Measuring Disease Progression in Early Parkinson Disease: the National Institutes of Health Exploratory Trials in Parkinson Disease (NET-PD) Experience

Abstract

Importance

Optimizing assessments of rate of progression in Parkinson Disease (PD) is important in designing clinical trials, especially of potential disease-modifying agents.

Objective

To examine the value of measures of impairment, disability, and quality of life in assessing progression in early Parkinson disease.

Design, Setting, and Participants

Inception cohort analysis of data from 413 early, untreated PD patients, who were enrolled in two multicenter, randomized, double-blind clinical trials.

Intervention

Participants were randomized into five treatments: 67 received creatine, 66 minocycline, 71 Coenzyme Q10, 71 GPI-1485, and 138 placebo. We assessed the association between the rates of change in measures of impairment, disability, and quality of life and time to initiation of symptomatic treatment.

Main Outcome Measure

Time between baseline assessment and need for the initiation of symptomatic pharmaceutical treatment for PD was the primary indicator of disease progression.

Results

After adjusting for baseline confounding variables Unified Parkinson Disease Rating Scale (UPDRS) II, UPDRS III, modified Rankin score (mRS), level of education, and treatment group, the rate of change of the following measurements was assessed: UPDRS II, UPDRS III, Schwab and England ADL (S&E), Total Functional Capacity (TFC), Parkinson’s Disease Quality of Life Questionnaire – 39 (PDQ39) ADL and Summary Index (SI), Short Form -12v2 Health Survey (SF12) Physical Summary (PS), and SF12 Mental Summary (MS). Variables reaching statistical threshold in univariate analysis were entered into a multivariable Cox proportional hazards model using time to symptomatic treatment as the dependent variable. More rapid worsening of UPDRS II (HR 1.15, 95% C.I. 1.08 – 1.22 for 1 scale unit change per 6 months), UPDRS III (HR 1.09; 95% C.I. 1.06 – 1.13 for 1 scale unit change per 6 months), and S&E (HR 1.29 95% C.I. 1.12 – 1.48 for 5 percentage point change per 6 months), was associated with earlier need for symptomatic therapy.

Conclusions and Relevance

In early PD, UPDRS II and III, and S&E can be used to measure disease progression, while the PDQ39 ADL, PDQ39 SI, TFC, SF12 PH, and SF12 MH are not sensitive to change.

Trial Registration

clinicaltrials.gov identifiers NCT00063193 and NCT00076492

Parkinson disease (PD) is a progressive neurodegenerative disease, which leads to significant morbidity, disability, and institutionalization.^1,2 Measuring disease progression in PD is challenging, and a variety of instruments have been employed for this purpose, including measures of impairment, disability, and quality of life. Yet, a consistently reliable, easy to use measure of disease progression remains elusive.³ As part of the National Institutes of Health Exploratory Trials in Parkinson Disease (NET-PD), two futility drug studies (FS-1 and FS-TOO - clinicaltrials.gov identifiers NCT00063193 and NCT00076492)^4,5 explored the potential of four compounds as disease-modifying agents in early, untreated PD. The primary outcome measure of these studies was the change in the total Unified Parkinson’s Disease Rating Scale (UPDRS) score between baseline and the time at which there was sufficient disability to warrant symptomatic treatment. Participants in these two studies were assessed with a number of impairment, disability, and quality of life measures at baseline, at predetermined intervals throughout the study, and at end-point. In a previously published analysis of the FS-1 and FS-TOO study populations we determined that, at baseline, the UPDRS III, UPDRS II, and the modified Rankin Scale (mRS), were independently associated with earlier need for symptomatic treatment.⁶ In the present paper, we examine the predictive value of the observed rate of change in the various assessment instruments used in FS-1 and FS-TOO, with regard to the time to symptomatic treatment. Our hypothesis was that while some instruments may be of greater value as baseline predictors, different instruments might prove more useful in monitoring disease progression in early PD.

METHODS

PARTICIPANTS

Data from the FS-1⁴ and FS-TOO⁵ studies was used for this analysis. Participants were 413 men and women 30 years of age and older, who had a diagnosis of PD for five years or less, and who, at the time of study entry, did not require symptomatic treatment for PD. Two of the three classic clinical signs of PD (tremor, rigidity, or bradykinesia) were required for the diagnosis, and had to be asymmetric. Other potential causes of parkinsonism, or prior brain surgery for PD were exclusion criteria. Women of childbearing potential had a negative pregnancy test at baseline and were required to use adequate birth control for the duration of the study. All participants gave written informed consent. The protocols and consents for the drug trials were approved by a National Institute of Neurological Disorders and Stroke Oversight Board, an independent Data Safety Monitoring Board and the institutional review boards of the participating sites. Both studies were designed as multicenter, randomized, double-blind trials powered to assess the futility of candidate drugs as disease-modifying agents in early, untreated PD.

BASELINE AND FOLLOW UP ASSESSMENTS

At the baseline visit, collected data included gender, age, level of education, race/ethnicity, and disease duration (from diagnosis). The UPDRS III was administered as a measure of impairment.⁷ The following scales were administered as measures of disability: UPDRS II;⁷ mRS;⁸ Schwab and England ADL Scale (S&E);⁹ Total Functional Capacity Scale (TFC);¹⁰ and, Parkinson’s Disease Questionnaire (PDQ) -39 ADL subscale.¹¹ PDQ39 summary index (SI), and the Medical Outcomes Study Short Form SF-12v2 Health Survey (SF12 – © 2000 QualityMetric Inc)¹² – Physical Summary (PS) and Mental Summary (MS) – were administered as indicators of quality of life. UPDRS II and III were obtained every 3 months during the study duration and at end-point. Besides at baseline, the mRS, S&E, TFC, PDQ39, SF12 were also administered at end-point, or at 12 months.

TREATMENTS

In each of the futility studies, participants were randomized in a 1:1:1 ratio to receive one of the 2 active experimental drug treatment arms or placebo. Of the 200 FS-1 participants 67 received creatine, 66 minocycline, and 67 placebo, while 71 of 213 FS-TOO participants received CoQ10, 71 received GPI-1485, and 71 placebo.

END-POINT DETERMINATION

The primary end-point for the analyses in this study was time to initiation of symptomatic treatment. Symptomatic treatments included levodopa, dopamine agonists, amantadine, anticholinergics, and selegiline. Cases were censored at twelve months of follow up, if they had not reached end-point by that time. Site investigators used their clinical judgment to determine when participants had reached a level of dysfunction sufficient to require symptomatic therapy in any one of three areas: ambulation, activities of daily living, or occupational status.^5,6,13

STATISTICAL DESIGN AND ANALYSES

We conducted the analysis using the two-stage model.¹⁴ First a linear model was fit to each individual using each measurement as the response variable, and time in 10 days increment as the sole covariate. A subject-specific rate of change for each measurement was obtained for each individual. The subject-specific slopes indicated the rates of change of each covariate per 6 months. Because mRS is a discrete ordinal scale a mRS “rate of change” was not considered for further analysis (see also eMethods), although the baseline mRS was included as a confounding variable. To reduce the number of covariates to be considered in the model, Cox proportional hazard models were used to test the association of each variable’s rate of change separately with the primary end-point according to the Hosmer and Lemeshow model development strategy:¹⁵ any variable associated with the time to symptomatic treatment at an α level of 0.15 was selected for inclusion in the second stage. Univariate models were subsequently adjusted for the four confounding baseline covariates previously identified (UPDRS II, UPDRS III, mRS, and level of education),⁶ and for treatment group assignment. Spearman rank correlations among all pairwise combinations of the selected variables were computed to assess potential co-linearity prior to the second stage analysis.

In the second stage, a multivariable Cox proportional hazard model with time to symptomatic therapy as the dependent variable was constructed using the selected variables, the four previously mentioned confounding baseline covariates, and the treatment group assignment. Backward variable selection was used to remove the variables meeting the exclusion criteria of α greater than 0.05. Scaled Shoenfeld residuals were computed to assess the proportional hazard assumption.¹⁶ The analysis was completed using R, version 2.15.3 (R Foundation for Statistical Computing, Vienna, Austria).

There were 47 participants (25 in the symptomatic treatment group and 22 in the group not reaching the end-point) whose baseline values on some of the variables were missing (eTable 1). After establishing that the subjects with missing baseline values did not differ significantly at baseline in terms of the variables of interest from the rest of the participants, we applied multiple imputation techniques using mi package in R while adjusting for each variable’s baseline value, the four previously mentioned confounding baseline covariates, and treatment assignment, to impute the missing values. A total of 62 subjects had missing values that needed to be imputed. A total of 13 participants (3.1%) were excluded from the multivariable analysis because the amount of missing data precluded analysis.

In assessing rate of change in PD, worsening corresponds to increases in UPDRS, PDQ-39, and subscales, but to decreases in S&E, TFC, and SF12 and subscales. Because these measures were expected to worsen, if anything, for most participants, we reversed the signs of the rates of changes for S&E, TFC, and SF12 and subscales, so that all slopes would represent rates of worsening, for ease of interpretation. In addition, the measurements of S&E were percentages ranging from 0 to 100 and incremented by 5, therefore we divided the S&E values by 5 before computing the rate of worsening.

RESULTS

BASELINE DEMOGRAPHICS AND DISEASE CHARACTERISTICS

Table 1 summarizes the baseline demographics of the patient population along with baseline impairment, disability, and quality of life measures. Two hundred out of a total sample of 413 participants (48.5%) started symptomatic treatment within 12 months from baseline.

Table 1.

Study Sample Baseline Characteristics (N=413)

Variable	Mean	SD
Age (Yrs)*	61.7	10.4
Education (Yrs)	15.2	3.2
Duration of PD Diagnosis (Yrs)	0.66	0.84
Male (%)	64.2
Non-Latino Caucasian (%)	91.0
UPDRS II	5.99	3.26
UPDRS III	15.89	6.74
SF12 PS	50.02	7.63
SF12 MS	51.40	9.09
Participants with mRS > 1 (%)	12.0
S&E	93.05	5.31
TFC	12.37	1.27
PDQ39-ADL	15.94	15.22
PDQ39-SI	15.78	11.73

Table 2 indicates that among the participants who needed symptomatic treatment, those with missing values had similar symptomatic treatment free survival time with those without missing values, and there was no difference in baseline parameters between participants with missing values vs. those without missing values within the two subgroups. The rates of worsening of S&E, TFC, PDQ39-ADL, PDQ39-SI, SF12 PS, and SF12 MS were computed for the participants without missing values while they were imputed for those with missing values.

Table 2.

Average baseline variables classified by missing values and need for symptomatic treatment (ST); SDs omitted for clarity

Need for ST	yes			no
Missing values	no	yes		no	yes
Variables	N=175	N=25	P	N=191	N=22	p
UPDRS II	6.93	6.88	0.94	5.23	4.18	0.12
UPDRS III	17.53	18.40	0.54	14.45	12.55	0.17
Education	15.63	15.80	0.78	14.70	15.27	0.44
mRS (% of >1)	17.7%	12.0%	0.67	6.3%	9.1%	0.96
ST free survival (days)	224.70	195.16	0.14	N/A	N/A

The rates of worsening of all variables met criteria for inclusion in the multivariate Cox model based on univariate models both unadjusted and adjusted for baseline confounding variables (eTable 2). The correlation between the rates of worsening of the different variables was not strong (the maximum correlation coefficient being 0.68). The final selected model retained rate of worsening in the following variables after adjusting for treatment assignment and the four confounding baseline variables: UPDRS II, UPDRS III, and S&E (Table 3). As seen in the final model, the effect of treatment assignment was not statistically significant. Proportional hazard assumption was satisfied for the final model. Rates of worsening of the UPDRDS II, UPDRS III, and S&E were associated with earlier need for symptomatic therapy. Rates of worsening of TFC, PDQ39-SI, PDQ39-ADL, SF12 PS, and SF12 MS did not retain statistically significant association with end-point in the multivariable model.

Table 3.

Final Cox Regression Model (n=400, 13 subjects are deleted due to missing values at baseline). Hazard ratios for rates of worsening were calculated for increases of 1 unit per 6 month interval in the rate of worsening of the corresponding measure. For the S&E scale 5 percentage points represented 1 unit.

Parameter	HR (95% CI)	P
Rate of worsening of UPDRS II	1.15 (1.08, 1.22)	<0.001
Rate of worsening of UPDRS III	1.09 (1.06, 1.13)	<0.001
Rate of worsening of S&E	1.29 (1.12, 1.48)	<0.001
Baseline UPDRS II	1.06 (1.01, 1.12)	0.03
Baseline UPDRS III	1.02 (0.99, 1.04)	0.17
Baseline mRS	2.22 (1.43, 3.42)	<0.001
Education (4 year increment)	1.32 (1.08, 1.60)	0.006
Creatine treatment	0.78 (0.46, 1.32)	0.35
Minocycline treatment	0.69 (0.41, 1.16)	0.16
GPI1485 treatment	0.88 (0.53, 1.47)	0.64
CoQ10 treatment	0.80 (0.48, 1.33)	0.39
Placebo	0.84 (0.51, 1.40)	0.51

DISCUSSION

This analysis complements our previous study of the value of measures of impairment, disability, and quality of life as indicators of disease progression in early Parkinson disease.⁶ We concluded that the rate of worsening of UPDRS II, UPDRS III, and S&E predicted earlier need for symptomatic therapy. Interestingly, when the rate of worsening of these variables was included in the model, the baseline UPDRS III was no longer statistically significant, while the baseline UPDRS II, and baseline mRS retained their previously reported associations. The combined findings of the two analyses are summarized in Table 4.

Table 4.

Predictive value of measures of impairment, disability, and quality of life in assessing early PD progression in the final multivariable model.

	Baseline			Rate of progression
	Impairment	Disability	Quality of Life	Impairment	Disability**	Quality of Life
Retained predictive value	UPDRS III *	UPDRS II mRS	_	UPDRS III	UPDRS II S&E	_
Did not retain predictive value	_	TFC PDQ39 ADL S&E+	PDQ39 SI SF12 PS SF12 MS	_	TFC PDQ39 ADL	PDQ39 SI SF12 PS SF12 MS

^*baseline UPDRS III score lost its significance when rate of progression variables were included in the model.

^**a “rate of change” could not be calculated for mRS, and, therefore, its sensitivity to rate of progression could not be assessed in this analysis.

Motor impairment and disability are the most consistent baseline predictors of an earlier need for symptomatic pharmacotherapy. Other important predictors include higher level of education,⁶ full time employment, lesser smoking history, and left-sided onset.^{17, 18}

A number of approaches have been employed in identifying the best suited measures of PD progression for use in clinical trials.³ The task is even more daunting when assessing patients in the earliest stages of the disease, when symptoms are generally mild, and impairment and disability can be minimal. Using the time between baseline and the need for symptomatic therapy as an end-point has been a common characteristic of many studies in early PD, particularly clinical trials of agents with disease modifying potential.^{4, 5, 13, 18} Our analysis now identifies the measures that are most likely to reflect the progression of impairment and disability that leads to the decision to treat.

Predictably, measures associated with physical impairment, such as the UPDRS II and III and the S&E were retained in the final model. These are conceivably the measures most reflective of increasing physical impairment among all the measures considered in this analysis. Mobility impairment and difficulties with mobility dependent ADL are indeed the most important determinants of emerging disability in progressing PD.¹⁹ Table 4 shows that among all the measures assessed in our analyses, UPDRS II was the most informative in assessing both baseline disease activity and subsequent disease progression. Certainly, this strong association between the compromise of ADL (UPDRS II) and the initiation of symptomatic treatment may simply reflect the fact that treating physicians use the decline in ADL as a trigger to discuss and recommend symptomatic treatment, and therefore may simply be the consequence of current standards in practice rather than an indication that this measure is the best way to assess disease progression.

Among the remaining measures of disability examined in this study, mRS demonstrated good predictive value at baseline; because it is an ordinal variable, it would be incorrect to calculate a “rate of change”, and therefore we were not able to assess its sensitivity to change in this particular analysis scheme. It should be recalled that the scale was developed to measure stroke outcomes,⁸ and, as such, its sensitivity to change when applied to a chronic progressive neurologic disease as opposed to a discrete insult to the nervous system, is unknown. Our analysis was unable to determine whether mRS is a reliable measure of early PD progression. TFC and PDQ39-ADL did not stand out either as a baseline predictor or as a measure of progression.

PDQ39-SI, a global PD-specific measure of health-related quality of life, and SF12 did not appear as responsive to disease progression in early disease. In the univariate models these measures of health-related quality of life demonstrated sufficient correlation to disease progression to be included in multivariable modeling, however, the multivariable analysis would suggest that they do not reflect disease progression as robustly as the measures that were retained in the final multivariable model. The implication may be either that changes in quality of life are not as potent a motive to initiate symptomatic treatment in early PD as progressing motor impairment, or that the available instruments are not as sensitive to change in the early disease stage. Additionally, “symptomatic treatment” as defined in the FS-1 and FS-TOO studies, referred to medication treatment of the motor symptoms of the disease, while these measures of global health may also be influenced by non-motor symptoms, and their relationship to an end-point so obviously tied to motor dysfunction may not be as robust.

A limitation of our analysis is that there were a relatively large number of subjects with missing data. We do believe that the algorithm we applied optimized the use of available data by imputing missing data, while minimizing the potential of bias that may have resulted from excluding cases with missing data. For completeness, we repeated the analysis, but restricted only to subjects without missing values (eTable 3), and arrived to essentially the same conclusions, with the only difference being that worsening of SF12 PS was now retained in the final model, while baseline UPDRS II was not. Since our data are derived from a clinical trial population, the applicability of our conclusions to clinical practice may be questioned, as subjects participating in clinical trials may exhibit somewhat different characteristics which may act as a “stealthy” source of bias, especially when it comes to instruments that are more dependent on subjective perceptions and attitudes. It is therefore conceivable, that instruments that may be more sensitive to change in the general population are not so in a clinical trial sample and vice versa. In any case the conclusions of our study should be helpful in designing similar clinical trials in early PD, and may serve as a starting point for evaluating these and newer measures in the general population. Although we forced treatment assignment to remain through the final model in our regression analysis, and found it to be not statistically significant, we cannot fully exclude that one or more of the treatments may have had a differential effect on the responsiveness of the outcome measures, thereby indirectly influencing the final model. We were unable to perform treatment subgroup analyses due to lack of sufficient statistical power. Our results are based on a 12-month long observation period, and while some outcomes were assessed repeatedly others were only assessed twice; it is possible that more frequent assessments over a longer observation span may have yielded different results. Finally, we should point out that our conclusions only apply to the outcome measures studied, and we cannot preclude the possibility that other, as yet not tested measures may prove more sensitive in the future.

In summary, although all measures of impairment, disability, and quality of life used in the FS-1 and FS-TOO studies track well with PD progression, UPDRS II and III seem to be the most reliable and responsive prognostic measures of disease activity at baseline, and of disease progression in early, untreated PD. The baseline mRS and longitudinal assessments with the S&E are similarly useful in predicting disease progression. These conclusions need to take into account the fact that the end-point used in this analysis was one heavily biased towards motor dysfunction.