Assessment of the evolution of health related quality of life in coronary patients. An application of generalized estimating equations to handle missing data and time‐dependent variables in longitudinal studies

Resumen

Objective The main aim of this thesis was to analyse the evolution of Health Related Quality of Life (HRQL) in Coronary Patients (CP), and to identify the clinical, demographic and psychological factors that influence this evolution from a longitudinal perspective, using Simple Generalized Estimating Equations (GEE) and Weighted Generalized Estimating Equations (WGEE), which permits the inclusion of time-dependent variables, the analysis of incomplete data depending on the mechanism of missingness, the exhaustive use of all the available information and the study of the effect of time on the outcome variable. To give visibility to these techniques, describe their methodology in a simple and accessible way to researchers, and show how to implement them in R. Methods A prospective observational study with repeated measures at baseline, 3 and 6 months follow-up was conducted in the Cardiology Services of the University Hospital Puerta del Mar (Cadiz) and the University Hospital of Puerto Real. The study population were patients admitted for acute coronary episode with confirmed diagnosis of Acute Myocardial Infarction or Unstable Angina according to clinical, biochemical and electrocardiographic criteria, and according to hospital discharge report. Patients with non-ischemic or non-cardiological precordial pain were excluded. The estimated sample size to detect a difference of 5 points in the SF-36 scores over time, with a 95% confidence level, and a power of 80%, was 247 individuals. Sociodemographic and clinical baseline data were collected after admission to the hospital, when the patient was clinically stable, by a trained interviewer, using a structured questionnaire and from the patients´ clinical records. Health Related Quality of Life (SF-36v1) and Mental Health (GHQ-28) were measured on 3 occasions (at baseline, 3 and 6 months after discharge). In the first paper, a descriptive analysis was performed, and the chi-squared and Student´s t-tests were used to compare the characteristics of patients that remained in the study with those that dropped out at 3 and 6 months. The Little’s test was used to check if the data were consistent with being “Missing Completely At Random” (MCAR). An ANOVA test for repeated measures was used to analyse the changes in SF-36 scores during the follow-up period. Partial eta-squared values were also used to estimate the effect size. To assess the evolution of each dimension of the SF-36 and to identify the predictive variables, a GEE model was performed for each one. The dependent variable in each model was the SF-36 scores of the corresponding dimension. To select the best model, we used a goodness-of-fit parameter which is a generalisation of the Akaike information criterion: the “corrected quasi-likelihood information criterion” (QICC), with the lowest possible values. In the second paper, the primary statistical analysis was the development of the code in R for the treatment of longitudinal databases with drop outs, the adjustment of Weighted GEE (WGEE) models and the performance of sensitivity analysis, as well as the simulation of the general case. To this end, the functions “geeglm”, “reshape”, “mice”, “complete” and “is.na” of the statistical package R were used. The analyses carried out in the third article were similar to those performed in the first paper. However, this time the analyses were performed on the entire sample (250 subjects from the two hospitals), the Mann-Whitney test was used for comparison between groups, and the Friedman test was used instead of the ANOVA to analyse the changes in SF-36 scores during the follow-up period. Besides, avoiding the assumption of MCAR, WGEE were used instead of GEE to assess the evolution of each dimension of the SF-36 and to identify the predictors of this evolution. In addition, a sensitivity analysis was performed to check the impact of an erroneous assumption about the missingness mechanism. Results In the first paper, as a first attempt to answer the research questions, we observed that Role Physical, Bodily Pain, General Health, Vitality and the Physical Component Summary of SF-36 improved over the follow-up. Being woman, older and having higher scores on GHQ-28 were associated to a decrease in HRQL throughout time. Previous history of coronary heart disease, co-morbidities, revascularisation, rehospitalisation, and episode of angina had a negative impact on HRQL, especially between 3 and 6 months after discharge. The main results in the second and third papers included the finding that the missingness was likely to be at random (rather than MCAR). All the dimensions of the SF-36 improved over time, except PF. Using WGEE, we found out that the factors related to the evolution of HRQL were: being woman (B=-23.9 in RE;B=-6.9 in MCS), older age (B=-0.5 in BP;B=-0.3 in VT), being single/separated (B=-14.5 in GH;B=-14.1 in SF;B=-23.3 in MH) and widow(er) (B=-23.2 PF;B=-29.8 in SF), hypertensive (B=-19.8 in RP;B=-8.9 in VT), worse mental health (B=-3 in PF;B=-2.8 in RP;B=-3.1 in BP;B=-1.2 in PCS;B=-3.8 in VT;B=-2.6 in SF), previous history of CHD (B=-12.5 in PF; B=-5.2 in PCS), and performing heart-healthy physical activities (B=13.9 in PF). Conclusions In view of the results obtained in the different studies included in the Doctoral Thesis, the following conclusions can be drawn: 1. The utility of GEE models to handle missing data and to analyse the evolution of HRQL with a longitudinal approach has been highlighted. 2. Simple GEE models are recommended, if the researcher is absolutely convinced that their outcomes are MCAR, and only in this case, as it is the simplest method leading to consistent and efficient results. 3. WGEE models are recommended for outcomes whose normality is not plausible, since consistency is guaranteed under MCAR and MAR if the missingness models are correctly specified. 4. As the missingness mechanism is uncertain, sensitivity analyses are needed to determine the robustness of an assessment by examining the extent to which results are affected by changes in the assumptions about the missingness mechanism, and if there are suspicions that the missingness might be not at random (MNAR), researchers are always encouraged to perform more comprehensive analyses to better understand the reasons for missingness. 5. Factors such as sex may partially explain the missingness in longitudinal studies, having an impact on the results. This highlights the importance of a correct treatment of the data before drawing clinical conclusions. 6. HRQL improved throughout time in CP, especially after 6 months, and the true magnitude of the effect of the diverse risk factors has been highlighted thanks to the use of WGEE. 7. Women have, on average, a worse evolution of Role Emotional and Mental Component Summary than men, and as age increases, Vitality decreases and Bodily Pain gets worse. The treatment and management of HRQL should be, therefore, personalised for each of these subgroups. 8. Single/separated people have, on average, a worse evolution of General Health, Social Functioning and Mental Health. Widow(er)s have, on average, a worse evolution of Physical Functioning and Social Functioning. It implies that, when treating these patients, the focus should be placed on the mental sphere in the case of single/separated people, and on the physical sphere in the case of widow(er)s. 9. People who perform heart-healthy physical activity have, on average, a better evolution of Physical Functioning. Hypertensive people have, on average, worse Role Physical and Vitality. Hence, healthier life and food habits should be promoted in these patients after the cardiac episode. 10. Previous history of coronary heart disease implies a worse evolution of Physical Functioning and Physical Component Summary. Knowing this, the recovery of these patients should be very focused on the physical aspects. 11. The evolution of mental health predicts changes of physical components of HRQL during the entire follow-up period, being the only factor significantly related to all the dimensions (except General Health). Consequently, as a standard, mental health should be a factor routinely considered when treating these patients. 12. In short, a global approach, including related factors such as age, marital status, performing physical activities, hypertension or mental health, is required in order to improve the HRQL in CP.