Machine learning identifies factors related to obesity, dyslipidemia, and inflammation as predictors of noncalcified coronary burden in patients with psoriasis, according to study results published in the Journal of the American Academy of Dermatology.
Managing risk for cardiovascular disease (CVD) and cardiac events in patients with psoriasis requires accurate assessment of risk factors and disease predictors. Machine learning algorithms improve the predictive power of clinical and imaging data and provide greater prognostic capacity for CVD risk stratification. Researchers aimed to apply machine learning to determine the top predictors of noncalcified coronary burden in patients with psoriasis using random forest algorithms.
Investigators collected data from 263 patient records (33% women) from the Psoriasis Atherosclerosis Cardiometabolic Initiative. Clinical data generated from blood draws and coronary computed tomography angiography were used to produce imaging data. Researchers analyzed a total of 62 variables by permutation through a random forest algorithm. They manually removed variables from the dataset and created decision trees using the machine learning algorithm, which subsequently assigned each variable an importance value (≤1) according to its predictive power. They then performed linear regression between each predictor variable and noncalcified coronary burden to calculate a correlation coefficient, β, which also indicated whether the association was positive or negative.
The strongest predictors of noncalcified coronary burden in patients with psoriasis were body mass index (importance value [IV] 0.66; β 0.64; P <.001) and visceral adiposity (IV, 0.64; β, 0.58; P <.001). Total adiposity (IV, 0.41; β, 0.54; P <.001), subcutaneous adiposity (IV 0.15; β 0.34; P <.001), and small low-density lipoprotein particle (IV 0.13; β 0.27; P <.001) were also positively associated with noncalcified coronary burden. Apolipoprotein A1 levels (IV 0.22; β −0.4; P <.001), high-density lipoprotein (IV 0.19; β −0.42; P <.001), and cholesterol efflux capacity (IV 0.11; β −0.28; P <.001) all showed significant negative associations with noncalcified coronary burden.
Investigators identified erythrocyte sedimentation rate among the top 10 predictors of disease burden by machine learning (IV 0.17) but did not show a significant association by linear regression (β −0.05; P =.52).
The researchers noted that the study was limited to analysis of a single baseline value for each patient and that further follow-up may improve stratification. In addition, validation with an external cohort was not performed.
The researchers believe that the study findings “highlight the importance of features related to obesity, dyslipidemia, and inflammation in predicting non-calcified coronary burden in psoriasis patients and also demonstrate how well-characterized datasets can be leveraged using machine learning algorithms to facilitate exploring the determinants of non-calcified coronary burden by [coronary computed tomography angiography].” They continued, “Further investigation into these top predictors of non-calcified coronary burden over time may provide insight into the treatment of inflammation and comorbidities in psoriasis to reduce [CVD] risk.”
Disclosure: Funding for the study was provided by the Colgate-Palmolive Company; Elsevier; and Genentech, Inc. Several study authors declared affiliations with the pharmaceutical industry. Please see the original reference for a full list of disclosures.
Munger E, Choi H, Dey AK, et al. Application of machine learning to determine top predictors of non-calcified coronary burden in psoriasis: an observational cohort study [published online October 30, 2019]. J Am Acad Dermatol. doi:10.1016/j.jaad.2019.10.060