Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Mahieux et al 1 report on an interesting study of predictors of dementia in Parkinson’s disease. We have certain reservations about the results because of the methodological limitations of the study.
The diagnosis of dementia was made by retrospective case note review or retrospective completion of questionnaires including some by non-specialists which raises questions about the validity of diagnosis. The researchers making the diagnosis of dementia were not blind to neuropsychological test score, which raises the possibility of observer bias. The study was uncontrolled so that the estimate of the incidence of dementia cannot be adequately interpreted because it is not known what the incidence in the general population would be using this method of case identification.
The findings of predictors of dementia are a little puzzling. Aithough four variables are reported as predictors, this seems to be misleading because for all but one of these the 95% confidence interval for the relative risk includes the figure one, indicating that the relative risk is not significant at the 5% level using a hypothesis test.2 The p values for two of these variables were indeed greater than 0.05, but although the p value is given for the WAIS picture completion subtest as 0.03 (table 3, not table 2 as stated), the 95% confidence interval for the relative risk includes one. If these two statistical tests give different answers, which is correct?
For only one predictor variable did the 95% confidence interval of the relative risk lie outside one. This variable, age at onset of Parkinson’s disease, was entered into the multivariate model, but it seems that age at inception to the study was not entered into the model. It is well known that dementia in Parkinson’s disease is associated with age.3 4 If age onset is correlated with age at entry to the study as has been found by others3 5then age at onset of Parkinson’s disease is confounded. Many studies have reported that dementia in Parkinson’s disease is associated with older age at onset, but those studies which have controlled for the effect of age have not found such an association.5 6
The authors reply:
We appreciate the interest of Hughes et al in our study on the neuropsychological prediction of dementia in Parkinson’s disease.1-1
The validity of the diagnosis of dementia was discussed in the paper. Hughes et al conjecture that the knowledge of initial neuropsychological tests scores would raise an observer bias. But the final diagnosis (dementia or not) was made by independent raters in 22 cases out of 86, and in every case we used a special case report form which did not even refer to the initial neuropsychological examination.
We agree that the incidence of dementia in our population cannot be interpreted as that of an inpopulation epidemiological study. We mentioned it as a mere additional descriptive characteristic.
We agree that the significance of the completion subtest was borderline in the multivariate analysis, as the confidence interval of the relative risk includes one, according to the calculation done by BMDP software. The level of significance is < 0.05, as mentioned in the table, again according to the test performed by the BMDP package. This minor discrepancy is not unusual and should not affect the interpretation of our results in such an exploratory clinical epidemiological study.
With regard to age and age at onset, these two variables correlate strongly. The two have been reported as predictive factors for dementia in Parkinson’s disease.1-2-1-5 Some authors consider that age at onset of Parkinson’s disease is a potentially more important variable.1-5-1-7 Moreover, the inclusion of age instead of age at onset in the multivariate analysis modifies the results only slightly. The WAIS-R completion subtest and age remain significant and independent predictors of dementia with respective risk ratios of 5.344 for age>68 years (95% CI 1.7–17.1; p<0.01) and 10.929 for a completion score <10 (95% CI 2.5—48.3).