Population Attributable fraction Estimation of the Cardiovascular Diseases–related factors: An Evidence from western part of Iran

Background: Cardiovascular diseases are the first cause of deaths and years of lost life due to disability, worldwide. The population attributable fraction has been widely used in literature to quantify an impact of removing risk factors in occurrence of diseases. This study was designed to estimate the population attributable fraction of prevalent cases of cardiovascular diseases associated with different risk factors, using data from people aged 35-65 years old in Ravansar County. Methods : Data of this study came from 9825 adults aged 35-65 years that were included in the study of Ravansar Non-Communicable disease (RaNCD). First, to identify the confounding variables, we did a comprehensive review of the available resources and then presented the relationship between different risk factors with directed acyclic graphs and for any risk factor. Sequential and average adjusted attributable fractions were used. To calculate 95% confidence intervals, the Monte Carlo simulation was conducted. All statistical analyses were performed using the ’averisk’ package in R version 3.4.4. Results: The age adjusted prevalence for cardiovascular diseases was 12.6% (95% CI: 11.9, 13.2%). Among the modifiable predictors of cardiovascular diseases, the highest amount of population attributable fraction after adjusting for age, sex, and other factors associated with cardiovascular diseases, in men were for hypertension 35.7% (95% CI: 30.2, 40.6%), dyslipidemia 12.2% (95% CI: 7.5, 17.6%), general obesity 5.7% (95% CI: -0.7, 11.9%) and for cigarette smoking 4.0% (95% CI: -6.3, 14.6%), and in women were hypertension 42.8% (95% CI: 36.5, 47.3%), general obesity 10.7% (95% CI: 6.2, 15.3%), central obesity 10.1% (95% CI: -7.0, 23.8%) and for dyslipidemia 9.7% (95% CI: 5.1, 14.5%). Conclusion : Due to the difference in the population attributable fraction in Our from the reduction in the prevalence of CVD in the community through appropriate interventions based on indigenous evidence of the

risk factors may sum up to more than 100% which is unconvincing [11]. Accordingly, some recent studies have addressed the need for new methods of computing PAF [12,13].
However, in observational data, estimating the average population attributable fraction using direct method is more promising than other existing methods of adjusted PAF calculation [13]. Accordingly, for PAF calculations in the recent studies, this new approach is used [12,13]. The aim of this study was to determine the PAFs of CVD-related factors in Ravansar. Methods:

Study population
For this study, data came from the baseline of the Ravansar Non-Communicable disease (RaNCD) cohort study. In RaNCD cohort study 10065 individuals aged 35-65 years were included, of which after clearing the data, 9825 (97.6%) were finally incorporated into this study. RaNCD is one of the 19 centers designed to study the prevalence and incidence of non-communicable diseases, including CVD, in the form of a large and nationwide prospective epidemiologic research studies in Iran (PERSIAN). The first phase of the study began in November 2014. Details of this study has been published elsewhere [14,15].

The study variables
In this study, following the review of available literature and the expert opinions, the risk factors for CVD were identified [6,7]. Age was divided into 2 groups; ≥ 45 and ≥ 55 years for males and females as high-risk age, and less than 45 and 55 years for male and females as low-risk age [16]. For cigarette smoking, those who used to smoke 100 cigarettes or more in the past or in the present were considered as smokers. Following self-reported variables, the marital status (single/ married), job status (unemployed/ employed), education status (illiterate/ literate (elementary and higher)), the use of hormone replacement therapy, contraceptive pills, menopause, substance abuse, alcohol consumption, family history of CVDs, and the area of residence, were used in analysis. In this study, dyslipidemia was defined as people who have one or more of the following; cholesterol ≥ 240 mg/ dl, triglyceride ≥ 200 mg/ dl, low-density lipoprotein ≥ 160 mg/ dl, high-density lipoprotein ≤ 40 mg/ dl and/ or people who used lipid-lowering drugs [16,17]. Hypertension also defined as systolic blood pressure ≥140 mmHg and/ or diastolic blood pressure ≥ 90 mmHg or use of lowering blood pressure medication [16,18]. People with fasting blood sugar ≥ 126 mg/ dl and/ or those who consumed anti-diabetic drugs were referred to as diabetic individuals [19,20]. Individuals with BMI ≥ 30 were considered as obese [18]. The waist to hip ratio (WHR) ≥ 0.95 in males, and ≥ 0.90 in women was considered as cut off for central obesity [21]. Depression was defined according to a person's self-report and/ or doctor's diagnosis and/ or use of antidepressants. Physical activity data were collected, using a standard questionnaire. The physical activity level was calculated as MET h/ wk [22]. Subjects were then divided into two groups: light activity (MET < 40) and heavy (MET ≥ 40), based on their physical activity. The socioeconomic status was calculated, based on durable goods. For the study population, using PCA [23]. For the purpose of this study, CVD defined as having ischemic heart disease, myocardial infarction and stroke, by self-report and/ or related drug use [24].

Data analysis
First, in order to identify the confounding variables, a comprehensive review of literature was conducted. Using current knowledge, several directed acyclic graphs (DAGs) was used to determine covariate adjustment sets for relationship of each risk factors and CVD. Figure 1, shows an example of DAG developed for association between WHR and CVD. In appendix 1, we presented the DAG for other risk factors.
For any risk factor, Sequential and average adjusted attributable fractions were used. To calculate 95% confidence intervals, the Monte Carlo simulation was conducted. in terms of the theoretical expression comprises the following steps: 1.
The risk factor has to be expressed in two distinct categories; in addition, the risk factor has to be eliminated from the study population, regardless of the real status of the subjects, by categorizing all as unexposed.

2.
A logistic model has been applied, using this modified dataset in order to estimate the anticipated probabilities for each subject.

3.
The total of all anticipated probabilities is the adjusted number of cases of the disease that would be expected if the absence of risk factors in the study population.

4.
Then, the AF is equal to the expected cases minus the observed cases, divided by the observed cases.
Any dichotomous risk factor used in the logistic regression model undergoes the same procedure. This is also applied for a sequential attributable fraction, which is the case of sequential elimination of the risk factor of the model. This process is repeated for all possible sequences of exit risk factors from the regression model, and the average estimated values of PAF are considered as the average PAF of the desired risk factor (average PAF). This procedure is named "average attributable fraction" by Eide and for dyslipidemia 9.7% (95% CI: 5.1, 14.5). In terms of numerical value, the total PAF attributed to the total number of CVD was higher in women, compared to men. The PAF to the total CVD was 93.4% and 97.7%, in men and women, respectively (Table2).
The PAF comparison for each gender-specific risk factor is shown in figure 2, which illustrates how the PAFs differ in the CVD-associated risk in men and women. The most differences in adjusted PAF for WHR, 6.8 units was observed between men and women ( Figure 2)

Discussion
The association between classic risk factors and CVD has been proven, but even for those risk factors with the same effect size, they can produce a completely different effect on the occurrence of disease due to the difference in the prevalence, so the calculation of PAF must be performed, using the population of the target community. The PAF of socioeconomic status was low for CVD. It was higher in men (6.7%) than in women (3.8%), which is not consistent with the results of the study conducted in Denmark (10% in men vs. 15% in women) [29], this difference is probably due to the use of different demographic groups, different computational methods, and ultimately different definitions and cut-off points for included variables.
In the present study, high-risk age in men (15.0%), in comparison with women (6.3%) had a higher PAF, which was consistent with the results of two population based cohort studies conducted in Iran (36.1% in men vs. 16.6% in women and 42% in men vs. 22% in women) [26, 27] and a study from China (11.4%) [12].
The family history of CVD, in the presence of other variables, had a higher PAF in men (1.5%), compared to women (0.7%), which was not statistically significant in both genders.
The results of this study are consistent with the results of the study from Sweden (9% in men vs. 3% in women) [28], but different from the results of two population based cohort studies conducted in Iran (3.9% in men vs. 7.6% in women and 2.5% in men vs. 6.8% in women) [26,27]; This difference is partly due to inclusion of different population.
In interpreting the results of studies on PAF, it should be considered that the reduction in the incidence of disease attributed to each risk factor is a theoretical concept. Because in practice, one can never reduce the exposure to a risk factor to zero in the community, or no one can design an intervention that, while maintaining the exposure to other risk factors, affects only one particular risk factor [35]. As a result, PAF values are more appropriate to prioritize the impact of risk factors at the community level, rather than planning, to achieve practically, this level of reductions.

Study Strength and Limitations
For the purpose of this study we used the prevalent cases. Thus, due to lack of knowledge regarding exposure-outcome associations of individuals in such studies, the identification of temporality between exposure and outcome is difficult. Therefore we cannot discuss about causality. Despite these limitations, there are several strengths in current study that are including use of population-based Information; large sample size, and high quality data. In addition, the accuracy of PAF estimation is dependent on the correct detection of confounders of each risk factor, so the directed acyclic graph was used.

Conclusion 13
Given the difference in the PAF's ranking of risk factors in men and women, prioritizing gender-specific preventive CVD interventions are recommended. In this regard, special attention should be paid to the control of high blood pressure, dyslipidemia, in men, and high blood pressure, general obesity, and dyslipidemia in women. Our results from a community based study provide necessary information needed to achieve the maximum reduction in the prevalence of CVD in the community through appropriate interventions based on indigenous evidence of the Country.  Figure 1 An example of DAG for association between WHR and CVD Figure 2 The comparison adjusted PAF of CVD

Supplementary Files
This is a list of supplementary files associated with the primary manuscript. Click to download. Appendix.docx