Advancing Multiple System Atrophy Diagnosis: Combining AI and Decision Analysis
Author(s)
Beate Jahn, PhD1, Igor Kuchin, MD MSc1, Gaby Sroczynski, MPH DrPH1, Marjan Arvandi, MStat PhD1, Julia Santamaria, MA1, Daniela Schmid, PhD2, Georg Goebel, Mag PhD3, Florian Krismer, MD PhD4, Anette Schrag, MD PhD5, Petra Schwingenschuh, MD PhD6, Alessandra Fanciulli, MD PhD4, Uwe Siebert, MPH MSc ScD MD7.
1Institute of Public Health, Medical Decision Making and Health Technology Assessment, Department of Public Health, Health Services Research and Health Technology Assessment, UMIT TIROL – University for Health Sciences and Technology, Hall in Tirol, Austria, 2Faculty of Life Sciences, Albstadt-Sigmaringen University, Sigmaringen, Germany, 3Institute of Clinical Epidemiology, Public Health, Health Economics, Medical Statistics and Informatics, Medical University Innsbruck, Innsbruck, Austria, 4Department of Neurology, Medical University Innsbruck, Innsbruck, Austria, 5Department of Clinical and Movement Neurosciences, University College London, London, United Kingdom, 6Department of Neurology, Medical University of Graz, Graz, Austria, 7Department of Public Health, Health Services Research and Health Technology Assessment, UMIT TIROL – University for Health Sciences and Technology; Harvard Chan School of Public Health; Harvard Medical School, Hall in Tirol, Austria.
1Institute of Public Health, Medical Decision Making and Health Technology Assessment, Department of Public Health, Health Services Research and Health Technology Assessment, UMIT TIROL – University for Health Sciences and Technology, Hall in Tirol, Austria, 2Faculty of Life Sciences, Albstadt-Sigmaringen University, Sigmaringen, Germany, 3Institute of Clinical Epidemiology, Public Health, Health Economics, Medical Statistics and Informatics, Medical University Innsbruck, Innsbruck, Austria, 4Department of Neurology, Medical University Innsbruck, Innsbruck, Austria, 5Department of Clinical and Movement Neurosciences, University College London, London, United Kingdom, 6Department of Neurology, Medical University of Graz, Graz, Austria, 7Department of Public Health, Health Services Research and Health Technology Assessment, UMIT TIROL – University for Health Sciences and Technology; Harvard Chan School of Public Health; Harvard Medical School, Hall in Tirol, Austria.
OBJECTIVES: Multiple system atrophy (MSA) is a rare, fatal neurodegenerative movement disorder. Differentiating between MSA and Parkinson's disease (PD) is prone to misdiagnosis. This study supports the development of an AI-based diagnostic tool incorporating clinical features and brain MRI to predict MSA or PD. We aim to optimize medical decision-making by identifying the optimal diagnostic cutoff along the receiver operating characteristic (ROC) curve, balancing sensitivity and specificity while considering the long-term consequences of correct and incorrect diagnosis.
METHODS: We developed a decision-analytic model to optimize AI-based MSA diagnostic tool. The decision population includes individuals presenting with parkinsonian symptoms at specialized clinics. The primary outcome is quality-adjusted life expectancy (QALE), synthesizing long-term benefits and harms. A probability-based decision-tree model quantified outcomes for correctly and incorrectly classified patients with MSA and PD across cutoffs along the ROC curve. As proof of concept, the model used published ROC curve data. MSA prevalence (12.5%) was based on tertiary referral center. We assumed a utility of 0.73 at diagnosis, with revision after five years to reflect disease progression. Utility decrements over time were derived from cohort studies and expert input, incorporating effects of personalized care and future disease-modifying treatment.
RESULTS: Our analysis provided a QALE of 8.75 QALYs for low-sensitivity (0.001)/high-specificity (0.999) and 8.67 QALYs for the inverse (sensitivity=0.999, specificity=0.001). The optimal cutoff (sensitivity=0.621, specificity=0.953) yielded 8.76 QALYs. Results were sensitive to assumptions about MSA prevalence and the potential effectiveness of a future disease-modifying treatment. Higher prevalence and availability of treatment favored cutoffs with higher sensitivity.
CONCLUSIONS: Our study presents a decision-analytic model enhancing the development of AI-driven diagnostic tools by incorporating long-term benefit-harm considerations in diagnostic decision-making. Next steps include integrating real-world diagnostic performance data and results from a Delphi study estimating the potential effects of future disease-modifying treatment on life expectancy and quality of life in MSA.
METHODS: We developed a decision-analytic model to optimize AI-based MSA diagnostic tool. The decision population includes individuals presenting with parkinsonian symptoms at specialized clinics. The primary outcome is quality-adjusted life expectancy (QALE), synthesizing long-term benefits and harms. A probability-based decision-tree model quantified outcomes for correctly and incorrectly classified patients with MSA and PD across cutoffs along the ROC curve. As proof of concept, the model used published ROC curve data. MSA prevalence (12.5%) was based on tertiary referral center. We assumed a utility of 0.73 at diagnosis, with revision after five years to reflect disease progression. Utility decrements over time were derived from cohort studies and expert input, incorporating effects of personalized care and future disease-modifying treatment.
RESULTS: Our analysis provided a QALE of 8.75 QALYs for low-sensitivity (0.001)/high-specificity (0.999) and 8.67 QALYs for the inverse (sensitivity=0.999, specificity=0.001). The optimal cutoff (sensitivity=0.621, specificity=0.953) yielded 8.76 QALYs. Results were sensitive to assumptions about MSA prevalence and the potential effectiveness of a future disease-modifying treatment. Higher prevalence and availability of treatment favored cutoffs with higher sensitivity.
CONCLUSIONS: Our study presents a decision-analytic model enhancing the development of AI-driven diagnostic tools by incorporating long-term benefit-harm considerations in diagnostic decision-making. Next steps include integrating real-world diagnostic performance data and results from a Delphi study estimating the potential effects of future disease-modifying treatment on life expectancy and quality of life in MSA.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
MSR14
Topic
Medical Technologies, Methodological & Statistical Research, Patient-Centered Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
Neurological Disorders, No Additional Disease & Conditions/Specialized Treatment Areas, Rare & Orphan Diseases