Comorbidities and characteristics of coronary heart disease patients: their impact on health-related quality of life

The study was performed as part of an international study entitled European Practice Assessment of Cardiovascular Risk Management (EPA Cardio) [18, 19], which has a cross-sectional observational design. Its protocol is described in detail elsewhere [18, 19]. In Slovenia, this study was descriptive and exploratory.

A total of 36 family medicine practices across Slovenia participated in this study. Our sample of practices was drawn randomly, using a table of random numbers, and stratified according to the size of practices (small-sized practices with two full-time family physicians or less working on the same location, and large-sized practices with more than two full-time family physicians working on the same location) and urbanisation/practice location (practices located in urban areas with 30,000 inhabitants and more, and those located in rural areas with less than 30,000 inhabitants). In participating practices, the family physicians randomly selected 30 patients with CHD, including those with myocardial infarction, angina pectoris or coronary interventions from the register available at every family medicine practice. At any visit, these patients were given a coded diagnosis (I20-I25) according to the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10). Patients with diabetes were excluded from this group to ensure homogeneity of the group. To continue, patients with a terminal disease, cognitive impairment and those who do not understand Slovenian language sufficiently were also excluded from the study. All patients gave written informed consent for participation in the study.

European Quality of Life - 5 dimensions (EQ-5D) [20, 21] was used to obtain the data which represented the main outcomes. EQ-5D is a standardised, generic instrument for describing and valuing health, and it is available in more than 50 languages. This instrument defines health in terms of five dimensions: mobility, self-care, usual activities, pain or discomfort, and anxiety/depression disorders.

Within each dimension, there are three levels: no problems, some problems or extreme problems. The EQ-5D score (indicating HRQoL) ranges from 0 to1 and can be calculated by applying scores from the EQ-5D preference weights elicited from the Slovenian population [22]. The maximum score of 1 indicates the best state of health. Published evidence supports the validity and reliability of the EQ-5D as an outcome measure within CVD [23].

The EQ Visual Analogue Scale (EQ-VAS) was used as a measure of general self-assessment of HRQoL. EQ-VAS is made up of a vertical scale ranging from 1 to 100, where the lowest endpoint is “Worst imaginable health state” and the highest endpoint is “Best imaginable health state”. These two instruments were both part of an extensive patient questionnaire [20, 21].

Using an extended patient questionnaire, which included the EQ-5D, EQ-VAS, MORISKY, PACIC and EUROPEP questionnaires, we obtained data on the following explanatory variables: demographic variables (gender, age, employment, education, marital status), duration of practice involvement, family physician visit frequency per year as well as data on comorbid conditions, which were self-reported, chosen from the list by patients themselves and confirmed by their physicians through an audit of patients’ records, namely angina pectoris, myocardial infarction, stent or coronary by-pass, stroke, and peripheral artery disease. Furthermore, we also obtained data from patients on medication adherence (MORISKY questionnaire) [24], their assessment of chronic illness care (PACIC questionnaire) [25, 26] and their self-rated satisfaction of family practice care and clinical behaviour of their family physician (EUROPEP questionnaire) [27, 28].

Gender, marital status, duration of practice involvement and presence of comorbid cardiovascular diseases were categorical variables. Age (years), education (years), family physician visit frequency medication adherence (sum of points in the questionnaire from 0 to 8), patients’ assessment of chronic illness care (mean value of points from 1 to 5), and patients’ self-rated satisfaction of family practice care (mean value of points from 1 to 5) were continuous variables.

The available sample size was multiplied by the design effect to account for the cluster research design [29]. For the design effect calculation we used equation 1 + (m-1)*ρ given by Donner et al. [30]. The intracluster correlation coefficient (ρ) was used from a large sample-study for cluster randomized trials and surveys at the primary care clinic level (a maximum value observed of 0.05) [31]. Considering the mean cluster size (m) of 11.8 ± SD 2.1 patients (range 8–15), the design effect was equal to 1.54. The next step included power calculation for the two main statistical methods used in the manuscript. Considering the design effect and using the G*Power software (version 3.1.9) [32], a total of 423 patients was calculated to have a >99% power to detect a significant association for linear regression (using alpha of 0.05, considering 16 predictors and assuming medium effect size of 0.15 [33]). Similarly, considering the design effect, and following Hsieh [34], a total of 423 patients was calculated to have a nearly 90% power to detect a significant association for logistic regression (using alpha of 0.05, medium odds ratio of about 2.5 to 1 [35], and the variance inflation factor of 1.22) . The variance inflation factor, which depends on the squared multiple correlation coefficient (R²) relating a specific predictor of interest to the remaining predictors, was calculated according to the instructions in Hsieh et al. [36]. Our study design did not involve one specific predictor of interest, therefore we calculated R² for each predictor in the model and used the maximum value obtained. For continuous predictors selected as dependent variables we applied linear regression to calculate standard R². In case categorical predictors were selected as dependent variables we applied logistic regression for Nagelkerke R².

We calculated descriptive statistics for patient characteristics. Demographic data were reported as frequencies (%) or mean (SD) values. Only patients with complete data on all explanatory variables were considered in the final model and included in the analyses. The characteristics of these patients were compared with those of patients who were excluded from the analysis using the chi-square test (gender), the independent-samples t-test (age) and the EQ-5D score (Mann-Whitney U test). Normal distribution of numerical variables was confirmed by the Shapiro-Wilks test.

The main study outcome was HRQoL as measured by two dependent variables of the EQ-VAS and EQ-5D index score, which were also reported as frequencies (%) or mean (SD) and median (interquartile range) values. EQ-5D index score was obtained by applying preference weights from the health value scale defined according to a 3-level EQ-5D scale for Slovenian population [22].

This study tested a total set of 16 potential explanatory variables as predictors of HRQoL: 9 variables representing patient characteristics and 7 variables for comorbidities. Linear regression was used to identify the associations between EQ-VAS, patient characteristics and vascular comorbidities; results were presented as B-coefficient with 95% confidence intervals, standard error and p-value. Contrary to the EQ-VAS the EQ-5D scores and its domains were classified as ordinal measurement scales as suggested by Alava et al. [37]. Ordinal logistic regression was used to identify associations between EQ-5D, patient characteristics and vascular comorbidities; results were presented as odds ratio with 95% confidence intervals and p-value. The same method was also used for the EQ-5D domains. For the purposes of logistic regression, to reduce the number of ordinal categories, the EQ-5D score was collapsed into 7 levels (1.0, 0.7, 0. 6, 0.5, 0.4, 0.3, 0.2 or lower). Multiple regression modelling included characteristics that were found to be significant by univariate regression only. Statistical analysis was performed using IBM SPSS software, version 21.0 (IBM Corp., Armonk, NY). Statistical significance was set at p < 0.05.

The weakness of this study was firstly the cross-sectional observational design, which does not assess the impact of changes in demographic and other factors on HRQoL in a specific time period. It also means that there is a risk of reverse causation or, better said, that the cross-sectional study shows associations between the outcome and explanatory variables but not causal associations.

EQ-5D is a simple and widely used questionnaire, but it presents potential problems for statistical analysis due to the fact that distribution is not normal. The distribution of our data was bi-modal, just as described by Alava et al. [37]. Several models have been used in the past to find the best statistical option, with the most frequently used one being a simple linear model. Comparison of the models led to contradictory dilemmas. On this basis we decided to analyse the data using ordinal logistic regression model due to non-linearity of the outcome variable [37].

There are several reasons as to why the sample might not be representative of the Slovenian CHD population: the low response rate, the exclusion of patients who did not understand Slovenian and those with diabetes, family physicians may have failed to include random sample of participants. The final number of patients in the sample was much smaller due to several missing values in the patient questionnaire, which was very long and cognitively demanding in some parts (e.g., PACIC). For the analysis, we used only a part of the instrument. Only patients who completed the questionnaire in full were included in the analysis. We focused on the quality of life in patients with CHD without diabetes, which was an exclusion criterion for the whole EPA Cardio study. However, in our analysis, the exclusion of patients with diabetes represented a limitation in the analysis of HRQoL in CHD patients with multimorbidity. In line with the study’s methodology, we excluded patients who did not understand the Slovenian. These were mostly physical labourers from a low socioeconomic background - a particularly vulnerable group of patients. In addition, a control group without CHD was not included in the study. This means that there is no general non-CHD population sample to compare our sample of CHD patients with, therefore we cannot conclude whether the values for separate EQ-5D dimensions are similar or not to those in the general population without CHD.

Another weakness of the study was that we could not find any correlation with self-assessment of HRQoL and chronic illness care. Both measures are self-reported, therefore a correlation is expected.

The main strength of the study was the random sampling of CHD patients from a computer-generated list available to every family medicine practice. Such patient selection is preferred over the inclusion of sequential visitors to the practice, where both frequent as well as rare visitors have an equal chance of participating in the study. Although, like in many studies, we can not be sure that all participating family physicians followed the inclusion protocol.

Additional strength was our focus on the HRQoL of CHD patients in a representative environment of family medicine practices throughout Slovenia. The obtained data can serve as a basis for further studies.