Patient Education and Counseling 87 (2012) 67–73

Self Management

A secondary analysis of the moderating effects of depression and multimorbidityon the effectiveness of a chronic disease self-management programme

Mark Harrison a, David Reeves a, Elaine Harkness a, Jose Valderas b, Anne Kennedy a,Anne Rogers a, Mark Hann a, Peter Bower a,*a National Institute for Health Research School for Primary Care Research, Health Sciences Research Group, University of Manchester, UKb Department of Primary Health Care, University of Oxford, Oxford, UK

A R T I C L E I N F O

Article history:

Received 14 January 2011

Received in revised form 5 May 2011

Accepted 10 June 2011

Keywords:

Multimorbidity

Chronic care

Depression

Health behaviour change

Randomized trial

Self-management

A B S T R A C T

Objective: Patients accessing the chronic disease self-management programme (CDSMP) often report

multiple long-term conditions (multimorbidity). Although multimorbidity often predicts poor

outcomes, CDSMP effectiveness may be enhanced in multimorbidity via synergies between self-

management for different conditions. This study assessed whether CDSMP benefits varied by patterns of

multimorbidity.

Methods: The study was based on a secondary analysis of an RCT. Patients with long-term conditions

(n = 629) were randomised to CDSMP or wait-list and completed baseline and 6 month assessments. We

identified four multimorbidity groups: (1) single physical condition; (2) multiple physical conditions; (3)

single physical condition plus ‘probable depression’; (4) multiple physical conditions plus ‘probable

depression’.

Results: Multimorbidity group significantly moderated the effect of CDSMP on vitality, health-related

quality of life, and mental well-being, with the greatest benefit found for patients with multiple physical

conditions plus ‘probable depression’.

Conclusion: The coexistence of depression and multiple physical conditions is associated with increased

illness burden, but such patients benefit more from the CDSMP. The mechanisms underlying this effect

are unclear, but it does not appear to be through self-management or self-efficacy.

Practice implications: The presence of multimorbidity in combination with depression may be a useful

criteria for referral to the CDSMP.

� 2011 Elsevier Ireland Ltd. All rights reserved.

Contents lists available at ScienceDirect

Patient Education and Counseling

jo ur n al h o mep ag e: w ww .e lsev ier . co m / loc ate /p ated u co u

1. Introduction

An ageing population and improvements in health care meansthat the global burden of disease is shifting to long-term conditions[1], and there is increasing interest in improving health bychanging patient behaviour through self-management support [2].Partly this derives from concerns about the costs of health careutilisation, and partly from data suggesting that self-managementis a core driver of improved outcomes in long-term conditions [3].In the United Kingdom, self-management is defined as:

‘the care taken by individuals towards their own health andwell being: it comprises the actions they take to lead a healthylifestyle; to meet their social, emotional and psychological

* Corresponding author at: National Institute for Health Research School for

Primary Care Research, Health Sciences Research Group, Manchester Academic

Health Science Centre, Williamson Building, University of Manchester, M13 9PL,

UK. Tel.: +44 161 275 7638; fax: +44 161 275 7600.

E-mail address: peter.bower@manchester.ac.uk (P. Bower).

0738-3991/$ – see front matter � 2011 Elsevier Ireland Ltd. All rights reserved.

doi:10.1016/j.pec.2011.06.007

needs; to care for their long-term condition; and to preventfurther illness or accidents’ [4]

Group-based interventions such as the chronic disease self-management programme (CDSMP) are a key source of self-management support [5–7]. These interventions are designed toteach a range of self-management skills and techniques, and toenhance patient self-efficacy in the management of long-termconditions and their functional and emotional sequelae. TheCDSMP can be used with patients with a variety of conditions.Secondary analyses have identified patient characteristics thatpredict benefit (i.e. moderators of treatment effect) [8,9]. Our recentanalysis found that patients reporting lower baseline self-efficacyand quality of life demonstrated more positive health outcomesfollowing the CDSMP [10].

1.1. The impact of multimorbidity

Many patients seeking support for self-management have morethan one condition (so-called multimorbidity) [11] and this is

M. Harrison et al. / Patient Education and Counseling 87 (2012) 67–7368

associated with poor outcomes [12,13]. However, key questionsabout the impact of multimorbidity remain.

First, what is the mechanism? There are a variety of plausibletheories of the mechanisms that might cause patients withmultimorbidity to have poor outcomes, including biologicalinteractions between disorders [14,15], but self-managementremains a potential candidate. Patients with multimorbidity haveto deal with multiple self-management activities with limitedresources of energy, time, attention and motivation [16,17] andpotentially face complex decisions about priorities among self-management tasks [18,19]. This may reduce their ability toconduct effective self-management and lead to poor outcomes.

The second key question concerns the types of multimorbidityand the combinations of conditions that are particularly associatedwith poor outcomes. For many long-term conditions (such asdiabetes and coronary heart disease), there are potential synergies,where several conditions benefit from the same self-management(e.g. exercise). A recent study found that patients with multi-morbidity who took part in a disease management programmereported significant improvements [20]. However, depression is acommon disorder which is often co-morbid with long-termconditions such as diabetes [21]. The co-existence of depressionand diabetes may lead to underregulation of diabetes (i.e. wheredepression lowers patient self-efficacy to undertake self-manage-ment) or mis-regulation (i.e. where improvements in depressionlead to changes in diet which in turn have a negative effect ondiabetes management) [22]. However, even here the evidence isunclear. Although depression is often associated with reduced self-management in observational studies [23], patients with long-term conditions who report comorbid depression have beenreported as demonstrating greater benefits after attending theCDSMP [24].

At present, the mechanisms by which multimorbidity impactson patient outcomes are unclear. We have previously demonstrat-ed that patients with high baseline levels of illness severity aremore likely to report favourable outcomes from the CDSMP [10].However, in this previous study, severity was assessed throughspecific research assessments and not through routine data.Therefore, this finding is less useful in everyday practice, becausesuch severity data are not routinely collected, and practitionerswould not be able to access such information easily to inform theirclinical decision making. This new analysis extends our previouswork by identifying whether assessments of multimorbidity basedon numbers of conditions are predictive of patient benefit in theCDSMP in the same way as research assessments of severity. Dataon number of conditions are routinely available in practice, andthus the present analysis will determine whether measures ofmultimorbidity provide a more clinically useful and practical aid toclinical decision making about referral to CDSMP.

In summary, we used secondary analysis of data from ourpublished CDSMP trial to test whether multimorbidity moderatedoutcomes from the CDSMP, to determine whether multimorbiditypredicted level of benefit from the course and could be used inroutine clinical decision making about referral to the CDSMP.

2. Methods

2.1. Design

The analysis was based on a trial of the cost-effectiveness of anAnglicised version of the CDSMP (the ‘Expert Patients Programme’– ISRCTN70532349) [7,25]. Volunteer patients (n = 629) with arange of self-reported long-term conditions were recruited fromcommunity settings, allocated to the CSDMP or a wait list controland 83% followed-up at 6 months. The course involved six 150 minweekly group sessions, each group consisting of 8–12 participants.

Groups were led by a pair of lay trainers or volunteer tutors andheld in non-health services premises. The intervention wasconducted according to a written manual, and included sessionson relaxation, diet, exercise, fatigue, managing pain and medica-tion, and communication with health professionals. Core aspects ofthe course are goal setting and action planning and the theoreticalmodel underlying the course can be categorised as ‘social learning’[26].

The main analysis of our trial found that CDSMP patientsreported greater vitality, self-efficacy, mental well-being, andhealth-related quality of life at 6 months. The analysis reportedhere was a secondary moderator analysis. A moderator is defined as‘a qualitative (e.g. sex, race, class) or quantitative (e.g., level ofreward) variable that affects the direction and/or strength of therelation between an independent or predictor variable and adependent or criterion variable’ [9]. We previously reported thatour trial participants with lower self-efficacy and health-relatedquality of life at baseline demonstrated more positive healthoutcomes after the course. The analysis reported in this paperfurther explored moderators to test whether multimorbiditymoderated (i.e. effected the strength and/or direction of) theeffects of the CDSMP (independent variable) on patient outcomes(dependent variables). This was a post hoc secondary analysis notregistered as part of the study protocol.

2.2. Outcome assessments

We selected for investigation six key domains of health andhealth behaviour of relevance to the CDSMP: physical health;mental health; quality of life; self-efficacy; vitality; and self-management behaviours. Vitality responds to both physical andmental health [27].

(a) Self-efficacy was measured using a validated scale consisting of19 items relating to disease management, symptom manage-ment, exercise, and management of depression rated along 10-point scales (from ‘not at all confident’ to ‘totally confident’).The mean score across the four domains was used as asummary score. The items were developed for the assessmentof health education interventions through multitrait scalingand the scale has acceptable reliability and validity [28].

(b) Vitality was measured using 5 items (e.g. ‘did you feel wornout’, ‘did you have enough energy’) rated on 6-point scales [28].This scale was based on a measure from the Medical OutcomesStudy [28,29].

(c) Disability was measured using 22 items (e.g. ‘are you able to getin and out of bed?’) rated on 4-point scales [28] and was basedon a modified scale from the Stanford Health AssessmentQuestionnaire [28,30].

(d) Mental well-being was measured using the 5-item MentalHealth Index (MHI-5) from the SF-36 [31].

(e) Health-related quality of life was measured using the EQ5D,which assesses five dimensions (self-care, mobility, anxiety/depression, usual activities and pain/discomfort) and providesa utility value based on a population tariff [32].

(f) Self-management behaviour was measured using the meanscore on reports of 6 activities of relevance to the CDSMP (e.g.diet, exercise, relaxation) rated on 3 point scales. This scale wasconstructed for the trial in the absence of a suitable existinggeneric self-management scale, but remains to be formallyvalidated.

Self-efficacy and vitality were primary outcomes (relating toself-care and health outcomes respectively) in the original trial [6].A measure of health services utilisation (sum of GP and practice

M. Harrison et al. / Patient Education and Counseling 87 (2012) 67–73 69

nurse consultations, accident and emergency attendances andoutpatient visits) was also a primary outcome in the original trialbut as no evidence was found for an impact on utilisation theoutcome has been excluded from the current secondary analyses.All assessments were rescaled 0–100, with high scores indicatingmore positive outcomes (i.e. more self-management, greatermental well-being, less disability).

2.3. Multimorbidity

Patients self-reported their existing long-term conditions viafree text in response to the question ‘What long-term medicalconditions do you have? (e.g. arthritis, diabetes)’. Patients used amix of diagnostic terms and symptoms. These were classified intocategories by an academic primary care practitioner working witha health services researcher (AK), taking into account bothhomogeneity and frequency of problems. Originally 25 categorieswere used, but we removed 2 relating to mental health (see below).The most frequently used categories were arthritis (40%), diabetes(20%), back and musculoskeletal problems (20%), heart andcirculatory conditions (15%), asthma (14%), digestive conditions(11%), hypertension (10%), chronic fatigue (10%) and ‘other’ (29%).All remaining categories were reported by less than 10%. Wecounted up the number of self-reported long-term physicalconditions, and categorised patients into those with 1 conditionand those with more than 1 condition for analysis. A secondacademic general practitioner (JV) independently rated a 25%random sample of these classifications to test reliability, andobtained 95% agreement, kappa of 0.87.

To derive a measure of mental health we used the MHI-5 scale,coding patients as having ‘probable depression’ if the MHI-5 scorewas �60 (high scores on this scale represented greater well-being).We based this cut-point on evidence from two different sources.First, a literature search for population-based studies of MHI-5identified four relevant studies [29–32] with recommended cut-points between �52 (for severe depression) and �74 (mild tosevere disorders), with sensitivity and specificity scores of around80% or better. None of these studies was UK-based, therefore wealso analysed MHI-5 data from the 2008 Welsh Health Survey(N ’ 12,000) as a predictor of patient self-reports of current mentalhealth treatment. Receiver Operating Curve analysis yielded a cut-point of 60, with sensitivity and specificity both 80%. We used thisfor the primary analysis, but tested the impact of our choice ofthreshold via sensitivity analysis, using the lowest thresholdreported from the review (a cut point of 52) to focus attention onthe more severe disorders.

For our analysis, we defined four patient groups, three of whichrepresented differing patterns of multimorbidity:

(a) single long-term physical condition (n = 98 at baseline),(b) multiple (2+) physical conditions (n = 169),(c) single physical condition plus ‘probable depression’ (n = 108),(d) multiple physical conditions plus ‘probable depression’

(n = 235).

Patient numbers were too limited for us to sub-divide numbersof physical conditions into more than two levels. We excludedpatients (n = 19) with no long-term physical condition or MHI-5score.

2.4. Analysis

Initial analysis explored differences between groups onbaseline variables, using ANOVA and chi-squared tests. Weanalysed the moderating effect of multimorbidity using multipleregression [9]. Independent variables were baseline values of the

outcome, treatment group, multimorbidity category, the interac-tion between treatment group and multimorbidity category, plustrial stratification variables (age, gender, ethnicity, accommoda-tion type, and type of condition). We excluded one stratificationfactor, general health, as this correlated considerably with thebaseline variables.

The test of the moderating effect of multimorbidity wasprovided by the interaction with treatment group. The interactioncan be interpreted as a test of whether the difference betweentreatment and control patients was the same across all fourmultimorbidity groups, after controlling for potential confoundingfactors. Where the interaction was not statistically significant, weremoved it from the model and reanalysed to obtain final estimatesfor the main effects. As highlighted earlier, in previous work wefound that for some outcomes, patients who reported worse scoreson outcomes at baseline made the biggest gains from CDSMP [10].Therefore to determine if a significant interaction might beattributable to baseline differences between multimorbiditygroups, we conducted a further analysis of each outcome thatincluded an interaction term between outcome scores at baselineand treatment group.

We plotted significant relationships in the form of subgroupmeans and computed subgroup-specific effect sizes (Cohen’s d),adjusted for stratification variables and baseline outcome scores.The multimorbidity groups demonstrated considerable differencesin outcome scores at baseline, particularly in mental well-beingscores. Since multimorbidity group itself was partially designatedusing these mental well-being scores, when deriving subgroupmeans and effects we adjusted for baseline outcome scores andother covariates within each multimorbidity category separately,rather than across all groups combined. This ensured that eachmultimorbidity treatment group was compared only againstdirectly equivalent control patients (i.e. with the same level ofphysical and mental morbidities and similar outcomes at baseline).Analyses used Stata (version 10), based on intention-to-treatsamples and alpha of 0.05.

3. Results

At baseline, there were significant differences between multi-morbidity categories in demographic variables (age, marital status,accommodation and employment) and self-reported main condi-tion (Table 1). There were also significant differences in self-management: patients with ‘probable depression’ reported lessuse of diet, complementary therapy or relaxation. All groups withmultimorbidity reported less exercise. There were also baselinedifferences in health measures: patients with a single conditionachieved the highest scores; patients with multiple physicalconditions and ‘probable depression’ the lowest. Patients withmultiple physical conditions scored higher on all outcomes thanpatients with a single physical condition plus ‘probable depres-sion’. These results highlighted the importance of controlling forthese baseline factors in the analysis.

The main results are given in Table 2, with plots in Fig. 1 andsub-group effect sizes in Table 3. We found a significantmoderating effect of multimorbidity on vitality, health-relatedquality of life and mental well-being. When we controlled forinteractions between baseline outcome scores and treatmentgroup, the results remained significant for vitality and health-related quality of life, but not for mental well-being. The pattern ofresults (Fig. 1; Table 3) shows that changes in vitality and health-related quality of life were similar for both CDSMP and controlpatients in all multimorbidity groups with the exception of thosewith multiple physical conditions and ‘probable depression’,where CDSMP patients improved considerably whilst controlpatients declined in score.

Table 1Baseline characteristics of multimorbidity groups.

1 physical condition

(n = 98)

Multiple physical

conditions (n = 169)

1 physical with ‘probable

depression’ (n = 108)

Multiple physical with

‘probable depression’ (n = 235)

p

Age (%, n)

�40 12% (12) 5% (8) 31% (34) 17% (39) <0.01

40–59 46% (45) 31% (52) 51% (55) 43% (100)

�60 42% (41) 65% (109) 18% (19) 41% (96)

Gender

Female (%, n) 69% (68) 65% (109) 77% (83) 71% (167) 0.48

Ethnicity

White (%, n) 96% (94) 95% (160) 93% (100) 95% (224) 0.86

Marital status

Lives alone (%, n) 18% (18) 31% (53) 20% (22) 32% (75) 0.02

Accommodation

Owner occupied (%, n) 82% (80) 71% (120) 69% (74) 62% (145) 0.03

Educational qualifications (%, n)

None 34% (33) 32% (54) 30% (32) 42% (99) 0.09

A-level/AS-level/NVQ/other 44% (43) 50% (84) 53% (57) 45% (106)

Degree 22% (22) 18% (31) 17% (18) 12% (29)

Employment (%, n)

Currently in paid work 30% (29) 14% (23) 31% (33) 15% (34) <0.01

Not in paid work (%, n) 70% (69) 86% (146) 69% (75) 86% (201)

Self-reported main condition (%, n)

Musculoskeletal 29% (28) 40% (68) 24% (26) 38% (89) 0.01

Diabetes 16% (16) 9% (16) 13% (14) 11% (26)

Heart condition 2% (2) 8% (13) 2% (2) 7% (17)

Other 53% (52) 43% (72) 61% (66) 44% (103)

Years with main condition (%, n)

�5 years 31% (30) 31% (53) 39% (42) 30% (70) 0.11

5–15 years 36% (35) 37% (62) 44% (48) 40% (95)

>15 years 34% (33) 32% (54) 17% (18) 30% (70)

Self management (mean, SD)

Diet 61.2 (34.1) 66.7 (33.1) 56.1 (35.7) 57.5 (32.1) 0.03

Complementary products 32.1 (38.5) 35.5 (36.0) 26.9 (34.6) 30.8 (33.9) 0.26

Complementary therapy 26.6 (36.4) 28.2 (36.8) 16.3 (28.5) 21.1 (32.3) 0.03

Exercise 53.6 (31.5) 42.4 (31.1) 40.7 (34.4) 38.9 (32.0) 0.01

Relaxation 55.2 (33.6) 50.0 (28.7) 36.4 (31.9) 37.6 (30.8) <0.01

Information seeking 60.7 (34.6) 60.2 (35.4) 66.2 (33.4) 60.3 (32.2) 0.44

Baseline health (mean, SD)

Vitality 46.0 (21.4) 38.0 (19.8) 28.5 (19.8) 25.3 (16.1) <0.01

Disability 79.6 (22.8) 73.5 (22.5) 73.3 (20.5) 68.8 (21.1) 0.01

Quality of life 61.3 (32.6) 55.6 (26.5) 40.5 (32.1) 36.7 (34.8) <0.01

Mental well-being 78.8 (10.0) 76.8 (10.1) 43.2 (12.9) 42.4 (14.2) <0.01

Self-efficacy 63.3 (20.7) 54.1 (19.0) 40.7 (20.8) 37.3 (19.3) <0.01

Patients with multimorbidity reported a median of 3 conditions in our different categories (IQR 2, 4).

Table 2Summary of p-valuesa for main effects and interactions.

Outcome Main effect of

multimorbidity group

Main effect of

treatment group

Interaction of

multimorbidity

and treatment group

Interaction controlling

for outcome at baseline by

treatment group interaction

Self-efficacy 0.27b <0.01b 0.67 –c

Vitality 0.14 0.12 0.01 0.01Health related quality of life 0.04 0.36 0.02 0.03Self-management 0.31b 0.07b 0.28 –c

Disability 0.28b 0.20b 0.30 –c

Mental well-being 0.61 0.01 0.03 0.18

a p-Values after adjustment for baseline values of the outcome, patient age, gender, ethnicity, disease condition, accommodation (figures in bold are statistically significant,

p<0.05).b p-Value after removal of non-significant interaction term.c Not analysed where the multimorbidity by treatment group interaction was non-significant.

Table 3Subgroup effect sizesa (95% CI) for significant moderators.

1 physical condition

(controls, n = 44;

treatment, n = 36)b

Multiple physical

conditions (controls,

n = 84; treatment,

n = 66)b

1 physical condition

plus ‘probable depression’

(controls, n = 38; treatment,

n = 42)b

Multiple physical plus

‘probable depression’

(controls, n = 101; treatment,

n = 97)b

Vitality 0.08 (�0.24, 0.40) 0.07 (�0.17, 0.30) �0.14 (�0.46, 0.18) 0.43 (0.23, 0.64)

Health-related quality of life 0.05 (�0.27, 0.36) 0.02 (�0.21, 0.25) �0.17 (�0.48, 0.14) 0.35 (0.16, 0.55)

Mental well-being 0.13 (�0.16, 0.42) 0.05 (�0.16, 0.26) 0.19 (�0.10, 0.48) 0.45 (0.27, 0.63)

a Effect sizes (Cohen’s d) for the difference between CDSMP and control groups, controlling for patient age, gender, ethnicity, disease condition, accommodation, and

baseline values of the outcome (within subgroups).b There was some minimal variability in sample size by outcome.

M. Harrison et al. / Patient Education and Counseling 87 (2012) 67–7370

Fig. 1. Interactions between multimorbidity group and effects of the CDSMP.

M. Harrison et al. / Patient Education and Counseling 87 (2012) 67–73 71

We found no significant interactions between multimorbidityand treatment group for self-efficacy, disability or self-manage-ment. In addition, we found no significant main effect ofmultimorbidity on these outcomes.

A sensitivity analysis exploring the effect of a different thresholdfor probable depression reduced the prevalence of depression in thesample from 57% to 42%. The main findings from the analysis ofmoderation were unaffected, and the significance of the interactionsreported in Table 2 did not change in relation to any outcome.

4. Discussion and conclusion

4.1. Discussion

In our secondary analysis, multimorbidity moderated theimpact of the CDSMP on three of six outcomes: vitality; health-

related quality of life; and mental well-being. Patients with thehighest level of multimorbidity burden who received CDSMP gainedsubstantially on these outcomes relative to controls, whereasoutcomes for CDSMP and control patients with other patterns ofmultimorbidity did not differ from outcomes for patients with singleconditions. For mental well-being, this might reflect baselinedifferences, but the effect endured in vitality and health-relatedquality of life even controlling for baseline differences.

Exploratory analyses such as those reported in this paper raisethe possibility of spurious findings because of multiple hypothesistesting, a problem accentuated when secondary analyses are notpre-specified. Furthermore, the original study was not powered todetect moderating factors, in line with most published in theliterature, and some of our group sizes were quite small. Even so,the 95% confidence intervals around the multimorbidity groupeffect size estimates (Table 3) were fairly narrow in most instancesand the positive effects on vitality and health-related quality of lifein the ‘multiple physical plus probable depression’ group were wellabove zero, whereas in other groups they were close to zero evenallowing for error. Nonetheless, secondary analyses can be prone togenerate findings that do not replicate and multimorbidity is arelatively recent area of research interest, as are investigations intomoderators of effect. Hence these results are clearly in need ofreplication in other CDSMP trials.

The sample included a heterogenous range of physical conditionsand very varied combinations of multimorbidity. We were unable toexplore specific combinations of physical conditions, or the effect of‘probable depression’ alone, and there were no measures of othermental health symptoms such as anxiety. Long-term physicalconditions were self-reported, involving a complex mix of diagnosticterms and symptoms. This has significant limitations, as patientsmay not always use accepted or easily interpretable terms for theirconditions, and their self reports may not agree with the views oftheir health professionals or with formal research classifications.Our coding in this respect was pragmatic, designed to maximise ourability to capture the data. It should be noted that these self reportsrepresented patient perceptions of separate conditions, which maybe as important as any formal classification system in determiningperceived burden and consequent self-management behaviour. Ourresults are clearly sensitive to the particular coding scheme used,and would benefit from replication using a more standardized andreliable system, triangulated with data from health professionals.However, the multimorbidity classification used in the analysis onlyconsidered whether patients reported one, or more than one long-term physical condition, and we would highlight that a secondindependent rater achieved high levels of agreement. Of course, thefact that our coding takes no account of the complexities ofinteractions is an additional weakness, as particular conditions maybe concordant or discordant in their management [33]. Our resultsmay also be specific to the lay-led CDSMP, and caution must beexercised in generalising to other self-management interventions.

We chose not to base our measure of ‘probable depression’ onpatient report. Only 13% of patients listed mental health in responseto the free text question on long-term conditions, which is likely tobe an underestimate given known underdiagnosis in primary care.Our preferred measure was to base assessment on the MHI-5, as thishas been shown to have high sensitivity and specificity for detectingclinical depression. Our results were also robust to a differentthreshold used to define ‘probable depression’ (we did not test ascore of greater than 60 as the resulting prevalence of ‘probabledepression’ was so high as to cast doubt on its validity).

4.2. Conclusion

Earlier, we highlighted that questions remained about themechanism by which multimorbidity impacts on outcomes in

M. Harrison et al. / Patient Education and Counseling 87 (2012) 67–7372

patients with long-term conditions, and the particular combina-tions of conditions more or less effected. We now consider therelevance of the current analyses to these questions.

Although robust data are not available, both self-managementand self-efficacy have been proposed as mediators of the effect ofCDSMP [34]. We found that self-efficacy and self-managementwere both lower at baseline in patients with multimorbidity, withthe co-existence of depression associated with the largest deficits.We then found that, despite these baseline deficits, patients withco-existing depression and multiple long-term conditions also hadthe greatest gains from the CDSMP intervention. However, theseadditional gains were found on vitality, health-related quality oflife and mental well-being, whilst self-efficacy and self-manage-ment did not change significantly. Our results are thus somewhatsimilar to those of a previous trial by Jerant et al., which found thatbaseline depression moderated the benefits of CDSMP, but Jerant’strial reported a moderating effect on self-efficacy [24] which wedid not replicate. Given that patients with co-existing depressionand multiple long-term conditions had poor self-management andself-efficacy at baseline, but reported better health outcomes post-intervention without a corresponding effect on those baselinevariables, it seems unlikely that self-efficacy and self-managementchanges are entirely responsible for this effect, even though theyare clearly core targets of the intervention. It is possible that thedelivery of CDSMP contains sufficient ingredients of therapeuticgroups that it is able to impact on depression through othermechanisms, such as improving self esteem or reducing isolation.These changes may then lead to improvements in health outcomes.We have previously found that measures of group processes arerelated to self-management and self-efficacy [35], but do notextend to health outcomes. The complexity of these mechanismshighlights the need for more detailed exploration of possiblemechanisms of change through formal analyses of mediation [36]which are currently absent in this literature.

Our analysis found that the moderating effect of multimorbiditywas largely confined to the group with multiple physical conditionsplus ‘probable depression’, which suggests that the benefits ofCDSMP are not influenced by the presence of multimorbidity per se,only when multiple physical conditions and ‘probable depression’coexist. With the exception of mental well-being, this differentialresponse appeared robust and was not attributable to groupdifferences at baseline. Our finding that the moderating effect ofmultimorbidity was only evident when ‘probable depression’ wascombined with more than one physical condition thus differs fromJerant’s trial which demonstrated that depression itself is amoderator of the benefits of CDSMP but did not analyse themoderating effect of multimorbidity per se [24]. Mean mental well-being scores (and standard deviations) were quite similar for bothsub-groups with ‘probable depression’ in the current study, so ourresults cannot be attributed to a higher burden of depressivesymptoms within the group with ‘probable depression’ and multiplephysical problems. Unfortunately, the nature of the trial samplemeans we lacked a group of patients with ‘probable depression’alone and so cannot make comparisons against such a group.

Making sense of these complex relationships clearly requiresmore trials reporting moderator analyses using a consistent analyticframework. However, the results are consistent in confirming theimportance of depression as a predictor of benefit from the CDSMP.Observational studies suggest that depression is associated withreductions in self-management [37,38], but identifying cause andeffect relationships in observational studies is difficult because ofconfounding, and methodological safeguards are not alwayssufficient [39]. This experimental study suggests that the negativeeffect of depression on self-management and self-efficacy does not

reduce the benefit that patients gain from CDSMP, but that thesepatients gain more from the intervention. This highlights the

interpretative advantages of experimental studies in helping todisentangle complex relationships between variables.

Multimorbidity represents both a threat and an opportunity inlong-term conditions. Our findings confirm previous researchwhich shows that although coexisting mental and physical healthproblems increase illness burden, the therapeutic benefits ofintervening with such patients are potentially magnified. The exactmechanisms underlying this moderating effect are unclear, but itappears that influences on self-management behaviour or self-efficacy are unlikely to provide a comprehensive explanation.

4.3. Practice implications

The current analysis represents an extension of previoussecondary analyses of our trial [10], which reported that patientswith lower baseline self-efficacy and quality of life demonstratedmore positive health outcomes following the CDSMP. Althoughthese results were interesting, their clinical implications arelimited, because self-efficacy and quality of life are not routinelymeasured in patients in primary care, and thus the results cannoteasily be used for clinical decision making. Our present analysisextended these findings to consider exactly which patterns ofproblems are most associated with benefit from the CDSMP. Thecurrent analyses also have the advantage that they have morepractical utility in a clinical context. The count of the number ofconditions that a patient has is a simple and accessible measure.Our results suggest that such a count needs to be combined withmeasurement of severity of depression, but such measurement isrecommended by clinical guidelines for some long-term condi-tions in the United Kingdom [40,41] and incentivised through theQuality and Outcomes Framework [42]. If the current results provereplicable, we would suggest that depression screening combinedwith a count of conditions would provide a useful assessment forpractitioners to prioritise referral to the CDSMP.

Acknowledgements

The original trial on which this analysis was based was fundedby the United Kingdom Department of Health. Further analysis wasfunded as part of the United Kingdom National Institute of HealthResearch (NIHR) School for Primary Care Research. The viewsexpressed in this publication are those of the authors and notnecessarily those of the NHS, the NIHR or the Department ofHealth. We thank Dr. Tom Blakeman for assisting with the codingof the long-term conditions.

References

[1] Murray C, Lopez A. The global burden of disease: a comprehensive assessmentof mortality and disability from disease, injuries and risk factors in 1990.Boston: Harvard School of Public Health on behalf of the World Bank; 1996.

[2] Wagner E, Groves T. Care for chronic diseases. Brit Med J 2002;325:913–4.[3] Bodenheimer T, Wagner E, Grumbach K. Improving primary care for patients

with chronic illness: the Chronic Care Model, part 2. J Amer Med Assoc2002;288:1909–14.

[4] Department of Health. Self care – a real choice: self care support – a practicaloption. London: Department of Health; 2005.

[5] Lorig K, Sobel D, Stewart A, Brown B, Bandura A, Ritter P, et al. Evidencesuggesting that chronic disease self management can improve health statuswhile reducing hospitalisation: a randomized trial. Med Care 1999;37:5–14.

[6] Griffiths C, Foster G, Ramsay J, Eldridge S, Taylor S. How effective are expertpatient (lay led) education programmes for chronic disease? Brit Med J2007;334:1254–6.

[7] Kennedy A, Reeves D, Bower P, Lee V, Middleton E, Richardson G, et al. Theeffectiveness and cost effectiveness of a national lay led self care supportprogramme for patients with long-term conditions: a pragmatic randomisedcontrolled trial. J Epidemiol Community Health 2007;61:254–61.

[8] Kraemer H, Wilson G, Fairburn C, Agras W. Mediators and moderators oftreatment effects in randomized clinical trials. Arch Gen Psychiatry2002;59:877–83.

M. Harrison et al. / Patient Education and Counseling 87 (2012) 67–73 73

[9] Baron R, Kenny D. The moderator–mediator distinction in social psychologicalresearch: conceptual, strategic and statistical considerations. J Pers Soc Psy-chol 1986;51:1173–82.

[10] Reeves D, Kennedy A, Fullwood C, Bower P, Gardner C, Gately C, et al. Pre-dicting who will benefit from an Expert Patients Programme self-managementcourse. Brit J Gen Pract 2008;58:198–203.

[11] Valderas J, Starfield B, Salisbury C, Sibbald B, Roland M. Defining comorbidity:implications for the understanding and provision of health services andhealth. Ann Fam Med 2009;7:357–63.

[12] Fortin M, Bravo G, Hudon C, Vanasse A, Lapointe L. Prevalence of multi-morbidity among adults seen in family practice. Ann Fam Med 2005;228.

[13] Wolff J, Starfield B, Anderson G. Prevalence, expenditures, and complications ofmultiple chronic conditions in the elderly. Arch Intern Med 2002;162:2269–76.

[14] Heckbert S, Rutter C, Oliver M, Williams L, Ciechanowski P, Lin E, et al.Depression in relation to long-term control of glycemia, blood pressure,and lipids in patients with diabetes. J Gen Intern Med 2010;25:524–9.

[15] Joyne K, Whellan D, O’Connor C. Depression and cardiovascular disease:mechanisms of interaction. Biol Psychiatry 2003;54:248–61.

[16] Bayliss E, Ellis J, Steiner J. Barriers to self management and quality of lifeoutcomes in seniors with multimorbidities. Ann Fam Med 2007;5:395–402.

[17] Lin E, Katon W, Rutter C, Simon G, Ludman E, Von Korff M, et al. Effects ofenhanced depression treatment on diabetes self-care. Ann Fam Med2006;4:46–53.

[18] Fried T, McGraw S, Agostini J, Tinetti M. Views of older persons with multiplemorbidities on competing outcomes and clinical decision making. J Am GeriatrSoc 2008;56:1839–44.

[19] Stack R, Elliott R, Noyce P, Bundy C. A qualitative exploration of multiplemedicines beliefs in co-morbid diabetes and cardiovascular disease. DiabetMed 2008;25:1204–10.

[20] Ose D, Wensing M, Szecsenyi J, Joos S, Hermann K, Miksch A. Impact of primarycare based disease management on the health related quality of life of patientswith type 2 diabetes and comorbidity. Diabet Care 2009;32:1594–6.

[21] Anderson RJ, Freedland KE, Clouse RE, Lustman PJ. The prevalence of comorbiddepression in adults with diabetes: a meta-analysis. Diabet Care2001;24:1069–78.

[22] Detweiler-Bedell J, Friedman M, Leventhal H, Miller I, Leventhal E. Integratingco-morbid depression and chronic physical disease management: identifyingand resolving failures in self-regulation. Clin Psychol Rev 2008;28:1426–46.

[23] McKellar J, Humphreys K, Piette J. Depression increases diabetes symptoms bycomplicating patients’ self care adherence. Diabetes Educ 2004;30:485–92.

[24] Jerant A, Kravitz R, Moor-Hill M, Franks P. Depressive symptoms moderatedthe effect of chronic illness self management training on self efficacy. Med Care2008;46:523–31.

[25] Richardson G, Kennedy A, Reeves D, Bower P, Lee V, Middleton E, et al. Cost-effectiveness analysis of a national lay led self care support programme forpatients with long-term conditions. J Epidemiol Community Health2008;62:361–7.

[26] Newman S, Steed L, Mulligan K. Self management interventions for chronicillness. Lancet 2004;364:1523–37.

[27] Hann M, Reeves D. The SF-36 scales are not accurately summarised byindependent physical and mental component scores. Qual Life Res2008;17:413–23.

[28] Lorig K, Stewart A, Ritter P, Gonzalez V, Laurent D, Lynch J. Outcome measuresfor health education and other health care interventions. Thousand Oaks:Sage; 1996.

[29] Stewart A, Hays R, Ware J. Health perceptions, energy/fatigue and healthdistress measures. In: Stewart A, Ware J, editors. Measuring functioning andwell-being: the Medical Outcomes Study approach. Durham, NC: Duke Uni-versity Press; 1992. p. 143–72.

[30] Fries J, Spitz P, Kraines R, Holman H. Measurement of patient outcome inarthritis. Arthritis Rheum 1980;23:137–45.

[31] Berwick D, Murphy J, Goldman P, Ware J, Barksy A, Weinstein M. Performanceof a five-item mental health screening test. Med Care 1991;29:169–76.

[32] Kind P. The EuroQoL instrument: an index of health-related quality of life. In:Spilker B, editor. Quality of life and pharmacoeconomics in clinical trials.Philadelphia: Lippincott-Raven; 1996.

[33] Piette J, Kerr E. The impact of comorbid chronic conditions on diabetes care.Diabet Care 2006;29:725–31.

[34] Lorig K, Seleznick M, Lubeck D, Ung E, Chastain R, Holman H. The beneficialoutcomes of the arthritis self-management course are not adequatelyexplained by behaviour change. Arthritis Rheum 1989;32:91–5.

[35] Harrison M, Fullwood C, Bower P, Kennedy A, Rogers A, Reeves D. Exploring themechanisms of change in the chronic disease self-management programme:secondary analysis of data from a randomised controlled trial. Patient EducCouns, doi:10.1016/j.pec.2010.10.026, in press.

[36] Emsley R, Dunn G, White I. Mediation and moderation of treatment effects inrandomised controlled trials of complex interventions. Stat Methods Med Res2010;19:237–70.

[37] Piette J, Richardson C, Valenstein M. Addressing the needs of patients withmultiple chronic illnesses: the case of diabetes and depression. Am J ManageCare 2004;10:152–62.

[38] Ciechanowski P, Katon W, Russo J, Hirsch I. The relationship of depressivesymptoms to symptom reporting, self care and glucose control in diabetes.Gen Hosp Psychiatry 2003;25:246–52.

[39] Bosco J, Silliman R, Thwin S, Geiger A, Buist D, Prout M, et al. A most stubbornbias: no adjustment method fully resolves confounding by indication inobservational studies. J Clin Epidemiol 2010;63:64–74.

[40] National Institute for Health and Clinical Excellence. Depression: the treat-ment and management of depression in adults. National Institute for Healthand Clinical Excellence; 2009 [update]http://www.nice.org.uk/nicemedia/pdf/Depression_Update_FULL_GUIDELINE.pdf.

[41] National Institute for Health and Clinical Excellence. Depression in adults witha chronic physical health problem: treatment and management. NationalClinical Practice Guideline, vol. 91. National Institute for Health and ClinicalExcellence; 2009 , http://www.nice.org.uk/nicemedia/pdf/CG91FullGuideli-ne.pdf.

[42] British Medical and Association. Investing in general practice: the new generalmedical services contract. British Medical Association; 2003.