PP39 Research using Ankle Brachial Pressure Index Statistically Reassessed

Abstract

Background Ankle Brachial Pressure Index (ABPI) is the ratio of ankle to brachial systolic blood pressure. It is widely used as a non-invasive, relatively simple method for the diagnosis of peripheral arterial diseases (PAD) in symptomatic patients. It is also widely used to estimate PAD prevalence. Furthermore, low ABPI – as an indicator of atherosclerosis – was also used to predict cardiovascular morbidity and mortality in many epidemiological studies. Several researchers think that ABPI is overvalued and its use in epidemiological studies had attracted some statistical pitfalls. This paper aims to investigate these pitfalls, and to suggest possible advance statistical solution for some of them. Methods Critical review of literature Results The use of ABPI was proposed by studies comparing ABPI with peripheral angiography and Doppler ultrasound. However, most of these studies either had small sample size (issues with representativeness and generalisability of the result), compared asymptomatic legs with symptomatic legs using independent sample t-test (inappropriate test as legs of the same patients are not independent), or chosen cases from old age group and controls from young age group (age is a potential confounder which might had distorted their results). There were no attempts to test for the potential association between ABPI and other variables such as gender, obesity, cholesterol level and physical activity. Nevertheless, several researchers tested the correlation between ABPI and brachial systolic blood pressure and claimed that ABPI is usually lower in patients with high brachial systolic blood pressure even in the absence of PAD. Some of these researchers used correlation coefficients, however, using Pearson correlation test is inappropriate as brachial pressure is mathematically coupled to the ABPI, thus a null hypothesis of zero correlation is erroneous and the conclusion based on it should be rendered invalid. Ordinary Least Square Linear Regression model will be unsuitable as well for the same reason. Bland-altman plot or Ordinary Least Product Regression would be more appropriate. Furthermore, most of the studies that used ABPI as a predictor of cardiovascular morbidity and mortality had introduced some major statistical flaws in their regression analysis such as collinearity, mathematical coupling and reversal paradox. Such statistical issues would have distorted those studies’ associations. Conclusion Caution should be taken when using ABPI in epidemiological studies. Covariate selection and adjustment for confounder should be done carefully in regressing analysis. Directed Acyclic Graph (DAG) is a good method for addressing these issues.