Temporal change in health-related quality of life: a longitudinal study in general practice 1999–2004

Analysis

Only those subjects with three consecutive values of EQ-5D_index available between 1999 and 2004 were included in the present analyses. Mean EQ-5D_index and EQ-5D_vas scores were calculated, and changes over time were determined by longitudinal analysis using a generalised estimating equation (GEE) regression model, which appropriately and efficiently assesses subject-specific temporal changes, implemented with STATA software. Significant changes and trends over time, with their associated confidence intervals (CIs), were derived from the coefficients of the GEE regression model, using the robust Huber–White sandwich estimator of variance. Adjustment was made for a number of possible confounding factors, namely age, sex, smoking status (current, ex, never), education beyond 16 years (yes/no) and degree level education (yes/no).

Summary of main findings

This study has shown a decline in HRQoL in a general population over a 5-year observation period. This remained significant even after adjusting for potential confounders: ageing of the population, sex, and smoking status, and applied whether a societal valuation of HRQoL (EQ-5D_index derived from a combination of five dimensions) or an individual valuation (EQ-5D_vas) was considered. This is an important and unexpected finding, as socioeconomic conditions and affluence in the UK have generally improved and HRQoL could be expected to be better as these factors improve. Browning et al found that HRQoL improved in relation to income and education, but did not find an association between poor HRQoL and unemployment.18

There are several possible reasons for a decline in HRQoL over time. First, improvements in medical care have led to more people surviving events such as strokes and myocardial infarctions, with the effect of more people being alive but with increased overall morbidity and, therefore, reduced HRQoL. Secondly, expectations and perceptions regarding HRQoL may have changed; this may partially explain the greater decline seen on EQ-5D_vas, as this measure records the individuals status in relation to ‘best imaginable health state’. This may be associated with the recent trend towards medicalisation of conditions (largely driven by the availability of treatments),19 for example, the lowering of blood pressure targets for hypertension means that more patients have a label for a disease that was not causing them any current symptoms. Thirdly, other factors that affect HRQoL but were not measured in these surveys may have changed, for example the prevalence of obesity.20

Strengths and limitations of the study

Selection bias must be considered when interpreting the study findings. Although the group included in the analyses was older and contained fewer smokers than the subjects excluded, the results were adjusted for these potential confounders. Response rates to the postal analysis from another paper21 on this study show responders in both 2001 and 2004 were older (mean 48.5 versus 37.6 years P<0.001 [2001]; 50.5 versus 38.8 years P<0.001 [2004]) and more likely to be female (54.9% versus 44.9% P<0.001 [2001]; 55.3% versus 48.5% P<0.001 [2004]). Early responders were older, and more likely to be female, but less likely to smoke than those who responded to reminders. There was no important association between respiratory symptoms, and associated feature prevalence and stage of response.21 However, if those who are most ill are least capable and therefore least likely to respond to repeated surveys (or indeed to any postal survey), the resulting response bias would in fact cause an underestimate of the decrease in HRQoL. Ceiling effects on EQ-5D (particularly due to the modelling algorithm for calculating EQ-5D_index), as reported elsewhere,22^–24 may also have contributed to this potential underestimate.

Comparison with existing literature

There would appear to be no other reports of the use of EQ-5D over three time points, which adds considerably to the reliability of assessments of trend.25 However, a Swedish study using EQ-5D_index also observed a smaller deterioration in health status over time in both men and women of all age groups.11

In addition to statistical significance, it is also essential to assess the importance of the changes found; given the relatively large population studied, small degrees of change could be statistically significant. Although some studies have addressed this problem and examined minimal important difference (MID)26 and clinically significant change27 in HRQoL, they have generally included only subjects who had one specific morbidity. Extrapolating figures from a study comparing the respiratory disease-specific St George's respiratory questionnaire (SGRQ),27 with EQ-5D, a clinically significant change in EQ-5D_index would be 0.08 and in EQ-5D_vas 9.08.28 The MID for EQ-5D_vas has not been widely studied, but one estimate is as low as 3.0.29 There are few reports, so far, of what changes are important in a general population, which includes a majority of healthy people in addition to those with a number of different single and multiple pathologies.

Implications for future research or clinical practice

The use of a generic measure of HRQoL, in this case EQ-5D, enabled the assessment of temporal change to be made irrespective of specific morbidity. HRQoL measures such as EQ-5D are often used to measure the benefit of healthcare interventions, and to ascribe economic value to different health states by conversion to utilities. This is a vital step in assessment of clinical effectiveness, and is recommended by bodies such as the National Institute for Health and Clinical Excellence (NICE).30 The same measure could be used to compare temporal change in HRQoL of groups of people with different morbidities and also multiple pathologies in one or several populations.

Differences in the rate of change between different morbidities and populations must be taken into account in assessing and planning healthcare needs and resource provision for the population, and it is therefore important to corroborate these unexpected findings of a decline in HRQoL with further research into the temporal change.