Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Improving Risk Scores for Dementia

Improving Risk Scores for Dementia Recently, Reitz and colleagues1 proposed a risk score for AD. Risk scores for AD are virtually nonexistent compared with the numerous etiological studies on individual risk factors for AD. We give 3 suggestions for improving the accuracy of the absolute risks assessed with the risk score. First, more advanced methods could have been used to estimate absolute risks from the interval-censored data.2 To establish the diagnosis of dementia, follow-up data were collected at sequential visits with 18-month intervals. As a result, the time of onset of dementia was not exactly observed. The authors assumed that the onset occurred at the end of each interval. This approach can lead to inaccurate estimates of absolute risks. Second, the presented absolute risks result from a logistic regression model rather than from the Cox model, ignoring the time to event character of the data. We disagree that regression coefficients from the Cox model are less practical for clinical use. In fact, the Cox model allows for risk estimates at different points in time, eg, 1-, 2- and 5- years risks. A third aspect that may have influenced the risk estimates is overfitting of the regression model. The number of estimated coefficients is relatively high given the number of participants with dementia (ie, 15 coefficients in the multivariable model with 92 cases of dementia). As a rule, no more than 1 coefficient should be considered per 10 patients with the disease. Therefore, the model is likely to be overfitted, and the predicted risks will be too extreme for new patients. Solutions for this typical prediction problem are considering fewer variables in the modeling procedure and shrinkage of the estimated coefficients, eg, by using bootstrapping techniques.3 Additionally, the authors state that the risk score can help clinicians; however, the score was developed in a community-based sample where the prior probability of developing AD will be much lower than in a clinical setting. Clinicians should be aware that the posterior probabilities presented in the article may not be applicable to their patients. Also, APOE genotype may not be available to clinicians. In conclusion, we appreciate the development of a risk score for dementia but the methods can be improved. This score should be tested in new patients and may need updating, as acknowledged by the authors. Correspondence: Dr Geerlings, Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, PO Box 85500, 3508 GA Utrecht, the Netherlands (m.geerlings@umcutrecht.nl). Financial Disclosure: None reported. References 1. Reitz CTang MXSchupf NManly JJMayeux RLuchsinger JA A summary risk score for the prediction of Alzheimer disease in elderly persons. Arch Neurol 2010;67 (7) 835- 841PubMedGoogle ScholarCrossref 2. Lindsey JCRyan LM Tutorial in biostatistics methods for interval-censored data. Stat Med 1998;17 (2) 219- 238PubMedGoogle ScholarCrossref 3. Harrell FE JrLee KLMark DB Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996;15 (4) 361- 387PubMedGoogle ScholarCrossref http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Archives of Neurology American Medical Association

Improving Risk Scores for Dementia

Archives of Neurology , Volume 68 (2) – Feb 14, 2011

Loading next page...
 
/lp/american-medical-association/improving-risk-scores-for-dementia-qWGsjWV0ko

References (4)

Publisher
American Medical Association
Copyright
Copyright © 2011 American Medical Association. All Rights Reserved.
ISSN
0003-9942
eISSN
1538-3687
DOI
10.1001/archneurol.2011.3
Publisher site
See Article on Publisher Site

Abstract

Recently, Reitz and colleagues1 proposed a risk score for AD. Risk scores for AD are virtually nonexistent compared with the numerous etiological studies on individual risk factors for AD. We give 3 suggestions for improving the accuracy of the absolute risks assessed with the risk score. First, more advanced methods could have been used to estimate absolute risks from the interval-censored data.2 To establish the diagnosis of dementia, follow-up data were collected at sequential visits with 18-month intervals. As a result, the time of onset of dementia was not exactly observed. The authors assumed that the onset occurred at the end of each interval. This approach can lead to inaccurate estimates of absolute risks. Second, the presented absolute risks result from a logistic regression model rather than from the Cox model, ignoring the time to event character of the data. We disagree that regression coefficients from the Cox model are less practical for clinical use. In fact, the Cox model allows for risk estimates at different points in time, eg, 1-, 2- and 5- years risks. A third aspect that may have influenced the risk estimates is overfitting of the regression model. The number of estimated coefficients is relatively high given the number of participants with dementia (ie, 15 coefficients in the multivariable model with 92 cases of dementia). As a rule, no more than 1 coefficient should be considered per 10 patients with the disease. Therefore, the model is likely to be overfitted, and the predicted risks will be too extreme for new patients. Solutions for this typical prediction problem are considering fewer variables in the modeling procedure and shrinkage of the estimated coefficients, eg, by using bootstrapping techniques.3 Additionally, the authors state that the risk score can help clinicians; however, the score was developed in a community-based sample where the prior probability of developing AD will be much lower than in a clinical setting. Clinicians should be aware that the posterior probabilities presented in the article may not be applicable to their patients. Also, APOE genotype may not be available to clinicians. In conclusion, we appreciate the development of a risk score for dementia but the methods can be improved. This score should be tested in new patients and may need updating, as acknowledged by the authors. Correspondence: Dr Geerlings, Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, PO Box 85500, 3508 GA Utrecht, the Netherlands (m.geerlings@umcutrecht.nl). Financial Disclosure: None reported. References 1. Reitz CTang MXSchupf NManly JJMayeux RLuchsinger JA A summary risk score for the prediction of Alzheimer disease in elderly persons. Arch Neurol 2010;67 (7) 835- 841PubMedGoogle ScholarCrossref 2. Lindsey JCRyan LM Tutorial in biostatistics methods for interval-censored data. Stat Med 1998;17 (2) 219- 238PubMedGoogle ScholarCrossref 3. Harrell FE JrLee KLMark DB Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996;15 (4) 361- 387PubMedGoogle ScholarCrossref

Journal

Archives of NeurologyAmerican Medical Association

Published: Feb 14, 2011

Keywords: dementia

There are no references for this article.