Limited generalizability and high risk of bias in multivariable models predicting conversion risk from mild cognitive impairment to dementia: A systematic review DOI Creative Commons
Robin J Vermeulen,

Vebjørn Andersson,

Jimmy Banken

et al.

Alzheimer s & Dementia, Journal Year: 2025, Volume and Issue: 21(4)

Published: April 1, 2025

Abstract Prediction models have been developed to identify mild cognitive impairment (MCI) cases likely convert dementia. This systematic review summarizes multi‐source prediction for MCI dementia conversion. PubMed and Embase were searched model development validation studies from inception up January 18 2024. Models assessed included predictors, predictive performance, risk of bias, generalizability. 62 included: 41 machine learning models, 11 regression 5 disease state indexes. The number predictors in the ranged 2 60; magnetic resonance imaging (MRI) scores most common sources. Performance measures indicate reasonable capabilities (area under curve [AUC] range: 0.58–0.98, accuracy 66.1–96.3%); however, are at high bias 47 lack external validation. Currently, no highly valid is available conversion due limited generalizability studies. Highlights Numerous predict likelihood individuals with MCI. seem a reasonably good performance predicting dementia, often lacking. There low that has externally validated accurately For more emphasis should be directed towards validation, generalizability, clinical applicability.

Language: Английский

Limited generalizability and high risk of bias in multivariable models predicting conversion risk from mild cognitive impairment to dementia: A systematic review DOI Creative Commons
Robin J Vermeulen,

Vebjørn Andersson,

Jimmy Banken

et al.

Alzheimer s & Dementia, Journal Year: 2025, Volume and Issue: 21(4)

Published: April 1, 2025

Abstract Prediction models have been developed to identify mild cognitive impairment (MCI) cases likely convert dementia. This systematic review summarizes multi‐source prediction for MCI dementia conversion. PubMed and Embase were searched model development validation studies from inception up January 18 2024. Models assessed included predictors, predictive performance, risk of bias, generalizability. 62 included: 41 machine learning models, 11 regression 5 disease state indexes. The number predictors in the ranged 2 60; magnetic resonance imaging (MRI) scores most common sources. Performance measures indicate reasonable capabilities (area under curve [AUC] range: 0.58–0.98, accuracy 66.1–96.3%); however, are at high bias 47 lack external validation. Currently, no highly valid is available conversion due limited generalizability studies. Highlights Numerous predict likelihood individuals with MCI. seem a reasonably good performance predicting dementia, often lacking. There low that has externally validated accurately For more emphasis should be directed towards validation, generalizability, clinical applicability.

Language: Английский

Citations

1