Mass spectrometry, machine learning and the path to better diagnostics for endocrine hypertension

NSW Clinical Mass Spec Forum presents... 

Mass spectrometry, machine learning and the path to better diagnostics for endocrine hypertension

The first Clinical Mass Spec Forum for 2025. Join us to learn about the latest developments in the clinical translation of novel mass spectrometry assays from experts in the field. 

Date: 30th April, 2025 | 5:30pm - 8:30pm

Location: In-person Room CB04.05.430, Building 4 University of Technology, Sydney                             

               Online: Zoom Link

Meeting Chair: Michael Fitzpatrick, Senior Hospital Scientist and Acting Lab Manager, NSW Health Pathology, Camperdown

Agenda

5:30 - 6:00pm 

Drinks and Networking

6:00 - 6:45pm 

ALDO+ Testing: LC-MS/MS based hypertension profiling on a global scale - Development path, clinical data, and vision 

Speaker: Marko Poglitsch, PhD

Marko Poglitsch holds degrees in Molecular Biology and Technical Chemistry, and obtained his PhD in Immunology at the Medical University of Vienna. Dr. Poglitsch joined the Vienna biotech scene in 2012, founding Attoquant Diagnostics, a company providing LC-MS/MS-based quantification of angiotensin peptides (RAS-FingerprintTM) to basic and clinical researchers worldwide. Ten years of research and development with Attoquant led to discoveries such as RAS equilibrium analysis and the AA2-Ratio as a novel biomarker for detection of primary aldosteronism, finally resulting in the IVD-registration of the RAAS Triple-ATM kit.
In 2024, Dr. Poglitsch stepped down from Attoquant CEO and co-founded aTENSION.life, a diagnostic company dedicated to battling uncontrolled hypertension on a global scale through a novel platform connecting laboratories, doctors and patients to mass spectrometry-based RAAS diagnostics.

Abstract

The talk will dive into the biochemistry of the Renin-Angiotensin-Aldosterone-System (RAAS), relevant to ALDO+ testing, such as angiotensin metabolism, dynamics, and function. Concepts relevant to sampling and testing requirements will be covered, followed by a detailed overview on the sample preparation process. Dr. Poglitsch also will provide information into the technical background of the LC-MS/MS assay, including hardware requirements, and insights into the product development process for ALDO+ testing (previously RAAS Triple-ATM). Clinical implications will be discussed, and a clinical data and performance overview will be provided, before finally sharing our experience with ALDO+ testing in Austria in more than 2000 hypertensive routine patients up to now.

6:45 - 7:30pm

Mass Spectrometry based multidimensional diagnostics with machine learning: validation for clinical decision support

Speaker: Professor Graeme Eisenhofer, PhD

Graeme Eisenhofer received his PhD in 1983 from the University of Otago, New Zealand, with clinical research on autonomic and neuroendocrine systems. He then moved to the NIH where he carried out basic and clinical studies mapping the pathways of catecholamine metabolism. In 1988 he moved to the Baker Heart Research Institute (Melbourne, Australia), where he worked on sympathetic nervous system function in health and disease. He returned to the NIH in 1991 before taking up a Professorship at the University Hospital in Dresden. Together with Dr. Jacques Lenders, Dr. Eisenhofer developed measurements of plasma metanephrines for laboratory diagnosis of phaeochromocytoma. He was also responsible for the first ever synthesis of 18F-fluorodopamine as a PET imaging agent for localising catecholamine-producing tumours. At Dresden, Dr. Eisenhofer leads laboratory and clinical research groups with a focus on adrenal disorders and endocrine hypertension. The work has most recently centred on applications of mass spectrometry-based measurements of steroids and catecholamine metabolites. For this, approaches are being developed that integrate multidimensional diagnostics with artificial intelligence to build clinical decision support systems for efficient and appropriate therapeutic interventions.

Abstract

With advances in clinical mass spectrometry that allow for simultaneous measurements of multiple analytes there is need for systems to facilitate interpretation of multidimensional data and conveyance to physicians in an easily digestible form. This can be achieved by integration of laboratory and artificial intelligence technologies within a clinical decision support system (CDSS), which for laboratory medicine must satisfy not only regulatory requirements for in vitro diagnostics but also those for a CDSS as a medical device. For medical device regulatory requirements, the CDSS should not only provide valid support of clinical decisions, but should also be robust, efficient, cost-effective and safe for the designated task. In diagnostics, applications of artificial intelligence usually involve supervised machine learning (ML) with algorithms for disease classification to generate models that are then applied to laboratory and other clinical data. Mass spectrometry-based steroidomics and catecholamine metabolite profiling for diagnostic stratification of patients with endocrine hypertension is one area of laboratory medicine for which the various technologies are now being integrated within a CDSS to convey ML model-derived interpretations to physicians. For this the technologies must be validated according to combinations of requirements for laboratory medicine and ML technologies. For deployment of ML models beyond the centre where models are developed it is critical to establish “generalisability”, or ability of ML-based models to correctly interrogate new data, which for ML applications also covers demonstration of robustness. Demonstration of robustness in ML refers to a model's ability to maintain high performance when faced with variations, perturbations or adversarial attacks in the input data. In part, such requirements for laboratory medicine can be met from conventional testing of reproducibility of generated ML-based probability scores within and between different laboratories, with comparisons to underlying test results. Need for “explainability” to ensure clinical trust in outputs of ML models may be facilitated by various tools, such as data visualisation aids, that may be incorporated within a CDSS. This and associated patient-specific interpretations and narrative reports may thereby assist clinical decision-making, which then requires final testing in controlled randomised trials against the routine clinical standard.

7:30 - 8:30pm

Refreshments and Networking