Kaplan-Meier plot of time to death for clinical subtype. PIGD indicates postural instability gait difficulty.
Kaplan-Meier plot of time to death for cognitive impairment.
Lo RY, Tanner CM, Albers KB, Leimpeter AD, Fross RD, Bernstein AL, McGuire V, Quesenberry CP, Nelson LM, Van Den Eeden SK. Clinical Features in Early Parkinson Disease and Survival. Arch Neurol. 2009;66(11):1353-1358. doi:10.1001/archneurol.2009.221
Copyright 2009 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2009
To examine the association between demographic and clinical features in early Parkinson disease (PD) and length of survival in a multiethnic population.
Clinical features within 2 years of diagnosis were determined for an inception cohort established during 1994-1995. Vital status was determined through December 31, 2005. Predictor variables included age at diagnosis, sex, race/ethnicity, as well as clinical subtype (modified tremor dominant, postural instability gait difficulty), symmetry, cognitive impairment, depression, dysphagia, and hallucinations. Cox proportional hazards regression analysis was used to identify factors associated with shorter survival.
Kaiser Permanente Medical Care Program, northern California.
Five hundred seventy-three men and women with newly diagnosed PD.
Three hundred fifty-two participants in the PD cohort (61.4%) had died in the follow-up period. Older age at diagnosis (hazard ratio [HR], 1.1; 95% confidence interval [CI], 1.09-1.12), modified postural instability gait difficulty subtype (HR, 1.8; 95% CI, 1.3-2.7), symmetry of motor signs (HR, 2.0; 95% CI, 1.1-3.7), mild (HR, 1.7; 95% CI, 1.3-2.2) and severe (HR, 2.7; 95% CI, 1.9-3.9) cognitive impairment, dysphagia (HR, 1.4; 95% CI, 1.1-1.9), and hallucinations (HR, 2.1; 95% CI, 1.3-3.2) were associated with increased all-cause mortality, after adjusting for age, sex, and race/ethnicity. None of the other factors altered mortality risk. In an empirical predictive analysis, most previous significant predictors remained associated with shorter survival.
Both motor and nonmotor features in early PD predict increased mortality risk, particularly postural instability gait difficulty, cognitive impairment, and hallucinations. These predictors may be useful in clinical practice and when designing clinical trials.
Parkinson disease (PD) is the second most prevalent neurodegenerative disorder in North America.1 Despite advances in therapy, the mortality rate in patients with PD remains higher than in the general population.2- 18 Several studies have examined predictors of mortality in PD. These studies found that increased mortality was associated with older age, male sex, dementia, and postural instability and gait difficulty (PIGD).1,3,6- 12,14- 16,19- 23 However, validity and generalizability are limited in these reports because of the use of prevalent cases,1,3,7- 10,14- 16,19- 23 decades-long accrual periods,6,12 and/or small sample sizes.11,21 To overcome limitations of prior studies, we investigated a large incident cohort of patients with newly diagnosed PD. We assessed demographic and clinical features present in the first 2 years after diagnosis for their effect on survival.
Details of the study population have been previously published.24 Briefly, our study database was drawn from the Parkinsonism Epidemiology at Kaiser Study, which identified patients with newly diagnosed PD from January 1, 1994, to December 31, 1995, from the Kaiser Permanente Medical Care Program (KPMCP) of northern California. The demographic and socioeconomic characteristics of the more than 3.0 million persons in the KPMCP population are generally representative of the northern California population.25 The KPMCP provides comprehensive vertically integrated health services. Specialty referral is encouraged.
Potential incident PD cases were identified through comprehensive clinical inpatient and outpatient databases, a computerized pharmacy system, and KPMCP physician referrals. The records of all potential cases were reviewed by a movement disorder specialist (C.M.T.) to ensure subject eligibility and determine diagnostic status. The incidence date was when diagnostic criteria were fulfilled. We found that neurologists see 91% of patients with newly diagnosed PD within 5 months of the primary care provider's first noting symptoms, with the median time being 27 days, assuring that the incidence date is close to the onset of clinical symptoms. All eligible cases met modified Core Assessment Program for Intracerebral Transplantation/Hughes diagnostic criteria at the time of diagnosis.26,27 Exclusion criteria were known causes of parkinsonism and atypical features, such as prominent oculomotor palsy, cerebellar signs, vocal cord paresis, severe orthostatic hypotension, pyramidal signs, amyotrophy, or limb ataxia at the time or within 24 months of diagnosis. Cases with dementia or cognitive impairment were not excluded. Interviewers trained and supervised by a neurologist (C.M.T.) collected additional clinical information, including scores on the Mini-Mental State Examination (MMSE) and an abbreviated Unified Parkinson Disease Rating Scale (UPDRS) (without tests of rigidity or postural instability).
Potential predictors were prespecified, including age at diagnosis, sex, and race/ethnicity. Symmetry of motor signs was defined as bilateral onset or absence of left-right discrepancy on neurological examinations.27 Clinical subtypes were classified as modified tremor dominant, PIGD, or mixed based on the presence of cardinal signs within 2 years of PD diagnosis, but not UPDRS score.28 Tremor-dominant PD is defined as the presence of resting tremor and absence of postural reflex impairment; PIGD is defined as the presence of postural reflex impairment and absence of resting tremor; and the mixed type is all other possible combinations. Cognitive status was based on the MMSE score (normal, 28-30; mildly impaired, 24-27; and severely impaired ≤23)29,30 or interviewer assessment if MMSE was not available. Depression, dysphagia, and hallucinations were determined using responses to individual questions on the UPDRS. Patients were classified as depressed if the UPDRS I question 3 score was 2 or greater; as dysphagic if the UPDRS II question 3 score was 1 or greater; and as having hallucinations if the UPDRS I question 2 score was 2 or greater.
Vital status follow-up began at the date of diagnosis and continued until death or December 31, 2005. Deaths were determined using multiple sources, including reports by family or friends, KPMCP medical records or clinical databases, linkage to the California death records, and search of social security records for deaths outside of California.
Data extraction and statistical analyses were performed using SAS, version 9.1 (SAS Institute Inc, Cary, North Carolina). The crude death rate was determined for the entire incident cohort. Cumulative survival was calculated by the Kaplan-Meier method. All baseline covariates were entered into the models as categorical indicator variables, except for age at diagnosis, which was continuous. Cox proportional hazard modeling with delayed entry (left truncation) was used to estimate hazard ratios (HRs) for mortality risk up to 11 years after diagnosis and associated 95% confidence intervals (CIs). This adjustment, known as delayed entry or left truncation, was necessary, because cases did not come under observation for mortality until the date of the interview, at which point predictors were measured. All models were adjusted for age at diagnosis, sex, and race/ethnicity. Other variables, including education, were examined but did not change our effect estimate and thus were not included in the final models. When examining hallucinations, we additionally adjusted for informant type and MMSE score. We also built a model based on the combination of stepwise regression, Akaike information criteria, and the best subset selection (referred to as the empirical predictive multivariate model).31 The sequence of models starting with a null model and ending with a full model that included all explanatory variables was determined by a stepwise selection method. The best model was then selected based on minimum Akaike information criteria. All tests of statistical significance were conducted at 2-tailed α = .05.
We identified a total of 588 incident PD cases from January 1, 1994, to December 31, 1995. Fifteen individuals subsequently presented with atypical parkinsonism within 2 years of diagnosis and were not included in this analysis. Of the remaining 573 individuals with typical PD, 352 (61%) were deceased as of December 31, 2005.
Most patients with PD were free from cognitive impairment, depression, dysphagia, or having hallucinations at baseline (Table 1). Non-Hispanic white individuals represented nearly 81% of the population.
Overall, covariates predictive of higher mortality were older age, PIGD, symmetry of motor signs, cognitive impairment, dysphagia, and hallucinations at or around the time of PD diagnosis (Table 2). Older age at diagnosis was the only demographic factor that showed an increased risk of PD mortality. The PIGD clinical subtype was associated with shorter survival (Figure 1). The PIGD subtype had a nearly 2-fold increased risk of death, but the tremor-dominant subtype had no elevated risk of death, when compared with the mixed subtype (Table 2). The severity of cognitive impairment correlated with the HR estimate (Figure 2). Severe cognitive impairment or dementia had the strongest impact on survival among all predictors in our study. Sex, race/ethnicity, and depression, however, did not appear to predict 10-year survival.
Sex, Hispanic ethnicity, a tremor-dominant subtype, dysphagia, and depression were not selected into an empirical predictive model based on a stepwise sequence procedure and minimum Akaike information criteria. Symmetry of motor signs was no longer a significant predictor in the extended model. Hallucinations became the strongest predictor of mortality in this analysis, followed by PIGD and cognitive impairment (Table 3).
Our study is unique in several ways. Unlike cases selected for clinical trials or recruited from specialty clinics, all of our PD subjects were identified from a health maintenance organization typical of the general population in northern California. We systematically collected information about demographic and clinical features in early PD, using accepted diagnostic criteria and standardized assessment tools. Such information is not often available in population-based studies. Our incident cohort includes patients who had shorter survival and would be more likely to be missed in a prevalence survey. In addition, all participants were identified within the same 2-year period, ensuring the background homogeneity of health-seeking behavior. In this study, we found that several motor and nonmotor features early in the course of disease predicted higher 10-year mortality risk in PD.
Cognitive impairment within 2 years of PD diagnosis was the strongest predictor; there was approximately a 2-fold increase in mortality risk. Although we only used MMSE score to define cognitive impairment and did not have behavioral information, poor performance on this screening test was associated with shorter survival, as has been reported in studies of dementia (HR, 1.9-2.5).11,15,22 Furthermore, we observed that the more severe the cognitive impairment, the higher the mortality risk (mild: HR, 1.7; severe: HR, 2.7). Similarly, in the Deprenyl and Tocopherol Antioxidative Therapy of Parkinsonism (DATATOP) study, a clinical trial of early PD, better MMSE scores predicted longer survival, though this population was limited to those with good cognition (MMSE score > 22) at enrollment.23 Current concepts of dementia with Lewy bodies and dementia associated with PD were not well formulated when we identified this inception cohort.32,33 Since cognitive impairment or dementia at baseline was not in our exclusion criteria, we more than likely included some patients with dementia with Lewy bodies or dementia associated with PD in this incident study. However, there was no way that we could retrospectively apply current criteria of dementia with Lewy bodies to differentiate it from PD. Survival time in our study might have been shorter had we included more cases with dementia with Lewy bodies. Conversely, had we conducted detailed neuropsychological testing, we might have detected cognitive impairment in additional cases.
Hallucinations early in PD were also strongly associated with a greater mortality risk. Dementia is a possible confounder, and in fact, cognitive impairment was common in our subjects with hallucinations (23 of 35 [66%]). However, hallucinations remained a significant predictor after adjusting for MMSE score, with a 2-fold increase in mortality risk. Our result strengthens and extends a prior report that patients with PD who have hallucinations are at greater risk of nursing home placement, which, in turn, predicts high mortality.34 Most prescription medications for PD can induce hallucinations. In this cohort, use of dopamine agonists was similar in those with or without hallucinations (data not shown). However, detailed information about duration or dosage of all anti-parkinsonian agents that were taken was not available. It is possible that PD patients with more disabling progression early in the course of the disease take more medications, leading to drug-induced hallucinations; these patients are likely to have higher mortality by nature of the disease.
As in the DATATOP cohort, symmetry of motor signs was associated with shorter survival. Symmetric motor signs have been associated with atypical parkinsonism in postmortem studies.23 Although we excluded all subjects with atypical features of parkinsonism in the first 2 years after diagnosis of PD, some patients might evolve into atypical parkinsonism later in the course of illness, which might have affected our findings. Further investigations of symmetric motor signs will be useful in elucidating this relationship.
Patients with the PIGD clinical subtype early in the course of illness had 80% higher mortality 10 years after diagnosis. This concurs with the findings of Marras et al23 in the DATATOP clinical trial population and of Rajput et al21 in their specialty clinic patients, confirming the prognostic significance of PIGD. Interestingly, because we did not have UPDRS data on all subjects, we defined the PIGD subtype according to the cardinal features of parkinsonism, rather than the previously reported formula derived from the UPDRS.28 However, we confirmed the results of others who used the more detailed definition. The simple definition we used for the PIGD subtype may be particularly useful in practice, when an entire UPDRS rating may not be possible. The PIGD subtype is among the most commonly observed predictors of poor prognosis, such as rapid progression and early death. A plausible mechanism underlying prognosis for PD motor subtypes came from a recent clinical-biochemical correlation study. Rajput et al35 found that dopamine loss was greater in akinetic/rigid or nontremor-dominant PD cases, and the ventral globus pallidum was particularly affected. This finding suggests that the PIGD subtype may be a clinical syndrome associated with more severe underlying neuronal injury.
This is the first report of dysphagia and greater mortality risk in PD. Dysphagia can give rise to aspiration pneumonia and thus increase mortality in PD. Several studies have shown that pneumonia is a major cause of death in PD,4,8,9,12,13,36 and aspiration pneumonia in particular has the highest mortality risk ratio among all comorbidities.37 We further assessed cause of death in 275 cases with death certificate data. Patients with baseline dysphagia were more likely to die of pneumonia than those with normal swallowing (13% vs 8%), supporting the link between dysphagia and subsequent pneumonia as a cause of death in PD. Swallowing difficulty is common in PD, but specific interventions for dysphagia may not be considered early in the course of the disease.38,39 Dysphagia is one of the few features identified that can be treated, at least in part. Our findings suggest that the early introduction of swallowing training may reduce mortality risk.
Depression was not associated with greater risk of mortality. In an earlier case-control study, depression did predict mortality risk.15 Because we did not use a clinical rating scale for depression but only the UPDRS self-report question, we may have misclassified many cases, explaining differences with other results. We did notice a trend toward shorter survival in those with depression on the UPDRS, suggesting a possible association. Future studies using validated depression rating scales at the time of PD diagnosis will be needed to further clarify the relationship between early depression and short survival.
In our PD cohort, mortality risk increased 10% for each 1-year increment in age at diagnosis (HR, 2.6, each 10-year increment). This result is consistent with the report by Parashos et al,11 who found a 2.8 HR for each 10-year increment. Other studies have also shown older absolute age or age at onset to predict shorter survival in PD cases; however, their comparison groups were controls without PD.3,6,8,9,12,15,17 From a practical perspective, age at diagnosis remains an important predictor for survival within a PD cohort, since the exact age at onset or disease duration is usually not available for most patients with PD in clinic populations.
Sex was not a significant determinant of PD survival. Although earlier studies examining sex-specific PD survival showed excess death rates for women,1,5,20 the difference was not apparent in larger studies.8,23 In unselected community-based cohorts, sex has not predicted survival.4,9,12 Women live longer than men in the general population, but this phenomenon does not seem to extend to patients with PD.
We found no difference in mortality risk of PD among white, Hispanic, Asian, or black individuals. Mayeux et al40 reported a greater number of deaths among black individuals with incident PD in northern Manhattan. However, their community PD registry was established by using information from various medical and nonmedical sources, where disease recognition and access to health care might be different for black individuals. In contrast, all of our subjects were under uniform medical insurance, minimizing this possible confounding effect. However, our investigation in this multiethnic population was limited by the predominance of non-Hispanic white individuals (80.6%). Replication in a population with larger subgroups is needed to determine if there is a difference in mortality by race/ethnicity.
Each clinical predictor was analyzed in a separate multivariate model, together with age, sex, and race/ethnicity. If PIGD is highly correlated with cognitive impairment, for instance, then this feature may not be a significant predictor after adjustment for cognitive status. We thus built an empirical predictive multivariate model to arrive at the most parsimonious set of clinical predictors. Most previous significant predictors remain associated with higher mortality, supporting the notion that these features in early PD are of independent prognostic value. The extended multivariate model was generated based on stepwise regression statistics rather than any known clinical-pathological mechanism; therefore, we should be conservative in interpreting the results. Although dysphagia and symmetry of motor signs were no longer significant predictors in the new model, rather than disregarding their predictive value, consideration of their correlation with other features will be important in future investigations.
Altman41 addressed the concern that variation in subsequent treatment might alter clinical course in PD, yet most published prognostic studies have not taken this into consideration. We did not investigate individual treatment plans in this population. However, because all subjects were identified within the same 2-year window, the treatment strategy for this incident cohort likely reflected standard therapeutic approaches, with little temporal variability. Moreover, all members of KPMCP have unrestricted access to neurology specialty care, and 529 of 573 of our cases (93%) had a PD diagnosis by at least 1 neurologist. This relative uniformity of care is likely to minimize variation in treatment. Importantly, no medication or surgical modality has yet been shown to prolong survival in PD.
In this multiethnic incident PD cohort, we conclude that several motor and nonmotor features in early PD can predict higher mortality risk, particularly older age at diagnosis, PIGD, cognitive impairment, and hallucinations. Our findings are in keeping with those of others, suggesting that these results are robust and generalizable. With effective clinical predictors, we can improve understanding of the disease process, refine risk stratification in designing clinical trials, and guide decision making in clinical practice.42 Finally, future trials directed specifically at treating these features may result in better survival for those affected.
Correspondence: Caroline M. Tanner, MD, PhD, The Parkinson's Institute and Clinical Center, 675 Almanor Ave, Sunnyvale, CA 94085 (email@example.com).
Accepted for Publication: May 23, 2009.
Author Contributions:Study concept and design: Tanner, McGuire, Nelson, and Van Den Eeden. Acquisition of data: Tanner, Leimpeter, Fross, Bernstein, McGuire, Nelson, and Van Den Eeden. Analysis and interpretation of data: Lo, Tanner, Albers, McGuire, Quesenberry, Nelson, and Van Den Eeden. Drafting of the manuscript: Lo, Tanner, and Van Den Eeden. Critical revision of the manuscript for important intellectual content: Lo, Tanner, Albers, Leimpeter, Fross, Bernstein, McGuire, Quesenberry, Nelson, and Van Den Eeden. Statistical analysis: Albers, Quesenberry, Nelson, and Van Den Eeden. Obtained funding: Tanner, Nelson, and Van Den Eeden. Administrative, technical, and material support: Albers, Leimpeter, Fross, McGuire, Nelson, and Van Den Eeden. Study supervision: Tanner, Nelson, and Van Den Eeden.
Financial Disclosure: None reported.
Funding/Support: This work was supported by the Michael J. Fox Foundation; grant R01-31964 from the National Institutes of Health; James and Sharron Clark; and the Research Program in Genes, Environment and Health, Kaiser Permanente.
Additional Contributions: We thank the patients who participated in this study.