Estimating the minimum important difference in the DEMQOL instrument in people with dementia

The Dementia-Related Quality of Life questionnaire is one way of measuring the quality of life of people diagnosed with dementia. The questionnaire has 29 questions, about topics including looking after yourself, health and well-being and relationships and scores range between 28 and 112. Higher scores indicate better quality of life. Until now, when using this questionnaire for research, we did not know what change or difference in scores a person with dementia would feel was important and which would cause doctors to think about a change in the person’s treatment or care. After using statistical techniques to assess this, we estimate that people with dementia would feel that a questionnaire score change of 5 or 6 points would be helpful and would cause their doctors to think about a change in the person’s treatment or care.

This information will help researchers to design future trials (they can work out how many people must be recruited) and analyse the results (has the activity or drug had a clinically significant effect on people?).

We also looked at the DEMQOL-Utility score (used by health economists to understand if activities or drugs are value for money) which ranges from 0.243 to 0.986 and is calculated using the Dementia-Related Quality of Life questionnaire scores. We found that a 0.02 to 0.05 points increase in the DEMQOL-Utility score would indicate that the person being tested was experiencing a clinically significant increase in quality of life.

Dementia is an umbrella term for a variety of diseases and conditions affecting the brain distinguished by a decline in memory, language, problem-solving and other thinking skills that affect a person's ability to perform everyday activities. There are a number of sub-types of dementia, but the five most common are Alzheimer’s disease, vascular dementia, dementia with Lewy bodies, frontotemporal dementia and mixed dementia. Alzheimer's is the most common cause of dementia [1]. As a person’s age increases, so does the risk of them developing dementia [2].

Worldwide, around 50 million people have dementia, and there are up to 10 million new cases every year [3]. Projections show that there will be 82 million people with dementia in 2030 and 152 million in 2050 [3]. Alzheimer’s Research UK reports that there are over 850,000 people currently living with dementia in the UK [4]. It is estimated that the number of people with dementia in the UK will rise to over one million by 2021 [1].

There is no treatment currently available to cure dementia or to alter its progressive course. Numerous new treatments are being investigated in various stages of clinical trials [3]. However, much can be offered to support and improve the lives of people with dementia and their careers and families and services are being encouraged to provide post-diagnostic treatment and support [5].

Given that we are currently unable to cure people with dementia, maintaining or enhancing quality of life (QOL) is often an important therapeutic goal [6]. Therefore, accurately measuring QOL is required to both guide clinical decision making and evaluate the impact of various interventions in the population of people with dementia. For this purpose, validated instruments have been developed to evaluate QOL in patients with dementia including the DEMQOL and DEMQOL-U. The DEMQOL is a patient-reported outcome measure (PROM) which is designed to enable the assessment of health-related quality of life of people with dementia [7]. The DEMQOL-U consists of items from the DEMQOL and generates a preference-based single index or score for use in economic evaluations [8].

Interpreting the numerical scores on these instruments can be challenging for a number of reasons. The scales and instruments used may be unfamiliar to many patients and clinicians, who may be uncertain of the meaning of the scale values and summary scores [9]. It is relatively simple to determine the statistical significance of a change in QOL but placing the magnitude of these changes in a context that is meaningful for patients, health professionals, and other stakeholders (pharmaceutical and medical device developers, insurance payers, regulators, governments) has not been so easy. Ascertaining the magnitude of change that corresponds to a minimal important difference would help to address this problem.

Jaeschke et al. 1989 defined the minimal important difference, from the patient perspective, as “The smallest difference in score in the domain of interest which patients perceive as beneficial and which would mandate, in the absence of troublesome side effects and excessive cost, a change in the patient’s management” [10].

Thus, individual change standards are needed to provide meaningful interpretation of treatment effects on QOL and to classify patients based on this standard as improved, stable or declined. Anchor-based and distribution-based methods can be used for estimating the MID in QOL instruments [9, 11, 12]. The aim of this study was to estimate the MID for the DEMQOL and DEMQOL-U instruments for comparisons between groups or within groups of people with dementia in a prospective cohort of people with dementia who were recruited as part of an RCT.

A total of 490 participants were recruited to the JtD study and had available baseline DEMQOL data, of these 386 also had 8-month data, and 344 had both baseline and 12-month DEMQOL data. Table 1 summarises the characteristics of the cohort. The median age of the cohort was 78.0 and 58% of the participants were male. The most prevalent diagnosis was Alzheimer’s (60%), the median length of time since dementia was diagnosed was 0.7 years (IQR 0.3–1.8 years). The inclusion criteria for this study were a MMSE score of 18 or more, so all participants in the cohort had either mild cognitive impairment (38%) or normal cognitive function (62%) at baseline according to the MMSE. The characteristics are presented separately for those that had baseline data, baseline and 8-month data, and baseline and 12-month data, the three groups appear to have similar characteristics, with a slightly lower proportion of participants with mild cognitive impairment in the group with 12-month data available. Table 2 shows how the mean DEMQOL and DEMQOL-U scores changed over time for the whole cohort and the sub-sample who completed all three assessments. Overall, there was little change in DEMQOL and DEMQOL-U scores over time.

All correlations between DEMQOL(-U) and the anchors at each timepoint are above the threshold of 0.3 and so the anchors have an appreciable association with the DEMQOL and DEMQOL-U (see Online Resource 2). DEMQOL-U has slightly higher correlation with Q29 than it does with EQ-5D. The correlation between change scores at 8 and 12 months for DEMQOL-U with both EQ-5D and Q29 was below the threshold of 0.3; EQ-5D was particularly low so the longitudinal anchor analysis for DEMQOL-U and EQ-5D is to be treated cautiously.

We have calculated distribution, anchor, and standard error of measurement-based estimates of the MID for the DEMQOL and DEMQOL-U instruments using a large cohort of 490 patients with dementia. We found that the anchored-based estimate of the MID for the DEMQOL is around 5–6 points; and it is 0.02–0.05 points for the DEMQOL-U for comparisons between groups or within groups of people with dementia.

The minimum detectable change (MDC) has been defined as the smallest change in score (at an individual level) that can be detected after allowing for measurement error of the instrument. There are several methods for estimating the MDC, usually involving the standard error of measurement (SEM) calculated from reliability coefficients such as test–retest and Chronbach’s alpha [11]. We used two methods two estimate the SEM (test–retest reliability and internal consistency) and calculated values of 6.4 and 4.7 for one SEM, respectively, from these methods.

Ideally, for a sensitive and reliable instrument, the MDC should be smaller than the MID [11]. The MID and MDC are important but they are different concepts measuring different things. The relationship between the MID and MDC is discussed by Turner et al. 2010 and de Vet (2010) [33, 34]. They agree that the MDC (and the associated SEM) cannot reliability replace the MID. They conclude that the MDC is a statistical property of the measurement, but the MID is the value of concern for interpretation and is based on the judgement of patients. The MDC (and the associated SEM) is of little relevance for interpretation [11].

We observed that the MID value from the anchor-based assessment of 5 was smaller than the SEM of 6.4 (based on test–retest reliability). In these circumstances, changes as large as the MID may be important for patients, but they cannot be distinguished from measurement error [34]. We also observed that the MID of 5 was slightly larger the then SEM of 4.7 (based on internal consistency reliability). In this situation, changes as large as the MID can be considered statistically significant and important to patients [34].

Using global QOL as an anchor is one accepted method and has been used in previous studies [9, 35, 36]. The advantages of using a global rating of QOL as an anchor are that they are relatively easy to obtain, patient-centred and can take into account a variety of information and determinants of well-being. However, in order to use an anchor-based method, there must exist some association (minimum correlation) between the QOL items and the chosen anchor [12]. A potential limitation of our study is that some of the correlations between the anchors used in our study, and the DEMQOL and DEMQOL-U scores were not that strong (the correlations ranged from 0.38 to 0.59), with the possible result that some of the anchors may not perform well in defining the MID. Nonetheless, these correlations were above the recommended minimum thresholds of 0.30 and the even higher value of 0.371 recommended by Hays et al. [37].

Other potential limitations associated with using DEMQOL Q29 as an anchor include recall bias and response shift [11]. Response shift occurs when a patient’s views, values or expectations change over time. Thus, a patient’s health might be seen to be getting worse, yet the patient with dementia may assert that their QOL has not changed, or even that it has improved. Alternatively, a patient’s health status may appear to be unchanging even though that same patient may report substantial changes in their QOL. It is difficult to assess response shift in people with dementia. The classical way of assessing response shift is to use the “then test” which may not be appropriate for patients with dementia. The then-test usually asked patients after completing a second assessment a question such as “We would like you to think back to the time of your first assessment. With hindsight, how would you now rate the way you felt then?” [11]. The difference between the values of the 1st assessment and the then-test provides an estimate of response shift.

Recall bias occurs when the respondent’s answer to a question is affected by the respondent’s memory. People tend to forget how extreme the past was. Response is likely to be influenced by the patient’s status at the time of recall. For these reasons, items with short recall periods or items that ask patients to describe their current or recent status are preferable [11]. The DEMQOL asks respondents questions about how they feel “right now” or “in the last week” so has a short recall period. However, response shift and recall bias remain a possibility with people with dementia, and this is a limitation. Indeed, given our patient population, our study is more likely to suffer from these possible threats to the validity of the results than other studies using similar anchor-based approaches.

We used a four-point global QOL anchor which resulted in seven-point change scale; others have used 14 points, which may be more sensitive [10]. The designation of what response on the DEMQOL Q29 anchor which suggests patients as fundamentally unchanged and what response suggests patients have experienced a small but important change is inevitably subjective.

The anchor, DEMQOL Q29, did not ask the participant to recall their overall health but instead rate it considering the previous week; this is a strength of the anchor and the global rating of QoL. However, a limitation of the study is that people with dementia had to self-report their quality of life and be able to do so again 12 months later, in which time their dementia may have progressed to moderate stages, and therefore, self-report could be less reliable for some participants. Only 70% (344/490) of participants completed the 12-month assessment; some of these may have been due to declining health.

This study was conducted in participants with mild/moderate dementia and in the relatively early stages of the disease, living in the community in the United Kingdom. Many (62%) of the participants were assessed as having normal cognitive function on the MMSE at baseline. The recommended MID estimates are most applicable to studies in a similar patient population—a single MID is unlikely to be relevant for all populations, and a different MID may be appropriate for later-stage dementia and participants living in residential or nursing care. A limitation of the study is that 21% of the participants were excluded from the longitudinal anchor analysis due to having missing outcome data (either having withdrawn from the study, lost to follow-up or death); however, Table 1 shows baseline characteristics to be similar for the subset that had available eight- and twelve-month data compared to all those that had baseline data.

Second, the follow-up duration was only 12 months, so the MID values that were determined herein may not be consistent with those in studies with longer follow-up durations. Thus, to validate the findings of this study, a longer follow-up and patients with different severities of dementia would be needed in future studies.

Our recommended MID should be confirmed based on evidence from other studies, and we recommend that other cohorts are used to estimate MID and the evidence is consolidated to provide a more robust estimate. Further research into the MID for the DEMQOL-proxy is needed for people with dementia who cannot self-report.