Computerized decision support is an effective approach to select memory clinic patients for amyloid‐PET
Hanneke F.M. Rhodius‐ Meester, Ingrid S. van Maurik, Lyduine E. Collij, Aniek M. van Gils, Juha Koikkalainen, Antti Tolonen, Yolande A.L. Pijnenburg, Frederik Barkhof, Elsmarieke van de Giessen, Jyrki Lötjönen, Wiesje M. van der Flier- Psychiatry and Mental health
- Cellular and Molecular Neuroscience
- Geriatrics and Gerontology
- Neurology (clinical)
- Developmental Neuroscience
- Health Policy
- Epidemiology
Abstract
Background
The use of amyloid‐PET in daily clinical practice is upcoming. Yet, ttranslation of the appropriate use criteria (AUC) to clinical practice is challenging, hampering successful implementation of amyloid‐PET. Tools that guide selecting patients for whom amyloid‐PET has most clinical utility are needed. Therefore, we developed and evaluated a computerized decision support approach to select patients for amyloid‐PET testing.
Method
We included 286 subjects (136 controls, 116 Alzheimer’s disease (AD) dementia, 26 frontotemporal lobe dementia (FTD), and 8 vascular dementia (VaD) from the Amsterdam Dementia Cohort, with available neuropsychology, APOE, MRI and [18F]florbetaben amyloid‐PET. In our computerized decision support approach, we first classified the subjects using only neuropsychology, APOE, and MRI data. Then, for uncertain subjects (probability of correct class (PPC) < 0.75) we tested the classification by adding amyloid positive (AD like) and negative (normal) visual PET read results, and assessed whether either of the values makes the diagnosis reliable (PPC≥0.75). If the threshold of PCC was reached, the actual visual PET read result was used in the final classification. We compared the proportion of PET scans and patients diagnosed with sufficient confidence (PPC ≥0.75) in the computerized approach with three scenarios: 1) without amyloid‐PET (only demographics, neuropsychology, APOE and MRI), 2) amyloid‐PET according to the AUC and 3) amyloid‐PET for all patients. The study was performed using five‐fold cross‐validation.
Result
The computerized approach advised PET in n = 76(27%) patients, leading to a diagnosis with sufficient confidence in n = 190(66%) patients. This approach outperformed the other three scenarios: 1) without amyloid‐PET, diagnosis was obtained in n = 147(51%), 2) applying the AUC resulted in amyloid‐PET in n = 113(40%) and diagnosis in n = 160(56%), and 3) performing amyloid‐PET in all patients resulted in diagnosis in n = 165(58%).
Conclusion
Our computerized data‐driven approach restricted the application of amyloid‐PET to 27% in a memory clinic cohort of controls, AD, FTD, and VaD, without compromising diagnostic accuracy. This approach can support clinicians in making a balanced decision in ordering additional biomarker testing during the dementia workup. Thus, innovative use of resources can be stimulated, which is essential when disease‐modifying drugs increase the demand for biomarker testing.