SF-36 data were mapped to eight domain scores, including the physical functioning domain (PF), using the published coding algorithms and imputation methods for missing items.[19 ] The best of the six grip measurements was used to characterise maximum muscle strength. Weight and chair rises variables were transformed to normal distributions using the loge transformation (geometric means and standard deviations, and proportional differences between groups, were therefore presented for these variables). Height and weight were highly correlated (r=0.46, p<0.001 for men; r=0.29, p<0.001 for women); to avoid multi-collinearity problems in modeling analyses we calculated a sex-specific standardised residual of weight-adjusted-for-height.
Variables were summarised using means and standard deviations, medians and inter-quartile ranges, and frequencies and percentage distributions. Prevalence of mobility disability was described using each of the 10 individual items comprising the PF domain and the overall PF score.
The internal validity of the PF domain was investigated using Cronbach’s alpha. The construct validity of the PF score as a disability measure was investigated by using linear and logistic regression models to explore the relationships between “low/poor” SF-36 physical functioning scores (defined as scores in the lowest sex-specific fifth of the distribution i.e. ≤ 60 for men, and ≤ 75 for women) and the objective measures of physical performance. We hypothesised that lower PF scores would be associated with poorer objective physical performance.[14 (link),15 (link)] Linear regression was used to measure associations between the continuously distributed physical performance variables (dependent variables) and the PF score (independent variable), and to derive estimates of the mean difference in dependent variable values between low and high PF groups. Logistic regression was used for the categorical balance time variable and yielded odds ratios for achieving maximal balance time between low and high PF groups. Analyses were conducted without and with adjustment for the potential confounding influences of age, height, weight-adjusted-for-height, walking speed, social class, smoking habit and alcohol intake.
As a separate exercise, normative summary statistics for the SF-36 PF score were produced by gender and five year age-bands using data from the Department of Health’s large and nationally representative Health Survey for England (HSE).[10 ] In 1996 this survey included the SF-36, from which we re-analysed the PF scores. This was accomplished by accessing the 1996 HSE dataset[28 ] from the ESRC UK data archive (www.data-archive.ac.uk ). The HCS was too small to provide these data itself, but the availability of nationally representative norms for the PF domain adds considerably to its usefulness as an epidemiological tool.
All analyses were carried out for men and women separately using Stata, release 8.0 (Stata Corporation 2003).
Variables were summarised using means and standard deviations, medians and inter-quartile ranges, and frequencies and percentage distributions. Prevalence of mobility disability was described using each of the 10 individual items comprising the PF domain and the overall PF score.
The internal validity of the PF domain was investigated using Cronbach’s alpha. The construct validity of the PF score as a disability measure was investigated by using linear and logistic regression models to explore the relationships between “low/poor” SF-36 physical functioning scores (defined as scores in the lowest sex-specific fifth of the distribution i.e. ≤ 60 for men, and ≤ 75 for women) and the objective measures of physical performance. We hypothesised that lower PF scores would be associated with poorer objective physical performance.[14 (link),15 (link)] Linear regression was used to measure associations between the continuously distributed physical performance variables (dependent variables) and the PF score (independent variable), and to derive estimates of the mean difference in dependent variable values between low and high PF groups. Logistic regression was used for the categorical balance time variable and yielded odds ratios for achieving maximal balance time between low and high PF groups. Analyses were conducted without and with adjustment for the potential confounding influences of age, height, weight-adjusted-for-height, walking speed, social class, smoking habit and alcohol intake.
As a separate exercise, normative summary statistics for the SF-36 PF score were produced by gender and five year age-bands using data from the Department of Health’s large and nationally representative Health Survey for England (HSE).[10 ] In 1996 this survey included the SF-36, from which we re-analysed the PF scores. This was accomplished by accessing the 1996 HSE dataset[28 ] from the ESRC UK data archive (
All analyses were carried out for men and women separately using Stata, release 8.0 (Stata Corporation 2003).
