The dynamics of quality: a national panel study of evidence-based standards

Cardiovascular disease

Cardiovascular disease is an umbrella term for a number of associated conditions. ELSA gathered information on achievement of quality indicators for IHD, hypertension and cerebrovascular disease (CVD) (stroke). A systematic review of 17 quality-of-care papers reporting on the RAND Corporation’s Assessing Care of Vulnerable Elders (ACOVE) quality indicators13 revealed that achievement rates are often high in primary care settings for indicated care for hypertension (72%) and stroke (69%).

Data from all 8409 general practices in the UK between 2005 and 2006, collected for the QOF, showed that quality indicator achievement for cardiovascular prescribing was generally high, with a mean achievement for all 26 indicators of 79.0% (standard deviation 4.4%). 14 This included antiplatelet therapy (89.75%), hypertension (74.83%) and beta-blockers (50.94%). An audit of 254 general practitioners (GPs) in the USA between 2007 and 2009, using quality indicators from the National Quality Forum and RAND Corporation, found that eligible patients with coronary artery disease were prescribed lipid-lowering medication on 76% of occasions and that beta-blockers were prescribed post myocardial infarction (MI) on 84% of occasions, though achievement for a quality indicator assessing prescriptions of antiplatelet therapy was lower at 48%. 15

Fürthauer et al. 16 assessed adherence to nine quality indicators by 58 GPs covering a range of cardiovascular conditions and interviewed patients and practitioners about the reasons for deviations from guideline recommendations. Approximately 70% of non-adherence was accounted for by physicians’ lack of knowledge and patients’ lack of awareness of the guidelines. Ludt et al. 17 sought to identify patient and practice characteristics that predicted good quality of care for prevention of cardiovascular disease using data from the clinical records of 7846 patients across primary care settings in 10 European countries. They did not find that practice factors predicted outcomes, but good risk factor control was more likely in those with higher health-related quality of life and lower for those whose marital status was single or who had achieved lower educational levels.

Diabetes

In 2009–10 the national audit on diabetes care18 reported that 1.8 million people were diagnosed with type 1 or type 2 diabetes and registered with an English general practice (5.24% adult prevalence). Despite 95.4% of patients with type 2 diabetes having consulted their GP at least once in the last year, the proportion receiving all processes included in a bundle of nine diabetes quality indicators19 was low (52.9%). Some processes [e.g. weight measurement 90.5% and glycosylated haemoglobin (HbA_1c) measurement 92.6%] were more frequently measured than others (e.g. retinopathy screening 78.9%, foot risk assessment 85.2% and renal risk assessment 54.4%). Similarly, De Wet et al. 20 reported compliance with 10 individual diabetes quality indicators and a ‘care bundle’ – a composite measure of the individual quality indicators – in nine Scottish general practices (56,948 patients). Achievement of individual quality indicators was high (ranging from 85.9% to 97.8%), but all-or-nothing receipt of the care bundle was lower (58.4%).

There is considerable variation in the delivery of core diabetes care between providers. 18 For example, practices located in more deprived areas perform less well relative to their wealthier counterparts. 21,22 Less is known about patient factors, even though we need to better understand why such variation arises in order to improve receipt of basic care processes. 20 The national diabetes audit18 reported that social deprivation, sex and time since diagnosis did not affect the likelihood of receiving a complete care bundle, although patient age (< 55 years or ≥ 85 years vs. 55–84 years) and ethnicity (‘non-white’ vs. ‘white’ patients) were associated. Kontopantelis et al. 22 analysed longitudinal data (2000–1 to 2006–7) from 23,920 patients with diabetes who were registered with one of 148 practices. Older patients (≥ 65 years vs. 17–39 years), those with comorbidities (three or more vs. none), and number of years living with diabetes (1–4 years vs. newly diagnosed) were more likely to receive a complete package, although no sex effect was observed.

Osteoarthritis

Osteoarthritis is the most prevalent chronic joint disease globally. 23 Key clinical practice guidelines for osteoarthritis include non-pharmacological interventions (e.g. exercise and weight optimisation), provision of information regarding the disease and treatments, and joint replacement for severe disease. 23

Previous work assessing quality of care for osteoarthritis7 found that indicated care was received on only 29% of occasions. A meta-analysis of 17 studies using ACOVE quality indicators found that achievement rates for quality indicators for osteoarthritis had an interquartile range of 29–41%. 13 A recent review of chronic disease management found similarly poor quality of care for osteoarthritis across the quality domains of effectiveness, efficiency, safety, timeliness and appropriateness and patient-centred care, despite the availability of several evidence-based clinical practice guidelines. 23 A longitudinal study following 268 patients newly referred to an orthopaedic specialist in the UK over 12 months found that information was provided regarding their condition, pain management and exercise on 42%, 35% and 43% of occasions, respectively. 24 There is a need for research to assess quality of care for osteoarthritis over a longer time period. Using basic patient descriptors taken from medical records, Li et al. 25 found that receipt of advice on exercise and weight optimisation was not provided equitably across sex, age, disability and levels of formal education in 1349 patients with osteoarthritis of the knee and/or hip.

Depression

Quality indicators of health-care processes for depression focus on the provision of pharmacological and psychotherapy treatments. A recent meta-analysis13 of studies using quality indicators developed for the ACOVE project found low achievement of quality indicators for depression in both primary care settings (31%) and in nursing homes (16%). Duhoux et al. 26 assessed adherence rates to 27 quality indicators for depression in 915 adults who consulted a GP, regardless of the reason for the consultation, who also met established criteria for a major depressive episode. Adherence to guidelines was high (> 75%) for around one-third of assessed quality indicators, but less impressive for over half of quality indicators (< 60% achievement). Only 52.2% of the sample received at least one minimally acceptable treatment for depression (i.e. pharmacotherapy or psychotherapy). Patient characteristics that were associated with receipt of indicated care included old age (≥ 65 years), having a comorbid anxiety disorder, having a family physician, having supplementary insurance cover and the increasing severity of the depressive episode.

Scope of the English Longitudinal Study of Ageing

The aim of ELSA is ‘to explore what it means to grow older in the 21st century and to understand which factors account for the variety of observed patterns’. 29 There is, therefore, an emphasis in the interviews on measuring change over time, with some measures being repeated exactly at each wave and others asking about change or giving respondents the opportunity to amend or update previous responses. A core set of modules on specific domains was included in all waves, with certain topics additionally covered at various waves. ELSA aimed to cover a broad array of salient domains, using standardised measures where possible. The modules common to all waves are listed in Table 1.

Data collection methods

Face-to-face interviews were conducted using computer-assisted personal interviews (CAPIs) and took place in respondents’ homes. A self-completion questionnaire was also given to respondents for them to complete and return at their convenience. If there were several respondents for a given household, interviews could be conducted at the same time using concurrent interview techniques, with modules requiring private answers still performed individually. Both the CAPI and self-completion instruments were piloted for each wave approximately 1 year before the main data collection, using a separate sample of households.

Additionally, a nurse visited the respondents in waves 2 (2004) and 4 (2008), and collected anthropometric data and blood samples. The blood samples were analysed for HbA_1c and many other biomarkers.

The CAPI instrument used in ELSA adopted the technique of ‘feeding forward’ data. That is, certain responses that individuals made in previous waves were used to improve consistency of their responses across waves, or to aid recall. For example, a respondent who had previously reported a certain diagnosis would be asked at subsequent waves to confirm that they previously had that diagnosis and whether or not they still had it. Previous responses were also used in the routing of the computerised interviews.

Sample design

Three cohorts of participants make up the ELSA sample. Recruits were sourced from HSE samples, as these were nationally representative. Households were eligible for ELSA if they contained at least one individual from one of the respective core HSE samples who was still living at a private residential address in England at the wave of ELSA at which that cohort was introduced. They also had to be age-eligible and had given permission to be recontacted in the future (though they were not specifically informed of ELSA at HSE). ‘Cohort 1′ was introduced at wave 1 (2002) and was selected from HSE samples in 1998, 1999 and 2001, with people being age-eligible if they were born on or before 29 February 1952. The HSE in 1999 originally included a ‘boost’ sample that represented ethnic minorities, but funding constraints meant that this sample were not included in ELSA. ‘Cohort 3′ was introduced at wave 3 (2006) to replenish the ageing sample with people aged 50–52 years and was taken from HSE in 2001–4. ‘Cohort 4′ was introduced at wave 4 (2008) to both replenish the youngest age range, as in cohort 3, and bolster the sample of people aged 52–74 years that had lost numbers due to participant attrition over the waves. It was sourced from people born between 1 March 1933 and 29 February 1958 who participated in HSE in 2006, which was chosen because it included a nurse visit with measures taken that could be compared with those obtained in the wave 4 nurse visits.

Recruits selected for one of these cohorts were termed ‘core members’ of their respective cohort and remained eligible to take part in ELSA at subsequent waves as long as they had not moved out of Britain or had died. At each wave, partners of core members who lived in the household could also be interviewed. From wave 2 onwards, core members who had moved out of the private sector (e.g. into a nursing home) since the previous wave were approached for an ‘institution interview’. Proxy interviews were attempted if cognitive impairment, physical or mental ill health prevented someone from participating in a face-to-face interview. The person who answered on behalf of the respondent was aged ≥ 16 years and knew enough about the individual’s circumstances to provide the required information. Not all of the core modules were included in the proxy interviews, and we excluded proxies for the purposes of all our analyses. Box 1 summarises the different types of sample member eligible for ELSA. Further details of the sampling design and fieldwork procedures can be found in the ELSA technical reports. 31–34

Participant response rates

The technical reports for the first four waves of ELSA provide a variety of in-depth response rates. 31–34 Here, we report the ‘individual response rates’ from initial HSE samples to the entry wave of ELSA for each of the three cohorts, the total number of core members at each wave and the ‘study response rate’ of cohort 1 core members (the longitudinal sample) across the five waves (Table 2). 67% of individuals determined to be eligible for ELSA from HSE in 1998, 1999 and 2001 took part in ELSA wave 1. 34 Of individuals from HSE in 2001–4 who were eligible for ELSA as part of cohort 3, 60% were interviewed at wave 3. 32 A response rate of 65% was achieved for cohort 4 core members sourced from HSE in 2006. 31 Of the 11,392 core members from cohort 1 interviewed at wave 1, only 5315 (47%) completed all subsequent waves up to and including wave 5. The replenishment samples from cohorts 3 and 4 meant that numbers of all core members ranged from 8780 in wave 2 to 9886 in wave 4.

Longitudinal and cross-sectional samples

To answer our research questions, cross-sectional samples from each wave of ELSA and a longitudinal sample of those who were interviewed at all waves were required. Proxy interviews were excluded, as we required accurate reporting of diagnoses and received care, as were interviews held in institutions, because these would not be representative of those living in private residences. The 5114 ELSA cohort 1 core members who were interviewed at all four waves, excluding proxies and those in institutions, formed our longitudinal sample (Table 5). This represents 45.5% of the original sample of 11,233 who responded at wave 1. The additions of cohorts 3 and 4 kept the cross-sectional samples above 8500 core members each. For analysis of the diabetes care bundle, we used a cross-sectional sample of all core members in 2010–11, who had also been non-proxy participants in 2008–9 (n = 8222).

Composite measures of quality

It is possible to combine individual quality indicators into composites that summarise quality of care at a higher level, such as for a condition or even for all health care in a system. There are both advantages and disadvantages of this approach. Summary measures are useful for conveying a complex range of information to policy-makers and the public, but important variations in individual health-care processes can be masked if combined without care. 36 Composites should, therefore, avoid overgeneralisation and have a specific focus and aim, such as formative quality improvement for specific conditions. 37

‘Care bundles’ are composites that comprise processes of care which should all be received by the eligible population for certain conditions, such as annual monitoring checks for patients with diabetes. The ‘gap’ in performance between individual indicators and the care bundle may be a more valid reflection of variation in expected care delivery. 20

There are a number of different methods for deriving composite measures, with different strengths and weaknesses. Below, we describe the main methods.

Quality indicators in the English Longitudinal Study of Ageing

In total, 32 quality of care indicators were used in ELSA. These quality indicators were adapted from a set of 236 quality indicators originally developed in the USA for the ACOVE project39 for conditions chosen for their prevalence, impact, effectiveness of treatment or prevention, importance in older people, feasibility of measurement and potential for quality improvement. They were based on systematic reviews of evidence of improved outcomes and focus on processes of care rather than health outcomes, as processes are under the control of the health-care system and are not subject to the array of other factors that influence health outcomes. Approximately half of the ACOVE indicators could be used in an interview survey without the use of clinical records, and these were rated for validity by an expert panel of clinicians for validity in patient interview surveys in England, using an adapted form of the RAND Corporation/University of California in Los Angeles appropriateness method for combining evidence with expert opinion. 9,40 The clinical panel reviewed the evidence for 119 quality indicators and scored them on the extent to which they reflected good practice in England. Two rounds of anonymous rating were used, with face-to-face discussion between the rounds. Indicators were excluded if the panel disagreed on their validity. The quality indicators were designed to reflect the minimum acceptable standard of care rather than the optimum, and the panel were informed that they were to be used in a national longitudinal panel survey.

The panel identified 17 of the 119 indicators that were not suitable for use in the UK. Forty-three of the remaining 102 quality indicators were not piloted in ELSA owing to difficulties in implementing them, or because they did not relate to conditions found to be a common cause of morbidity in HSE 2000. A further 27 were not used in ELSA owing to problems found with their implementation in the pilots or main survey. 7 The remaining 32 indicators were incorporated in wave 2 of ELSA, with many also repeated at subsequent waves. These covered 13 conditions: CVD (stroke), depression, diabetes mellitus, falls, hearing problems, hypertension, IHD, osteoarthritis, osteoporosis, pain management, smoking cessation, urinary incontinence and cataracts.

Description of quality indicators and their generation from English Longitudinal Study of Ageing questions

The quality indicators for each condition are described below. Each quality indicator starts with an ‘if’ clause that determines the eligible population, followed by a ‘then’ clause that states the indicated care. The quality indicators specify that they are for those with the relevant condition who are aged ≥ 50 years or over. Some of the quality indicators specify that treatment should be offered ‘at least once’ or relate to treatments that are usually continuous in practice. For these quality indicators, participants who were eligible at the current wave and had reported receiving the indicated treatment at a previous wave were deemed to have also passed the indicator at the current wave. The derivation of each quality indicator follows its description. Appendix 1 shows the translation of each quality indicator and patient-centred question into its corresponding ELSA questions.

Availability of quality indicators across waves

There were changes in the questions included in ELSA between the waves. As a result, not all quality indicators could be generated at all waves. Table 6 displays the availability of the quality indicators. All quality indicators were generated at wave 2. Wave 5 was the next most complete wave, where all quality indicators except those for depression could be generated. Apart from wave 2, depression quality indicators were available only at wave 4, and then only two out of the three could be derived. For diabetes, CVD and osteoarthritis, there were at least two waves at which quality of care could be compared using all quality indicators.

Potential covariates

Possible covariates were selected by a panel consisting of academics, clinicians, researchers and Patient and Public Patient Involvement in Research (PPIRes) members of the team. The choices were based on clinical relevance, feasibility to extract the data from ELSA and previous work. Covariates were grouped into five domains: demographics, health and disability, health behaviours, physical measures, and psychological and social measures. Table 7 lists all covariates that were used in any of the analyses in this work. Any relevant descriptors, explanations and groups are also shown.

Standardised scales

Standardised scales available to use as covariates are described below.

Definition of comorbidities

In our regression analyses we used a measure of comorbid conditions and these were defined as having two or more comorbid conditions as opposed to one or none. This is a commonly used definition of multimorbidity. 48 This variable was created by counting the number of comorbid conditions that every participant had out of a possible 16 conditions, and excluding those conditions closely related to the specific condition. For example, those with osteoarthritis often also had the more general diagnosis of ‘arthritis’, and this was not counted as a condition comorbid to osteoarthritis. For diabetes, hypertension, congestive heart failure, diabetes and high cholesterol were not considered comorbid conditions; for depression, Alzheimer’s disease and dementia were not considered comorbidities. For IHD, hypertension, prevalent angina, prevalent heart attack, heart murmur, arrhythmia, CVD and high cholesterol were not considered comorbid conditions.

Excluded variables

It was agreed by the clinical screening panel that ethnicity should be tested for association with health care; however, owing to < 2% of the sample being non-white this could not be used in the regression analyses. Other covariates that were not included in analyses were country of birth (not specific enough as recoded as UK or non-UK), geography (not available), internet use (missing data), employment status (majority not employed as they were past retirement age) and fasting glucose measurements (not used in QOF as a measure of diabetes). Previous work on ELSA accounted for the clustering and stratification of the participants in the HSE. On advice from the ELSA data manager at the National Centre for Social Research (NatCen), we did not include these variables as the ELSA sample is now effectively unclustered owing to non-response and people moving.

Quality indicator achievement

The entire ELSA sample for this study consisted of all participants who had been interviewed at any of the four waves (n = 16,773). The primary outcome was achievement of each quality indicator, which was the number of times the indicator was achieved divided by the number of times it was triggered, expressed as a percentage, with possible values between 0% and 100%. The quality indicator achievement at the level of each condition (cardiovascular disease, depression, diabetes or osteoarthritis) was the number of times any indicator for that condition was achieved divided by the number of times any indicator for that condition was triggered, expressed as a percentage, with possible values between 0% and 100% (see Table 9).

The percentage change in achievement was calculated between wave 2 and the most recent year of follow-up, and the statistical significance of changes over time were tested with Pearson’s chi-squared test for comparing proportions (see Table 10). Some data points were paired within individuals between wave 2 and the most recent year of follow-up, and some were not, as different individuals were eligible for indicators at different waves. For this reason we did not use McNemar’s test for paired data, although McNemar’s test would have had greater power and precision had it been appropriate to use. We excluded ‘don’t know’ responses from both the numerator and the denominator when calculating quality scores.

We used correlation analysis to investigate whether or not quality of care for one condition was associated with quality of care in another.

Transitions in receipt of care across time

We explored the extent to which participants with diabetes, cardiovascular disease, osteoarthritis and depression transitioned from non-receipt of indicated care to subsequently receiving it, and vice versa, across the available waves of ELSA. To achieve this, we used cross-tabulations of whether or not an indicator was passed at one wave with whether or not it was subsequently passed at the wave at which it was next available. Some quality indicators were only available at one wave other than the baseline (wave 2), either wave 4 or wave 5 (see Availability of quality indicators across waves). As a result, owing to participant attrition and changes in individuals’ health status, sample sizes for these quality indicators were too small for comparisons across waves. Additionally, sample sizes were too small for osteoarthritis referral and osteoarthritis paracetamol was not suitable, as if paracetamol for pain relief was not the first prescribed drug, this could never be rectified. We therefore report the proportions of individuals who successfully transition from non-receipt of care to receiving their care, who transition from receipt of care to subsequent non-receipt of care, and those who do not transition, for stroke antihypertensives, hypertension treatment, diabetes HbA_1c, diabetes proteinuria, diabetes ACE inhibitor, diabetes foot check, all-or-nothing receipt of the care bundle for diabetes, osteoarthritis exercise and osteoarthritis education.

Modelling for longitudinal quality of care

For the modelling we used a longitudinal data set comprising only those participants who had completed interviews at all four waves (n = 5114) (see Table 13). We calculated quality indicator achievement for these 5114 participants in exactly the same way as we calculated it for the entire data set, described in Quality indicator achievement – that is, the quality indicator achievement at the level of each condition was the number of times any indicator for that condition was achieved divided by the number of times any indicator for that condition was triggered, expressed as a percentage, with possible values between 0% and 100% (see Table 15).

For the regression, modelling we needed to calculate quality indicator eligibility by participant rather than by condition, as the covariates were linked to participants rather than to conditions (cardiovascular disease, depression, diabetes or osteoarthritis). Each participant might have been eligible for more than one indicator within each condition, and so we reshaped the data into ‘long’ format in Stata statistical software, with each participant having a separate record for each indicator. These person-indicator level records of quality achievement were the unit of analysis for the modelling, using ‘svy’ commands in Stata ‘long’ format. This was done for each condition separately, and the outcome was expressed as the odds of achievement of any quality indicator that a participant was eligible for, within the relevant condition, with adjustment for participant characteristics. Intraperson correlation of outcomes was accounted for using robust adjustment with Stata, with each participant’s unique identifier included in the regression equation as a cluster variable. All data were weighted for historical non-response and non-response since the last wave (see Table 15).

Age group (50–64, 65–74 and ≥ 75 years) and sex were included in all regression models, and the following were examined as potential explanatory variables, and included in final models if they were significantly associated with the outcome before adjustment for other variables: wealth, marital status, education, socioeconomic status, NHS or independent sector delivery of joint surgery, general health, eyesight, hearing, chronic pain, limiting illnesses, instrumental activities of daily living (IADL) (scale of 1–7), ADL (scale of 1–6), GP access, the Rose Angina Scale,44 cognitive impairment, comorbidities, smoking, eating five portions of fruit and vegetables a day, alcohol consumption, physical activity levels, body mass index (BMI), blood pressure, blood glucose (HbA_1c) levels, self-perceived social status, CASP-19, social detachment, and depression (CES-D).

We categorised BMI according to the World Health Organization criteria,21 blood glucose [HbA_1c ≥ 7.5% (Diabetes Control and Complications Trial: DCCT) or 59 mmol/mol (International Federation of Clinical Chemistry and Laboratory Medicine: IFCC)] and cholesterol (≤ 5 mmol/l or less) according to the QOF criteria,8 and depression according to the CES-D (a score of 4 or more was classified as depression). 22

A single fasting plasma glucose was used to estimate the prevalence of diabetes, rather than the two separate measurements recommended by the World Health Organization. This approach is not recommended for clinical care, but is acceptable at population level and has been used in other studies. 50 It may overestimate diabetes51 but will also not identify the small number of people whose diabetes can be detected only on an oral glucose tolerance test. 52

All characteristics that were significantly associated with quality of care at the 5% level for the conditions using univariate regression analyses were used in the final models.

Predicting non-receipt of indicated care for diabetes care bundles

We used the care bundle approach as an alternative method to model quality of care for diabetes, to explore which characteristics were predictive of not receiving indicated care for diabetes 2 years later. For the diabetes care bundle analyses we used a data set comprising ELSA core members at wave 5 who also responded at wave 4 (n = 8222) (see Table 17). The latest outcomes were deemed to be the most clinically relevant to model, so we used wave 5 outcomes and covariates from wave 4. A covariate of particular interest was whether or not participants had previously received the indicated care.

We therefore had two outcomes: all-or-nothing receipt of the care bundle and whether or not the diabetes ACE inhibitor quality indicator was passed. Separate logistic regression models were developed for each of the two outcomes using the following process: short-listed covariates were first entered into univariate regressions, adjusted for age and sex, and the models were weighted for non-response using the wave 5 cross-sectional sample weights (see Weighting). All characteristics that were significantly associated with quality of care at the 5% level for either outcome were then entered into a forced entry logistic regression, adjusted and weighted as above, for both outcomes (see Table 17).

Patient characteristics associated with the illness burden, diagnosis and treatment pathway

To examine whether or not factors such as wealth are associated with quality of health care received through the care pathway, we defined burden of illness, diagnosis and treatment for four conditions from data generated from the four waves of ELSA. We used the sample of ELSA core members who had been interviewed at any of the four waves (n = 16,773).

Achievement of quality indicators

The eligibility and achievement rates of individual quality indicators for each wave of data for every participant (n = 16,773) show that quality indicator pass rates varied substantially within conditions, with the depression suicide indicator being achieved by 46.5% in wave 2 and treatment by 78.9% at the same wave (Table 9). There was also variation between pass rates at different waves; the greatest percentage increase in quality indicator achievement over the waves was seen in osteoarthritis exercise, where there was a 20% improvement in passing the quality indicator, while the worst performing indicator was depression treatment, with a 17% drop between the two waves where it was measured.

We used quality indicator achievement at condition level to see whether care for specific conditions had increased or decreased over the 6 years of the study (Table 10). A statistically significant difference between wave 2 (2004–5) and wave 5 (2010–11) was seen only for osteoarthritis quality indicators. Although the achievement rates for osteoarthritis show a small increase, they are still very low (36% in 2010–11). We used the chi-squared test for associations, as the number of people with conditions at both time points was small (n = 185 for cardiovascular disease, n = 6 for depression, n = 349 for diabetes and n = 366 for osteoarthritis).

Those who reported ‘don’t know’ were not included in the analyses. The numbers reporting this were extremely low and in wave 2 (total n at wave 2 = 9432) only four did not know if they had been asked about osteoarthritis and paracetamol use. This dropped to two at wave 5 (total n at wave 5 = 10,274). The offer of antihypertensive medication after a stroke was not known by 19 people at wave 2 and by two at wave 5.

Associations of care between conditions

We assessed whether or not quality of care for one condition was associated with quality of care for another condition at each wave that data were available (see Chapter 2, Availability of quality indicators across waves, and Table 6). The only statistically significant association was at wave 2, with quality of care for diabetes and cardiovascular disease correlated (r = 0.13, p = 0.035). No associations between quality of care for other diseases were found at wave 2 (osteoarthritis and diabetes, r = 0.16, p = 0.18; osteoarthritis and cardiovascular disease r = –0.044, p = 0.72; depression and osteoarthritis r = –0.17, p = 0.48; depression and cardiovascular disease r = 0.20, p = 0.41; and depression and diabetes r = 0.43, p = 0.28). At wave 3 quality of care for osteoarthritis and diabetes were not associated (r = 0.078, p = 0.27). At wave 4, depression and osteoarthritis and depression and diabetes care showed no associations (r = 0.18, p = 0.11 and r = 0.037, p = 0.85, respectively). Likewise, there was no relationship between quality of care at wave 4 for osteoarthritis and diabetes (r = 0.075, p = 0.23). At wave 5, similarly, there were no associations found between quality of care in the different diseases (cardiovascular disease and diabetes r = 0.14, p = 0.074; cardiovascular disease and osteoarthritis r = 0.057, p = 0.3; and osteoarthritis and diabetes r = 0.039, p = 0.47)

Achievement of diabetes care bundle quality indicators

This analysis used participants who were core members at wave 5 and had also responded at wave 4 (n = 8222). Eligibility and non-achievements rates of the individual indicators that made up the diabetes care bundle and the one indicator (diabetes ACE inhibitor) were calculated. At least one of the three monitoring checks included in the care bundle was not received in 288 out of 907 cases in 2010–11 [weighted rate = 32.2%, 95% confidence interval (CI) 29.1% to 35.5%]. This is equivalent to achievement of 67.8% for all three quality indicators in the bundle. The HbA_1c test, proteinuria test and foot check quality indicators had similar, low rates of non-achievement (weighted 2010–11 range = 17.4–22.8%), although the ACE inhibitor quality indicator had much higher non-achievement (weighted 2010–11 score = 44.5%). The non-achievements of the individual quality indicators and the care bundle varied little between waves in the study (Table 11).

Transitions in quality of care across time

Transitions from receiving indicated care in one wave to not receiving care in a subsequent wave, and also from not receiving care to subsequently receiving care, were calculated for wave 2 to wave 3, wave 3 to wave 4, and wave 4 to wave 5 for the quality indicators available at every wave (see Chapter 2, Availability of quality indicators across waves) (Table 12). Different patterns were seen for different conditions. Two quality indicators for cardiovascular disease could be derived at every wave: the quality indicator for stroke antihypertensives and the quality indicator for hypertension treatment. The percentage of eligible participants who did not receive the indicated care for stroke antihypertensives in either the earlier or subsequent wave of each wave pairing ranged from 65.4% in waves 2–3 to 73.1% in waves 4–5. Only 26.9–34.6% transitioned to receiving their care at the subsequent wave. For hypertension treatment the percentage of those who were not offered an antihypertensive who then transitioned to subsequently passing the quality indicator was much higher (59.6–69.0%), though 31.0–40.4% continued to not receive the indicated care. In contrast, only 1.5–2.8% of those who were previously offered antihypertensive medication transitioned to not receiving this care at the subsequent wave.

A similar pattern to that seen with hypertension was observed in diabetes HbA_1c, proteinuria and foot check and the care bundle of these three indicators. Of eligible participants who did not receive all indicated care in the bundle at the earlier wave, 51.7–54.5% did transition to receiving the care at the subsequent wave, while 45.5–48.3% continued to not receive the full bundle. Of those who received all three annual monitoring checks at the earlier wave, only 19.4–24.2% subsequently transitioned to not receiving the complete bundle, as 75.8–80.6 received all the indicated care at both waves across the wave pairings. The diabetes quality indicator for ACE inhibitors showed a slightly different pattern. Achievement of this quality indicator was lower (see Table 9), and a greater proportion (63.8–66.4%) of eligible participants who were not offered the indicated medication at the earlier wave continued to not receive the care at the subsequent wave. In comparison, only 17.0–22.6% of those who were offered the indicated medication at the earlier wave subsequently transitioned to not pass the indicator.

As described above (see Achievement of quality indicators), achievements of the osteoarthritis quality indicators were low. Most of the eligible participants who did not receive their indicated care continued to subsequently not receive their care (osteoarthritis exercise 83.8–92.1%; osteoarthritis education 85.0–87.7%). Only a small minority transitioned to passing the quality indicators at the subsequent wave (osteoarthritis exercise 7.9–16.2%; osteoarthritis education 12.3–15.0%).

Demographics for the longitudinal data set

In all four waves 5114 participants were interviewed, and the number with at least one condition of interest in any wave ranged from 1998 to 2961 (Table 13). The mean age of participants was 65 years (range 52 to > 90 years) in 2004–5 and just over half were women. This sample was used to address research question 2: what characteristics are shared by people aged ≥ 50 years who received or did not receive specific items of recommended NHS or non-NHS care for cardiovascular disease, depression, diabetes and osteoarthritis in a national population-based study followed up for 6 years? A detailed breakdown of the whole ELSA sample by covariate and wave is given in Appendix 5.

Associations between quality of care and patient characteristics

We found few statistically significant associations between quality of care and most participant characteristics (Table 16 and Figure 1). This analysis was conducted as described in Chapter 2 (see Statistical methods and Modelling for longitudinal quality of care). For each condition, univariate models tested each of the covariates listed in Table 7 for association with quality indicator achievement at the 5% level of statistical significance. For all covariates that were associated at the 5% level for each condition, we ran a multivariate model that also included age and sex (see Table 16).

FIGURE 1.

Direction and size of associations between quality of care and participant characteristics by condition and logistic regression ORs. OA, osteoarthritis; Q, quintile.

We found univariate associations between CVD and wealth, BMI and CASP-19. In the multivariate model only increased wealth was associated with poorer care for CVD. Hypertension care was associated in the univariate analyses with age, BMI and mean arterial blood pressure. In the final multivariate model only increasing age was associated with better care for hypertension. IHD care showed univariate associations with age, cognitive impairment, physical activity level, mean arterial blood pressure and cohabiting status, and in the final multivariate regression model, only cohabiting status and increased age were significantly associated with quality of care.

Diabetes care was associated with socioeconomic classification, wealth, cognitive impairment, physical activity and cohabiting status in the univariate analyses. In the multivariate model, only being cognitively impaired and living alone were associated with quality of care for diabetes.

Osteoarthritis care was associated with age, pain, BMI, limiting illness, ADL, IADL, frequency of alcohol consumption, physical activity level, CASP and cohabiting status at the univariate 5% level, but only obesity, limiting illness and pain remained significantly associated with better quality of care for osteoarthritis in the multivariate model.

We did not model depression as the relatively small number of participants with depression in the longitudinal data set (see Table 15) led to very small numbers of participants with depression and specific covariates. Instead, we included depression as a covariate in the regression models for the other condition, and, notably, depression was not associated with worse quality of care. The full results from the multivariate regression models, including statistically non-significant results, are reported in Appendix 6.

Demographics for the diabetes care bundle

The sample for modelling associations between participant characteristics and achievement of the diabetes care bundle consisted of ELSA core members at wave 5 who had also responded at wave 4 (n = 8222). At wave 5, 907 (11.0%) of this sample had diabetes and were eligible for at least one quality indicator in the diabetes care bundle and 759 (9.2%) had diabetes with additional cardiac risk factors and were eligible for the diabetes ACE inhibitor quality indicator. Participants in these two groups tended to comprise more males than the whole sample, had more respondents in the 75-and-older age group and fewer in the youngest group, were more likely to be from the lowest socioeconomic classification and had a greater proportion with no educational qualification (Table 17).

Demographics

The sample of participants aged 50 years or over who had responded to at least one wave of ELSA from wave 2 in 2004–5 until wave 5 in 2010–11 and were core members was used to address research question 3: are factors such as wealth, or patient education, physical activity, obesity, coexisting other chronic conditions or practical factors such as good eyesight or marital status independently associated with the quality of health care received, from symptoms to diagnosis and, finally, to treatment?

There were 12,765 core members in the sample, and 8729 in wave 5, of whom slightly over half were women, and slightly over half were aged ≥ 60 years (Table 20). Diagnosed diabetes, angina and osteoarthritis were more prevalent with increasing age, but the opposite trend was seen with depression, with 10% of the youngest age group reporting a diagnosis compared with only 3% in the oldest age group. The highest prevalence of all conditions was seen in the poorest quintile. For example, the prevalence of depression was 4% in the wealthiest quintile and 11% in the least wealthy.

There was very little variation in the prevalence of illness burden, diagnosis and treatment between the waves for each of the four conditions (Table 21).

Modelling with illness burden, diagnosis and treatment

The hypothetically poorest participant had higher odds of illness burden than the hypothetically richest for each of the four conditions at all four time points (ORs ranging from 3.9 to 16.0) (Table 22). The odds were generally smaller for diagnosis (ORs 0.6 to 5.3) and appropriate treatment (ORs 0.2 to 3.3). The substantially higher illness burden in less wealthy participants was not matched by higher levels of diagnosis and treatment (Figure 2).

FIGURE 2.

Illness burden, medical diagnosis and treatment of angina, depression, diabetes and osteoarthritis, comparing the least wealthy with the most wealthy: overall ORs (adjusted for age and sex) and 95% confidence bars (logistic regression).

For angina, the overall OR for meeting the criteria for illness burden was 7.6, indicating that the hypothetically least wealthy individual was almost eight times more likely to have angina symptoms (defined by the Rose Angina Scale) than the wealthiest. The OR for reported medical diagnosis was 4.5, suggesting that some less wealthy people with angina symptoms had not received a diagnosis of angina, as the expected OR for equitably distributed diagnosis would be 7.6. The OR for treatment was 3.2, and again the expected OR for equitably distributed treatment would have been around 8.

For depression, the overall OR for illness burden was 6.4, while for medical diagnosis it was 3.3 and for treatment it was 2.6, again suggesting that some poorer people with illness burden of depression were less likely to have received a diagnosis or indicated health care.

For diabetes, the overall ORs for illness burden were 4.2 and 4.0 for diagnosis, suggesting that diabetes diagnosis was distributed equitably. However, the OR for treatment was 0.9 and not statistically different from 1, again suggesting that some less wealthy people with medically diagnosed diabetes had not received treatment, as the expected OR for equitably distributed treatment would have been around 4.

We also calculated the ORs of receiving a diagnosis in a subsequent wave, only for those who were estimated to have had the relevant long-term condition in a previous wave, and then the likelihood of receiving treatment, only for those who had received a medical diagnosis in a previous wave. This is a secondary analysis as the numbers of participants that could be followed over time in this manner was small, particularly for treatment in angina and depression, and so these results should be interpreted with great caution. The substantial inequalities in the estimated prevalence of conditions by wealth are similar to Table 23, as expected, and subsequently the numbers are too small for the results to be statistically significant. The general pattern, however, is consistent with less medical diagnosis and treatment in less wealthy participants. Taking osteoarthritis as an example, ORs were 12.7 (95% CI 9.1 to 17.8) for estimated prevalence in wave 3, 0.7 (95% CI 0.3 to 1.8) for diagnosis in wave 4 and 0.7 (95% CI 0.3 to 1.6) for treatment in wave 5 (Table 23).

Quality indicator achievement

Our findings are broadly consistent with the findings of a systematic review by Askari et al. 13 of 17 papers reporting on achievement of the ACOVE quality indicators, which showed large variation between conditions, and notably low rates for osteoarthritis quality indicators. This review only found a few conditions with a stable pass rate range over multiple studies, and noted that caution is required in making direct comparisons of quality indicator achievement between studies. Most indicators were reported in only a small number of studies and were drawn from different countries and diverse populations (e.g. nursing home residents or primary care settings).

For the cardiovascular disease indicators reported here, the proportion of eligible patients who reported receiving care ranged between 64% for MI beta-blocker and 100% for INR test in wave 2. By wave 5 these figures ranged between 73% for MI beta-blocker and 94% for INR test. Improvements were observed in the achievement rates for some indicators between waves 2 and 5 (stroke antihypertensives +7%, hypertension treatment +8%, MI beta-blocker +12%). However, achievement rates of the remaining indicators stayed constant across time with the exception of the INR test, which declined by 6% from the 100% attainment noted at wave 2. Askari et al. 13 reported comparable achievement rates in primary care settings for hypertension (72% vs. 75% prevalent hypertension treatment at wave 5) and stroke (69% vs. 85% achievement in ELSA) indicators.

For the diabetes indicators, achievement rates in wave 2 ranged between 48% for ACE inhibitor and 83% for diabetes HbA_1c testing. By wave 5 these figures ranged between 57% for ACE inhibitor and 89% for diabetes cholesterol test. Most diabetes indicators remained largely unchanged across time, with the exception of the ACE inhibitor one, which increased by 9% between waves 2 and 5. Our findings are broadly consistent with figures reported for 2009–10 in the recent National Diabetes Audit18 which examined all available clinical records in the UK. This audit found achievement rates of 93%, 74% and 85% for HbA_1c checks, renal risk checks and foot checks, respectively. This compares with wave 5 data, where 82% patients reported having HbA_1c levels checked, 79% had a proteinuria check and 83% received a foot check.

Receipt of care for depression at wave 2 was 46% (suicide), 79% (treatment) and 76% (review), and at wave 4 was 58% (suicide) and 62% (treatment). Although depression is a condition included in the QOF for UK general practice, the QOF indicators do not map perfectly to the criteria used in ELSA and no direct comparisons can be made.

Achievement of the osteoarthritis indicators was lower than for any cardiovascular disease, depression or diabetes indicators, ranging from 26% (education) to 45% (paracetamol) at wave 2, and from 32% (education) to 46% (exercise) at wave 5. Although a prevalent condition in later life, osteoarthritis is not included in the QOF. For both depression and osteoarthritis we found no national audit data that could be compared with our sample. The review by Askari et al. 13 found very similar rates of achievement of osteoarthritis indicators (interquartile range 29% to 41%), which they described as ‘notably low’.

Transitions in quality indicator achievement over time

We explored transitions in quality indicator achievement across time by tabulating whether or not patients received their care at a given wave with subsequent receipt of that care by the next wave. This allowed us to investigate the rates of transitions from poor-quality care to good-quality care and vice versa, as well as the rates of continued receipt of poor- or good-quality care. Not all quality indicators were suitable for this analysis as some indicators (e.g. depression) referred to the management of incident disease within specific time scales.

For other quality indicators it was not possible to transition from receiving care at one wave to subsequently not receiving it, as the indicated care either only needed to be provided once (osteoarthritis exercise and education), or referred to the provision of long-term treatment (stroke antihypertensives). However, there were relatively low rates of transitions from not receiving the care to subsequently receiving it, especially for the two osteoarthritis quality indicators. The majority of eligible people who did not receive indicated care continued to receive poor quality of care subsequently.

A different pattern emerged from the follow-up for the hypertension quality indicator, the three quality indicators for annual monitoring checks for diabetes (HbA_1c test, proteinuria and foot check) that compose the diabetes care bundle, and the quality indicator assessing the prescription of ACE inhibitors to those with diabetes and additional cardiac risk factors. For these quality indicators, more participants who did not receive their care at a given wave transitioned to subsequently receiving it than to continuing to not receive it. There were comparatively low rates of patients transitioning from receiving good quality of care to not receiving their care (e.g. 2% to 3% for hypertension and 19% to 24% for the diabetes care bundle) in relation to those who continued to receive poor-quality care (e.g. 31% to 40% for hypertension and 45% to 48% for the diabetes care bundle). Overall, our results suggest that those participants who received poor care tended to persistently receive poor care, and that time alone did not rectify deficits in care.

Participant characteristics associated with receipt of indicated health care

The results of multivariate logistic regression modelling for individual conditions found that very few of the wide range of personal characteristics that we measured were significantly and independently associated with eligible individuals achieving indicated care. At condition level, some participant characteristics were associated with achievement of care. For example, achievement of diabetes care was higher for patients without cognitive impairment, or for patients who lived alone than those cohabiting. In contrast, care achievement for patients with osteoarthritis was associated with participants reporting severe pain or having a limiting illness, or being obese compared with their healthier counterparts.

Modelling achievement of the care bundle for diabetes

This analysis included 8222 ELSA participants at wave 5 who also responded at wave 4, of whom 11% had diabetes. Despite low non-achievement for the individual diabetes quality indicators (17% to 23%) in wave 5, we found that at least one of the three bundled annual monitoring checks was not achieved in one-third of eligible patients in 2010–11 (33%). Our figure is somewhat lower than data from the National Diabetes Audit in 2009–10,18 which found that 47% of patients did not receive all nine National Institute for Health and Care Excellence quality indicated diabetes care processes.

Only a limited number of variables were found to be independently and significantly associated with receipt of incomplete care. Previous (2008–9) non-achievement of the care bundle was the best predictor of subsequent non-achievement 2 years later, increasing the odds of this by over threefold. Patients who had previously reported in 2008–9 that they did not know ‘everything about managing their diabetes’ or who were cognitively impaired were more likely to subsequently report not having a complete diabetes care bundle. Individuals who were single or had low health literacy were more likely to report not having been offered an ACE inhibitor. In contrast, people in the oldest age group, or classified as obese according to their BMI category, were more likely to have been offered an ACE inhibitor.

The National Diabetes Audit modelled, in a cross-sectional sample, the impact of patient characteristics (available from routine records) on the likelihood of receiving a complete bundle of care. 18 Common to both studies is the finding that only a limited number of patient characteristics explained variability in the receipt of incomplete care. Patient social deprivation, sex and duration of diabetes were not found to affect the likelihood of receiving complete care, but age and ethnicity were. Significant geographical variation was observed across the 151 primary care trusts that contributed data.

The variations in indicator achievement by condition were much larger than those by participant characteristics. This, together with the findings from the diabetes care bundle analysis, suggests that the major causes of variations in health-care quality may lie elsewhere than with the patient, although ELSA could not account for all patient factors which contribute to variation and are difficult to capture in a survey, such as patient behaviour. We were unable to model health service performance data as they are not in ELSA, although participants were asked to consent to linkage of ELSA data to NHS data (e.g. Hospital Episode Statistics), which opens the possibility for future research on linked data sets.

Distribution of illness burden, diagnosis and treatment

This analysis suggested that barriers to equitable receipt of health care according to need may exist predominantly at the stage at diagnosis rather than treatment. For angina, depression, diabetes and osteoarthritis, the hypothetically poorest participant always had higher odds of illness burden than the hypothetically richest. Poorer participants usually had no higher odds for subsequent receipt of diagnosis or treatment. This serial cross-sectional analysis of four waves of ELSA included all eligible participants in each wave in order to maximise the sample size. This approach means that some participants with a diagnosed condition would no longer have had illness burden, if they were being successfully treated. Examples would be diabetic participants whose blood sugar levels were being successfully controlled by treatment, and participants with successfully treated depression. We checked our main results with the secondary (longitudinal) analysis, which assessed subsequent diagnosis in those with illness burden, and subsequent treatment in those with illness burden, and found a similar picture of inequalities in illness burden not being fully addressed by appropriately greater levels of diagnosis and treatment for poorer participants.

Care-seeking behaviour and patient preferences may differ with wealth. Given the same information, patients may be less likely to seek treatment than their doctors53,54 and pessimism about availability of treatment may make older people reluctant to seek help. 55 People from poorer backgrounds may view living with illness burden (such as pain or emotional problems) as a normal part of ageing rather than a disease. 56 It is possible to imagine that the response of the primary care physician may also vary according to the wealth of the patient, however subconsciously. For example, a physician might initially consider symptoms such as aches and pains to be part of the tapestry of life, or the natural ageing process, when they occur in a patient from a deprived background. The same symptoms in a wealthier, well-educated and articulate patient might arguably be considered more unusual and, therefore, more likely to lead to a diagnosis. Our results fit with research showing that while people from deprived groups were more likely to have Rose angina, they were not more likely to receive a GP diagnosis of CHD. 57 Comorbidity is more common in deprived populations, and may make diagnosis of all conditions harder for doctors within the constraints of a short consultation. 58

Wave 1

Coded 0 if they mention having no problem (i.e. refusals and don’t knows are excluded)

make w1_dim01_d = 0 if w1_hedim01=96

label variable w1_dim01_d ‘Hypertension’

make w1_dim02_d = 0 if w1_hedim01=96

label variable w1_dim02_d ‘Angina’

make w1_dim03_d = 0 if w1_hedim01=96

label variable w1_dim03_d ‘Heart attack’

make w1_dim04_d = 0 if w1_hedim01=96

label variable w1_dim04_d ‘Congestive heart failure’

make w1_dim05_d = 0 if w1_hedim01=96

label variable w1_dim05_d ‘Heart murmur’

make w1_dim06_d = 0 if w1_hedim01=96

label variable w1_dim06_d ‘Abnormal heart rhythm’

make w1_dim07_d = 0 if w1_hedim01=96

label variable w1_dim07_d ‘diabetes or high blood sugar’

make w1_dim08_d = 0 if w1_hedim01=96

label variable w1_dim08_d ‘Stroke’

replace w1_dim01_d = 1 if (w1_hedim01=1 or w1_hedim02=1 or w1_hedim03=1 or w1_hedim04=1 or w1_hedim05=1 or w1_hedim06=1 or w1_hedim07=1)

replace w1_dim02_d = 1 if (w1_hedim01=2 or w1_hedim02=2 or w1_hedim03=2 or w1_hedim04=2 or w1_hedim05=2 or w1_hedim06=2 or w1_hedim07=2)

replace w1_dim03_d = 1 if (w1_hedim01=3 or w1_hedim02=3 or w1_hedim03=3 or w1_hedim04=3 or w1_hedim05=3 or w1_hedim06=3 or w1_hedim07=3)

replace w1_dim04_d = 1 if (w1_hedim01=4 or w1_hedim02=4 or w1_hedim03=4 or w1_hedim04=4 or w1_hedim05=4 or w1_hedim06=4 or w1_hedim07=4)

replace w1_dim05_d = 1 if (w1_hedim01=5 or w1_hedim02=5 or w1_hedim03=5 or w1_hedim04=5 or w1_hedim05=5 or w1_hedim06=5 or w1_hedim07=5)

replace w1_dim06_d = 1 if (w1_hedim01=6 or w1_hedim02=6 or w1_hedim03=6 or w1_hedim04=6 or w1_hedim05=6 or w1_hedim06=6 or w1_hedim07=6)

replace w1_dim07_d = 1 if (w1_hedim01=7 or w1_hedim02=7 or w1_hedim03=7 or w1_hedim04=7 or w1_hedim05=7 or w1_hedim06=7 or w1_hedim07=7)

replace w1_dim08_d = 1 if (w1_hedim01=8 or w1_hedim02=8 or w1_hedim03=8 or w1_hedim04=8 or w1_hedim05=8 or w1_hedim06=8 or w1_hedim07=8)

Wave 2

For dim variable use the same coding as wave 1

For dia variable use the same coding as wave 1

For dib variable use the same coding as wave 1

For opt variable use:

make w2_opt1_d=0 if w2_heopt1=96

label variable w2_opt1_d ‘w2 glaucoma’

make w2_opt2_d=0 if w2_heopt1=96

label variable w2_opt2_d ‘w2 diabetic eye disease’

make w2_opt3_d=0 if w2_heopt1=96

label variable w2_opt3_d ‘w2 macular degeneration’

make w2_opt4_d=0 if w2_heopt1=96

label variable w2_opt4_d ‘w2 cataract’

make w2_opt96_d=0 if w2_heopt1=96

label variable w2_opt96_d ‘w2 no eye dis’

replace w2_opt1_d=1 if (w2_heopt1=1 or w2_heopt2=1)

replace w2_opt2_d=1 if (w2_heopt1=2 or w2_heopt2=2)

replace w2_opt3_d=1 if (w2_heopt1=3 or w2_heopt2=3)

replace w2_opt4_d=1 if (w2_heopt1=4 or w2_heopt2=4)

replace w2_opt96_d=1 if (w2_heopt1=96)

Wave 1

Wave 2

NOTE: there was an error at w2 whereby not everyone with a cardiovascular disease at w1 was fed forward to w2. This makes ascertaining prevalence rates for w2 CVD conditions tricky. We decided that we would use report of the condition at w1 or w2 for prevalence of the typically chronic CVD conditions: hypertension, heart attack, heart murmur, heart failure, diabetes and stroke. For angina and abnormal heart rhythm, we will need to use the feed forward variable ‘confirms diagnosis from previous wave’ (hediac) and ‘still has condition from previous wave’ (hedias) for the conditions respectively, because many reporting the condition at the previous wave may not still have it at w2.

Wave 3 onwards

Wave 2

dk=don’t knows

ref=refusals

Patient-centered care

Subsequent waves

Wave number is named X, one wave before is X-1, 2 waves before is X-2 and three waves before is X-3.

NB: not all indicators can be generated at every wave.

Patient-centred care

• Diabetes training: waves 2 and 4.

• Diabetes knowledge: waves 2 and 4.

• Cholesterol explanation: wave 2 only.

• Cholesterol preferences: wave 2 only.

• Blood pressure explanation: wave 2 only.

• Blood pressure choice: wave 2 only.

• Osteoarthritis treatment purpose: waves 2 and 4.