were invited to undergo pharmacogenomic testing. Medical records were audited to compare switching decisions across patients with different CYP2C19 phenotypes. Differences in voriconazole exposure and voriconazole-related adverse effects were also explored. Results: Among 194 patients, most were normal or intermediate metabolizers (69%); 21% were rapid, 7% ultrarapid, and 3% poor metabolizers (underpowered). Switching to alternative antifungal therapy (32%; 62/194) mainly occurred due to adverse effect incidence and was not associated with CYP2C19 phenotype (P = 0.9041). C-reactive protein levels were significantly higher in patients who switched therapy (P < 0.001). All ultrarapid metabolizers on standard voriconazole 400 mg/day had subtherapeutic concentrations and only those on higher doses (500–1200 mg/day) achieved therapeutic concentrations. Conclusions: CYP2C19 phenotype was not predictive of switching, which is a multifactorial prescribing decision likely influenced by therapeutic drug monitoring, inflammation and clinical status. Our observations on voriconazole dosing and exposure support a complementary prescribing approach: pharmacogenomic testing to identify patients who require atypical voriconazole dosing regimens and subsequent therapeutic drug monitoring to guide dose adjustments.

Objectives: Osteoarthritis (OA) is the most prevalent chronic musculoskeletal disease. Early identification of individuals at risk of knee structural progression is essential for targeted interventions. This study aimed to develop and validate a machine/deep learning (ML/DL)-based prognostic model for knee OA progression using serum metabolomics. Methods: Baseline serum metabolomic factors from two independent cohorts were analysed: Tasmanian Older Adult Cohort (n = 180) for model development and Licofelone trial (n = 137) for external validation. Participants were categorized by likelihood of knee structural progression using MRI and X-ray data. Four metabolomic scenarios were evaluated: metabolites alone (104), plus their ratios or inverse ratios (each 5460) and all combined (10 816), alongside age, sex, BMI. Metabolomic profiling used high-throughput omics. Batch effects were corrected via ComBat. Feature selection combined variable clustering with elastic net regularization. Top-selected features of each scenario were trained using five ML/DL models, with performance assessed by AUC, accuracy, sensitivity and specificity with 95% CIs. Results: The metabolomic-only scenario, including sex and age, yielded the best performance. Feature clustering reduced to 22, with the top 10 used for training/testing. The Artificial Neural Network with domain-adversarial component outperformed other algorithms with domain shift. The final model, including six metabolites (SM (OH) C22:2, proline, citrulline, LysoPC a C18:0, glutamate and C12-DC), plus sex and age, demonstrated excellent predictive performance (test: AUC 0.98 [0.95, 1], accuracy 0.89 [0.84, 0.94]; validation: AUC 0.89 [0.83, 0.95], accuracy 0.85 [0.79, 0.91]). Conclusion: This validated metabolomics DL framework (https://github.com/AFSHINJAM/KOA_Prediction_Metabolomics_Ratios) introduces a novel approach for predicting knee OA structural progression, enabling personalized risk assessment and early intervention.

Background: Population-wide reference intervals (RIpop) are commonly used in laboratory medicine but may not reflect an individual’s tightly regulated homeostatic interval. Personalized reference intervals (RIper) could enhance diagnostic precision by accounting for individual variability. A parametric empirical Bayes (PEB) framework stabilizes individual estimates using population parameters, enabling reliable RIper even from a limited number of individual results. Methods: We applied the PEB framework to estimate RIper for 9 biomarkers: albumin, creatinine, phosphate, cortisone, cortisol, testosterone, androstenedione, 17-hydroxyprogesterone, and 11-deoxycortisol. The PEB parameters tested were derived from both routine Laboratory Information System (LIS) data and a local biological variation (BV) study. Using serial samples from healthy adults, we assessed the proportion of results flagged with a 95% prediction interval and compared RIper to conventional RIpop and reference change values (RCVs). Results: LIS parameters were based on data from 1986 to 185 488 patients. PEB-based RIper were consistently narrower than RIpop while maintaining or reducing the proportion of flagged results. For example, albumin flagging decreased from 4.7% (RIpop) to 0.3% (RIper), phosphate from 5.4% to 3.7%, and cortisone from 7.1% to 3.9%. Conversely, 17-hydroxyprogesterone increased from 0.0% to 5.5% but remained close to the expected 5%. PEB thresholds were narrower than standard RCV estimates by correcting for regression toward the mean without increasing flagged results. Conclusions: The PEB framework effectively provides personalized cutoffs for laboratory tests even when few individual patient results are available. PEB parameters can be established using LIS or BV data, offering a feasible and cost-effective implementation pathway.

Background: Traditional markers modestly predict chronic kidney disease progression in Aboriginal and Torres Strait Islander people. Therefore, we assessed associations of cardiometabolic and inflammatory clinical biomarkers with kidney disease progression among Aboriginal and Torres Strait Islander people with and without diabetes. Objectives: To identify cardiometabolic and inflammatory clinical biomarkers that predict kidney disease progression in Aboriginal and Torres Strait Islander people. Design: Prospective observational cohort study Setting: Northern Territory, Australia Participants: Aboriginal and Torres Strait Islander participants of the estimated glomerular filtration rate (eGFR) study with (n = 218) and without diabetes (n = 278) Measurements: Baseline biomarkers (expressed as 1 standard deviation increase in logarithmic scale), plasma kidney injury molecule-1 (pKIM-1) (pg/ml), high-sensitivity troponin-T (hs-TnT) (ng/L), troponin-I (hs-TnI) (ng/L), and soluble tumor necrosis factor receptor-1 (sTNFR-1) (pg/ml) were assessed in 496 adults. Annual change in eGFR (ml/min/1.73 m ² ) and a composite kidney outcome (first of ≥30% eGFR decline with follow-up eGFR 2 , initiation of kidney replacement therapy or kidney disease-related death) over a median of 3 years. Methods: Linear regression estimated annual change in eGFR (ml/min/1.73 m ² ). Cox proportional hazards regression estimated hazard ratio (HR) and 95% CI for developing a combined kidney health outcome. Results: In individuals with diabetes, but not those without diabetes, higher baseline hs-TnT (−2.1 [−4.1 to −0.2], P = .033) and sTNFR-1 (−1.8 [−3.5 to −0.1], P = .039) predicted mean (95% CI) eGFR change, after adjusting for age, gender, baseline eGFR, and urinary albumin-to-creatinine ratio. Baseline variables explained 11% of eGFR decline variance; increasing to 27% ( P < .001) with biomarkers. In diabetes, hs-TnT and hs-TnI were significantly associated with increased risk of kidney health outcomes. Limitations: Limitations included potential chronic kidney disease misclassification from single creatinine and albumin measurements, limited adjustment for covariates due to a small sample size, and short follow-up restricting long-term outcome assessment. Conclusions: Cardiovascular, kidney, and inflammatory biomarkers are likely associated with kidney function loss in diabetes, with particularly prominent associations for cardiac injury markers.

The reference change value (RCV) is calculated by combining the within-subject biological variation (CV_I) and local analytical variation (CV_A). These calculations do not account for the variation seen in preanalytical conditions in routine practice or CV_I in patients presenting for treatment. As a result, the RCVs may not reflect routine practice or align with clinicians’ experiences. We propose a novel RCV approach based on routine patient data that is potentially more clinically relevant. This study used the refineR algorithm to determine RCVs using serial patient data extracted from a local Laboratory Information System (LIS). The model was applied to biomarkers with a range of result ratio distributions varying from normal to log-normal. Results were compared against conventional formula-based RCVs using CV_I estimates from a state-of-the-art biological variation study. Monte Carlo simulations were also used to validate the LIS data approach. The RCVs estimated from LIS data were: 11-deoxycortisol (men): −70%/+196%, 17-hydroxyprogesterone (men): −49%/+100%, albumin: −10%/+11%, androstenedione (men): −47%/+96%, cortisol (men): −54%/+51%, cortisone (men): −32%/+51%, creatinine: −16%/+14%, phosphate (women): −23%/+29%, phosphate (men): −27%/+29%, testosterone (men): −38%/+60%. The formula-based RCV estimates showed similar but slightly lower results, and the Monte Carlo simulations confirmed the applicability of the new approach. RCVs may be estimated from patient results without prior assumptions about the shape of the ratios between serial results. Laboratories can determine RCVs based on local practice and population.

Background: Reference change values (RCV) are used to indicate a change in analyte concentration that is unlikely to be due to random variation in the patient or the measurement. Current theory describes RCV relative to a first measurement result (X1). We investigate an alternative view predicting the starting point for RCV calculations from X1 and its location in the reference interval. Methods: Data for serum sodium, calcium, and total protein from the European Biological Variation study and from routine clinical collections were analyzed for the effect of the position of X1 within the reference interval on the following result from the same patient. A model to describe the effect was determined, and an equation to predict the RCV for a sample in a population was developed. Results: For all data sets, the midpoints of the RCVs were dependent on the position of X1 in the population. Values for X1 below the population mean were more likely to be followed by a higher result, and X1 results above the mean were more likely to be followed by lower results. A model using population mean, reference interval dispersion, and result diagnostic variation provided a good fit with the data sets, and the derived equation predicted the changes seen. Conclusions: We have demonstrated that the position of X1 within the reference interval creates an asymmetrical RCV. This can be described as a regression to the population mean. Adding this concept to the theory of RCVs will be an important consideration in many cases.

Analytical performance specifications (APS) are used for decisions about the required analytical quality of pathology tests to meet clinical needs. The Milan models, based on clinical outcome, biological variation, or state of the art, were developed to provide a framework for setting APS. An approach has been proposed to assign each measurand to one of the models based on a defined clinical use, physiological control, or an absence of quality information about these factors. In this paper we propose that in addition to such assignment, available information from all models should be considered using a risk-based approach that considers the purpose and role of the actual test in a clinical pathway and its impact on medical decisions and clinical outcomes in addition to biological variation and the state-of-the-art. Consideration of APS already in use and the use of results in calculations may also need to be considered to determine the most appropriate APS for use in a specific setting.

Analytical performance specifications (APS) are used for the quantitative assessment of assay analytical performance, with the aim of providing information appropriate for clinical care of patients. One of the major locations where APS are used is in the routine clinical laboratory. These may be used to assess and monitor assays in a range of settings including method selection, method verification or validation, external quality assurance, internal quality control and assessment of measurement uncertainty. The aspects of assays that may be assessed include imprecision, bias, selectivity, sample type, analyte stability and interferences. This paper reviews the practical use of APS in a routine clinical laboratory, using the laboratory I supervise as an example.