Dietary and exercise change following acute cardiac syndrome onset: A latent class growth modelling analysis

Introduction

The onset of cardiac disease can trigger a number of changes in diet and levels of physical activity. Diet may improve in the short-term (Bennett et al., 1999), although old habits may return over time. Luszczynska and Cieslak (2009), for example, found only 20 per cent of cardiac patients met guidelines for fruit and vegetable intake immediately following a cardiac rehabilitation programme: this fell to 12 per cent at 1-year follow-up. In the context of physical activity, Bennett et al. (1999) found significant spontaneous changes in mild–moderate activity levels in the 3 months following myocardial infarction (MI). However, Lear et al. (2003) reported minimal changes from baseline on measures of leisure time exercise and treadmill performance 1-year following MI despite participants taking part in a rehabilitation programme.

Critical to any rehabilitation process is an understanding of the factors that influence such changes. These may include disease status and demographic variables such as gender, age and socio-economic status. More psychological determinants include beliefs about the nature or threat value of the disease, the emotional response to disease onset and the coping efforts made in response to them (e.g. Leventhal et al., 2013).

The majority of studies investigating the role of these factors in patients with heart disease have focused on the influence of mood. These have shown depressed individuals are less likely to attend and complete cardiac rehabilitation and more likely to be re-admitted to hospital in the absence of cardiac complications than those who are not depressed (e.g. McGrady et al., 2009). The impact of depression on diet and exercise appears modest, although good-quality longitudinal data are lacking. Depression predicts low levels of engagement in leisure activities and exercise at 1-year, but not 3-year follow-up post-MI (e.g. Allan et al., 2007; Söderman et al., 2007). The relationship between anxiety and behaviour has received less attention, although it may impact negatively on adherence to exercise (Kuhl et al., 2009).

From a more cognitive perspective, the self-regulation model (Leventhal et al., 2013) suggests patient beliefs will also influence responses to illness. According to this model, individuals appraise the nature of an illness in terms of its identity, cause, timeline, consequences and the degree to which it is treatable or manageable by either medical intervention or the individual themselves. Individuals who consider their cardiac condition to be well understood, controllable, symptomatic and to have severe consequences, for example, are most likely to attend cardiac rehabilitation (French et al., 2006). Less positive were findings that illness perceptions explained only 2 per cent of the variance in cardiac patients’ exercise and dietary choices (Byrne et al., 2005). Finally, while coping efforts have been associated with mood (e.g. Garnefski et al., 2009) and disease severity (Chiavarino et al., 2012), we are aware of no studies that have considered the role of coping in relationship to health behaviours in cardiac patients.

All the studies considered so far have explored behavioural responses to illness through analysis of changes in mean scores of the relevant variables. While clearly productive, this approach is limited in that it can show trends across the entire population under study. However, individual participant scores will necessarily diverge from these changes both in magnitude and, potentially, even direction. These differing trajectories of change may be predicted by a range of participant characteristics. This study reports a first attempt to explore this issue in cardiac patients. It used latent class growth modelling (LCGM) analysis (Nagin, 2005) to determine whether clusters of participants with different behavioural trajectories on measures of physical activity and diet could be identified and their psychological characteristics determined. Predictor variables were baseline measures of illness beliefs, mood (anxiety and depression) and coping in response to the acute event. Clinical and demographic variables were also included. The exploratory nature of the analyses precluded strong hypotheses.

Method

Results

Descriptive statistics

A total of 150 participants completed time 1 (T1) questionnaires: 106 completed time 2 (T2) questionnaires and 97 completed time 3 (T3) questionnaires. Between T1 and T2, 55 of the 106 completers had received additional surgery (stent and coronary artery bypass) and 43 attended a cardiac rehabilitation programme which comprised a preliminary individual visit by a nurse in hospital followed by a 6-week series of group meetings held on the hospital campus, including an exercise programme and educational programme focusing on risk factors for coronary heart disease (CHD) including diet.

Comparisons on baseline questionnaires between completers and non-completers at T3 revealed completers were older (mean = 63.33 (SD = 10.16 years) years versus mean = 59.55 (SD = 10.97) years: t₁₄₈ = 2.303; p < .05) with significantly higher scores on the Coping with Acute Coronary Syndrome Questionnaire (CACSQ)-problem-focused coping scale (mean = 8.40 (SD = 1.80) versus mean = 7.75 (SD = 1.67): t₁₄₆ = 2.175; p < .05). Missing data analyses showed that the pattern of missing was random (Little’s missing completely at random (MCAR) test: χ²(414) = 398.95, p = .694). Missing values were therefore imputed with expectation–maximization (EM) algorithm.

Baseline and subsequent scores on all psychological measures are reported in Table 1. At baseline, participants appeared to have an understanding of the chronic nature of their disease. In addition, they believed they had significant personal control over their condition: more so than that ascribed to their medical care. They also evidenced significant concern over their nature of the illness and a strong negative emotional response. This was reflected in their Hospital Anxiety and Depression Scale (HADS) scores. Forty-one percent of participants scored ⩾8 (the criterion for being a probable ‘case’) on the HADS anxiety sub-scale. For the depression sub-scale, the equivalent figure was 18 per cent. By 1-month follow-up, the equivalent figures were 35 per cent and 14 per cent, and at 6-month follow-up, 29 per cent and 12.5 per cent. Participants reported relatively frequent low-intensity physical activity, occasional medium-intensity activity and very infrequent high-intensity activity. They also appeared to be engaging in similar levels of emotion- and problem-focused coping.

OPEN IN VIEWER

Table 1.

Mean and standard deviation for variables under study.

Variable	Mean	SD
Illness beliefs
Consequences	4.07	3.11
Timeline	6.01	3.35
Personal control	4.75	2.96
Treatment control	1.20	1.43
Identity	3.70	2.85
Concern	8.36	2.11
Illness comprehension	3.20	2.78
Emotional affect	6.15	3.00
Mood
Anxiety	7.92	4.54
Depression	5.29	3.99
Coping strategies
Emotion-focused	10.16	5.66
Problem-focused	11.49	10.43
Dietary index
Time 1	12.13	5.12
Time 2	15.36	4.26
Time 3	15.05	4.49
Low-intensity physical activity
Time 1	5.11	1.63
Time 2	5.54	1.40
Time 3	5.40	1.43
Medium-intensity physical activity
Time 1	2.70	1.49
Time 2	3.38	1.83
Time 3	3.71	1.83
High-intensity physical activity
Time 1	1.62	0.98
Time 2	1.56	1.00
Time 3	1.85	1.18

SD: standard deviation.

Latent classes of change

Dietary index

For dietary index scores, a four-class solution seemed to be optimal as this had the lowest value of BIC (1994.41), the highest value of entropy (.717) and significant improvement over a three-class model (conditional bootstrap of LL difference associated with increase in number of classes = 30.28, p < .001). As can be seen in Figure 1(a), the classes differed in terms of intercepts (Wald test = 166.59, p < .001), but not in terms of both linear (Wald test = 6.22, ns) and quadratic (Wald test = 5.12. ns) slopes.

OPEN IN VIEWER

Figure 1.

Results of latent class growth curve modelling: (a) trajectories for dietary index, (b) trajectories for low-intensity physical activity, (c) trajectories for medium-intensity physical activity and (d) trajectories for high-intensity physical activity.

Based on the posterior probabilities, 56 per cent of the sample was categorized within the first trajectory, 26 per cent within the second trajectory, 12 per cent within the third and 6 per cent within the fourth. The best fit model was of no change for Class 3 and equal quadratic slopes for Classes 1, 2 and 4 (BIC = 1976.33, entropy = .722). The first, second and fourth trajectories are similar (linear slopes = 4.05, p < .001; quadratic slopes = −0.58, p < .001) and characterized by an immediate improvement in dietary behaviour, followed by a period of stability or minor falls.

Physical activity

Data relating to physical activity were analysed using the same methods. The BIC and entropy together with examination of slopes and intercepts indicated three-class solutions for low- and medium-intensity physical activity and a four-class solution for high-intensity physical activity (see Figure 1(b) to (d)).

Trajectories for low-intensity physical activity are presented in Figure 1(b). Class 1 contained more than half the participants (54%) and was characterized by significant increases in activity in the first month followed by more modest increases over the following 5 months (linear slope = 0.86, p < .05; quadratic slope = −0.13, p < .05). Members of classes 2 (31% of the sample) and 3 (15% of the sample) had similar trajectories, with no significant changes over time, but with different baseline levels.

The trajectories obtained for medium physical activity are plotted in Figure 1(c). Again, Class 1 had the highest membership (51%). These participants reported the highest frequency of this level of exercise at baseline and marked increases in the month following discharge from hospital. This was followed by continuing, but slower increases (linear slope = 1.30, p < .001; quadratic slope = −0.17, p = .001). Two other classes, 2 (35% of the sample) and 3 (14% of the sample), represent stable and low activity of that type.

High-intensity physical activity was rarely reported (see Figure 1(d)). A steady increase was observed only among patients with the highest starting point (class 4: 22% of the sample, linear slope = 0.11, p < .05). In Class 1 (47%) and Class 2 (20%), there was no change within time and these classes can be characterized by very low levels of this level of activity. By contrast, for Class 3 (11%) a significant drop in activity was noted during the first month (linear slope = −1.17, p < .001; quadratic slope = −0.17, p < .001) which did not recover of the longer time period.

Predictors of class membership

Dietary index

The results of all multinomial logistic regressions are presented in Table 2. Only two predictors were finally entered into the diet model: gender ( χ 3 2 = 15 . 07 , p = .002) and depression ( χ 3 2 = 11 . 14 , p = .011). Comparing to the reference group (Class 3 ‘the best and stable’), the group starting with middle values of dietary index has a higher ratio of men to women and also higher depression at the baseline. These relationships are even more pronounced in the groups starting with the lowest and the higher values of dietary index. Thus, the stable group had more equal gender proportions and the lowest depression levels soon after the cardiac event.

OPEN IN VIEWER

Table 2.

Summary of multinomial logistic regression for dietary and physical activity class membership (resultant models).

Class	Predictor	B	SE	Wald χ2	p	Odds ratio	95% Confidence interval
							Lower bound	Upper bound
Dietary index (Class 1: high and stable as a reference category)
Class 3: increase and middle starting point	Gender (female = 0, male = 1)	2.21	.76	8.49	.004	9.16	2.06	40.62
	Depression	.33	.14	5.64	.018	1.39	1.0	1.83
Class 2: increase and the lowest starting point	Gender	3.09	.92	11.17	.001	21.90	3.58	133.86
	Depression	.38	.15	6.56	.010	1.46	1.09	1.95
Class 4: increase and higher starting point	Gender	3.14	1.42	4.99	.025	23.10	1.47	362.94
	Depression	.44	.17	6.48	.011	1.55	1.11	2.18
Medium-intensity physical activity (Class 2: no activity as a reference category)
Class 3: seldom	Control over heart problems	−.30	.14	4.89	.027	.74	.57	.97
Class 1: increase	Control over heart problems	−.16	.08	4.22	.040	.85	.73	.99
High-intensity physical activity (Class 2: no activity as a reference category)
Class 1: seldom	Problem-focused coping	−.06	.02	6.98	.008	.94	.89	.98
Class 4: increase	Problem-focused coping	−.06	.03	3.77	.052	.95	.89	1.00
Class 3: decrease	Problem-focused coping	−.06	.04	2.16	.142	.94	.87	1.02

SE: standard error.

Physical activity

For low-intensity physical activity, no significant relationships were obtained. For medium-intensity physical activity, the derived model comprised only one significant predictor, the perceived degree of control over the heart problem ( χ 2 2 = 7 . 35 , p = .025). The pattern seems to be linear, so the highest control was noted for the non-active patients, lower for infrequently active and lowest for patients who had increased their activity after their cardiac event.

Finally, for high-intensity physical activity, problem-focused coping was the only predictor entered into the model ( χ 3 2 = 9 . 16 , p = .03). Participants with low levels of high-intensity physical activity reported less problem-focused coping than participants who reported no activity of that kind (the reference class). Interestingly, a similar tendency can be observed for two other classes but without reaching the level of statistical significance.

Discussion

This study was the first to consider the combined role of mood, illness beliefs and coping strategies in predicting health-related behavioural change following an acute cardiac event. In addition, it shifted analysis from one involving mean changes across a whole cohort to one identifying differential trajectories of change within the cohort.

Not surprisingly, perhaps, the greatest dietary changes occurred in the first month following discharge from hospital. Four sub-groups of patients were identified on the diet measure. The majority (56%) scored highly on the dietary scale and did not change over the period of the study. The three sub-groups with lower healthy eating scores at discharge each had similar trajectories of change, with dietary improvements in the first month following discharge from hospital followed by maintenance up to 6 months follow-up. Their trajectories of change were almost perfectly parallel. However, at no time did they reach the baseline levels of the highest health eating group. Accordingly, while these improvements are encouraging, more change is clearly possible.

Comparing to the reference group (Class 3 the best and stable), the other groups had a higher ratio of men to women and also higher depression scores at baseline. The finding that men were eating less healthily than women is consistent with previous findings (e.g. Wardle et al., 2004). That depressed individuals were more likely to improve their dietary habits was perhaps more surprising, although the changes made were modest. It is possible, also, that any dietary changes were the choice of others rather than the identified patients, particularly if they were male: married men’s dietary intake is frequently determined at least, in part, by their partners (e.g. Homish and Leonard, 2008).

Three sub-groups were identified on the measure of low-intensity physical activity. Of these, only the group with the lowest levels of activity post-discharge subsequently reported a modest increase. The others did not change their levels of physical activity. We were unable to discriminate between these groups on any of the variables measured. Medium-intensity physical activity involves exercise likely to benefit most individuals with heart disease (Wise, 2010) and requires active planning. It was therefore disappointing to find that only the group already engaging in the highest frequency of medium-intensity exercise reported any increase over time. The two sub-groups engaging in lower levels of physical activity at baseline, approximating nearly half the sample, showed no evidence of change. Clearly, it is difficult to motivate long-term exercise abstainers to change their behaviour.

Interestingly, the variable that discriminated between change and no change groups was their perceived control over heart disease, and in the opposite direction to that expected. Higher levels of control were associated with lower levels of change in physical activity. It might be expected that participants would come to believe that exercise was an important factor in gaining control over their disease. However, this did not seem to have been the case. For some participants, a perception of control over their disease as a result of other factors may have indicated they did not need to exercise. For others, the concept of managing disease may simply have involved resuming a previous unhealthy lifestyle in the absence of any immediate health cost or symptoms. More consideration could usefully be given to exploring the meaning ascribed to these concepts, perhaps through a more qualitative methodology. Certainly, some caution needs be given to the interpretation of single-item scales such as the Brief Illness Perceptions Questionnaire (BIPQ) (see Broadbent et al., 2011; Van Oort et al., 2011) despite its extensive use among cardiac patients.

As expected, only a minority of participants engaged in high-intensity physical activity with any regularity. Nevertheless, this group (22% of the sample) actually increased the frequency of this level of exercise. By contrast, less frequent high-intensity exercisers either maintained their low levels of exercise or reduced them to virtually 0. Importantly, membership of these groups was not predicted by any of clinical (medical) variables, nor by the BIPQ measure of illness identity, which can be seen as a proxy for the frequency of continuing symptoms. Accordingly, there was no evidence that these differences were a consequence of participants’ medical status or continuing symptoms. Of note also was that participants in the high-intensity/increase activity group were less likely to engage in problem-focused coping than those in the other groups, suggesting this response may have been relatively automatic, based around habitual responses, rather than a clearly planned process. It may also be that the preferred response to the onset of cardiac disease is to change diet rather than physical activity. Such changes have the advantage of being relatively easy to implement and may be considered less risky than increasing exercise levels.

Overall, the impact of psychosocial and medical variables on behaviour change was disappointingly low, and in accord with Byrne et al.’s (2005) findings of their explaining only 2 per cent of the variance in behaviour. Only gender, control beliefs, depression and coping appear to have had even a modest impact on behaviour; and this was inconsistent and in ways that require some speculative interpretation. In many ways, they are supportive of the social cognitive perspective that past behaviour predicts future behaviour (e.g. Ouelette and Wood, 1998), even in the face of a significant event such as the onset of cardiac disease.

The study was necessarily of an explorative nature. Its sample size lacked statistical power to detect differences between sub-groups and to confidently predict their membership. Accordingly, larger studies are needed to extend our findings and determine their reliability. However, the study has shown that distinct trajectories of behaviour change can be identified within a sample of cardiac patients, and that variables that predict group membership can also be determined. The finding of several trends within the overall behavioural data was important and indicates that studies of mean change may hide significant variance in outcome. If one takes the example of moderate physical activity, these apparent shifts may be actively misleading. The overall analysis indicated a shift towards more exercise: a good overall outcome. The LCGM indicated that this change was almost entirely situated within a group of individuals already exercising at reasonable levels, and there was little or no change among those with lower baselines: a rather more modest outcome. The understanding of personal trajectories of change should also have clinical relevance. Larger studies may reveal more subtle outcomes and perhaps allow more nuancing of interventions. However, we have already identified one key clinical imperative. It cannot be assumed that individuals with unhealthy lifestyles will be motivated to change them, or at least not sufficiently to achieve change. Similarly, those with healthy lifestyles appear likely to sustain or even improve their lifestyle despite the occurrence of disease. Accordingly, measurement of pre-event health behaviours should form an important part of any discharge interview, facilitating a particular rehabilitative focus on those with the most unhealthy lifestyle before the onset of ACS (Al-Khalili et al., 2007; Moreno, 2013). This would ensure that those with the least health resources receive the most impactful intervention and support in increasing them, preventing a potential downward spiral of health (Hobfoll, 2002).