Article Text

other Versions

Research paper
Demographic and motor features associated with the occurrence of neuropsychiatric and sleep complications of Parkinson's disease
  1. Renato Puppi Munhoz1,2,
  2. Hélio A Teive1,
  3. Hariklia Eleftherohorinou3,
  4. Lachlan J Coin3,
  5. Andrew J Lees4,
  6. Laura Silveira-Moriyama4,5
  1. 1Department of Neurology, Hospital de Clínicas, Federal University of Paraná, Curitiba, Brazil
  2. 2Associacao Paranaense dos Portadores de Parkinsonismo, Curitiba, Brazil
  3. 3Department of Epidemiology and Public Health, Centre for Biostatistics, Imperial College St Mary's Campus, London, UK
  4. 4Reta Lila Weston Institute, UCL Institute of Neurology, London, UK
  5. 5Department of Neurology, FCM, University of Campinas, Campinas, UNICAMP, Brazil
  1. Correspondence to Dr Laura Silveira-Moriyama, Reta Lila Weston Institute of Neurological Studies, UCL Institute of Neurology, London WC1N 1PJ, UK; laura.moriyama{at}


Objective To determine whether four key neuropsychiatric and sleep related features associated with Parkinson's disease (PD) are associated with the motor handicap and demographic data.

Background The growing number of recognised non-motor features of PD makes routine screening of all these symptoms impractical. Here, we investigated the hypothesis that standard demographic data and the routine assessment of motor signs is associated with the presence of dementia, psychosis, clinically probable rapid eye movement (REM) sleep behavior disorder (cpRBD) and restless legs syndrome (RLS).

Methods 775 patients with PD underwent standardised assessment of motor features and the presence of dementia, psychosis, cpRBD and RLS. A stepwise feature elimination procedure with fitted logistic regression models was applied to identify which/if any combination of demographic and motor factors is associated with each of the four studied non-motor features. A within-study out-of-sample estimate of the power of the predicted values of the models was calculated using standard evaluation procedures.

Results Age and Hoehn&Yahr (H&Y) stage were strongly associated with the presence of dementia (p value<0.001 for both factors in the final selected model) while a combination of age, disease duration, H&Y stage, dopamine agonists and catechol-O-methyltransferase (COMT) inhibitors was associated with the presence of psychosis. Disease duration and H&Y stage were the significant indicators of cpRBD, and the lack of significant motor asymmetry was the only significant feature associated with RLS-type symptoms but the evidence of association was weak.

Conclusions Demographic and motor features routinely collected in patients with PD can estimate the occurrence of neuropsychiatric and sleep-related features of PD.


Statistics from


Non-motor features are well recognised symptoms of Parkinson's disease (PD). Hyposmia and rapid eye movement (REM) sleep behavior disorder (RBD) may precede the motor signs while dementia and psychosis are common in the late stages of the illness.1 A smaller proportion of patients with PD have sensory/motor symptoms which have similarities to akathisia and restless legs syndrome (RLS).2 The evaluation of motor features of PD is part of routine neurological practice and it is done in research settings through the use of standardised scales such as the Unified Parkinson's Disease Rating Scale (UPDRS)3 and the Hoehn and Yahr (H&Y) staging.4 Non-motor features impact significantly on the quality of life of most PD patients5 ,6 and require specific symptomatic therapy.

We performed a cross-sectional study of a large group of patients with PD to evaluate if data routinely collected in neurological consultations (age, disease duration, details of treatment and estimates of disease severity) could indicate which subjects are at increased risk of presenting with four key neuropsychiatric and sleep related features of PD: psychosis, dementia, REM sleep behavior disorder (RBD) and RLS-type symptoms.

Subjects and methods


All subjects with parkinsonism followed for at least 2 years by the Associação Paranaense dos Portadores de Parkinsonismo between March 2004 and July 2008 were invited to participate in the study. The Associação Paranaense dos Portadores de Parkinsonismo is a large non-profitable organisation with more than 2000 parkinsonian patient members in the state of Parana, in Southern Brazil. Initially 1008 patients with PD were assessed by a movement disorder specialist (RPM) who reviewed the diagnosis in each case and excluded from the study subjects who presented with limitations to the assessment of motor signs other than parkinsonism (previous stroke, stereotactic functional neurosurgery, chronic skeletal disorders, etc). Only the 775 patients who fulfilled the Queen Square Brain Bank criteria for PD7 were included.


Patients who agreed to participate provided an informed consent and were asked to come to their next follow-up visit after a 12 h (overnight) withdrawal of dopaminergic drugs.

Evaluation of motor features

Each patient was examined using a protocol that included H&Y staging4 and the UPDRS part III3 with separate rating for each side of the body and an additional item for arm swing scored similarly to UPDRS items (0: normal, 1: mild reduction, 2: moderate reduction, 3: marked reduction, 4: no arm swing). Rating of gait and balance problems was done using the previously validated Postural Instability and Gait Disorders (PIGD) subscore that includes items 13, 14, 15, 29 and 30 of the UPDRS.8 Classification of motor subtypes: All subjects were classified into early-onset PD (age of onset 45 years old or less), or late-onset PD (age of onset >45 years old). After the examination and interview, subjects were subjectively classified by the examiner into subtypes of parkinsonism according to predominant phenomenology: tremor-dominant (TD)—prominent resting tremor plus mild bradykinesia; rigid-akinetic (RA)— no resting tremor; and tremor, rigidity and bradykinesia (TRB) subtype. For the classification into symmetric and asymmetric motor features, we used items 20 (resting tremor), 22 (rigidity), 23, 24, 25 (bradykinesia) and the arm swing item for those on wheelchair or bedridden. To be considered asymmetric, each cardinal sign was considered separately as body sides were compared: resting tremor and rigidity had to have a minimum of 1-point difference in side-to-side comparison, while in the case of bradykinesia, the sum of the scores of the three UPDRS items had to have at least a 3-point difference. For patients with the TRB subtype, at least two of the three cardinal signs had to be asymmetric. To avoid bias for symmetry, in cases with RA subtype, the item for resting tremor (#20) was not included and rigidity and bradykinesia had to be asymmetric. Finally, in cases with TD subtype, the item for rigidity (#22) was included only if rated ≥1. In cases presenting with what we considered as motor asymmetry, side of asymmetry using these UPDRS items had to be congruent with side of asymmetry observed in the arm swing item. Additionally, patients were asked about disease laterality at disease onset and at the time of assessment.

Evaluation of neuropsychiatric and sleep related features

Each patient underwent a structured interview that included basic demographic and historical disease data, as well as specific questions regarding the presence of neuropsychiatric and sleep related symptoms. The diagnosis of dementia was determined according to diagnostic and statistical manual of mental disorders (DSM)-IV criteria.9 The presence of psychosis was determined by asking subjects about current visual or other sensory perceptual disturbances.5 Clinically probable RBD (cpRBD) was ascertained clinically by the criteria proposed by the American Sleep Disorders Association, which does not require polysomnography.10 This clinical classification has been used and validated in recent studies on this parasomnia.11 RLS-type symptoms were diagnosed according to the recommendations of the International Restless Legs Syndrome Study Group for the diagnosis of RLS.12

Statistical analysis

To investigate which demographic and motor features were better associated with the presence of the non-motor features of interest we applied a backward elimination procedure on logistic regression models. The four output variables that were tested were dementia, psychosis, cpRBD and RLS. The 16 covariates that were considered for inclusion in the models were the continuous variables of age, gender, disease duration, H&Y, PIGD, the number of medications used to treat PD; the nominal variables of the presence or absence of motor asymmetry, classification into early or late onset PD, use or no use of L-dopa, dopamine agonist, amantadine, selegiline, catechol-O-methyltransferase (COMT) inhibitor and anticholinergics; and two indicator variables to compare the TD and the RA forms with the more common TRB form as reference value. The stepwise feature elimination procedure starts by fitting a logistic model that has as covariates all the 16 demographic/motor features and as output the categorical variable of interest, such as dementia. In successive steps, it then eliminates the ‘worst’ included variable in an attempt to optimise a certain measure of goodness of fit of the model to the data. If at any point any of the eliminated covariates is worth being included again in the absence of others it gets chosen to be reincluded in the model. The procedure terminates when the optimisation criterion, in our case the Akaike information criterion has reached its minimum. The procedure was applied on the entire dataset using R and in subsets of the datasets to reduce the effects of overfitting. At all times, the results converged.13 See online supplementary material appendix 1 for more details on our model selection.

In order to provide an estimate of the power of the selected variables to identify the non-motor features we performed 10-fold cross validation. We fitted 10 models on 10 random 80/20 splits of the data. The 80% was each time used to train a model and the 20% of the data was used to test the predictive performance of the fitted model. The average sensitivity and specificity across all 10 trials were computed to produce the average receiver operating characteristic (ROC) plot and the area under the ROC curve according to standard procedures.14

To further explore the correlation patterns between the clinical features in the patients we performed additional analyses which are described in the online supplementary material appendix 2 and include measures of association between the studied features in pairs, and also their association with disease severity. All analyses were conducted at the end of the data collection for the total of 775 patients. No intermediate analyses or adjustments to the protocol were performed.


The stepwise elimination on fitted logistic regression models analyses showed that the combination of motor and demographic features that were more strongly associated with dementia were age (p<10−9, β=0.1, SE=0.01) and H&Y stage (p<10−16, β=1.7, SE=0.17); the ones that were best associated with psychosis were age (p<0.001, 95% CI for β=1.0 to 1.1), disease duration (p<10−4, β=0.07, SE=0.02), H&Y stage (p<10−16, β=1.3, SE=0.15), use of anticholinergic (p=0.04, β=0.61, SE=0.3) and use of COMT inhibitor (p=0.02, β=0.73, SE=0.32); the ones that were best related to cpRBD were disease duration (p=0.001, β=0.05, SE=0.01); age (p=0.02, β=0.01, SE=0.007) and H&Y stage (p=0.006, β=0.3, SE=0.11); and the one that was best related to RLS was symmetry of motor features (p=0.006, β=−0.85, SE=0.31). An estimate of the predictive performance of these models as provided by the 10-fold cross validation procedure is shown in the ROC curves in figure 1.

Figure 1

ROC plots showing the average power of the logistic models predicted values to identify each of the studied features. ROC plots that show the average performance of the 10-fold cross-validation models fitted to predict dementia (top left panel), psychosis (top right panel), RBD (bottom left panel) and restless legs syndrome (RLS) (bottom right panel). The x-axis shows the False Positive Rate (FPR) and the y-axis the True Positive Rate (TPR). At the bottom right corner of each plot the Average area Under the Curve (AUC) indicates the average classification performance of the models that is how well can the fitted models classify the outcome. An AUC=1 indicates a perfect classifier, an AUC=0.7 indicates a fair classifier and an AUC=0.5 indicates a random classifier. As it is shown, prediction of dementia was excellent, of psychosis was very good, of RBD was good and for RLS was poor. RBD, sleep behavior disorder; ROC, receiver operating characteristic.

The overall prevalence of RLS-type symptoms was 11.4% (88 out of 687 patients with PD screened, 95% CI=10.5% to 15.5%). In patients with what we considered symmetric motor features, the frequency of RLS was 21% (15 out of 70 patients with PD with symmetric features) while in the patients with asymmetric features, the frequency was 10.4% (73/705 patients, χ² p=0.005). The frequency of cpRBD, psychosis and dementia increases as the H&Y stage progresses, as can be seen in figure 2.

Figure 2

Stacked bar graph showing subjects with and without each of the studied feature. Stacked bar graph showing the number of subjects with (black subdivision of the bars) and without (grey subdivision of the bars) each of the studied features. Each bar represents one Hoehn and Yahr stage. Clinically probable REM sleep behavioural disorder (cpRBD) is the most common feature, and the proportion of subjects with cpRBD is higher in more advanced disease stages. The presence of restless legs syndrome (RLS) is not higher in advanced disease stages.

The continuous variables of age, age of onset, disease duration, H&Y and PIGD scores all correlated with each other. Strengths of the correlation are presented in online supplementary figure SF1 in appendix 1. Dementia and psychosis were strongly associated with each other independently of the H&Y stage. RLS was associated with cpRBD (p=0.024), but not with dementia (p=0.6), psychosis (p=0.7) or H&Y stage (p=0.4). Dementia, psychosis and cpRBD were all associated with H&Y stage (p<0.001), and were associated with each other when controlled for H&Y stage. Refer to online supplementary table ST1 in appendix 1 for details.


There is an extensive literature on the factors relating to which baseline clinical features indicate the future development of dementia in various cohorts of PD patients. Older age and disease severity have been consistently shown to predispose to dementia,15–20 and neuropathological data on longitudinally followed patients with PD show that older patients develop dementia and die sooner, having much shorter disease duration and more cortical pathology.21 This can be confirmed by clinical impression, which demonstrates that some patients with very long follow-up and a benign form of PD present with no dementia, while others with shorter disease duration but rather aggressive course of PD tend to present with dementia sooner in disease course. To our knowledge, no previous study has assessed whether the age and disease severity could be used in a cross-sectional fashion to indicate patients with high possibility of presenting with dementia. We show here that a combination of only those two variables is strongly associated with concurrent dementia. Age and H&Y stage can be obtained almost instantaneously during a PD consultation, making this a potential practical marker to assist clinicians with screening of dementia in the future.

Age, disease severity and also disease duration were all indicators of the presence of psychosis, in line with previous studies.22 ,23 Of interest, they were also indicative of cpRBD. Although RBD has been extensively discussed as an early non-motor feature of PD, few studies have examined the relationship between RBD and the disease course in established PD. Onofrj et al22 followed up 80 patients with PD for 8 years and demonstrated that although at baseline only 5 out of 80 (6.2%) had confirmed RBD, after 8 years 27 (34%) had developed RBD. Lavault et al23 followed up 61 patients with PD for 2 years and report that patients with suspected RBD had more motor disability at baseline, and also that with disease course, some patients without RBD developed this feature, while in others, the features had disappeared. Benninger et al24 compared 13 patients with PD with confirmed RBD with 13 without, and failed to show that patients with RBD presented worse gait or postural features on detailed examination. Lee et al25 examined 447 consecutive patients with PD and showed that age, disease duration, motor disability and motor subtype were predictors of cpRBD. Yoritaka et al26 clinically screened 150 patients for the presence of RBD and found that patients with RBD were older, had more dyskinesia and sleep attacks, but not worse H&Y stage. Our study demonstrates a strong association between cpRBD and disease duration and disease severity. Although RBD can happen early in disease course in a minority of patients, it is likely that RBD is a common feature in moderate to advanced stage PD. The large number of subjects makes our clinical finding robust, but the clinical screening for RBD is a limitation and further studies using polysomnography are warranted. In addition, if we were to use the clinical screening for disease severity and duration as a simple clinical screening tool for cpRBD, it would be advisable to calculate more accurate population-based estimates of predictive power on a two-stage larger scale study.

The poor performance of the model to predict RLS indicates that asymmetry might not be an appropriate screening tool for RLS but there was evidence of association between the RLS-type features with the presence of symmetric motor features, which was unexpected and of interest. The association between RLS and PD is a matter of controversy,2 ,27 and the overall frequency of RLS in our sample was 11.3%, which is similar to figures of previous studies of the same mean age group in the general population.28 However, in the PD group with symmetric features, the frequency of this RLS was twice as frequent (21%) as in the group with asymmetry (10.4%). The higher frequency of RLS in the group with symmetric motor features cannot be attributed to differences in dopaminergic treatment, gender or age, as these factors were accounted for in the logistic regression. In agreement with our data, a previous study failed to show association between presence of RLS and severity or duration of PD, age, gender or type of treatment used.2 In our study, the diagnosis of RLS was defined using a questionnaire. This represents a weakness because this diagnostic tool may include up to 16% of cases of ‘RLS mimics’, which include cramps, positional discomfort and local leg pathology.29 We have no reason to believe, however, that our cases of symmetric PD are more prone to any of these conditions. It is possible that this association can be explained by a common anatomical substrate which increases the risk of RLS symptoms and of symmetric motor features. Discreet anatomical changes in the substantia nigra have been found in subjects with RLS30 ,31 and more symmetric motor subtypes of PD show higher neuronal losses in medial and lateral parts of substantia nigra with more severe gliosis, extraneuronal melanin deposits and neuroaxonal dystrophy.32 Another possibility would be that patients with symmetric parkinsonism are more likely to have atypical parkinsonism which has been misdiagnosed as PD. To our knowledge only one report screened for RLS in various forms of parkinsonism.33 This study in fact showed a higher prevalence of RLS in progressive supranuclear palsy (PSP) when compared with PD, but the diagnosis of RLS and PSP was done clinically, and the study only included a very small number of patients (16 with PD and 14 with PSP), therefore, further studies are needed to clarify this possible link. Finally, another caveat is the fact that, although we have used an adaptation of the method proposed by Lee et al34 to distinguish between symmetric and asymmetric parkinsonism, this differentiation is still controversial and limits the conclusions on the RLS finding.

Strengths of our study include the large number of patients and the evaluation by a single researcher. A main limitation is the lack of polysomnography to confirm RLS or RBD, although the fact that our findings are in line with previous works in the literature is reassuring. Another limitation which is intrinsic to the observational design of this study is that local practice influences the results significantly, and it is possible that the lack of association between dopaminergic medicine and psychosis for instance is due to the fact that those medications are decreased in this particular study centre when psychosis is detected. Reproduction of similar analysis in other cohorts would be interesting to validate our findings. Accumulation of similar studies may also help provide practical guidelines as to which patients should be screened for these and other neuropsychiatric and sleep related features, so they can be properly identified and managed.


We thank the patients and supporters of the Associação Paranaense dos Portadores de Parkinsonismo. Dr Silveira-Moriyama received financial support from a Reta Lila Weston Fellowship and a visiting professorship from University of Campinas, UNICAMP.


View Abstract
  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:


  • Correction notice This paper has been amended since it was published Online First. There is an error in the Results section of the abstract. The first sentence of the results has now been changed, previously it was written “Age and Hoehn&Yahr (H&Y) stage were strongly associated with the presence of dementia (page=0.03, pHY=4×10−14 in the final selected model)”. It now reads: “Age and Hoehn&Yahr (H&Y) stage were strongly associated with the presence of dementia (p value<0.001 for both factors in the final selected model).”

  • Contributors RPM: drafting and revising the manuscript; study concept and design; analysis or interpretation of data. HAT: revising the manuscript; analysis or interpretation of data. HE: drafting the manuscript analysis or interpretation of data. LJC: revising the manuscript. AJL: drafting and revising the manuscript; study concept; analysis or interpretation of data. LS-M: drafting and revising the manuscript; study concept and design; analysis or interpretation of data.

  • Disclaimer RPM has received personal compensation for educational activities with Boehringer Ingelheim, Novartis and Roche. HAT has received personal compensation for educational activities with Allergan, Boehringer Ingelheim, Ipsen, Novartis and Roche. HE has nothing to disclose. LJC has nothing to disclose. AJL is part of the Advisory Boards of Novartis, Teva, Meda, Boehringer Ingelheim, GSK, Ipsen, Lundbeck, Allergan, Orion, BIAL, Noscira and Roche. He has received honoraria from Novartis, Teva, Meda, Boehringer Ingelheim, GSK, Ipsen, Lundbeck, Allergan, Orion, BIAL, Noscira and Roche, as well as grant support from PSP Association, Weston Trust—The Reta Lila Howard Foundation. LS-M receives support from Reta Lila Weston Trust for Medical Research, University of Campinas (UNICAMP), Parkinson's UK and Parkinson's Foundation. Received honoraria from Teva Lundabeck and travel grants from UCB, Genus and Abbott. Part of this material was presented as an oral platform at the 62nd annual meeting of the American Academy of Neurology.

  • Competing interests None.

  • Ethics approval Federal University of Paraná, Curitiba, Brazil.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request permissions

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.