Design and subjects

We examined all-cause mortality in a prospective cohort study with randomly selected, age matched controls of women 65 years of age or younger who had lived in the greater Stockholm area (5). All female patients were hospitalized for an acute CAD event (110 (37%) AMI, 182 (63%) UAP) between February 1991 and February 1994. Controls were randomly selected from the city census and were matched by age and catchment area. Age at entry (mean, SD) was 56 ± 7 years and equal in the patient (n = 292) and control (n = 300) cohort by matching.

The final available update of mortality status of participants (collection time between 15 September 2019 and 10 February 2020) linking the unique 10-digit person identification numbers to the community health care registers provided a follow-up time median 26.1 years, range 25.3–28.3 years. The status update was successful in 286 (97.9%) of 292 patients and 299 (99.7%) of 300 controls. Seven subjects moved away or were not registered in Sweden.

Data sources

The dataset of the Fem-Cor-Risk Study contained 292 female patients with CAD from Stockholm, Sweden, and 300 matched healthy female controls. Data about 160 variables were obtained from medical examination and survey responses shortly after study entry [5,11]. For this study, we examined two main criteria related to social strain and behavioral risk factors.

Social strain

1. Lack of social integration
   For the measurement a modified version of the Interview Schedule for Social Interaction with two main dimensions was used [14,15,29,34]: Social Integration (SI) measured the relationships to friends, neighbors, work associates and acquaintances. This includes a sense of belonging, practical help and support. Attachment (AT) measured close emotional ties.
2. Stress experience
   • Marital stress was measured by a structured interview using the Stockholm Marital Stress Scale [8]. One point was given for each question answered in a way that reported stress, and a point was given for every problem reported (infidelity, substance use/abuse, economic problems, health problems, or other unspecific problems). Higher scores indicate higher marital stress.
   • Work-stress was measured using the Swedish version of the Karasek demand–control questionnaire, which has been tested for consistency and reliability in the Swedish population [35]. Strain at work was derived from the ratio of psychosocial job demands to control. A large ratio indicated increased job stress, which occurs when demands exceed control.

Behavioral risk factors

1. Physical activity (PA) was assessed according to the World Health Organization criteria and graded into (I) reading, watching TV, or other sedentary leisure activities; (II) walking, cycling, or other forms of physical activity; (III) exercises to keep fit, heavy gardening, etc., for at least 4 h a week; and (IV) hard training or participation in competitive sports several times a week [36]. In the analyses, physical exercise in leisure time was dichotomized into sedentary lifestyle (I) and non-sedentary lifestyle (II, III and IV). Q-shaped patterns (I and III/IV vs. II) and linear (L-) patterns (I vs. II vs.III/IV) were examined separately. In an additional measurement, patients reported their present physical activity (posture) during the 24 h ECG measurement at leisure time 7 times a day (apply ECG, morning, after lunch, afternoon, after dinner, before sleeping, awakening). A measurement of physical activity in the morning and after lunch was specified as “Posture_d2”.
2. Smoking status was categorized as current, or non and previous smoker (1 year prior to the date of study recruitment).
3. Depressive symptoms were assessed using a short and convenient self-rating questionnaire, originally developed by Pearlin [37]. Nine common symptoms of depressive feelings are presented; the recent experience (preceding month) of each symptom is checked, and the number of symptoms is added to a score from 0 to 9. The scale had an adequate internal consistency (Cronbach’s alpha = 0·85) and was significantly correlated (r = 0.70; p < .0001) with the Beck Depression Inventory in a pilot study of Swedish women [11]. Vital exhaustion (VE) was measured by an earlier version of the Maastricht Questionnaire [11,20]. The scale yields 18 items and includes questions on fatigue, irritability, depressed affect, and personal accomplishment. The scale demonstrated excellent internal reliability (Cronbach’s alpha 0.93).
4. Disturbed sleep was measured by Karolinska Sleep Questionnaire (KSQ) [26,38]: The women were asked about mental fatigue, not feeling well-rested on awakening, disturbed or restless sleep, premature awakening, and difficulties falling asleep. The response alternatives were “never,” “rarely,” “some of the time,” and “most of the time.” A disturbed sleep index was computed by simply adding the item scores (range 0 to 15). A higher score indicated more frequently disturbed sleep. Additionally, we used a variable, sleep quality, in women who responded with “some of the time” or “most of the time. These women received 1 score point in each of three sleep questions: disturbed sleep, premature awakening and difficulties falling asleep. Patients who responded with “never” or “rarely” received no points in this examination (range 0 to 3). Cronbach’s alpha for the sleep quality index of the KSQ was reported to be .80 [38].

Covariates: Age at baseline examination, body mass index, menopausal status and educational level were assessed by standardized methods (S1 File). The severity of heart failure (Killip classification), systolic blood pressure, serum levels of cholesterol, HDL, and triglycerides, as well as CRP, coagulation factors, uric acid, alkaline phosphatase, AST, ALT and DHEAS, were assessed at clinical examination.

Coronary angiography and further measurements of the study related to left ventricular dysfunction are described in Supplemental material, S1 File.

In the ECG 24-h-HRV assessment, we included only HRV recordings >50% (“HRV-filter”) [22]. ECG recordings were excluded if they showed more than 10% nonsinus rhythm (seven patients), or less than 50% of the original ECG recording was available for analysis (eight patients) due to technical artifacts, atrial fibrillation or frequent arrhythmia.

Further psychological questionnaire data related to anger, hostility, and self-esteem were included in the Cox Boost model and described in the Supplemental material, S1 File.

Data were recorded in an updated dataset of Fem-Cor-Risk-study [5] with a follow-up of median 26.1 (range 25.3–28.3) years. The control cohort was analyzed separately using the same model-building strategy without cardiac markers of CAD and findings of coronarangiography.

The only endpoint was survival time characterized by all-cause mortality.

Data analysis

The first step of our analysis shows survival curves—all-cause mortality—of patients and controls. The differences in estimated survival probability are obvious, with different shapes for the two groups. The density plot of the matched ages shows agreement between patients and controls. We opted to reject the idea of a common model for analyzing risk factors, as controls and patients are different. In addition, several cardiological markers of CAD are not available for healthy controls.

Our main interest was the question of whether baseline psychosocial status might be among well-established clinical conditions predictors of all-cause mortality, even after a very long follow-up period of up to 26 years. The complexity of estimable logistic regression-based risk factors of survival rates in given periods, like 5-, 10-, and 15-year survival, is limited. Therefore, different models have been applied to the same data in the past [8]. Using individual survival times for a regression approach would allow assessing the effects of many potential risk factors simultaneously. Variable selection is mandatory, given the very rich set of baseline covariates of more than 80 per individual.

We used a method for boosted regression (COX Boost) [39] for variable selection, arriving at a sparse pragmatically constructed prediction model of survival (S2 File).

In an additional explorative analysis, a subgroup of patients stratified by age and social integration was analyzed further. Mortality predictors were calculated after 15 years, ten years after behavioral follow-up data were published by the Fem-Cor-Risk-Study [8,11]. A different impact of these predictors in the Cox Boost analysis could show the strengthening or weakening of the influence of a baseline predictor within the observation time of 15 to 26 years.

A. Related to our hypotheses, the following results emerged in CAD patients

1. Social strain
   • Lack of social integration and individual survival or mortality over 26 years are presented in Table 3A. We found an interaction between SI and age upon study entry (HR = 0.99, 95% CI 0.99–1.0, p = 0.042). The hypothesis related to lack of social integration seems to be supported by this analysis: older women died if they were socially not integrated. However, we found that younger women survived if they had less social integration (Fig 2). These results were only partly in line with our hypothesis: Patients with age around 65 and a lack of social integration (score = 10) showed a mortality of 63%, but patients aged around 45 with low social integration (score = 10) showed a low mortality (11%; Figs 2 and S2). The situation we found was much more differentiated than hypothesized. The variables attachment and household size were not selected in the model.
     

     Download:

     • PPT
       PowerPoint slide
     • PNG
       larger image
     • TIFF
       original image

     Fig 2. Age and lack of social integration among older women with CAD.

     There is an interaction between social integration and age. The predicted risk ratio for mortality increases with the degree of social isolation in older women (65 years) and decreases in younger women (45 years). This result could be plausible, as a lack of social integration could be an additional strain for older women.

     https://doi.org/10.1371/journal.pone.0277028.g002

     

     Download:

     • PPT
       PowerPoint slide
     • PNG
       larger image
     • TIFF
       original image

   • Stress experience
     Stress was no significant predictor for all-cause mortality at the 26-year follow-up:
     1. Marital stress was higher in patients with all-cause mortality. Due to missing values of patients without a partner, this variable was adjusted by comparing its regression with the total model, which was highly significant (HR 2.66, 95% CI 2.14–3.31, p = 0.001). An HR coefficient of 1.03, 95% CI 0.97–1.1, showed a tendency with risk, but a small overlap with no effect levels. In bivariate analyses without consideration of age, married patients had a higher survival than widowed, divorced, and single patients (Table 1; not adjusted). The small subsample of 99 patients living alone without partner and with depressive symptoms showed a high but not significant association with all-cause mortality.
     2. Work stress (HR 1.0, 95% CI 0.96–1.03) and homemaker status (HR 1.01, 95% CI 0.69–1.47) did not predict all-cause mortality after 26 years.
2. Behavioral risk factors
   1. Practising moderate exercise compared to a sedentary lifestyle (WHO criteria) was highly predictive for survival (HR 0.54, 95% CI 0.37–0.79, p = 0.002) (Fig 3A). In the sub-analysis of the women with regular-intensity exercise (n = 13), we found they were younger, smoked less, and had lower blood glucose than the sedentary and moderate exercise group but died earlier than patients in the moderate exercise group (Table 1).
      

      Download:

      • PPT
        PowerPoint slide
      • PNG
        larger image
      • TIFF
        original image

      Fig 3. Predicting survival for exercise and non-smoking in women with CAD.

      HRs do not allow quantifying effects on survival directly, therefore, predicted survival curves are displayed for a selection of covariates from the final model. For all plots, orange codes the baseline survival curve, cyan codes survival for relevant values of the covariate considered. a) Exercise. Moderate Exercise seems to be beneficial, showing an increase in survival of 20% appr. b) Smoking. Smokers might have a reduced chance of survival during follow up of up to 17% appr.

      https://doi.org/10.1371/journal.pone.0277028.g003

      
      PA (posture) during 24 h ECG showed the same result as the WHO criteria but did not reach predictor quality for survival (HR 0.92, 95% CI 0.84–1.01, p = 0.073) (Table 3A).
   2. Smoking after 26 years was a significant predictor of all-cause mortality (HR 1.56, 95% CI 1.03–2.36) (Table 3A and Fig 3B).
   3. Depression did not become a significant predictor for all-cause mortality (HR 1.01, 95% CI 0.94–1.08). Vital exhaustion (VE) did not reach the list of the most important 20 predictors in that model.
   4. Disturbed sleep was not a significant predictor of all-cause mortality (HR 1.06, 95% CI 0.71–1.59). Sleep quality was not included in the model.

As expected, other well-known baseline cardiovascular risk factors showed an HR between 1.18 and 2.85 and were significant predictors for all-cause mortality (Table 3A). DHEAS as an indicator of estrogen protection predicted survival (HR 0.67, 95% CI 0.46–0.99) in this model (Table 3A). A calculation of predictors of the 15-year follow-up showed similar results compared to the 26-year follow-up. Exercise (HR 0.42, 95% CI 0.24–0.74, p = 0.003), became significant and SI × age showed a statistical tendency (HR 0.99, 95% CI 0.98–1.0, p = 0.058). Interestingly at 15-year follow-up, current smoking was not significant.

B. Predictors in age-matched healthy controls

This study made it possible to compare these 26-year follow-up predictors of all-cause mortality in CAD patients with predictors of mortality in age-matched healthy women (S1 Fig). A high proportionality (global score and standard deviation of normal-to-normal R-R interval (SDNN)), possibly due to the low number of mortality cases in this presumably healthy group, made this model less convincing than the patient model. Age was again a significant predictor (HR 1.13 95% CI 1.08–1.17; p = 0.001). Related to our hypotheses we found for SI in this group no associations with all-cause mortality as for work stress. Cardiac indicators possibly associated with behavioral stress SDNN (HR 0.98, 95% CI 0.96–1.0, p = 0.076) and HRV without dysrhythmias (HR 0.20, 95% CI 0.81–0.48; p = 0.001) emerged as predictors in the model. Related to behavioral risk factors, PA became again a signinificant predictor (HR 1.62 (95% CI 1.10–2.40; p = 0.016) (Fig 4). Current smoking (HR 1.59, 95% CI 0.93–2.72; p = 0.089) and depressive symptoms (HR 1.89, 95% CI 0.89–3.99; p = 0.097) were included in the predictive model but were not found to be significant (Table 4). Vital exhaustion and sleeping disturbances became not part of the model.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Predicting survival for exercise in healthy women.

Moderate exercise seems to be beneficial, showing an increase in survival of 10% appr.

https://doi.org/10.1371/journal.pone.0277028.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Model for controls after 26 years.

https://doi.org/10.1371/journal.pone.0277028.t004

Female CAD patients

In this study on CAD females we could confirm that „few social contacts“, „low physical exercise”and „smoking”remained significant in the model, which was able to predict all-cause mortality in CAD patients after 26 years.

1. We found a significant interaction of social integration with age. As hypothesized and confirmed in other studies [15,29], low social integration or social isolation in older female patients (>55 years) increased all-cause mortality. This was consistent with the Berlin Aging study. “Living alone” (present more in women than in men) and feelings of “emotional loneliness” in the significantly older study group (mean age 84.5 years) predicted total mortality [43]. A large review also came to similar conclusions [33]. Against our hypothesis and unexpected, in younger patients, a lack of social integration did not predict mortality. Instead an unfavourable outcome in the long-term follow-up was predicted by frequent social interactions. Whether frequent social activities are dangerous to a special form of CAD in younger women [44] after an acute coronary event or a gender-specific aspect of the female reproductive system are involved, we could partly clarify looking on DHEAS, a precursor of oestrogen associated with the ageing process and in our study an important predictor for survival. Consistent with the finding that low oestrogen levels do not protect female CAD patients [44], low DHEAS and low social integration led to high mortality in patients over 55 years of age. Patients ≤55 years with low social integration and presumably higher DHEAS levels than in patients >55 years, had low mortality. The buffer hypothesis of the primary objective in the FemCorRisk study that parts of the reproductive system protect against social strain, seems to be supported by this finding.

In contrast to a review [1], another long-term follow-up study in the USA [4] and the finding of an association between low social education and coronary stenosis narrowing in the Fem.Cor-Risk angiography study [12], low education, possibly an indicator of social strain, did not predict all-cause mortality in our model. In the individual analysis, we found no effect in healthy controls and a very small effect contrary to our expectation among patients: patients from lower social class showed marginally lower mortality. This finding is in contradiction to a review on Nordic countries (including Sweden) that demonstrated relatively high levels of mortality inequalities in Sweden, although the socio-economic differences in the Swedish population were not as pronounced as in other countries [45]. In addition, effects differ between men and women and influences from other risk factors [46] that are related to the influence of education on mortality in different countries. Indeed, the importance of the educational level seems to diminish with age among women and is negligible at the age of 85 [47] Another possible explanation could be that the cut-off in education was set between mandatory education and upper-secondary school. Statistics indicate that the difference in life-length is more pronounced between those with upper-secondary education or lower and higher education. Also, the finding that low education does not predict all-cause mortality in these Swedish CAD patients could be related to several influencing factors: The data of CAD patients admitted to any of the 10 coronary care units in Stockholm for an acute coronary event and the healthy controls included in this study were representative for Stockholm County and may differ to the general population. Although, while this study included the major confounders, potential effects of unmeasured characteristics and direct and indirect selection mechanisms may also play a role.

Behavioral risk factors:

1. The exploratory analysis demonstrated that PA was associated significantly with survival in female CAD patients 26 years after the first cardiac event. In the 9-year follow-up of this sample, a sedentary lifestyle was already a significant predictor of total mortality [27]. In several studies in male and female CAD patients and in epidemiological surveys, a sedentary life style was at risk for further cardiovascular events [28]. A large meta-analysis in women demonstrated the benefits of PA in relation to all-cause mortality [48]. The 26 years long-term follow-up in 287 women with CAD could confirm the fact that PA is beneficial compared to a sedentary life style. PA was associated with a significantly reduced risk of overall mortality in both men and women (HR 0.72, 95% CI 0.62 to 0.84 and HR 0.81, 95% CI 0.72 to 0.92, respectively[48]. When analysing in this study [49] the dose–response association of continuous minutes of weighted moderate to vigorous physical activity per day at baseline, authors observed a reduction in the risk of incident CHD which was specially marked between 20 and 40 min of physical activity per day. In our study, moderate exercise was more beneficial than in the very small group of CAD patients who did regular-intensity exercise. A high dose of “physical activity” in women with CAD could be unfavorable. Despite the low sample size in this group this result is interesting. Then there is an ongoing discussion about the proper amount of exercise training in cardiac rehabilitation [28].
   Our results argue for cautious use of this preventive measure in women with AMI or UA.
2. Smoking showed a high predictive power for all-cause mortality after 26 years in these female CAD patients. Smoking after 15 years was not a significant predictor. These differences in probabilities in 15 and 26 years of follow-up indicate a higher power of this predictor with the length of the observation time. Current smokers had a higher risk than former smokers, but we did not calculate the packages/year of cigarettes in our study. As in a 20-year follow-up [50], the severity of smoking and the level of left ventricular dysfunction seemed to influence the proportion of all-cause mortality.

Due to the length of time since the first cardiovascular event, other hypothesized social strain and behavioral risk factors did not reach significant predictor qualities. However, several non-significant HRs with a low 95% confidence interval and a statistical trend point to the direction, strength, and precision a variable might predict in this explorative analysis:

Stress experience: After selecting only married or patients with a partner, this follow-up found that marital stress was included in a significant model but did not reach statistical significance. So the predictor qualities of marital stress diminished during long-term follow-up. An association between work strain and 26-year mortality in CAD patients was weak. Work strain was included in the model as a hypothesized predictor but did not show significant predictor qualities. Because patients were enrolled in this study with a median age of 56 years, most patients had already retired after 26 years. In former studies with shorter follow-up times, work strain had a higher impact [6,7,51]. In a national representative cohort of French employees, the associations of current exposure to job strain and mortality were found to be higher among women (HR = 1.15 (95% CI: 1.06–1.25) [52].

Depression entered the model as a possible predictor of mortality. Many studies reported depression as an important predictor of MACE or atherosclerotic coronary progression [15,18,19]. The low HR value of the prediction could be due to a decreasing effect over time or the kind of measurement: In our study, depressive symptoms were measured by the Pearlin scale [37] and not by an instrument to detect diagnoses of severe depression. The somatic depression subtype measured with this scale did not become a part of the model. One review found that only some depressive symptoms could predict mortality in women 56–69 and > 70 years of age [53]. In long-term follow-up studies, depression was a predictor in male but not female CAD patients [54]. In our study with unselected female CAD patients, depressive symptoms became not significant as a predictor for 26-year mortality. VE was not chosen for the final model in our study. There could be a gender effect; our study focused on women, other studies mostly examined males and only in a smaller part female [54]. The low prediction quality could also be due to a diminishing effect over time.

Sleep disturbances influenced the development of cardiovascular diseases in CAD patients [26] and in the healthy Fem-Cor-Risk control sample after 9 years [32]. Our hypothesis that sleep disturbances influence mortality at very long-term follow-up could not be confirmed. Additional collection of sleep data during a longitudinal study, as well as the inclusion of other variables (depression, fatigue, snoring, work strain), would enhance the quality of the predictor analysis.

Healthy female controls

In the study on healthy females we could confirm that „low physical exercise”remained significant in the model, which was able to predict all-cause mortality in healthy controls after 26 years. Smoking and depressive symptoms became part of the model, social class (low income, education), few social contacts, high body weight, and high alcohol consumption were not included in the model [46]. Due to the small number of deaths and the large deviations from the proportional hazard assumption of the model in the healthy controls after the 26-year follow-up period, the results should be treated with caution.

Some results in controls are in good agreement with the CAD-patients:

Behavioral risk factors

1. PA: In concordance to the results in CAD patients, a sedentary lifestyle or regular intensive exercise versus moderate activity was a significant behavioral predictor for all-cause mortality in healthy women. In a meta-analysis of 59 studies in women [48], PA was associated with a lower risk for all-cause mortality (HR 0.71 (95% CI 0.65–0.78).
2. The prediction of current smoking after 26-year follow-up was higher than former smoking, but both became not significant. In a meta-analysis [48] in healthy women, current smoking showed a higher risk (HR of 2.22 (95% CI 1.92–2.57)) for all-cause mortality. The severity of smoking seems to play a role in this context.
3. In the healthy control group, depression also appeared as a predictor for mortality in the model (HR 1.89, 95% CI 0.89–4.0; p = 0.097)), but this was not significant. In a sample of healthy males, depression predicted fatal ischemic coronary disease events after 20.9 years follow-up (adjusted HR 1.07, 95% CI 0.99 to 1.15; p = 0.08) [54].

Limitations

1. The basis of this examination was the complex data pool of the Fem-Cor-Risk study and all-cause mortality data. A statistically pre-planned analysis and different surveys within the follow-up period with time-dependent covariate information would have strengthened their explanatory power. When examining survival time for all-cause mortality in our cohorts, we received limited information about cardiac death, recurrent MI or PCI during the first years of follow-up [8,22,27]. Assuming we had the complete information, a much more appropriate method of analysis would have been possible. For healthy controls several cardiac markers of CAD were not available We opted to reject the idea of a common model for analyzing risk factors, as controls and patients models are different.
2. For the calculation in this exploratory study, we used an established machine-learning-based statistical method (Cox Boost multiple regression on survival-times [39]) to include the complete set of baseline variables. We did not provide formal testing between patients who survived and died, avoiding misinterpretation of apparent relations.
3. In the first year, the severity of CAD (left main stenosis and 3-vessel disease) is an important risk factor for cardiovascular events [59]). The impact of social and behavioral cardiovascular risk factors in the Fem-Cor-Risk sample was presented after three to five years [5,8,27,30] in a situation where patients faced acute disease limitation. Due to life changes and intervening influences in the long follow-up period, the impact of these factors became blurred.
4. d. This study started with the morphological knowledge about coronary arteries at that time. Today, endothelial dysfunction, reduced microcirculation and myocardial perfusion measured by SPECT, or coronary computed tomography angiography [60] can specify more clearly patients with and without obstructive CAD on different vessels [61]. We could not show whether most of these specifications or the effect of new treatments are related to the behavioral factors studied here.
5. e. Psychological and sociological rating instruments known to be potentially important for prognosis were used at baseline. New psychological constructs, for example, anxiety, negative affectivity or substance abuse, were not applied but might have more predictive power.

Our results show at first time in female CAD patients strong predictors of all-cause mortality at 26 years. These results should be replicated and extended by longitudinal studies during the follow-up period and by collecting data on cardiovascular events and cardiac mortality. Future studies will show whether new cardiac or behavioral diagnostic procedures and treatments will elicit additional predictors of survival.

We were able to present with the support of a novel statistical method based on machine learning new findings on the course of CAD women and their healthy controls and open the window for long-term socio-behavioral studies on cardiac disease progression. This could initiate new research on long-term interactions between social strain, behavioral, cardiac and endocrinological factors, including the identification of their psychobiological pathogenetic mechanisms.