In many publications, researchers compare measures using a screening procedure between extreme conditions of interests (those highly affected vs. healthy) without including patients for which the test is most likely to be used on; those for which the situation is not clear.
In your analysis, you compare AD patients to healthy patients without including those with MCI. However, in practice, we are likely to use the instrument on patients with the entire spectrum of severity of dementia. Your results are therefore only valid if we can previously exclude those with MCI and then use the test.
How is this clinically feasible?
Article Predictors of cognitive decline in Alzheimer's disease and m...
I do not understand your question. The AD group does not include the MCI-Ad group
Indeed, but neither does the control group. This means you are comparing AD to a healthy control. You have artificially increased the AUC, specificity and positive predictive values. This is due to the fact you excluded from your analysis those most likely to be considered as false positive.
This does not mean your test is unable to distinguish ADs from non-ADs, it just means your predicting values are probably exaggerated.
When organizing a diagnostic study, it is important to select a representative sample of patients. By this, I mean patients for which the test will be applied to.
The error found in your study is quiet frequent in publications and unfortunately editors and reviewers are often not trained enough in epidemiology to notice them. This is unfortunate as it prevents authors from correcting their interpretation or adding important limitations in the discussion section.
1. The purpose of the article was: MCI-Ad, often have global high scores at baseline. The global score of MMSE and CAMCOG not discriminate either cognitive impairment. We wanted to know discriminant CAMCOG subscales impairment at baseline for all cases. The result was: Subscales orientation and recent memory and learning.
2. The greater certainty of diagnosis between groups (although not 100%) was that of a proven diagnosis after 5 years.
3. These results are valid and are contrasted and tested in daily clinical practice
The diagnosis was contrasted after 5 years in ALL CASES (ALSO FOR THE CONTROL GROUP)
Thank you professor Lluis Conde-Sala for these precisions. Compared to many other published epidemiological studies in psychology or psychiatry, I do think your study is well designed and its longitudinal aspects does give it value.
Again, I am not contesting your overall results, I am just pointing out that predictive values for AD at five years are exaggerated given the population who developed MCI were excluded from that analysis.
I am also sorry to insist on the fact that this does limit external validation of the AUC values for detecting AD. For them to be applicable we would need to be able to previously rule out all patients who would develop MCI during the following five years. I believe the AUC should compare patients who develop AD at five years compared to all other participants who do not including those who develop MCI.
Thanks for your interest and your comments. I attach supplementary data with 2 options
Thanks for your interest and your comments. I attach supplementary data with 2 options
Perfect, this is exactly the information I was hoping you could provide. I am grateful for your professionalism and scientific integrity. Not many researchers are willing to take the time to answer picky comments from unknown colleagues.
Here is a summary of the results provided in the attached file (for those who do not have microsoft office with them).
When comparing sub-tests ability to detect AD versus MCI and healthy population, we obtain the following AUC.
CAMCOG Baseline AUC=0.688
MMSE Baseline AUC=0.744
OR+MEM Baseline AUC=0.865
These results indeed confirm that combining OR to MEM has a very good ability to distinguish ADs from non-ADs. These results are applicable for patients attending primary care physicians and show that this combination can serve as a screening test to detect AD.
Thank you again for these precisions and complementary information.
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India