PAGE. Abstracts of the Annual Meeting of the Population Approach Group in Europe.
PAGE 17 (2008) Abstr 1256 [www.page-meeting.org/?abstract=1256]
Oral Presentation: Lewis Sheiner Student Session
Thuy Vu[1,2], Jay Nutt and Nick Holford
 Department of Pharmacology and Clinical Pharmacology, University of Auckland, New Zealand  Center for Drug Development Sciences, University of California at San Francisco, CA, USA  Department of Neurology, Oregon Health and Science University, Portland, OR, USA
There are two reasons for studying the time course of disease status as a predictor in time to event (T2E) analysis. Firstly, it is well understood that pharmacologic treatments may influence both the time course of disease progress and a clinical event such as death. Secondly, the two outcome variables (i.e., disease status and clinical event) are highly correlated e.g. the probability of a clinical event may be increased by the worsening disease status. Despite these reasons, separate analysis for each type of outcome measurements is usually done and only baseline disease status is often used as a time-constant covariate in T2E analysis. We contend that more useful information can be gained when time course of disease status is modeled as a time-dependent covariate providing some mechanistic insight for the effectiveness of treatment. Furthermore, an integrated model to describe the effect of treatment on the time course of both outcomes would provide a basis for clinicians to make better prognostic predictions of the eventual clinical outcome.
Using Parkinson's disease (PD) as an example, we achieved three specific aims that linked the time course of disease progress to clinical events: (1) described disease progression using time courses of total Unified Parkinson Disease Rating Scale (UPDRS) and its subscales (bradykinesia, tremor, rigidity, postural instability/gait disorder [PIGD], and activities of daily living [ADL]); (2) evaluated the influence of predicted disease status as a continuous time-dependent covariate in T2E models of death, disability, dementia and depression; (3) assessed the prognostic value of early disease status measurements on the probability of clinical events. The working hypothesis was that the UPDRS subscales would progress at different rates and their relative responses to anti-parkinsonian treatments would not be similar. Consequently, the predictive power of these subscales on clinical events would also be different.
Treatment information, patient demographics, disease status assessments, and clinical events (i.e., survival times, and times to first event of disability, dementia, and depression) were obtained from the DATATOP cohort, which enrolled and followed 800 PD patients for 8 years.
(1) Quantitative models for the time course of Disease Status: Models for disease progress and pharmacodynamics were developed to describe the time course and quantify treatment responsiveness for each of the subscales using a nonlinear mixed effects approach as previously described for total UPDRS measurements . A Gompertz model was used to describe the asymptotic natural disease progression. Symptomatic effects were modeled as an offset from the baseline disease status and the disease-modifying effects were included as either a shift in the progression half-time or in disease status asymptote.
(2) T2E models to describe relationship between Disease Status and Clinical Events: Parametric distributions (i.e., exponential, Gompertz, and Weibull) were used to describe the baseline hazard function. To link time course of disease status to clinical events, we included the predicted time course of disease status (i.e., UPDRS and its subscales) in the hazard function. Individual parameter estimates (IPP) were obtained from the disease progress models in (1) to compute the predicted time course of disease status. Age, smoking status, and sex were also evaluated as potential predictors.
(3) Prognostic value of early measurements of Disease Status on the probability of Clinical Events: Clinical assessments of disease status up to 1 year were used to obtain IPP with the disease progress models in (1). Individual probability of clinical events at 5 years was computed using predicted disease status based on 1 year measurements. The probability distributions at 5 years based on 1 year or 8 year series of UPDRS measurements were visually inspected for predictive agreement.
(1) Quantitative models for the time course of Disease Status: The natural disease progression for all subscales was best described by a Gompertz model. Total UPDRS, PIGD, rigidity, bradykinesia and ADL progressed at similar rates (half-time range 2-5 y), whereas tremor progressed at a much slower rate (half-time of 11 y). Responsiveness to levodopa is lower for the PIGD subscale (ED50 ~ 1300 mg/d) but is similar for the other subscales (ED50 range 7-10 mg/d) Levodopa and deprenyl showed a disease modifying effect by decreasing the disease status asymptote for UPDRS subscales and by increasing the half-time to asymptote for total UPDRS.
(2) T2E models to describe relationship between Disease Status and Clinical Events: Except for dementia, time was an important explanatory variable in the base models for clinical events of death, disability and depression. Age had an independent effect on the hazards for death and dementia. Among the subscales, time course of ADL best explained the increased hazard of death in our patient population. Time course of total UPDRS was a better predictor than those of UPDRS subscales for T2E models of dementia, depression and disability. Visual predictive check plots were adequate in describing the goodness of fit.
(3) Prognostic value of early measurements of Disease Status on the probability of Clinical Events: Probability distributions of survival and dementia at 5 years predicted from 1 year measurements were well correlated with the predictions based on the full 8 year follow up (r=0.81 for death; r=0.9 for dementia). The correlation was not as good for disability (r=0.74) and depression (r=0.78).
We have shown that hazards for 4 clinical events in PD are not constant over time. They are clearly influenced by PD progress, as measured by total UPDRS or its subscales. With the integrated models of time course of disease progress and clinical events, the differences in probability of clinical events could be explained by the symptomatic and/or protective effects of anti-parkinsonian medications on PD progression. The use of early disease status measurements may have clinical application in predicting the probability of clinical events and giving patients better individual prognostic advice.