Abstract
Background:
Data on predictors of decline in PD are largely based on de-novo populations and limited to the use of motor outcomes that fail to capture the full scope of disease.
Objective:
Determine the clinical predictors of decline in early treated PD using a novel multi-domain measure.
Methods:
Data from NINDS Exploratory Trials in PD Long-Term Study 1 (NET-PD LS1), a multicenter Phase 3 study of creatine in early treated PD, were analyzed. Functional decline was defined by a global outcome metric (GO) that consisted of: Schwab and England ADL scale, PD 39-item Questionnaire, Unified PD Rating Scale, Ambulatory Capacity Score, Symbol Digit Modalities Test, and Modified Rankin Scale. Univariate and multivariate models were used to test the association of predictors of interest with a standardized rank-sum of the GO.
Results:
765 of 1741 participants completed five-year assessments and were included. Older age at disease onset (p < 0.0001), higher baseline levodopa equivalent dose (p = 0.01), and worse Scales for Outcomes of Parkinson’s Disease Cognition score (p = 0.001) at baseline were the strongest predictors of functional decline in multivariate analysis. PD symptom subtype was not a significant predictor of outcome (p = 0.42). The full model was only a modest predictor of change in GO (R2 = 0.186).
Conclusions:
This is the largest study to systematically assess predictors of functional decline in early treated PD over several years, and the first to use a multi-domain outcome measure of decline. Older age at disease onset and worse cognition, and not PD subtype, were predictors of decline.
Keywords
INTRODUCTION
Parkinson’s disease (PD) is a progressive neurodegenerative disease characterized by the development of motor and non-motor symptoms that can have variable degrees of impact on activities of daily living, quality of life, and overall disability. Establishing predictors of functional decline at early stages of disease is important for providing patients with information about prognosis to aid in making social or occupational life decisions, but also for stratification of patients for treatment and study recruitment. The National Institute of Neurological Disorders and Stroke (NINDS) 2014 PD research recommendations specifically emphasize the importance of characterizing the clinical features and biomarkers that could be used to predict risk of faster progression as this may have implications for response to targeted therapies. Disease-specific biomarkers are the gold standard for assessing disease progression, but despite significant progress in the field of biomarker research, no validated biomarkers of PD progression exist at this time; therefore, clinical predictors of functional decline are needed.
While a number of studies have assessed clinical predictors of functional decline in PD [1–14], the existing body of data as a whole has limited utility in its application to the clinical and research settings. Most studies have assessed change in functional ability by one-dimensional outcomes such as motor rating scales, thereby potentially limiting the applicability of the findings to that specific domain of PD. While multiple validated tools exist to assess specific domains of impairment in PD (for instance, the Movement Disorders Society Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) includes motor and non-motor functional domains), studies of disease progression and functional decline are limited because there is no well-accepted global measure of disease progression that encompasses the spectrum and impact of motor and non-motor dysfunction, quality of life, and disability. In addition, most studies have been limited to “de novo” subjects – those who are newly diagnosed and unexposed to medications [7, 13]; it is unclear whether data from these studies can be generalized to patients who still have early PD but already receive dopaminergic therapy which constitutes the majority of the clinical population. Additionally, many studies that comment on predictors of decline are restricted to small cohorts and lack extended follow-up with which to make firm conclusions about progression. One exception to this is the CamPaIGN study [15] which assessed progression to postural instability, dementia, and death in a naturalistic setting in a cohort of 142 PD patients over a 10-year period. Two large interventional multicenter clinical trials are also important in this regard–Deprenyl and Tocopherol Antioxidative Therapy of Parkinsonism (DATATOP) and Parkinson Research Examination of CEP-1347 Trial (PRECEPT) [7, 13]. In these studies, a greater severity of parkinsonism, as measured by UPDRS scale at baseline, was the most consistent predictor of disease progression. However, in both studies, progression was defined by the need to initiate dopaminergic therapy. Additional predictors of decline have included older age at disease onset and postural-instability-gait disorder predominant (PIGD) subtype of disease [7], and these two factors were singled out by American Academy of Neurology (AAN) practice parameters in 2007 as the strongest predictors of more rapid rate of progression in PD [16].
We used data from a large cohort of PD patients enrolled in a longitudinal study of creatine (Long-term Study 1; LS-1) conducted by the NINDS Exploratory Trials for Parkinson’s Disease (NET-PD) program to conduct an exploratory analysis of predictors of functional decline in PD, with decline defined by a novel global outcome measure (GO). Our primary aim was to determine what baseline variables of demographics and disease severity are the best predictors of functional decline in PD, as determined by the GO.
METHODS
LS-1 was a multicenter, double-blind, placebo-controlled Phase 3 study of creatine in subjects with early treated PD. All subjects were within 5 years from diagnosis and were required to be on a stable regimen of dopaminergic therapy (levodopa or a dopamine agonist) for at least 90 days but no more than 2 years at the time of enrollment. Investigators were allowed to adjust dopaminergic therapy at any point in the course of the study. Subjects were recruited from 45 sites in the US and Canada. The detailed study design and characteristics of participants enrolled in the trial are published elsewhere [17, 18]. The primary objective of the NET-PD LS1 study was to test the hypothesis that daily administration of creatine (10 gm/day) is more effective than placebo in slowing clinical decline in PD at 5 years. Enrollment occurred from 2007 to 2010, and the study was terminated for futility in September 2013. This analysis was based on the final locked database as of May 5, 2014.
The LS1 study primary outcome measure makes use of the global statistical test (GST) to compare the clinical decline between treatment groups [17]. The global outcome(GO) is the standardized version of the GST [18, 19] which is comprised of 5 measures: 1) The Schwab and England Activities of Daily Living scale (S&E) [20]; 2) Parkinson’s Disease 39-item Questionnaire (PDQ-39) [21]; 3) Unified Parkinson’s Disease Rating Scale (UPDRS) Ambulatory Capacity, the sum of 5 items from the UPDRS (items 13, 14, 15, 29, and 30) [22]; 4) Symbol Digit Modalities (SDM) Test [23]; and 5) Modified Rankin (mRS) [24]. The components of the GO were selected by the study steering committee to reflect the major domains of PD disability, impairment, and quality of life specifically focusing on the domains that have been shown to correlate with longer term disability. Change from baseline to 5 years was measured, except for the mRS, which used the 5 year cross-sectional value.
Study statistical analysis: In order to compute the GO, the measures were reverse coded, as necessary, such that higher values reflected worse functioning. Then, these five measures were ranked with higher ranks representing a worse outcome, and the five ranks were summed for each person resulting in a rank-sum measure. Since there is no standardized definition of “progression” in PD, we defined progression based on the standardized form of the rank-sum labeled as the Global Outcome (GO). For the purpose of this analysis only subjects who had 5 year outcome data were included in the analysis. With 765 subjects having completed 5 year data (the rest being unable to complete primarily due to early termination of the study), the formula for standardizing the rank-sum for each subject as a summed-rank per person-years follow-up was:
The following baseline dependent variables were assessed for correlation with the change in the GO: Race, gender, age at symptom onset, duration of diagnosis at time of study entry, baseline Body Mass Index (BMI), baseline Beck Depression Inventory (BDI) score [25], baseline Total Functional Capacity (TFC) score [26], baseline levodopa equivalent dose (LED) calculated based on the formula reported by Tomlinson et al. [27], family history of PD, predominant PD symptom type (postural-instability gait predominant (PIGD) subtype or other) calculated based on the formula by Jankovic et al. [28] (and treating “indeterminate” ratio as missing), use of MAO-B inhibitors, baseline Scales for Outcomes of Parkinson’s Disease Cognition (SCOPA-COG) score (which includes 10 items sensitive to cognitive deficits in PD, with higher score reflecting better performance) [29], baseline EuroQOL-5D (EQ-5) score [30], and baseline total UPDRS score. The dependent variables were based on the LS1 parameters available in this database and which have been suggested as potentially important variables of disease progression in previous smaller studies.
Linear mixed models were used to test the association of predictors of interest with disease progression, using baseline and 5-year assessment data, both in univariate and multivariate models. For each model, we controlled for treatment by including it as a covariate, and included site as a random effect. We used univariate models to check the effect of variables, and investigated the significant variables altogether in a multivariate model. R2 for mixed models was used to quantify the contribution of each covariate in the multivariate mixed models [31]. SAS Proc Mixed procedure was used for analyses.
To assess whether the GO is influenced primarily by any one of its five component variables, an additional analysis was conducted. Specifically, in order to establish which of the 5 baseline variables that comprise the GO are the best predictors of the GO, the baseline value of each component was correlated with the GO. The partial correlation coefficients, which have a range [–1, 1], with higher absolute values meaning stronger correlations, were used to identify the correlation between one covariate and the GO while controlling for the effect of the other covariates. The change in each of these variables over time was also correlated with the GO, with the exception of the mRS. The change in mRS could not be considered in this study because the baseline assessment was performed in relation to PD disability while the 5 year follow-up assessment was performed in relation to global disability. Therefore, mRS is the only component which is reported as a 5-year cross sectional value rather than a delta value as per the original analysis plan, and this is reflected in the GO calculation.
RESULTS
The study enrolled 1741 patients with PD. A subset of 765 subjects for whom 5 year data was available (and in some cases up to 7 years) was included in the analysis. This included 382 subjects from the creatine treatment group and 383 subjects from the placebo treatment group. Due to the fact that the study was terminated early, out of the 1741 patients enrolled, 732 subjects (365 of those randomized to creatine and 367 to placebo group) could not reach a five year milestone based on their date of randomization and were not included in the analysis (considered administrative drop outs). Of those who were expected to have a 5 year visit, 244 (127 in the creatine group and 117 in placebo group) dropped out before that visit [17]. A comparison of the demographic and baseline disease characteristics between the analyzed cohort and drop-outs revealed significant differences (at the p < 0.05 level) in the following variables, all of which were included in the analysis as covariates: baseline BDI, TFC, SCOPA-COG, EQ-5D, LED, PD symptom type, race, gender, and total UPDRS. This comparison table is provided as Supplement A. Table 1 provides baseline demographic and disease characteristics of the cohort for all discrete and continuous variables analyzed. The mean age of subjects at baseline was 62.5 (SD 9.4) and the duration of PD since diagnosis at time of study entry was 1.7 (SD 1.1) years. 47.3% of subjects were categorized as PIGD subtype.
The estimates and p-values for the univariate analysis for each variable with the GO are presented in Table 2. Positive estimates indicate that covariates are positively correlated with GO; higher values of covariates with positive estimates indicate faster decline. Similarly, higher values of covariates with negative estimates imply slower decline. Variables that were found to have a significant association with the higher GO (i.e. faster decline) at a significance level of p < 0.05 included: non-white race, older age at symptom onset, older age at baseline, higher baseline Beck Depression Inventory score, worse baseline Total Functional Capacity score, requiring more PD medications at baseline (higher LED), worse cognition (lower baseline SCOPA-COG), not taking an MAO-B inhibitor, and worse motor dysfunction (higher baseline UPDRS score). Duration of disease since diagnosis (p = 0.43), PD symptom subtype (p = 0.06), and baseline Body Mass Index (BMI) (p = 0.53) were not statistically significant. Table 3 shows that the variables that remained significant predictors (at 0.05 level) of faster decline in the multivariate model included: older age at symptom onset (p < 0.0001), higher baseline LED (p = 0.01), and worse cognition (lower baseline SCOPA-COG score, p = 0.001). Variables that were no longer significant in the multivariate model were race (p = 0.13), BDI score (p = 0.81), TFC score (p = 0.07), UPDRS score (p = 0.16), EQ5 score (p = 0.06), and use of MAO-B inhibitor (p = 0.07). The full model was only a modest predictor of change in GO (R2 = 0.186), and the relative contribution of each variable to the model is shown in Table 3 as percent of the total R2. Age at symptom onset had the strongest contribution to GO (rank of change in R2 = 1).
Table 4a shows partial correlation coefficients between GO and each baseline component variable of the GO. In this analysis, the following baseline variables were significant: Symbol Digit Modality (p < 0.0001), ambulatory capacity score (p = 0.007), and PDQ-39 score (p = 0.04), while baseline Schwab and England was not (p = 0.66). Since the change in each of the variables contributes to the GO test calculation, a similar analysis was conducted to evaluate the correlation between the GO and its components as they change from baseline to study termination. Table 4b provides partial correlation coefficients and p-values of each covariate. In this analysis all variables showed a weak correlation with GO, although mRS value had the strongest magnitude of partial correlation coefficient (0.419, p < 0.0001), followed by the change in SDMT (–0.390, p < 0.0001), while S&E had the weakest relationship (0.014, p = 0.69).
DISCUSSION
The strongest baseline predictors of five-year decline in a multi-domain global outcome (GO), in this large cohort of patients with early treated PD enrolled in the NET-PD LS1 study, were: older age at disease onset, being on a higher dose of PD medications as measured by LED, and worse cognitive status at baseline as measured by SCOPA-COG. PD symptom subtype was not found to be a significant predictor of decline. To the best of our knowledge this is the largest study that has systematically assessed predictors of functional decline in a relatively early PD population over a period as long as five years. Unlike prior studies looking at predictors of disease progression [1–8, 14], this study used a multi-domain measure of disease progression to better capture relevant PD-related parameters that might predict functional decline. AAN practice parameters published in 2007 concluded that the two strongest predictors of a more rapid rate of progression in PD were older age at onset (defined as age 57–78), and rigidity (as opposed to tremor) as a presenting symptom [16]. We confirmed the importance of age of symptom onset, found cognitive function to be an important predictor of decline, but did not confirm the influence of PD subtype.
PD subtype
Gait instability and falls are often difficult to treat pharmacologically; the PIGD subtype of PD has typically been considered a poor prognostic feature. Indeed, based on the existing evidence AAN practice parameters identified the bradykinetic-rigid subtype of disease to be associated with greater overall decline when compared to the tremor-predominant subtype of PD [16]. Our data, which included a higher number of PIGD categorized subjects than most published studies, did not support this conclusion [32, 33]. The most frequently referenced study associating PIGD subtype with faster progression, and the one which AAN practice parameters are largely based on, followed patients for 2 to 3 years, but presented “projected slopes” of decline going up to 10 years for progression of UPDRS as a function of PD subtype [4]. They found a steeper slope of decline for patients with the PIGD subtype (n = 149) when compared to TD subtype (n = 77). A limitation of that analysis is the “projected” results and the fact that the groups were not risk adjusted for any factors such as disease stage or medications.
Dose of dopaminergic therapy
We expected that a higher burden of motor disability would predict a greater overall decline. Insomuch as higher medication dosage may reflect a higher burden of motor disease we did find this to be the case. However, the significance of LED did persist even after controlling for multiple disease variables that would comprise the higher burden of disease, including UPDRS score. Whether or not this reflects a direct effect of levodopa and other dopaminergic therapy use on the disease course cannot be determined from this analysis. An alternative explanation is that higher LED at baseline may signal more levodopa-resistant symptomatology, although this was not assessed in this analysis.
Cognition
Worse cognition, as measured by SCOPA-COG, was found to be a significant predictor of decline in this cohort. This was true even though the mean baseline SCOPA-COG score for our cohort was fairly high 30.3 (SD 5.4, with a maximum score of 40). Consistent with earlier reports [1, 8], recent studies [15, 35] have presented findings in support of a relationship between cognitive impairment and progression of PD motor scores, and support the notion that cognitive dysfunction in PD may reflect a more advanced pathological disease state. The significance of baseline cognitive status was not unexpected in the current study and adds to the existing body of literature on the relationship between cognitive status and functional decline in PD.
Depression
Depression, as measured by BDI, was not significant in the multivariate analysis. Depression is often seen as an early or even pre-motor disease manifestation in PD [36]. Previous studies reported depression as a major factor driving quality of life impairment [8, 38] and time to initiation of dopaminergic therapy [8]. One reason we may not have seen this association in our study is that the mean BDI score for this cohort was low (6.4, SD 5.5), with usual cutoff for depressive symptoms being >13 [39]. This reflects a low prevalence or successful treatment of depression in this particular cohort of early PD patients.
MAO-B Inhibitors
The impact of MAO-B inhibitor utilization on disease progression is still debated. In our study, while there was a significant correlation between MAO-B use and slower functional decline in the univariate analysis, this was not confirmed in the multivariate model. Our results are in accord with the recently reported data from the long-term open label follow up of the ADAGIO rasagiline study [40].
The strengths of this study are the large size of the cohort, the relatively long follow-up period, the use of an “early” as opposed to strictly “de novo” population, and the use of a multi-domain outcome measure to capture the scope of the disease. Another strength was the use of a composite measure encompassing 5 domains of disability, quality of life, motor and cognitive impairment to assess PD progression. This is in contrast to other studies that have relied on single measures such as the need for dopaminergic therapy [28] to assess disease progression. While the GO measure is not validated, it represents an attempt to look at disease progression globally in order to capture the multi-factorial nature of PD. Ideally, predictors of disease progression should be based on a set of objective biomarkers which would allow not only prediction of the course of the disease but also stratification into specific treatment paradigms. While there is tremendous interest in biomarkers of PD progression, none have been validated so far [13, 42].
This study has a number of limitations. The major limitation of the study is that the GO measure is unvalidated and not routinely utilized in clinical care. While GO is not validated, it is calculated as a rank sum of several measures which have been validated and routinely used in this population and thus should have a higher power than any individual component. In an attempt to find a simple global measure of disability in PD we have previously published an analysis of baseline cross-sectional data from this cohort on mRS. We found that the mRS correlated with motor impairment, non-motor dysfunction, and quality of life measures, suggesting its potential use as a global measure of disability in PD [43]; however, 5-year mRS, while significant, had only modest correlation with motor and non-motor measures (Table 4b, partial correlation coefficient = 0.419) and as such is not promising as a global outcome measure in PD. We demonstrated that as a composite score the GO actually correlated poorly with each of the component variables of GO taken individually, and while requiring validation, it theoretically should be more powerful than any of the validated variables including mRS. The study is also limited to analysis of the variables available in the NET-PD dataset which were extensive but not all-inclusive. Disease progression as assessed by GO may be influenced by factors not considered in our model. The phenotypic expression of PD may be related to variables unaccounted in our dataset such as exercise, spectrum of non-motor symptoms, environmental factors, neuropsychiatric comorbidities, or inherited mutations, which may uniquely impact disease progression and were not included in the dataset. That could explain why our model was only a modest predictor of change in GO (R2 = 0.186). Finally, limitations surrounding the statistical methods include the assumption of a parametric distribution for all variables and outcomes measured and the possibility of overfitting noise in the model due to the number of variables assessed. Due to these limitations the data should be interpreted with some caution.
The analysis was run on the subset of study subjects who completed a 5 year study visit. Of the 1009 subjects expected to complete the 5 year visit, 244 (24.2%) dropped out early and were not included in the analysis. We recognize that the early drop outs could represent subjects with more severe disease and as such the data might not be representative of the full cohort, however, we have run an analysis of the comparison between the baseline characteristics of the completers versus early drop outs and included the variables that were statistically significant as the covariates in the analysis. An alternative way would have been to model the disease progression in the early dropouts which introduces its own bias.
In conclusion, our study demonstrates that older age at disease onset, worse cognitive status, and higher dose of dopaminergic therapy at baseline are the strongest predictors of five-year functional decline as measured by the GO in this cohort of subjects with early treated PD. PD symptom subtype was not a significant predictor of decline. This is the first study to assess progression over five years in the largest cohort of early PD patients reported to date using a global outcome, a novel composite measure of multi-dimensional motor and non-motor disease severity and disability. These findings can guide clinicians in providing prognostication for early PD patients and help with the stratification of decisions for treatment and future studiesrecruitment.
AUTHOR ROLES
Danny Bega, MD was involved in project conception, organization, and execution. He was also involved in statistical design and review. He was involved in writing the manuscript.
Soeun Kim, PhD was involved in project conception, organization, and execution. She was also involved in statistical design and review. She was involved in review and critique of the manuscript.
Yunxi Zhang was involved in project conception, organization, and execution. She was also involved in statistical design and review. She was involved in review and critique of the manuscript.
Jordan Elm, PhD was involved in project conception, organization, and execution. She was also involved in statistical design and review. She was involved in review and critique of the manuscript.
Jay Schneider, PhD was involved in project conception, organization, and execution. He was also involved in statistical design and review. He was involved in review and critique of the manuscript.
Robert Hauser, MD was involved in project conception, organization, and execution. He was also involved in statistical design and review. He was involved in review and critique of the manuscript.
Andy Fraser, MD was involved in project conception, organization, and execution. He was also involved in statistical design and review. He was involved in review and critique of the manuscript.
Tanya Simuni, MD was involved in project conception, organization, and execution. She was also involved in statistical design and review. She was involved in review and critique of the manuscript.
FINANCIAL DISCLOSURES
Danny Bega, MD has no conflicts to disclose.
Soeun Kim, PhD has no conflicts to disclose.
Yunxi Zhang has no conflicts to disclose.
Jordan Elm, PhD has research grant support from NIH and was a consultant for Teva.
Jay Schneider, PhD reports grants from National Institutes of Health, Michael J Fox Foundation, and the Parkinson Council. He reports no conflicts of interest.
Robert Hauser, MD has received consulting or advising fees from Pfizer, UCB Biosciences, Chelsea, Impax Pharmaceuticals, Lundbeck, AstraZeneca, AbbVie, Acadia, Biotie, Eli Lilly, Allergan, Neurocrine, TEVA, Auspex Pharmaceuticals, and Novartis.
Andy Fraser, MD has no conflicts to disclose.
Tanya Simuni, MD has received consulting fees as a speaker and consultant, honorariums and educational grant support from GE Medical. In addition to consulting fees as a speaker and consultant, honorariums and educational grant support, she also received research funding from TEVA. Acadia, Abbvie, Eli Lilly, Harbor, IMPAX, Lundbeck, Merz, Inc., Navidea, Pfizer, and US World Meds have provided consulting fees in her role as a consultant. She has also received funding from Lundbeck as a speaker. She is an advisory board consultant and speaker with IMPAX. In addition to consulting fees, Dr. Simuni received an honorarium from Allergan where she was a presenter at a regional conference and is a consultant. She received both an honorarium and consulting fees from Ibsen in her role as a speaker and consultant. The National Parkinson Foundation has provided Dr. Simuni with research funding for studies and she is a consultant for the foundation. Dr. Simuni is a consultant and speaker for UCB Pharma and they have provided Dr. Simuni with an honorarium and consulting fees. Dr. Simuni has received research funding from TEVA, Auspex, Biotie, and Civitas. She is the site investigator for studies sponsored by these companies. The NIH and the Michael J. Fox Foundation has provided research funding for investigator-initiated research projects. The Michael J. Fox Foundation has provided research funding for the MJFF PPMI study where Dr. Simuni is the investigator.