Effects of data preprocessing on results of the epidemiological analysis of coronary heart disease and behaviour-related risk factors

Abstract

Tuomainen (2021) Effects of data preprocessing on results of the epidemiological analysis of coronary heart disease and behaviour-related risk factors,

Epidemiological analysis of coronary heart disease and its main risk factors: are their associations multiplicative, additive, or interactive?

“…In this study, we used the fasting serum low-density lipoprotein cholesterol (S-LDL-C) concentration instead of the fasting serum total cholesterol (STC) concentration to detect high cholesterol levels. In our previous paper, we had used STC [ 26 ], but preliminary analyses of this study revealed that S-LDL-C could provide even stronger associations with CHD. Based on S-LDL-C concentration we categorised study participants as follows: S-LDL-C < 3.4 mmol/L indicating below borderline concentrations, S-LDL-C 3.4 − 4.1 mmol/L indicating borderline high concentrations, and S-LDL-C > 4.1 mmol/L or cholesterol-lowering medication indicating high concentrations.…”

“…We also excluded measures of physical activity since the proportion of physically inactive men in the KIHD cohort is practically zero and, consequently, it is impossible to estimate the impact of sedentary lifestyle on the risk of CHD. Moreover, based on our previous analyses physical activity levels very poorly predict the risk of CHD in this cohort [ 26 ]. Other studies also have found the contradictory nature of physical activity, as in some study populations physical activity, not inactivity, is associated with an increased risk of CHD [ 9 ].…”

“…We used the covariates as categorical variables and investigated the Schoenfeld residuals to test the proportional hazards assumption [ 40 ]. Based on our previous study the type of covariate, categorical or continuous, has probably no statistically significant effect on Cox model results [ 26 ] but including continuous covariates in the model requires the assessment of nonlinearity by means of martingale residuals [ 41 ]. Second, we added statistically significant covariate interactions to the Cox model.…”

“…On the other hand, this study focussed only on White Finnish males without detailed information regarding their genetic risk for CHD, which limits possibilities to generalise its findings [ 57 ]. Moreover, this study refers specifically to CHD, ICD 10 codes I20 − I25, as an endpoint, and because the main CHD risk factors are to some extent outcome sensitive [ 26 ], i.e. explanatory powers and the order of importance among risk factors slightly differ across endpoints, such as CHD, an acute myocardial infarction (AMI), and a fatal AMI, the present results are not directly generalisable to AMI and sudden cardiac death, although them are typical manifestations of CHD.…”

“…explanatory powers and the order of importance among risk factors slightly differ across endpoints, such as CHD, an acute myocardial infarction (AMI), and a fatal AMI, the present results are not directly generalisable to AMI and sudden cardiac death, although them are typical manifestations of CHD. Our previous paper provides analyses to verify this outcome sensitivity [ 26 ]. To control possible bias originated from inaccuracies related to exclusions of study participants based on the ergometer test we executed statistical analyses and reported their results also concerning the dataset without exclusions.…”