Quality of life of Australian chronically-ill adults: patient and practice characteristics matter

This study was part of a larger study examining the impact of the organizational capacity of general practices in Australia to manage chronic diseases. It was conducted in 27 Divisions (local primary care support organizations) in five states and in the Australian Capital Territory between December 2003 and October 2004. The data on Division characteristics showed that the 27 were representative of the 103 Divisions approached except that recruited general practices from 27 Divisions tended to be larger and to have a lower population to general practitioners ratio than the Australian average [16]. In each practice, clinical management software was used to select a random sample of 180 patients aged 18 years or more currently being prescribed medication for three common chronic diseases: asthma, type 2 diabetes, and hypertension/ischaemic heart disease. Practices were permitted to remove patients from the list who were deceased or otherwise inappropriate to invite. A total of 12,544 patients attending 96 practices were invited to participate. Completed surveys were received from 7606 patients (a response rate of 61%). A priori sample size calculations on the SF-12 physical component score confirmed that after adjustment for clustering (previous studies on SF-36 indicated a cluster effect (ICC = Intra-cluster correlation) of 0.011 for the PCS-36 [14]) predicted that an average of 50 patients from each of 100 practices would have sufficient power (1-β = 0.8 and α = 0.05) to detect an effect size of 0.10 between patients with good and poor general health assuming that about half of the patients were in good general health.

Ethics approval for the study was obtained from the University of New South Wales (UNSW) Human Research Ethics Committee and University of Adelaide Human Research Ethics Committee. Both practice staff and patients provided written informed consent.

The standard SF-12 version 2 is a 12-item questionnaire measuring physical and mental health [6, 12]. The adoption of the SF-12 version 2 over the original version 1 form for all new studies including population surveys is recommended [17]. It is an abbreviated form of the SF-36 Health Survey, which is one of the most widely used instruments for assessing HRQOL [12]. Both instruments produce eight dimensions of health (physical functioning (PF), role physical (RF), bodily pain (BP), general health (GH), vitality (VT), social functioning (SF), role emotional (RE), and mental health (MH)) [18, 19]. They also produce two summary scores – the Physical Component Summary (PCS) and the Mental Health Component Summary (MCS) – and have been validated for use in the USA, UK and many other European countries for large scale health measurement and monitoring [12, 19]. For ease of interpretation, scores are standardized to population norms, with the mean score set at 50 (SD = 10): higher scores indicate better health. The SF-12 has been shown to have good validity and reliability [17]. Previous research has supported the use of the standard SF-12 in Australian settings, rather than development of an 'Australian' short-form [20, 21]. The SF-12 is an instrument that can be administered in three minutes with a small trade off between brevity and precision [21].

The same sample of patients completed the General Practice Assessment Survey (GPAS) version 2 [22] along with the SF-12. The patient characteristics including self-reported general health and chronic medical condition/conditions were collected using the GPAS. Patient satisfaction was also assessed through the GPAS. The GPAS is a multi-item self-report questionnaire which measures several dimensions relating to patients' assessment of general practice. The psychometric properties of the GPAS have been evaluated [23].

The dependent variables were PCS-12 and MCS-12. Because patients do not register with general practitioners (GPs) in Australia, it was not possible to determine the "list size" of practices accurately and thus the number of general practitioners in a practice was used as a measure of the practice size. Geographical area was defined by using the Rural, Remote and Metropolitan Area (RRMA) classification [24] as urban (all metropolitan centers with populations ≥ 100,000) or rural (rural centers and all other areas with populations of less than 100,000). There were no practices in the sample which were zoned as remote. The socio-demographic characteristics of respondents studied were gender, age, self-reported general health status in the last 12 months, home ownership, education, employment, marital status, country of birth, disease and overall satisfaction with care (Table 1). Home ownership can be considered as one marker of economic status [25]. For some respondents, their specific chronic disease or diseases were not known and therefore 'unknowns' were included in the analysis as a separate category to minimize the data loss.

Summary physical (PCS-12) and mental (MCS-12) components were constructed using the standard SF-12 version 2 US algorithm empirically derived from the data of a US general population survey [17]. To confirm the dimensions as documented by Kontodimopoulos et al. [26] and Ware et al. [17], we carried out a factor analysis using SPSS statistical software (version 15; SPSS, Chicago, IL, USA) with principal components analysis using the varimax rotation [26]. The number of factors was determined by the scree test and eigen values > 1. The two principal components were then rotated into simple orthogonal structures, a procedure previously implemented in similar studies [26]. It was hypothesized that two factors would be obtained (Table 2) known as physical health and mental health. In addition, items originally belonging to the PF, RP, BP and GH domains were hypothesized to load (or correlate) higher on the physical health factor, whereas the MH, RE, SF and VT items were hypothesized to relate most strongly to the mental health factor. However, VT and SF have been shown to load on both physical and mental components [26].

First, we examined the association between the independent variables and physical or mental health component scores in univariate analyses with analysis of variance using SPSS (Table 1). The analysis of variance was conducted to compare unadjusted scores. The Pearson χ² – test was used to compare proportions analyzed and missing.

Multilevel regression models were used with two dimensions (physical and mental component scores) as continuous dependent variables and general practice and patient characteristics, including the hypothesized interaction between gender and employment (based on the previous studies [15, 27, 28]), as the independent variables. Multilevel analysis (with MLwiN Software [29]) adjusted for clustering of patients (level 1) within practices (level 2) [11, 14, 30]. Initially, we fitted a baseline variance component model (no independent variables) for each of the response variables followed by the main model. The main model expands the baseline model by including patient and practice characteristics with the hypothesized interaction [15, 27, 28] as fixed effects. The interaction effect of independent variables was included in the model if their regression coefficients were significant (Table 3) and they showed a significant improvement to the model without the interaction.

Parameter estimates were tested by the t value, determined by dividing the estimated coefficients by their standard errors (Table 3) [29]. Because the two models were nested, we used -2 log likelihood, known as the "change in the deviance", which has a chi-square distribution to test whether the difference between the two models was statistically significant (Table 4).

The baseline variance component model explained how the total variance was partitioned into variance between patients and practices (Table 4). The variance explained was estimated using the baseline model and main model [31]. Differences in the modeled variance indicate how much better a model can account for the variance at a specific level [32]. The formulas to calculate the proportion of variance explained are given by Snijders and Bosker [31] and Sixma et al. [32].

Factor analysis suggested a two-factor solution (Table 2). These two factors account for approximately 68.1% of the variance in the twelve items of the SF-12.

Correlations between physical and mental summary scores were very low with 0.054 (principal components analysis with the varimax rotation gives uncorrelated factors). The overall mean of PCS-12 and MCS-12 of these chronically-ill respondents were 42.4 (SD = 11.8) and 49.1 (SD = 11.1) respectively.

Table 1 shows the characteristics of respondents and practices (independent variables). Almost one-half of the respondents were patients from large practices and 40% of respondents were from rural areas. The mean age was 60 years (range 18–96). The majority (53%) was female and nearly 80% owned their own homes. Only 34% of respondents were employed and 40% were retired. Seventy-four per cent were born in Australia, 14% in USA, UK, Canada or New Zealand and the remaining 12% in non-English-speaking countries.

The multilevel regression included only data from the questionnaires for which information on all relevant variables was available, resulting in a final sample size of 6997 (92%) patients from 96 practices. Pearson Chi-Squared tests indicated that proportions of practice size, practice location, gender and country of birth were similar between the records used in multilevel analyses and missing data (data not shown). There were small but significant differences between the proportions of records analyzed and the total (including missing) for other characteristics: 0.7% (age), 0.5% (general health status), 0.3% (home ownership), 0.4% (education), 0.7% (employment), 0.3% (marital status) and 0.4% (disease).

Table 3 shows the results of the multilevel regression analyses for each of the response variables.

Patient characteristics including self-rated general health and chronic medical conditions were collected independently using GPAS²² for the same respondents (Table 1). Patients' assessment of overall satisfaction with care was also assessed through the GPAS. PCS-12 declined with age, but in contrast MCS-12 increased with age. Patients with better self-reported general health status rated both PCS-12 and MCS-12 higher than those with poor general health (Table 3). Both self-reported PCS-12 and MCS-12 were positively related to home ownership. Well-educated patients tended to rate PCS-12 higher than less well-educated patients, but there was no association with MCS-12.

Patients who were employed or retired were likely to have higher PCS-12 and MCS-12 than unemployed. Gender interacted with employment in predicting both PCS-12 and MCS-12 with unemployment being more associated with poorer health in males than in females (Figure 1).

Patients who were married or cohabiting tended to have higher MCS-12 than those who were not. Marital status did not have any effect on PCS-12. The number of chronic medical conditions was negatively associated with both MCS-12 and PCS-12. Results also showed an association between general satisfaction with care and MCS-12 but not PCS-12. Patients born in Australia were likely to have higher MCS-12 than those born in non-English-speaking countries but country of birth was not associated with PCS-12.

Patients from smaller general practices (1–3 GPs) reported lower PCS-12 and MCS-12 compared with those from larger practices. Practice location had no relationship with either PCS-12 or MCS-12.

Ninety seven per cent of the total variance in PCS-12 was at the patient level, the remaining 3% variance (Intra-cluster correlation (ICC) = 0.03) was at the practice level. For MCS-12, the corresponding figures were 99% at patient level and 1% (ICC = 0.01) at practice level (baseline model in Table 4). At the patient level (level 1) 42% and 21% of the variance respectively among patients for PCS-12 and MCS-12 were explained by the independent variables used in the analysis (Table 4). At the practice level (level 2), 73% and 49% of the variance among practices for PCS-12 and MCS-12 were explained by the variables used in the analysis (Table 4).

Based on the results of the analysis reported here, the SF-12 and its component scales appear to be valid and useful tools to use in identifying differences in quality of life of the chronically-ill Australian population on the basis of social determinants of health [7]. Known group comparisons based upon differences in general health, age, socio-economic status, and number of medical conditions yielded support for the construct validity of the SF-12 in this data [8, 10, 42]. Further, our data showed an association between general satisfaction with care or marital status with mental health but not with physical health confirming the results of previous studies [43‐45]. In our sample it appeared that the dimensions were discriminative enough to distinguish between respondents with a single illness and with two or more illnesses or low and high socio-economic status or younger and older respondents. Further, there was strong association between SF-12 summary scores and self-rated general health status collected independently using GPAS for the same respondents (clinically significant large effect sizes of 1.27 and 0.79 for PCS-12 and MCS-12 respectively). This ability to discriminate between groups means that clinicians can use scores better to understand the functional status and health care needs of at-risk subgroups, and also enables policy makers to measure clinical effectiveness [10]. The ability to detect previously hypothesized differences or associations between variables showed the construct validity of SF-12 in Australia [6]. Further, the results suggest that the SF-12 has construct validity when applied to an Australian primary care population with chronic illness.