A classification system for identifying persons with an unknown cardiovascular disease (CVD) status for a...

The need for classification systems for cardiovascular disease (CVD) that is population-specific is important towards understanding the clinical disease and diagnostics associated with the disease. This paper presents the form and validation results of this classification system.

The survey data used was captured from 778 participants, 526 persons with no prior CVD, and 252 who reported prior CVD. Binomial logistic regression and Discriminant analysis were utilised to develop classification models. This classification system provided a general measure of severity of disease by utilising scores estimated from two algorithms developed from 13 routine physiologic measurements, along with demographic information of age and ethnicity, inter alia, and previous health status.

For each model, specific score ranges were identified, which gave the best classification for those with a prior CVD incident (higher scores) and for others labelled as non-CVD (lower scores). The two classification models (Logistic Regression Model and Discriminant Analysis Model) developed had high area under the receiver-operating characteristic (AUROC) values (98% & 99%) and sensitivity (86 and 90%), which improved discrimination between Non-CVD and CVD participants and, more importantly, correctly classified a greater proportion of CVD participants. New to this type of research was the estimation and detailed evaluation of a range of scores, labelled non-differentiating, which fell in the middle of the spectrum and which contained the higher-end scores for the non-CVD individuals and the lower-end scores for CVD patients, all of whom were incorrectly classified, based on their prior history.

The classification system of scores is able to differentiate the CVD status of individuals, with good predictability, and could assist physicians with recommending different treatment plans. The two models in this classification system each individually outperformed the three established models in terms of the strength of their correct classifications of individuals with or without prior reported CVD incidents. More importantly, they have smaller non-differentiating ranges than the three known models and, in that range, the two new models have lower CVD/non-CVD ratios suggesting they are more likely to misclassify non-CVD individuals compared to CVD patients, which is a more benign misclassification. Further, when used in combination, the two models increased the sensitivity, in classifying individuals of different ethnicities, beyond that of either one used independently or of any of the three standard European/North American models. These efforts will be instrumental in advancing personalised CVD management strategies and improving health outcomes across diverse populations.