Abstract
In a prospective study, we recently discovered 8 clinical predictors of dementia in Parkinson’s disease. Here, we validate these dementia predictors using two additional prospective cohorts (n = 134). After a 3.6-year follow-up, 35/134 developed dementia (7.2% per year). When confirming individual variables, 5/8 were significantly associated with dementia in the validation cohort. These included age, male sex, baseline RBD, orthostatic hypotension, and MCI. Bilateral onset, hallucinations and falls/freezing did not significantly predict dementia; however, point estimates of OR were all >1. In all cohorts, the strongest determinant for dementia development was the co-existence of RBD, MCI and orthostatic hypotension at baseline.
Dementia is a devastating yet difficult-to-predict complication of Parkinson disease (PD). Identifying people with PD who are at risk of developing dementia can be important in optimizing care, by avoiding medications that cause hallucinations, screening actively for potential cholinesterase use, etc. It also can help understand heterogeneity of disease mechanisms and create cohorts of patients who are potential participants in clinical trials aimed at dementia prevention.
Several possible predictors of dementia in PD have been previously identified [1–6]. Whereas some predictors, such as age and hallucinations have been consistently reported in other studies, others have been controversial. In a recent 4.5-year prospective cohort study [1, 7], we capitalized upon a deeply-phenotyped cohort to identify an array of dementia predictors in PD. REM sleep behaviour disorder (RBD) and mild cognitive impairment (MCI) were the strongest predictors, as well as orthostatic systolic blood pressure drop, impaired color vision, bilateral disease onset, male sex, reduced quantitative motor scores (Purdue Pegboard test, alternate tap test), history of falls and freezing, and predominant gait involvement [1]. A key for any cohort study with multiple outcomes is to confirm findings on a different cohort. Therefore, we aimed to replicate our recently-reported dementia predictors on two additional cohorts; one based in Montreal, which was gathered from studies on non-motor PD [8] and insomnia in PD [9], and another based in Totorri Japan [10].
METHODS
Two validation cohorts were combined:
1. A Montreal cohort of PD patients who had participated in two previous studies; a non-motor screen for PD [8] and a study of insomnia in PD [9]. All were enrolled between 2009 and 2011. These included 67 participants. 8 had also participated in the primary cohort, and hence were excluded from the validation cohort. Of the remaining 59, follow-up cognitive status was available in 53 (90%).
2. A Tottori (Japan) cohort consisting of 82 participants from a cohort study that focused on sleep and dementia [10]. 133 patients at the Tottori University Hospital were enrolled between October 2004 and January 2011, of whom follow-up information was available in 82 (61.7%).
In both cohorts, all were free of dementia at baseline based on the MDS PD dementia diagnostic criteria [11]. To maximize generalizability, no other exclusion criteria were applied and all patients in the baseline study were eligible for the follow-up assessment. All patients had provided written consent to participate, and the protocol was approved by each centers’ research ethics board.
We assessed 8 baseline variables that predicted dementia in our previous study [1], including age, male sex, orthostatic hypotension (systolic drop >10 mmHg lying to standing), REM sleep behavior disorder (RBD: diagnosed with polysomnography in Totorri and on expert interview by RP in Montreal, mild cognitive impairment (MCI: defined as education-adjusted Montreal Cognitive Assessment [12] <26 with subjective cognitive change (Montreal cohorts only)), hallucinations, bilateral symptom onset (according to patient interview, supplemented by examination documenting no unilaterality at diagnosis) and falls and/or freezing.
The primary outcome was final cognitive status categorized as dementia versus no dementia. Dementia was defined according to Level I MDS criteria as the presence of cognitive impairment on a global bedside test (i.e. MMSE <26), plus substantial impairment in activities of daily living due to cognitive impairment [11]. Analysis was with IBM SPSS version 22.0. To validate individual dementia predictors, binary logistic regression analysis with adjustments for baseline age, sex, disease duration, and duration of follow-up were performed for each variable. No adjustment for multiple comparisons was performed.
RESULTS
A total of 199 patients had participated in the validation cohort. Follow-up cognitive status could be confirmed for 135/199 (68%). On examination of those lost to follow-up in the Montreal cohort (Japanese data not available), 10 patients were not included in the follow-up. Of these, 2 died, 2 had a diagnosis reassigned (both to MSA), and 6 were lost to follow-up. 6/10 (60%) were over 70 (compared to 36% of the whole cohort), and 6 were male. The number of dementia risk factors in those lost to follow-up was similar 3.65 / 8 (46%), compared to 3.2 / 8 (40%) in the whole cohort. After a mean 4.2-year follow-up (range = 1–7), 30/134 (22%) developed dementia. Results of validation are shown in the Table 1. When confirming individual variables, 5/8 predictive variables were significantly associated with dementia in the validation cohort. These included age, male sex, baseline RBD, orthostatic hypotension, and MCI. Bilateral disease onset, hallucinations and falls/freezing did not significantly predict dementia; however, OR point estimates were all >1. Bilateral onset was strongly predictive in the Montreal cohort (OR = 15 (95% CI 1.8–125.0)), but not in the Tottori cohort (OR = 0.4 (0.1–1.5)). When analysis of RBD was restricted to those cases with polysomnographic confirmation (i.e. Tottori cohort), the odds ratio increased to from 5.4 to 9.8 (95% CI 2.0–49.7). Variables were correlated; if all variables were added to the model, systolic drop became non-significant (OR = 2.8 (0.76–10.6); this is consistent with a close correlation between RBD and OH, as previously described [7, 13] (see Supplemental Table 1).
DISCUSSION
In this validation analysis of dementia predictors, we broadly confirmed the results of our previous cohort study. In both cohorts, the strongest determinant for dementia status conversion was the co-existence of RBD, MCI and orthostatic hypotension at baseline. A high correlation between these variables has been reported, and may mark a diffuse degenerative process with poor prognosis [7].
Three variables from our primary cohort (hallucinations, bilateral disease onset, and falls/freezing) were not confirmed in the second cohort (although OR ranged from 1.2–1.8). This may simply reflect between-cohort heterogeneity, particularly given that centers from three different populations in two different continents were used. Note that hallucinations and gait have been clearly associated with dementia risk in other studies [14]; the fact that these variables were relatively uncommon at baseline suggests that lack of power may explain the absence of significant correlation.
There are some limitations to our study. 32% of our population was lost to follow-up, and is possible that lost patients may have had different dementia risk than those included. Power was limited to assess some variables, especially given the 22% dementia conversion in the confirmatory cohort (somewhat lower than the original cohort). We also did not adjust for multiple comparisons (n = 8); if correction was performed, the study would have been further underpowered; note, however, that on the full regression model, 4/8 variables were significant. Hence, confirmation of our predictive variables and validation of our scoring system in a larger cohort would be useful. Indeed, many dementia predictive studies are ongoing and will produce valuable information soon. In both studies we assessed dementia predictors over a medium term (follow-up period of 4-5 years); obviously, over long term, a much higher proportion of patients will develop dementia. In the initial study and in one of the validation cohorts, RBD diagnosis was based on a gold-standard polysomnogram, which will generally not be available in a routine clinical setting. When the tool assesses RBD only with clinical interview, resulting misclassification would reduce sensitivity/specificity (note that the second validation cohort used only clinical interview for RBD diagnosis, with some attenuation of the OR for RBD as a predictive marker). The original cohorts were not created to assess dementia predictors; they were selected here because the essential information had been documented as part of baseline study protocols. However, baseline cognition in the Tottori cohort was tested only with MMSE; since this cannot reliably define MCI, cognitive data from the Tottori cohort could not be added to the validation analysis. Finally, the diagnosis of dementia in the confirmatory Montreal and Tottori cohorts was made according to level I MDS dementia diagnostic criteria; a more detailed neuropsychological evaluation (level II criteria) might have resulted in higher dementia conversion in this group.
On the other hand, our study has some strengths. We selected validation cohorts from very different populations, increasing generalizability of findings. The analysis used prospectively-acquired data, minimizing information bias. Finally, the predictors we selected are easily acquired in the office setting, allowing practical use in most clinical settings.
In summary, the risk of dementia in PD can be predicted using relatively simple clinical variables. This may provide the opportunity to design risk prediction tools to help decision-making, individualize symptomatic therapy, and even select suitable candidates for clinical trials against dementia.
DISCLOSURES/CONFLICTS OF INTEREST
This study was supported by the Fonds de Recherche du Québec - Santé and the Canadian Institute of Health Research; the authors have no financial relationships with any organizations that might have an interest in the submitted work in the previous 3 years. RB Postuma received personal compensation for travel and speaker fees from Biotie, Biospective, Boehringer, Roche, Novartis Canada and Teva Neurosciences, and is funded by grants from the Fonds de Recherche du Québec - Santé, the Parkinson Society of Canada, The Webster Foundation, and by the Canadian Institutes of Health Research. JF Gagnon is funded by grants from the Fonds de Recherche du Québec - Santé, the W. Garfield Weston Foundation and by the Canadian Institutes of Health Research. He holds a Canadian Chair on Cognitive Decline in Pathological Aging. J. Anang, T. Nomura, S. Rios Romenets and K. Nakashima report no disclosures.