Distribution-based criteria for change in health-related quality of life in Parkinson's disease

Abstract

Background and objective

To be useful, results from health-related quality of life (HRQoL) measures must be interpretable. The objective of this article is to examine statistical (distributional) approaches to interpretability. The standard error of measurement (SEM) and the standard error of the difference (S_diff) are used in data on individuals with Parkinson's disease to calculate the minimum change scores required to be statistically meaningful for each dimension of an instrument to assess HRQoL in Parkinson's disease, the PDQ-39.

Methods

Data was collected from both a community and a clinic study; in both studies the PDQ-39 was administered at baseline and follow-up.

Results

The patterns of SEMs and S_diffs were similar both across time periods and between samples, for all dimensions except Social Support.

Conclusions

The results suggest that, for example, six points change on a 0–100 transformed scoring of the Mobility dimension may be considered on distributional grounds a minimum meaningful change. The demonstrated consistency across occasions and types of sample of SEMs and S_diff for the majority of the dimensions of the PDQ-39, is evidence of the theoretically claimed advantage of this measure of sample independence, and supports use of this distributional approach to minimum meaningful change.

Introduction

Parkinson's disease is a progressive neurodegenerative disease with wide-ranging impact on individuals. As a result, it is increasingly accepted that conventional clinical scales need to be supplemented by patient-assessed measures of health status and health-related quality of life (HRQoL). Of the range of disease-specific instruments that have been developed to assess these dimensions of outcome, the PDQ-39 is the most widely used instrument with the most extensive supportive evidence of measurement properties such as reliability, validity, and responsiveness [1], [2], [3].

A common problem in the use of health status or (HRQoL) measures in any field of health care is that the scores of instruments are unfamiliar compared to clinical scales, so that it is difficult for users to determine what constitutes a meaningful change in an instrument's score. Methods are required that improve the interpretability of such data [4], [5]. One approach to this problem, termed “distributional.” focuses upon statistical features of the data produced by an instrument [6]. One “distributional” method of assessing change in health status instruments to emerge is the “effect size” that relates data on change produced by an instrument to variance, usually in baseline data of that instrument [7].

A potentially stronger alternative approach to effect size is based on the standard error of measurement (SEM) [8], [9]. The SEM of an instrument reflects its reliability in addition to variance. It is calculated so that an SEM of zero means that an observed score is the true score. In relation to change scores the SEM estimates the extent to which observed change is true change or measurement error. Essentially, any change score above the SEM is considered likely to be a meaningful change in the specific sense of being a statistically probable change. A unique theoretic advantage claimed for the SEM is that it is sample independent, a potential advantage over the effect size.

One potential limitation of the SEM is that it is based on information about scores at a single point in time, whereas, usually, there is access to longitudinal data to assess measurement properties of HRQoL instruments. The standard error of the difference (S_diff) is based on SEM at both time points in a longitudinal study and may, therefore, provide a more accurate estimate of the measurement error of an instrument being used to assess change [10]. The S_diff also is the denominator in the calculation of the Reliable Change Index, which has been advocated as a statistic that determines the magnitude of change score necessary for a measure to be considered statistically reliable [11].

The focus of distributional approaches to change is assessments of whether given change scores are beyond what might occur by chance. The alternative to “distributional” approaches to identifying minimal meaningful changes in an instrument is to use “anchor-based” approaches. This alternative method relates change scores in instruments to other external patient- or clinician-derived evidence of change (hence, “anchor”), for example, changes over time in an instrument observed in respondents considered independently to have changed according to clinical scales or transition questions (e.g., how do you rate your health compared to last assessment: better, same or worse?) [12]. This approach would, therefore, for example, calculate the change scores in an instrument among respondents who separately rate themselves as changed on a transition item and consider such change scores the minimal required to count as meaningful.

This article uses the SEM and the S_diff in relation to longitudinal data on individuals with Parkinson's disease derived from both community and clinic studies to calculate the minimum change scores required to be beyond measurement error for each of the dimensions of the PDQ-39. The minimal change scores for PDQ-39 derived from SEMs and S_diffs were then compared to change scores observed in patients responding to retrospective transition questions based upon an anchor-based approach. The purpose of the article is to determine the extent to which there is convergence across samples and methods in identifying minimal meaningful changes for the PDQ-39.