Quality of Life Research

, Volume 22, Issue 1, pp 13–26

Worsening trends and increasing disparities in health-related quality of life: evidence from two French population-based cross-sectional surveys, 1995–2003

Authors

    • Biostatistics and Epidemiology Unit, Assistance Publique-Hôpitaux de Paris, Hôtel-DieuNancy-Université, Université Paris-Descartes, Université Metz Paul Verlaine
  • Stéphane Rican
    • Space, Health and Territories LaboratoryUniversity of Paris Ouest-Nanterre
  • Joël Coste
    • Biostatistics and Epidemiology Unit, Assistance Publique-Hôpitaux de Paris, Hôtel-DieuNancy-Université, Université Paris-Descartes, Université Metz Paul Verlaine
Article

DOI: 10.1007/s11136-012-0117-7

Cite this article as:
Audureau, E., Rican, S. & Coste, J. Qual Life Res (2013) 22: 13. doi:10.1007/s11136-012-0117-7

Abstract

Purpose

To investigate time trends in health-related quality of life (HRQoL) in France and to report existing and changing demographic, socioeconomic, and geographic disparities.

Methods

Data were drawn from two independent national cross-sectional surveys conducted in 1995 and 2003, including 3,243 individuals aged 18–84 in 1995 and 22,743 in 2003. HRQoL was measured with the 8 subscales of the French version of the SF-36.

Results

After multiple linear regression, a significant decrease was observed between 1995 and 2003 in all scales scores, from −0.11 adjusted standard deviations for Social Functioning (95% CI: −0.15 to −0.08) to −0.23 for Vitality (−0.26 to −0.19). Increasing age, female gender, divorce/widowhood, lowest educational levels, chronic conditions, and living in the Northern region were identified as independent predictors of lower HRQoL scores. Testing interactions showed significantly greater differences between 1995 and 2003 for subjects aged 75–84 and for least educated subjects (Physical Functioning, General Health). The Gini index increased for all scales.

Conclusions

We report evidence of worsening trends and possibly increasing demographic, socioeconomic, and regional disparities in HRQoL between 1995 and 2003 in France. Monitoring HRQoL in populations can provide unique and sensitive data, complementary to classical indicators based on mortality and morbidity.

Keywords

Health disparitiesHealth-related quality of lifeSurveillancePopulation-based study

Abbreviations

BP

Bodily pain

GH

General health

HRQoL

Health-related quality of life

MH

Mental health

PF

Physical functioning

RE

Role limitations relating to mental health

RP

Role limitations relating to physical health

SES

Socioeconomic status

SF

Social functioning

SRH

Self-rated health

VT

Vitality

Introduction

Monitoring the health of populations is of crucial importance for several reasons, including the identification of temporal trends in health status and sociodemographic or geographic determinants. Such information is essential for recognizing unmet population health needs, planning programs or assessing the effectiveness of health policies. Reliable and meaningful indicators are required for these purposes; generally, indicators based on mortality or morbidity have been used [13]. However, interest has been growing in tracking health-related quality of life (HRQoL) or self-rated health (SRH) measures in populations, because such measures have been shown to be associated with numerous diseases and chronic conditions [47] and in longitudinal studies to be predictive of subsequent hospitalization or mortality [810]. In addition, given its multidimensional and patient-centered construction, tracking HRQoL could reveal important complementary information, only partially provided by classical approaches based on more “objective” indicators.

Consequently, several national surveillance systems have included HRQoL or SRH measures in their routine data collection [1113]. Nevertheless, reports assessing temporal trends of HRQoL in the general population remain scarce and mostly derived from simple self-rated measures, such as the mean number of unhealthy days [12, 14] or general health assessed through a single 5-grade Likert scale [11]. Currently, the results available reflect the possibility of worsening trends in some western countries [11, 14]: reports from Spain and the United States indicate that this trend was more marked in particular social groups, especially the least educated [15] and the most deprived [11]. This suggests SRH disparities may be widening.

The French situation regarding overall population health and disparities has long been described as paradoxical: despite good global health indicators, such as life expectancy, all past and recent reports indicate that health disparities are wider in France than in other Western European countries [1620]. In addition, whereas constant improvements have been observed over recent decades in global indicators such as mortality by cancer [21] and child mortality [3], the prevalence of chronic conditions, including diabetes [22, 23], hypertension [24], and obesity [25], has conversely been steadily increasing in France as in other industrialized countries.

It is unknown to what extent and even whether the evolution of HRQoL has reflected these worsening patterns in France. Therefore, we aimed to investigate trends through time of overall HRQoL and to report existing and changing demographic, socioeconomic, and geographic disparities in France. To do this, we used two large population-based cross-sectional surveys conducted in 1995 and 2003 both assessing HRQoL through a validated multidimensional instrument of measure.

Materials and methods

Population surveys

Two population-based surveys, both representative of the French population, included the same self-reported health-related quality of life questionnaire and collected information on demographic, socioeconomic, and health status. Both surveys randomly sampled households on the basis of data from the latest available national census of population and housing. Although subjects aged 15 and over (1995) and of all ages (2003) were included in the initial surveys, in our analysis, we considered only adult subjects aged below 85 to avoid any possible inconsistency that might occur in the extremely elderly. There was no incentive for participating in either survey, and both received the approval of the appropriate committees for observational studies in France: CNIS (Conseil National de l’Information Statistique) and CNIL (Commission Nationale de l’Informatique et des Libertés).

The SOFRES Health Survey (1995)

In 1995, the SOFRES polling agency conducted a national survey originally designed to be the norming survey for the SF-36 questionnaire in France [26]. A two-stage stratified sampling design was used, surveying one individual randomly chosen in each household selected. Data were collected by postal mail and included sociodemographic characteristics, demand for health services and health status. Of the 4,000 postal mails sent, 3,308 (82.7%) were returned; 3,243 (81.1%) of the subjects had completed at least one scale of the SF-36 questionnaire and were aged between 18 and 84 years. An oversample of 348 subjects aged over 65 was used to improve accuracy for this age group.

The Decennial Health Survey (2003)

In 2003, the French National Institute of Statistics and Economic Studies (INSEE) conducted the latest Decennial Health Survey, a national survey of households performed every decade since 1970 [27]. The sampling design was multistage and stratified on region and size of urban unit and surveyed all individuals in the households selected. Information was collected on sociodemographic characteristics, demand for health services and health status, using a combination of face-to-face interviews with specifically trained interviewers and self-administered questionnaires collected after 3 monthly visits. The initial sample included 40,796 subjects of all ages. Of the 30,544 subjects aged 18–84 years—of whom 8,896 were oversampled for Paris, North, Eastern Parisian Basin, and Mediterranean Basin regions—29,663 effectively received the questionnaire (97.1%), of whom 25,539 (86.1%) completed and returned the self-reported questionnaire on the last visit; 22,743 (76.7%) of these subjects had completed at least one scale of the SF-36 questionnaire.

Health-related quality of life measurement

The Medical Outcomes Study (MOS) 36-item short-form (SF-36) questionnaire is a validated, generic self-administered questionnaire [28, 29] and is widely used in population-based studies to measure health-related quality of life. The same French SF-36 version was used in both the 1995 and 2003 surveys and was developed and validated as part of the International Quality of Life Assessment (IQOLA) project [30]. The SF-36 includes 8 subscales: Physical Functioning (PF), Role limitations relating to Physical health (RP), Bodily Pain (BP), General Health (GH), Vitality (VT), Social Functioning (SF), Role limitations relating to mental health (RE) and Mental Health (MH). Scales were scored according to the documented procedure [31], with imputation of missing values from the mean of non-missing items of the same scale, when more than half of the items were available [26]. Scores are normalized (0–100 range; higher values indicating better-perceived health) and can be standardized and adjusted for gender and age to be expressed in standard deviations (SD) from the reference normative 1995 French data.

Outcomes and predictors

The primary outcome measures were the eight SF-36 scale scores, expressed either as crude scores (0–100) or as age- and sex-adjusted standardized scores. The predictors assessed were the year of survey, age, gender, marital status, education, occupation, self-reported chronic conditions, region and size of urban unit. Age was categorized into seven groups from 18 to 84 years. Marital status was classified into four categories (married/living with partner, single, divorced/separated, widowed). Socioeconomic status was investigated through the highest educational level achieved (no diploma, primary school, secondary, lower tertiary and upper tertiary level) and the occupational status (10 independent categories: Farmers, Tradesmen/Craftsmen, Senior executives/Intellectual professions, Middle-level professions, Service staff/Employees, Manual workers, Retired, Unemployed, Inactive for health reasons, Others). Self-reported status as concerns chronic conditions was assessed for five conditions defined in the same way in both samples (diabetes, cancer, hypertension, myocardial ischemia [angina/infarction] and congestive heart failure). To evaluate geographical effects, nine areas were defined, hereafter referred to as regions, aggregated from the 22 administrative Régions of metropolitan France. These areas are large but nevertheless representative of known French health and socioeconomic contrasts. Additionally, the size of the urban unit was studied, a French administrative subdivision based on agglomerations of one or more adjacent municipalities (rural, 2,000–20,000 inhabitants, 20,000–200,000, more than 200,000 and the Paris metropolitan area). No data regarding the race or ethnicity were available in either survey, in accordance with the French constitutional law prohibiting the collection of such data.

Statistical methods

Univariate analysis

Results are presented as means (±SD) for continuous data and percentages for categorical data. Unpaired t tests and Chi-square tests were used to compare demographic, socioeconomic, and general health characteristics between the two samples—excluding the oversample of subjects aged over 65 from the 1995 survey—and to examine differences in SF-36 results both expressed as crude and standardized scores, including oversamples, and using calibration weights to adjust for non-response and sampling bias of the 2003 survey.

Multivariate analysis

Multivariate linear regression was used to assess the effect of independent predictors on non-standardized SF-36 scores. The strategy for regression modeling involved four successive steps, in accordance with our hypotheses. First, we entered the year of survey, region, age, and gender. Second, we entered socioeconomic variables (education, marital status, occupational status) and size of urban unit, followed by self-reported chronic conditions. Two- and three-way interaction terms were then tested using global tests of interaction, and if these were significant, individual terms of interaction were tested, assessing whether region and year effects on HRQoL were modified by age, gender, and socioeconomic variables, and whether the region effect was modified by year. F-partial tests were performed at each step to determine whether any variable should be kept or dropped. The results are presented as regression coefficients, directly conveying the expected effect of one unit of predictor on the score. For the Decennial survey, similar results were yielded when using multilevel modeling to take into account the clustering effect of household, and when using calibration weights to adjust for non-response, so that only results from unweighted standard fixed effects models are shown.

Summary measures of disparity

We computed two summary measures of disparity at the individual level: the Gini index based on the Lorenz curve [32] (with x-axis as the cumulative proportion of individuals ordered by level of health and y-axis as the cumulative proportion of health in these individuals) and the Concentration index derived from the concentration curve [33] (x-axis as the cumulative proportion of individuals ordered by socioeconomic level and y-axis as the cumulative total proportion of health in these individuals). The Gini index measures the overall level of health inequality, ranging from 0 (no inequality) to 1 (one individual healthy vs. all others unhealthy), whereas the Concentration index measures the extent of health inequalities related to socioeconomic status (SES), i.e., education-related inequality in the present study, ranging from −1 (health declining with increasing SES), through 0 (perfect equality across SES groups) to +1 (health declining with decreasing SES). Standard errors and confidence intervals were computed using bootstrap with 400 replications [3436].

A two-tailed p value of <0.05 was considered to be significant. All statistical analyses were performed using Stata, version 10.0 (StataCorp, College Station, TX, USA). To illustrate the differences in regional standardized scores between the two surveys, maps were built using the Geographical Information System (GIS) software ArcGis v.9.3 (ESRI Corp., Redlands, California, USA).

Results

Univariate analysis

Demographics, socioeconomic characteristics, health status, and regional distribution of the two samples are presented in Table 1. All characteristics were significantly different between the two surveys, except for gender, diabetes, and cancer history. Notably, the sample included in the 1995 survey tended to be slightly older (47.0 [±17.7] vs. 46.0 [±17.2]), less rural (21 vs. 24%) and included more retired subjects (28 vs. 22%). Table 2 summarizes the differences between the two surveys in all crude SF-36 scale scores, stratified by gender and age. With few exceptions, all scale scores decreased significantly between 1995 and 2003, regardless of gender (e.g., PF in 18–24 year-old subjects: 96.7 [1995] vs. 94.1 [2003] for men; 94.9 vs. 92.7 for women) and age (e.g., PF: 96.7 vs. 94.1 for 18–24 year-old men and 68.4 vs. 61.4 for 75–84 year-old men). Figure 1 shows standardized scores by region, illustrating similar decreases across the whole of France. The northern and to a lesser extent eastern regions presented lower scores than other regions in 1995 and 2003.
Table 1

Demographic, geographic, and socioeconomic characteristics of the two samples

 

1995

2003

(N = 3,243)

(N = 22,743)

Age, years (SD)

47.0 (17.7)

46.0 (17.2)

Age group (%)

 

 

 18–24

10

11

 25–34

21

17

 35–44

19

21

 45–54

12

20

 55–64

14

14

 65–74

19

11

 75–84

5

6

Women (%)

54

53

Region (%)

  

 Paris

18

21

 North

8

10

 East

9

7

 Eastern Parisian Basin

8

16

 Western Parisian Basin

10

7

 West

12

11

 South-West

10

8

 South-East

13

8

 Mediterranean Basin

12

11

Size of urban unit of residence (%)

  

 Rural

21

24

 2,500–20,000

17

16

 20,000–200,000

23

19

 ≥200,000

39

41

Matrimonial status (%)

  

 Married/in couple

72

72

 Single

15

18

 Divorced/separated

6

5

 Widowed

7

5

Education level (%)

  

 No diploma/primary school

21

23

 Secondary

54

51

 Lower tertiary

21

19

 Upper tertiary

4

7

Occupational status (%)

  

 Farmers

1

2

 Self-employed tradesmen/craftsmen

2

3

 Senior executives/intellectual professions

6

10

 Middle-level professions

15

15

 Service staff/employees

16

19

 Manual workers

10

14

 Retired

28

22

 Unemployed

5

6

 Inactive for health reasons

1

2

 Others

17

7

Chronic conditions (%)

  

 Cancer

1.7

2.0

 Diabetes

3.3

3.4

 Hypertension

16.2

12.7

 Myocardial ischemia (infarction/angina)

2.3

0.7

 Congestive heart failure

3.4

0.2

Table 2

Crude SF-36 scale scores according to gender and age

 

PF

RP

BP

GH

VT

SF

RE

MH

1995

2003

1995

2003

1995

2003

1995

2003

1995

2003

1995

2003

1995

2003

1995

2003

Male

Age

 18–24

96.7

94.1

95.8

90.6

86.8

84.7

82.4

76.1

67.9

63.2

89.8

85.8

93.1

87.2

72.9

71.2

 25–34

95.7

93.8

91.2

90.5

83.6

81.3

76.1

74.9

64.5

61.6

86.5

85.9

90.6

89.0

72.1

70.7

 35–44

94.4

91.2

90.8

87.5

81.0

78.3

73.1

71.0

64.3

60.9

86.4

83.9

89.1

88.3

71.0

69.3

 45–54

92.3

86.2

88.6

83.3

79.0

72.8

72.3

66.6

66.1

59.1

86.7

82.1

90.0

85.8

71.5

67.7

 55-64

82.3

81.8

79.8

77.6

70.2

69.2

66.0

64.2

60.2

59.1

83.3

81.5

83.0

81.8

71.1

69.6

 65–74

76.6

73.3

74.7

69.3

67.1

66.4

62.4

59.1

58.2

56.4

79.7

81.1

80.0

76.6

71.2

69.7

 75–84

68.4

61.4

62.0

53.1

63.6

58.5

58.6

54.1

54.4

49.3

76.4

73.1

66.7

60.0

70.0

65.3

Female

Age

 18–24

94.9

92.7

89.5

88.4

79.4

78.5

71.8

72.1

60.3

57.2

79.7

79.3

81.6

79.4

64.9

65.5

 25–34

92.4

90.3

86.0

84.8

80.2

76.5

75.1

72.4

60.6

55.4

82.8

79.2

86.1

82.2

68.2

65.0

 35–44

91.0

89.2

88.9

85.2

78.3

74.1

74.6

69.6

61.9

56.3

84.1

79.4

85.5

83.4

67.6

64.4

 45–54

86.2

83.3

85.5

81.9

72.0

69.1

69.1

66.0

58.9

55.5

79.8

78.6

83.4

80.6

64.8

63.2

 55–64

77.9

76.5

77.8

75.5

66.2

64.2

65.1

62.0

58.2

55.3

78.3

77.6

77.6

76.8

66.2

63.3

 65–74

71.6

65.6

68.7

62.7

62.4

58.7

61.2

56.6

55.1

51.1

76.5

76.5

73.2

67.0

65.4

62.1

 75–84

60.6

51.2

52.8

53.0

62.2

52.8

61.1

51.9

49.6

45.3

73.3

70.7

62.8

61.3

60.7

59.0

Results in bold are statistically significant at the 5% level

https://static-content.springer.com/image/art%3A10.1007%2Fs11136-012-0117-7/MediaObjects/11136_2012_117_Fig1_HTML.gif
Fig. 1

Regional distribution of standardized SF-36 scale scores (France, 1995–2003)

Multivariate analysis

Multivariate linear predictors of all SF-36 scale scores are presented in Tables 3 and 4, showing the main effects coefficients of the final model (Table 3: age, gender, year of survey, and region; Table 4: socioeconomic covariates, size of urban unit, chronic conditions) and global interaction test results (Table 4).
Table 3

Multiple linear predictors of non-standardized SF-36 scale scores (I)

 

PF

RP

BP

GH

VT

SF

RE

MH

β

p

β

p

β

p

β

p

β

p

β

p

β

p

β

p

Female

−3.46

<10−4

−3.09

<10−4

−4.85

<10−4

−1.51

<10−4

−4.39

<10−4

−4.21

<10−4

−5.06

<10−4

−4.89

<10−4

Age

                

 18–24

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

 25–34

−2.75

<10−4

−3.03

0.001

−1.60

0.02

0.66

0.20

1.12

0.03

−1.38

0.03

1.44

0.14

1.21

0.02

 35–44

−3.26

<10−4

−2.35

0.01

−3.13

<10−4

−3.14

<10−4

0.35

0.52

−1.95

0.002

0.77

0.43

−1.93

<10−4

 45–54

−5.93

<10−4

−3.25

0.001

−6.08

<10−4

−5.21

<10−4

0.08

0.88

−2.17

0.001

0.36

0.72

−2.13

<10−4

 55–64

−9.67

<10−4

−6.14

<10−4

−8.83

<10−4

−6.57

<10−4

0.10

0.87

−1.82

0.01

0.72

0.52

0.70

0.24

 65–74

−16.79

<10−4

−14.73

<10−4

−12.69

<10−4

−10.57

<10−4

−3.34

<10−4

−3.41

<10−4

−5.70

<10−4

−1.43

0.05

 75–84

−28.23

<10−4

−27.04

<10−4

−17.47

<10−4

−14.36

<10−4

−8.47

<10−4

−8.29

<10−4

−15.23

<10−4

−3.86

<10−4

Year

                

 1995

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

 2003

−3.05

<10−4

−3.44

<10−4

−2.99

<10−4

−3.54

<10−4

−3.96

<10−4

−2.24

<10−4

−3.47

<10−4

−2.31

<10−4

Region

                

 Paris

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

 North

−1.60

0.001

−1.48

0.05

−1.71

0.004

0.81

0.07

0.34

0.45

0.72

0.17

−1.58

0.05

−1.48

0.001

 East

0.84

0.13

0.27

0.77

0.13

0.84

0.20

0.69

0.01

0.99

0.60

0.32

0.97

0.30

0.03

0.95

 Eastern Parisian Basin

0.63

0.20

0.49

0.53

0.40

0.48

0.18

0.67

0.02

0.96

0.54

0.30

0.15

0.85

0.10

0.81

 Western Parisian Basin

0.25

0.67

0.56

0.54

0.14

0.84

0.21

0.68

0.21

0.69

1.25

0.04

1.04

0.28

0.45

0.39

 West

1.33

0.009

0.12

0.89

0.68

0.26

1.31

0.004

0.76

0.10

1.21

0.03

1.91

0.02

1.20

0.01

 South-West

1.84

0.001

0.35

0.69

0.92

0.15

0.15

0.76

0.86

0.08

1.07

0.07

1.81

0.04

1.60

0.001

 South-East

1.38

0.008

0.82

0.32

0.32

0.60

0.06

0.89

0.29

0.54

0.33

0.56

0.06

0.94

0.63

0.17

 Mediterranean Basin

0.45

0.35

−1.64

0.03

0.08

0.89

0.64

0.14

1.35

0.002

0.38

0.45

0.06

0.94

0.80

0.06

Results in bold are statistically significant at the 5% level

Table 4

Multiple linear predictors of non-standardized SF-36 scale scores (II) and results for global tests of interaction

 

PF

 

RP

 

BP

 

GH

 

VT

 

SF

 

RE

 

MH

 

β

P

β

p

β

p

β

p

β

p

β

p

β

p

β

p

Marital status

 Married

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

 Single

0.10

0.81

0.11

0.87

2.02

<10−4

0.77

0.04

1.11

0.003

2.13

<10−4

2.34

0.001

1.29

<10−4

 Divorced

1.67

0.002

3.34

<10−4

−1.14

0.08

−0.87

0.08

1.52

0.002

5.57

<10−4

7.93

<10−4

5.07

<10−4

 Widowed

2.42

<10−4

−1.66

0.08

−0.58

0.40

0.80

0.14

−0.71

0.18

2.80

<10−4

3.44

<10−4

3.91

<10−4

Education

 No diploma/primary

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

 Secondary

7.11

<10−4

5.98

<10−4

5.01

<10−4

3.56

<10−4

3.42

<10−4

1.65

<10−4

5.70

<10−4

2.47

<10−4

 Lower tertiary

9.29

<10−4

7.12

<10−4

6.58

<10−4

5.39

<10−4

4.33

<10−4

2.76

<10−4

7.18

<10−4

4.55

<10−4

 Upper tertiary

10.60

<10−4

9.80

<10−4

8.92

<10−4

6.52

<10−4

4.96

<10−4

4.15

<10−4

9.17

<10−4

5.40

<10−4

Occupation

 Farmers

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

 Tradesmen/craftsmen

2.62

0.03

2.49

0.19

5.50

<10−4

3.76

<10−4

0.80

0.46

3.12

0.01

0.77

0.70

0.20

0.85

 Senior executives

4.04

<10−4

2.97

0.08

5.28

<10−4

3.26

0.001

2.36

0.02

2.52

0.03

1.47

0.41

1.32

0.17

 Middle-level professions

3.35

0.001

1.32

0.42

3.76

0.002

1.89

0.04

0.40

0.66

1.48

0.18

0.96

0.57

1.03

0.26

 Employees

3.20

0.002

2.45

0.13

3.52

0.003

1.16

0.20

0.48

0.60

1.00

0.36

1.19

0.48

0.20

0.83

 Manual workers

0.61

0.55

0.19

0.91

0.07

0.96

−0.01

0.99

−0.71

0.44

0.38

0.73

0.89

0.60

−0.08

0.93

 Retired

1.59

0.14

−0.14

0.94

2.61

0.04

0.59

0.54

1.76

0.07

1.04

0.37

−2.43

0.17

1.69

0.08

 Unemployed

−0.69

0.53

3.49

0.05

−0.73

0.58

2.47

0.01

−0.56

0.57

3.51

0.003

7.11

<10−4

4.41

<10−4

 Inactive for health reasons

26.00

<10−4

38.10

<10−4

22.75

<10−4

25.81

<10−4

17.18

<10−4

20.82

<10−4

30.69

<10−4

12.47

<10−4

 Others

1.54

0.16

1.52

0.39

3.20

0.01

1.08

0.28

0.70

0.48

0.83

0.48

−3.42

0.06

0.34

0.73

Size of urban unit

 Rural

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

Ref

 

 2,000–20,000 inhabitants

0.80

0.05

0.04

0.96

−0.35

0.47

−0.58

0.11

−0.13

0.71

−0.54

0.21

−0.82

0.22

−0.15

0.69

 20,000–200,000 inhabitants

−0.60

0.12

0.14

0.82

−0.54

0.24

1.09

0.002

−0.27

0.44

1.09

0.01

−1.11

0.09

0.68

0.05

 >200,000 & Paris

1.04

0.01

−0.21

0.72

−0.27

0.54

1.04

0.002

0.83

0.02

0.85

0.04

−1.03

0.10

−0.39

0.26

Comorbidities

 Cancer

5.09

<10−4

15.1

<10−4

6.63

<10−4

9.75

<10−4

5.62

<10−4

6.60

<10−4

11.12

<10−4

3.49

<10−4

 Diabetes

8.46

<10−4

8.88

<10−4

6.90

<10−4

8.76

<10−4

5.02

<10−4

5.07

<10−4

8.24

<10−4

2.42

<10−4

 Hypertension

3.12

<10−4

3.26

<10−4

2.73

<10−4

3.58

<10−4

1.73

<10−4

1.37

0.001

2.55

<10−4

1.62

<10−4

 Myocardial ischemia (infarction/angina)

9.17

<10−4

17.33

<10−4

9.22

<10−4

8.51

<10−4

7.89

<10−4

8.11

<10−4

16.02

<10−4

6.31

<10−4

 Congestive heart failure

13.34

<10−4

15.03

<10−4

9.70

<10−4

11.64

<10−4

9.96

<10−4

11.57

<10−4

10.10

<10−4

5.07

<10−4

Interactions*

 Year * age

 

<10−4

 

0.03

 

0.01

 

0.0001

 

NS

 

NS

 

NS

 

NS

 Year * education

 

0.0005

 

NS

 

NS

 

0.03

 

NS

 

NS

 

NS

 

NS

R² (step 1: age, gender, year, region)

 

0.202

 

0.090

 

0.099

 

0.107

 

0.046

 

0.029

 

0.046

 

0.035

R² (step 1 + 2 : marital status, education, occupational status, size of urban unit)

 

0.268

 

0.126

 

0.140

 

0.164

 

0.076

 

0.060

 

0.077

 

0.066

R² (step 1 + 2+3: comorbidities)

 

0.281

 

0.138

 

0.148

 

0.186

 

0.087

 

0.068

 

0.085

 

0.071

R² (step 1 + 2+3 + 4: interactions)

 

0.282

 

0.139

 

0.149

 

0.187

 

0.087

 

0.068

 

0.085

 

0.071

* Levels of significance for global test for interaction; not significant interaction tests not shown: year * region, year * gender, region * education, region * gender, region * age

The year of survey was a strong independent predictor for all scales, showing a significant decrease between 1995 and 2003 of between −2.2 for SF and −4.0 for VT (see Table 3). The decrease, expressed as effect size, was −0.11 [SF] (95% confidence interval −0.15 to −0.08), −0.11 [RE] (−0.15 to −0.07), −0.11 [RP] (−0.15 to −0.07), −0.12 [MH] (−0.16 to −0.09), −0.13 [BP] (−0.17 to −0.10), −0.16 [PF] (−0.21 to 0.12), −0.21 [GH] (−0.25 to −0.17) and −0.23 [VT] (−0.26 to −0.19). Results for regions showed various patterns, depending on the scale considered. Taking the Paris region as the reference, one region presented distinctively lower scores (northern region, for PF [−1.60], RP [−1.48], BP [−1.71] RE [−1.58] and MH [−1.48]) and two regions had generally higher scores (western region, for PF [+1.33], SF [+1.21], VT [+0.76], MH [+1.20] and RE [+1.91] and the south-west, for PF [+1.84], MH [+1.60] and RE [+1.81]). Age had a strong effect on scores: increasing age was directly and linearly related to physically oriented and general scale scores (PF, RP, BP, and GH; e.g., PF from 0.00 [18–24 year] to −28.33 [75–84 year]), whereas mentally oriented scale scores presented a threshold at 65 years (MH, RE, VT, and SF; e.g., RE <−1.44 for ages below 64 year and from −5.70 to −15.23 above 65y). Gender was strongly associated with SF-36 scores, with lower scores for women (from −1.51[GH] to −5.06 [RE]).

Farmers were used as the reference group for occupational status (see Table 4). Scores were consistently higher for senior executives (PF [+4.04], BP [+5.28], GH [+3.26], VT [+2.36], SF [+2.52]), lower for the unemployed (for RP [−3.49], GH [−2.47], SF [−3.51], RE [−7.11], MH [−4.41]) and consistently lower for people inactive for health reasons (from −12.47 [MH] to −38.10 [RP]). Scores increased progressively with increasing educational level (highest vs. lowest level: from +4.15 [SF] to +10.60 [PF]). Divorced people had lower scores than married people in all scales but BP and GH (from −1.52[VT] to −7.93[RE]) and widowed people had lower scores for scales PF (−2.42), SF (−2.80), RE (−3.44) and MH (−3.91). Self-reported comorbidities were all strongly associated with HRQoL, regardless of the scale (e.g., RP: from −3.26 [Hypertension] to −17.33 [Myocardial ischemia]).

Global tests of interaction were performed (see Table 4): the interaction between year of survey and age was significant for PF, RP, BP, and GH; and the interaction between year of survey and education was significant for PF and RP. Detailed interaction terms were as follows: differences between 1995 and 2003 were significantly greater for subjects aged 65–74 (−5.2[PF], −6.8[RP] and −5.0[GH]) and 75–84 (−9.6[PF], −8.3[BP] and −8.2[GH]) than for subjects aged 18–24 (−1.5[PF], −2.5[RP], −1.4[BP] and −1.9[GH]); and for subjects with the lowest educational level (−5.8[PF] and −4.8[GH]) than those with the highest educational level (−2.1[PF] and −2.3[GH]). Significant interactions were found between year of survey and occupational status in all scales, showing greater decreases for persons inactive for health reasons. The global test for interaction between year and region was not significant, but the Northern region nevertheless showed significantly larger differences between 1995 and 2003 than the reference Paris region for PF (−4.9 vs. −2.5; p = 0.05) and GH (−5.0 vs. −2.3; p = 0.04). No significant interaction was found between the year of survey and gender, marital status or size of urban unit, nor between region and age, gender, marital status, size of urban unit, education or occupational status.

Summary measures of disparity

Summary measures of disparity and details on the evolution of age- and sex-adjusted standardized scores according to age and education level are presented in supplementary Table 1A and 1B. Increasing trends between 1995 and 2003 were observed in all scales for the Gini index (e.g., PF from 0.108 [1995] to 0.128 [2003]; MH from 0.141 to 0.150) and in physically oriented scales for the Concentration index (e.g., PF from 0.037 to 0.048).

Discussion

Using two large representative samples of the French population, we report evidence for a substantial decrease in HRQoL between 1995 and 2003, affecting all scales of the SF-36 questionnaire. Notable demographic, socioeconomic, and regional disparities emerged from the analysis of the main effects explaining HRQoL. Moreover, we found significant interactions between the year of survey and education, and between the year and age, indicating probable worsening trends in self-reported health disparities.

General decrease

A significant decrease in HRQoL was observed over an 8-year period. Although there have been numerous single cross-sectional studies, few reports are available on time trends in general populations of western countries [11, 14]. Surveillance-based data from the Centers for Disease Control (CDC) showed similar worsening trends from 1993 to 2001 in the United States: mean unhealthy days (number of days during the preceding 30 days for which physical or mental health was not good) increased 14% during the period [12]. Examination of the evolution of some hallmark health indicators over the period studied reveals a contrasted situation in France. On the one hand, there has been overall progress as concerns certain mortality indicators such as cancer mortality rates, most likely due to multiple factors involving variable improvements in prevention, early diagnosis or treatment [21]. On the other hand, the prevalence of one particular potentially chronic and invalidating condition—obesity—has almost doubled during the 1990s [25], and the mean annual increase in the prevalence of diabetes between 2000 and 2005 was 5.7% [23]. The French economic context over the period provides some additional clues and included contrasting situations: the period 1995–2000 was marked by economic growth and declining unemployment, the opposite applied between 2000 and 2003 with economic growth stagnating and unemployment rising. Overall, our findings illustrate the essential benefits of tracking HRQoL and the specificity of such measures expressed by the populations themselves. Given the predictive value of HRQoL, monitoring time trends in HRQoL may provide some unique and sensitive public health information, only partially captured by classical objective indicators. In a general context of widening disparities and weakening social ties, further research may help describe and understand the complex mechanisms involved in the observed decline in HRQoL.

It has recently been pointed out that monitoring SRH in populations may not be reliable for showing temporal trends. Salomon et al. [37] analyzed the results from four national surveys conducted between 1971 and 2007 in the United States and found discrepancies in progression, particularly affecting the lowest education levels. However, the amplitude of variation was small in any given period and could have been the consequence of differences in survey designs or of random fluctuations [38]. Moreover, SRH was generally assessed as a simple global question. By contrast, the results of our study relied on a validated and comprehensive measurement instrument, expected to provide a more faithful reflection of HRQoL complexity. Using the same dataset, we have demonstrated the properties of invariance of the SF-36 questionnaire, showing a satisfactory temporal stability when using a Rasch model to take account of the differential item functioning (DIF) between age, gender, geographical area, and year sample (manuscript in preparation) [39]. Indeed, we believe that the magnitude of the decrease we identify and its persistence in all groups after multiple adjustments cannot be explained merely by differences in survey design or continuous aging of the population.

Widening disparities

The literature describing factors associated with health disparities is substantial [18, 40]. Likewise, disparities in quality of life have been extensively studied over the 20 past years: HRQoL measures have been found to be poorer for the elderly, women, ethnic minorities, and people with the lowest SES [17, 4042]. Work in the fields of geography and sociology has led to growing epidemiological interest in studying the role of contextual and geographical factors in health disparities [43].

Consistent with previous findings [31], HRQoL in our study was strongly associated with demographic characteristics. Older subjects and women reported lower scores than younger subjects and men, respectively. Notably, physical and general scales (PF, RP, BP, and GH) were more consistently and substantially affected by increasing age than mental scales (MH). Likewise, self-reported comorbidities were strong predictors of poor HRQoL, consistent with the extensive literature available [4, 5].

Low educational and occupational statuses were strongly predictive of poorer HRQoL. The association between individual socioeconomical status (SES) and HRQoL has been extensively studied [41, 4449]. Most reports demonstrate similar trends, regardless of the socioeconomic factors considered (income, occupation, education). Current understanding of this phenomena points at complex relationships between health perception, access to care and health behavior [50, 51].

Concern has recently been increasing over the interpretation of disparities in SRH as a function of SES. Numerous studies report significant association between SRH and “objective” health status, and some reports show an interaction between education and SRH, suggesting a poorer predictive ability of SRH for the least educated subjects [52, 53]. From similar observations, it has been suggested that SRH may underestimate the extent of health inequalities between socioeconomic groups and hence be unsuitable for assessing any such disparities [54, 55]. However, defining the true health status that could be used as the gold standard for comparisons is extremely difficult if not impossible, self-reported morbidity, biomarkers, and even mortality each having their own limitations in this context. Instead, health-related quality of life should be considered as a unique and complementary marker, reflecting the patient’s perception, and indeed, it is this perception that should be viewed as important.

We compared aggregated administrative regions and found disparities. Although the regions used were large, this geographic classification appeared meaningful and consistent with the French context, for historical and political reasons. Our results were consistent with data previously published concerning the spatial distribution of both mortality and morbidity: cancer incidence and mortality rates [19, 20] as well as prevalence of each diabetes [56] and obesity [25] are indeed higher in the Northern region, which showed the poorest HRQoL scores in our study. In addition, we found evidence of a possible larger decrease in HRQoL in the Northern region than elsewhere between 1995 and 2003, strengthening the notion of increasing disparities at multiple levels. Both contextual factors and individual factors contribute to health variation [43]. In the field of HRQoL, understanding of the role of geographical factors has been steadily improving, with several studies reporting various results, depending on the unit of analysis chosen. Some studies focused on administrative subdivisions such as regions [57] and households [58, 59], others used comparisons between, for example, rural/urban or central/periphery [48], and some addressed specific areas with common characteristics [11, 60]. Most studies report significantly poorer SRH in the most deprived areas, defined according to a socioeconomic deprivation index composed of multiple socioeconomic variables (including income, unemployment rates, housing) [61, 62]. The interpretation of findings regarding rural and urban areas is not straightforward and seems to differ between countries [48, 63].

Testing interactions between predictors and the time of survey indicated widening gaps between the least and most educated and between the youngest and oldest age groups, with non-significant tendencies for rural/urban and divorced/married subjects. In addition, computing summary measures of health disparity at the individual level showed a trend toward greater overall health inequality in all scales (Gini index) and education-related health inequality (Concentration index) for physically oriented scales. These findings are consistent with reports drawn from surveillance-based data in the United States and Spain, where worsening trends in SRH were especially found among the most deprived groups [11], the least educated of the middle-aged and older adults [15] and in some ethnic minorities [12]. Remarkably, similar trends were reported in France for “objective” indicators during the nineties, indicating widening gaps in mortality between the most and least deprived areas [64], and in the increase in prevalence both for diabetes between regions [56] and for obesity between social groups and regions [25]. Note that the northern and eastern regions present the highest rates, agreeing with our results showing a more marked decrease in HRQoL in these areas. Finally, a significant interaction with time was found in the oldest group (>75) for PF (physical functioning) but not for RP (role limitations relating to physical health), a finding more in line with the improvements in old age disability and limitations observed during the 1990s [65, 66]. However, a cautious interpretation should be made of our results regarding the evolution of inequality in HRQoL, because assessing average disparities between groups or overall inequality between subjects prevents the clear understanding of the underlying heterogeneity. Likewise, using only two time-points precludes the demonstration of complex temporal fluctuations, such as those of the Gini index for the Health and Activities Limitation Index (HALex) reported in the United States, showing an increase between 1990 and 1995 [67], a decrease between 1997 and 2000 and a regular increase thereafter (2000–2007) [68].

Limits

This study has some limitations, mainly related to differences between the two population-based surveys used, and particularly as concerns sampling and the methodology used for data collection. First, there were significant differences regarding sociodemographic characteristics, despite both surveys having been initially designed to include a representative sample of the French population at the time, and despite the sets of calibration weights specifically calculated to address this issue. Second, the 1995 survey was a mailing postal survey, whereas the 2003 survey involved both face-to-face interviews to collect information on past health history and self-report questionnaires. However, the mode of assessment of the primary outcome was essentially identical, because SF-36 forms were all self-completed, using the same validated translated French version (v.1.3) with the same instructions provided. Likewise, all sampled households were asked and accepted to participate in both surveys before questionnaires were sent or given by interviewers, and response rates of subjects having effectively received the questionnaire were comparable (81% in 1995 vs. 77% in 2003), limiting the risk of volunteer bias. In addition, fitting multivariate linear regression models contributed to controlling for initial differences in sociodemographic characteristics.

A larger range of descriptors would have helped better explain differences in health-related quality of life, the limiting factor being the SOFRES survey. For example, no information was available for chronic conditions other than those previously described [4, 5], or for body mass index [69, 70], income, environmental factors [71, 72], social ties [73], or season of the report [74] although these descriptors are significantly associated with HRQoL. This limitation arises from the different objectives pursued by the two surveys: the Decennial Health Survey is a national survey aimed at accurately monitoring health trends in diverse fields on a regular basis, whereas the SOFRES Survey was specifically designed to validate the French version of the SF-36 questionnaire and establish national reference values. Finally, diseases were self-reported in both surveys, and given the lack of a confirmed diagnosis, this data collection method could lead to possible biases.

Conclusion

In conclusion, this study shows a significant decrease in health-related quality of life in France between 1995 and 2003. Although the decrease affected all groups of the population, there was clear evidence of possible worsening disparities impacting the most fragile categories. Our findings are in accordance with the spatial distributions and trends reported in France during the same period for other health indicators, such as prevalence of both diabetes and obesity, and disparities in mortality. These observations need to be confirmed, and further research is required to elucidate the complex mechanisms underlying this apparent decrease. Monitoring HRQoL in populations can provide global, unique, and sensitive data, complementary to classical indicators based on mortality and morbidity, and is thus potentially helpful for planning programs and evaluating the effects of interventions. In the European context of growing health disparities [18, 75], health policies should not ignore such findings, and indeed should consider implementing such validated health measures, directly expressed by populations, in routine surveillance systems.

Supplementary material

11136_2012_117_MOESM1_ESM.pdf (67 kb)
Supplementary material 1 (PDF 67 kb)

Copyright information

© Springer Science+Business Media B.V. 2012