Patient activation in older people with long-term conditions and multimorbidity: correlates and change in a cohort study in the United Kingdom

Patient activation is defined as how well a patient understands their own role in the healthcare process; and their level of knowledge, skill, and confidence in managing their own health [11]. Hibbard et al. (2005) describes four stages or levels of patient activation which range from those who are passive recipients of care, unable to self-manage, to those who are effective self-managers but may need some additional support during times of stress or crisis [12].

Improving patient activation is increasingly seen as an important component of new strategies to reform healthcare and improve health outcomes [13]. There is a growing body of evidence that patients who are more ‘activated’ have better health outcomes as well as an improved care experience [8]. In a US sample of mostly female patients with moderate to severe depression, level of patient activation was found to predict remission or improvement from depression over 12 months [14]. Depressed patients who were better activated were also more likely to engage in healthy behaviour changes such as quitting smoking and losing weight, and more likely to attend smear tests and mammogram screening [14].

NHS England (the body of the UK Department of Health which plans and oversees delivery of the English NHS) has identified patient activation as a potentially useful tool in the development of services to meet the needs of patients with long-term conditions [15]. Despite the interest in patient activation, there is relatively little data from UK patients. We use an existing cohort of patients aged 65+ with long-term conditions to describe levels of patient activation in the cohort, and to explore factors that predict change in activation over time.

We used the 13 item version of the Patient Activation Measure (PAM) [13]. The PAM consists of 13 statements relating to patients beliefs about health care, confidence in their management of health related tasks, and self-assessed knowledge of their condition [13]. For each statement patients are required to say how much they either agree or disagree on a response scale of 1–5, where 1 represents “strongly disagree”, 4 represents “strongly agree” and 5 indicates that the statement is “not applicable” to them. A standardised spreadsheet in excel is used to score the PAM. Responses are used to generate a continuous score from 0 to 100 where higher scores indicate that the patient is more activated [13]. Where participants have answered that a statement is not applicable to them the data is treated as missing. A total score is generated where participants have answered at least 10 out of the 13 questions.

The continuous PAM scores are then categorised into four levels for descriptive purposes using the standardised excel spreadsheet. Those who fall into Level 1 are defined as passive recipients of care who do not understand that they can play an active role in their own healthcare. Level 2 includes patients who lack the basic knowledge and confidence to effectively self-manage (for example they may not understand the treatment options available to them or what their medications do). Level 3 includes those who have a basic knowledge about their health but they lack the confidence and skills to engage in positive self-management behaviours. Level 4 is for patients who have the knowledge and confidence to self-manage but who may need support during times of personal stress or health crisis [12].

The PAM has been found to be a valid and reliable measure in people with long-term conditions, such as those attending cardiac rehabilitation,[11] and in those patients with multimorbidity [16]. We used the continuous PAM score in the analyses.

Analysis was restricted to patients for whom a valid PAM score could be derived. In the baseline questionnaire there was an error in the presentation of the PAM questionnaire in that the ‘not applicable’ option was omitted from eight out of the 13 items. However, ‘Not applicable’ was only chosen by 5 % of respondents where it was available as an option and therefore the impact of this is likely to be small.

There was missing data for some patients across some independent variables. To address this, for cases where no more than one independent variable across the full set was missing, logistic regression (linear, binary logistic, ordinal logistic or multinomial as appropriate), was used to impute the missing information, using the other eight variables, plus the baseline PAM score, as predictors. In so doing, 513 extra patients were included in the analyses. Missing data on the PAM questionnaire was not imputed.

Descriptive statistics are used to characterise the distribution of PAM scores and PAM levels, and changes in activation scores over time. Multiple linear regression was used for all analyses. Initially, we investigated associations between baseline PAM scores and independent variables. We examined Variance Inflation Factors (VIF) across all models to eliminate collinear variables. All VIF were below 10 and no variables were dropped from the analyses on the basis of VIF. Both raw and standardised regression coefficients are presented to determine the relative importance of each independent variable in addition to the absolute effect. Prospective multiple regression analyses considered whether these independent variables were predictive of change in PAM scores over the 6 months between the baseline and follow-up questionnaires. We used analysis of change scores as this approach is less bias than analysis of outcome at 6 months for use with observational data [35]. All regression analyses were clustered by general practice.

We repeated the analyses outlined above for the subset of patients with at least two long-term conditions (multimorbidity). The MULTIPleS scale was included as an additional independent variable in these analyses. Mean item imputation was used in order to calculate a MULTIPleS score for those patients with two or fewer items missing out of the original sixteen MULTIPleS items.

The flow of participants is shown in Fig. 1. At baseline 4377 out of 12,989 (33.6 %) people returned a questionnaire. At 6 month follow up, 4225 were eligible to be mailed and 3390 (80.2 %) were returned.

Participant characteristics at baseline are shown in Table 1. The majority of participants were female, White British, and retired or not otherwise working. Symptoms of depression and anxiety were reported by 42.6 % of participants and 60.1 % reported having 2 or more long-term conditions.

A PAM score could be computed for a total of 3636 (83.1 %) participants at baseline. The mean PAM score was 60.7 (sd 15.4) with the majority of participants reporting PAM Level 3 (n = 1685, 46.3 %). Table 2 shows participant characteristics by PAM level.

The overall correlation between baseline and follow up PAM score was r = 0.43, p = 0.00. For those participants who had a complete PAM score at baseline total PAM scores varied between baseline and follow-up, decreasing or increasing by up to 50 points (Table 3). For over half of participants (52 %, n = 1392), PAM level remained stable between the baseline and 6 month follow-up questionnaires, increasing for 27 % (n = 718) of participants and decreasing for 21 % (n = 571). Participants who scored at PAM level 4 on the baseline were those more likely to remain at level 4 than to decrease by 6 month follow-up. Those who scored PAM levels 2 and 3 at baseline were the groups most likely to increase their scores at follow-up with 40 % (n = 188) at level 2 reaching level 3 at follow up and another 10 % moving up 2 levels to level 4 (Table 3).

At baseline, patient activation was significantly lower in older patients, those with depression, and those with poor health literacy. Patient activation was higher in those with good quality of life, living alone and with better social support (Table 4). Depression had the strongest association with patient activation. Any self-reported impairment with health literacy was associated with lower activation scores; there was little difference in magnitude of effect between rarely needing help when reading medical literature and always needing help.

In the subset of participants who self-reported 2 or more long-term conditions, depression, being older, impaired health literacy, HRQoL, social support, number of self-reported comorbidities, and a greater perceived impact of multimorbidity, were all significantly associated with patient activation scores at baseline. Living alone and being retired or not working were not significant for those with multimorbidity (Table 4).

Multivariate regression analysis of change in patient activation scores over 6 months across the whole sample did not identify any significant predictors of patient activation. In the subsample of patients with multimorbidity being depressed and being retired were significant predictors of change in patient activation scores (Table 5).

Although pilot data is being collected on the use of the PAM, there is only limited published data on mixed samples of less than 400 patients [36]. Using a large cohort of older patients with long-term conditions, we explored demographic and health related factors associated with patient activation scores. We found that the most prominent factor associated with patient activation is depression in both the group of patients with one long-term condition and the subgroup of patients’ with multimorbidity. Other factors associated with patient activation were: older age, being retired, poor health literacy, health-related quality of life, and social support. The number of self-reported comorbidities and perceived impact of multimorbidity were also important for those with more than one long-term condition. However, the effects of all factors (except depression) were small implying that their impact on patient activation may not be clinically significant.

We explored predictors of change in patient activation scores over 6 months. We did not identify any significant predictors across the whole sample. However, in the subset of patients with multimorbidity both depression and being retired or not working significantly predicted change in patient activation scores.

Study strengths included a large sample size, comprehensive measurement of patient-reported factors, and longitudinal measurement allowing assessment of predictors of change in patient activation scores.

The cohort achieved a 34 % response rate. This is similar to other studies in these populations using similar methods, [25, 34, 37] but leaves significant potential for non-response bias. The lack of data on non-respondents makes it difficult to assess direction and magnitude of bias. The data are therefore not a strong basis for assessments of prevalence, and data such as the proportions at each level of patient activation should be used with caution. Assessments of the relationships between measures may be less vulnerable to non-response bias. Completion of follow up was over 80 % and thus the potential for bias here is less, although some scales (such as MULTIPleS) did suffer additional missing data. The cohort is also limited to older patients and those living in one area in the UK, which means that the results may not generalise. Long-term conditions were self-reported and we did not confirm diagnoses with medical records.

Previous work in a cross sectional study found lower patient activation scores to be associated with symptoms of depression and poor quality of life [38]. Our study extends this by showing that depression predicts change in patient activation scores over 6 months. A longitudinal study found that patients with symptoms of depression, even when subclinical, were less likely to improve their activation scores or to engage in self-management behaviours over 6 months [39]. A large longitudinal study of patient activation in US patients with moderate to severe depression found that those with more severe depression tended to be less activated and that higher patient activation at baseline predicted greater reduction in depression over 1 year [14]. Those patients with the highest patient activation scores at baseline had over twice the odds of remission from depression.

In a cross sectional survey of US patients with multiple sclerosis, patients’ level of education, as well as depression, was found to be associated with patient activation [40]. Lower levels of education have previously been shown to be associated with poor health literacy [41] which we found to be significantly associated with patient activation in our sample of older patients. A large Danish cohort study of over 29,000 patients has shown that people with long-term conditions have greater difficulties in understanding health information than the general population and have greater problems choosing and engaging with health care [42, 43].

We found that the total number comorbid long-term conditions was associated with patient activation in the short term, but did not predict change in activation prospectively. A smaller US study of 850 patients with multimorbidity over the age of 65 also found that number of comorbid conditions was not related to patient activation [16]. Although a crude count of the numbers of conditions may not predict activation, we found that patient experience of multimorbidity, as measured by MULTIPleS, was associated with patient activation and had the second strongest association after depression.

Social support had a strong association with patient activation across all patients. In a large cross sectional survey of Danish patients with Type 2 diabetes greater contact with friends was associated with higher patient activation and having a poor functional social network was associated with lower patient activation [44].

We found that living alone was associated with higher patient activation in the short term. A systematic review of studies on chronic illness self-management in people living alone found that patients who live alone often ensure that they actively attend to their health needs as they are required to manage their health independently and often have no immediate help available to them [45]. Furthermore, those who live alone have been shown to value social support outside the home and may actively seek and maintain better social connections than those with greater support in the home [45].

We found patient activation scores at 6 months were moderately correlated with baseline scores, with fairly significant movement between levels. It is complex to interpret this change data, as it is not known what clinical or other health and social care interventions were received by patients. The entire local health economy was undergoing a major re-organisation for this age group, but linking that to changes in patient scores would be challenging. The data do suggest some caution in using patient activation scores for very specific targeting of interventions, given patient movement between levels of activation.

Our data suggest a cluster of characteristics that are associated with low activation, mainly depression but also other demographic, social, educational, and health factors. If effective interventions are developed in the UK to improve patient activation, these patients, and particularly those with depression, may be usefully prioritised.

Additionally, the changeable factors which are associated with patient activation may suggest the potential content of interventions. For example, interventions or programmes will need to provide assistance with health literacy (through education) and depression symptoms (through web-based support or cognitive-behavioural therapy interventions), and may usefully involve a component around social support, particularly around increasing functional social support. The associations with patient experience of multimorbidity (MULTIPleS) highlight the potential importance of the assessment of this issue in patients [46].

This study is unique because it presents the first assessment of depression and patient activation in older adults with long-term conditions. We found that depression was significantly associated with patient activation and, for those patients with multimorbidity, it predicted change in patient activation scores over 6 months. However, the opposite pattern of findings has been shown in the literature; patient activation predicted remission, response and changes in depression scores over 12 months in a large US sample of patients with moderate to severe depression [14]. Thus, an unresolved research challenge is to fully understand whether lower patient activation precedes depression or depression leads to lower levels of patient activation. The examination of the temporal links between depression and patient activation as part of a longitudinal study with multiple follow-up assessment points is a fruitful approach to disentangle the relationship between depression and patient activation. This knowledge has the potential improve the effectiveness of self-management interventions by timely and appropriate management of depression and patient activation.

Moreover, much of the research to date on patient activation has assessed the degree to which it is associated with key outcomes, such as quality of life and costs. Such might suggest two more specific causal functions of activation.

First, activation may be a mediator of other outcomes, [8] driving other outcomes. Intervening to improve activation may have effects on other outcomes further down the causal pathway. For example, a self-management intervention that raises a patient’s level of activation may help them to better manage their long-term condition in the short-term, leading to longer-term reductions in costs and quality of life. Methods to assess mediation are available, although they are complex, and often most usefully done in the context of a trial [47].

Secondly, activation may be a moderator of other outcomes [8]. If this is the case, then levels of activation will influence the benefit that patients report from other interventions, and assist in targeting. For example, a low-intensity web-based self-management intervention may demonstrate good outcomes with patients with a higher level of activation, but may be ineffective with those with lower activation, where a more intensive intervention with greater professional input may be required. Again, displaying such effects is possible but complex, and a rigorous demonstration may require subgroup analyses in a large trial or pooled analyses across several trials [48]. The high levels of change in patient activation may limit is utility as a moderator.