Abstract
Background:
Psoriasis is associated with elevated risk of heart attack as well as increased accumulation of subclinical non-calcified coronary burden by coronary computed tomography angiography (CCTA). Machine learning algorithms have been shown to effectively analyze well characterized datasets.
Objective:
In this study, we used machine learning algorithms to determine top predictors of non-calcified coronary burden by CCTA in psoriasis.
Methods:
The analysis included 263 consecutive patients with 62 available variables from the Psoriasis Atherosclerosis Cardiometabolic Initiative. The random forest algorithm was utilized to determine top predictors of non-calcified coronary burden by CCTA. We evaluated our results using linear regression models.
Results:
Using the random forest algorithm, the top 10 predictors of non-calcified coronary burden were: body mass index, visceral adiposity, total adiposity, apolipoprotein A1, high-density lipoprotein, erythrocyte sedimentation rate, subcutaneous adiposity, small low-density lipoprotein particle and cholesterol efflux capacity. Linear regression of non-calcified coronary burden yielded results consistent with our machine learning output.
Limitation:
We were unable to provide external validation and did not study cardiovascular events.
Conclusion:
Machine learning methods identified top predictors of non-calcified coronary burden in psoriasis. These factors were related to obesity, dyslipidemia, and inflammation demonstrating that these are important targets to treat comorbidities in psoriasis.
Keywords: Psoriasis, machine learning, random forest algorithm, atherosclerosis, coronary artery disease, cardiometabolic disease
CAPSULE SUMMARY:
Inflammation and dyslipidemia are known to play major roles in the development of atherosclerosis.
Our machine learning methods identified top predictors of coronary artery burden in psoriasis patients, which were markers related to obesity, dyslipidemia, and inflammation, demonstrating that these are potentially important comorbidities to treat in psoriasis.
INTRODUCTION:
Psoriasis, a chronic inflammatory disease, is associated with elevated rates of acute coronary syndrome, stroke, and cardiovascular mortality.(1, 2) Patients with psoriasis develop accelerated atherosclerosis which leads to an increase in coronary artery disease and its complications such as incident myocardial infarction. (3, 4) For example, patients with psoriasis have greater coronary artery disease burden which is predominantly non-calcified on coronary computed tomography angiography (CCTA). (3, 4) Given the acceleration of myocardial infarction risk, characterization of non-calcified coronary burden via CCTA may yield valuable information prior to cardiovascular events.(3, 5)
Preventive cardiology hinges upon the ability to accurately assess CV risk and predict prospective major adverse cardiac events. Conclusions derived from large prospective studies demonstrate our ability to calculate the long-term likelihood of major adverse cardiovascular events and to elucidate important risk factors of cardiovascular disease. Current statistical models rely on only a few variables in order to predict complex outcomes. Machine learning algorithms open the opportunity to map multiple complex data variables to clinical outcomes that are crucial for the advancement of our understanding of cardiovascular disease risk factors. Thus, machine learning is potentially well-equipped to use multiple variables to predict a complex outcome.
By combining clinical and CCTA data via machine learning, performance from machine learning algorithms have been shown to be a superior predictor of 5-year all-cause mortality than current clinical or CCTA data alone.(6) Another study demonstrated that a risk score developed by utilizing machine learning algorithms had greater prognostic accuracy for cardiovascular disease risk stratification from CCTA readings than the standard CCTA integrated risk scores.(7) Similarly, machine learning algorithms can be applied to better understand the relationship between various clinical variables and non-calcified coronary burden in psoriasis. Thus, we hypothesized that machine learning algorithms would be able to accurately determine the top predictors of non-calcified coronary burden in psoriasis.
METHODS:
Patient Population
The machine learning algorithm was developed using 263 consecutive patient records (January, 2013 to January, 2018) with 92 phenotypic variables measured at baseline from the Psoriasis Atherosclerosis Cardiometabolic Initiative, an ongoing prospective trial to understand the relationship between psoriasis and cardiometabolic diseases (Supplementary methods).(8)
CCTA
Acquisition:
All participants underwent coronary computed tomography angiography (CCTA) on the same day as blood draw, using the CT scanner (320-detector row Aquilion ONE ViSION, Toshiba, Japan). Guidelines implemented by the NIH Radiation Exposure Committee were followed. Scans were performed with prospective EKG gating, 100 or 120kV tube potential, tube current of 100–850 mA adjusted to the patient’s body size, with a gantry rotation time of 275ms. Images were acquired at a slice thickness of 0.5 mm with a slice increment of 0.25 mm. One-year visit scans were performed using the same scanner and protocol.
Analysis:
All scans were read in a blinded fashion to patient characteristics, visit date, and treatment. Coronary characteristics were analyzed across each of the main coronary arteries > 2 mm using dedicated software (QAngio CT, Medis; The Netherlands) (Figure 1).(3, 9) Automated longitudinal contouring of the inner lumen and outer wall was performed and results were manually adjusted when clear deviations were present.(10) Results of the automated contouring were also reviewed on transverse reconstructed cross-sections of the artery on a section-by-section basis at 0.5-mm increments. Lumen attenuation was adaptively corrected on an individual scan basis using gradient filters and intensity values within the artery. Intra-rater reliability was high, with intra-class correlation coefficient = 0.900, 95% CI [0.903–0.919].
Figure 1: Psoriasis and non-calcified coronary artery disease by coronary computed tomography angiography.
Image depicting psoriasis patient undergoing a CT scan (right panel). A 3D reconstruction of the coronary artery obtained from coronary computed tomography angiography scan of the psoriasis patient (middle panel). Planar reconstruction of the coronary artery obtained from coronary computed tomography angiography. The orange line represents the outer border (total volume) and the yellow line represents the inner border (lumen volume) (left panel).
To account for variable coronary artery lengths, coronary burden (in cubic millimeters) was divided by the corresponding segment length (in millimeters), yielding a coronary burden index.9 Total burden was defined as the sum of non-calcified and dense-calcified coronary burden. Non-calcified coronary burden was obtained after adaptively correcting for lumen attenuation and depicted based on Hounsfield Units derived by the software.
Machine Learning Algorithm
Removing excessive variables in large datasets improves model accuracy, performance, and interpretability, and reduces overfitting.(11) One method of removing excessive variables in a dataset is through the use of random forest ensembles. In addition to the algorithm’s high predictive performance, random forests are particularly well suited for our dataset because of some important characteristics: 1) the construction of decision trees that are unique to every dataset, 2) the capacity to handle both categorical and continuous variables, and 3) the ability to process missing values or invalid/erroneous datasets (Supplemental Table 1).(12) In this method, we begin by manually removing variables in the dataset, and then we grow an ensembles of decision trees to measure variable importance by permutation. The variable importance value that is outputted by the machine learning algorithm indicates the predictive power of that variable in determining non-calcified coronary plaque burden.
29 variables out of the initial 92 variables were deemed to be redundant (e.g. basophil count vs. absolute basophil count) and were thus removed prior to analysis by the random forest algorithm. The importance of the remaining 62 variables with respect to non-calcified coronary burden was determined by permutation within a random forest algorithm of 200 regression trees. Using the method introduced by Breiman et al., we measured the accuracy of each phenotypic variable for predicting non-calcified coronary burden.(13) After determining the variables importance values of each predictor variable, a simple unadjusted linear regression was performed between the predictor variable and non-calcified coronary burden. All machine learning algorithms were implemented using MATLAB (2018a version) Statistics and Machine Learning toolbox.
Statistical Analyses
Skewness and kurtosis measures were considered to assess normality. Data were reported as mean with standard deviation for parametric variables, median with interquartile range for non-parametric variables and as percentages for categorical variables. We conducted linear and logistic regression between non-calcified coronary burden and the predictor variables. For predictor variables with binary outputs, non-calcified coronary burden was dichotomized by median non-calcified coronary burden value of our cohort in order to perform logistical regressions if necessary. P-value <0.05 was considered statistically significant. Statistical analysis was performed using STATA-12 software (STATA inc., College Station, Texas).
RESULTS:
At baseline, patients with psoriasis were middle-aged, predominantly male, low cardiovascular risk by Framingham risk score, and with mild-to-moderate skin disease (Table 1). 29 of the initial 92 variables available in the dataset were manually removed from the machine learning algorithm due to redundancy of the variables (e.g. neutrophil count vs. absolute neutrophil count). The target variable (non-calcified coronary burden) was removed and the remaining 62 variables were ranked for importance using the random forest algorithm.
Table 1.
Description of psoriasis participants at baseline visit.
| Variable Demographics and clinical history | Total (N=263) |
|---|---|
| Age, years | 47.6 ± 14.0 |
| Gender, males | 175 (67) |
| Hypertension | 73 (28) |
| Hyperlipidemia | 120 (46) |
| Type 2 diabetes mellitus | 24 (9) |
| Body mass index | 28.8 ± 5.8 |
| Current smoker | 19 (7) |
| Statin use | 77 (29) |
| Lipid and cell characterization | |
| Total cholesterol, mg/dL | 172.6 ± 42.1 |
| HDL cholesterol, mg/dL | 56.7 ± 17.7 |
| LDL cholesterol, mg/dL | 104.0 ± 36.2 |
| Triglycerides, mg/dL | 115.6 ± 70.4 |
| Framingham risk score | 2 (1–6) |
| High sensitivity c-reactive protein, mg/L | 1.5 (2.8) |
| Cholesterol efflux capacity | 0.99 ± 0.18 |
| Psoriasis characterization | |
| Psoriasis area severity index | 6 (3–10) |
| Systemic/biologic treatment | 67 (26) |
| Coronary characterization | |
| Non-calcified coronary burden, mm2 (×100) | 1.10 ± 0.41 |
| Adipose characterization | |
| Visceral adiposity, cm3 | 15364.8 ± 9128.2 |
| Subcutaneous adiposity, cm3 | 18808.1 ± 10484.6 |
Values reported in the table as Mean ± SD or Median (IQR) for continuous data and N (%) for categorical data. HDL: High-density lipoprotein, LDL: Low-density lipoprotein.
The top 20 variables ranked by random forest algorithm are listed in Table 2: body mass index, visceral adiposity, total adiposity, apolipoprotein A1, high-density lipoprotein, erythrocyte sedimentation rate, subcutaneous adiposity, small low-density lipoprotein particle, cholesterol efflux capacity, absolute immature granulocyte count, total cholesterol, waist-to-hip ratio, apolipoprotein B, very low-density lipoprotein particle, absolute monocyte count, high sensitivity c-reactive protein (CRP), large very low-density lipoprotein particle, large medium high-density lipoprotein particle, large medium very low-density lipoprotein particle and white blood cells (Table 2). The importance values outputted by the random forest algorithm indicate how important that variable is in predicting non-calcified coronary burden with the highest possible importance value of 1.0. Of note, the importance values are absolute values and does not suggest whether the variables have a positive or negative relationship with non-calcified coronary burden. By running an unadjusted linear regression between non-calcified coronary burden and each of these top predictors, these top variables with non-calcified coronary burden yielded similar results consistent with our machine learning outputs (Table 3). The standardized correlation coefficient (Beta-coefficient) for each variable in order of variable importance is shown in Table 3, which in contrast to Table 2, shows the whether the predictor has a positive or negative association with non-calcified coronary burden. The correlation coefficient of these predictor variables with non-calcified coronary burden, shown in Table 3, was obtained from an unadjusted linear regression model. Apolipoprotein A1, HDL lipoprotein, cholesterol efflux capacity, and large medium high-density lipoprotein particle had a statistically significant negative association with non-calcified coronary burden. Erythrocyte sedimentation rate, absolute immature granulocyte count, total cholesterol, apolipoprotein B, very low-density lipoprotein particle, absolute monocyte count, large very low-density lipoprotein particle and large medium very low-density lipoprotein particle did not have a statistically significant correlation coefficient in the unadjusted regression models. The remainder of the predictor variables had statistically significant positive correlations with non-calcified coronary burden (Table 3).
Table 2.
Top 20 predictors of non-calcified coronary burden using random forest algorithm in psoriasis.
| Variable (N=263) | Importance |
|---|---|
| Body mass index | 0.66 |
| Visceral adiposity | 0.64 |
| Total adiposity | 0.41 |
| Apolipoprotein A1 | 0.22 |
| High-density lipoprotein | 0.19 |
| Erythrocyte sedimentation rate | 0.17 |
| Subcutaneous adiposity | 0.15 |
| Small low-density lipoprotein particle | 0.13 |
| Cholesterol efflux capacity | 0.11 |
| Absolute immature granulocyte count | 0.11 |
| Total cholesterol | 0.10 |
| Waist-to-hip ratio | 0.09 |
| Apolipoprotein B | 0.09 |
| Very low-density lipoprotein particle | 0.06 |
| Absolute monocyte count | 0.06 |
| High sensitivity c-reactive protein | 0.06 |
| Large very low-density lipoprotein particle | 0.05 |
| Large medium high-density lipoprotein particle | 0.04 |
| Large medium very low-density lipoprotein particle | 0.04 |
| White blood cells | 0.04 |
Table 3.
Unadjusted linear regression of top 20 predictors using machine learning with non-calcified coronary burden in psoriasis.
| Variable (N=263) | Beta (P-value) |
|---|---|
| Body mass index | 0.64 (<0.001) |
| Visceral adiposity | 0.58 (<0.001) |
| Total adiposity | 0.54 (<0.001) |
| Apolipoprotein A1 | −0.40 (<0.001) |
| High-density lipoprotein | −0.42 (<0.001) |
| Erythrocyte sedimentation rate | −0.05 (0.52) |
| Subcutaneous adiposity | 0.34 (<0.001) |
| Small low-density lipoprotein particle | 0.27 (<0.001) |
| Cholesterol efflux capacity | −0.28 (<0.001) |
| Absolute immature granulocyte count | 0.03 (0.69) |
| Total cholesterol | −0.15 (0.06) |
| Waist-to-hip ratio | 0.33 (<0.001) |
| Apolipoprotein B | 0.03 (0.69) |
| Very low-density lipoprotein particle | −0.05 (0.50) |
| Absolute monocyte count | −0.04 (0.60) |
| High sensitivity c-reactive protein | 0.16 (0.02) |
| Large very low-density lipoprotein particle | 0.07 (0.32) |
| Large medium high-density lipoprotein particle | −0.18 (0.01) |
| Large medium very low-density lipoprotein particle | 0.08 (0.28) |
| White blood cells | 0.15 (0.03) |
Data represented as standardized beta correlation coefficient (p-value).
DISCUSSION:
In our study, we demonstrate that machine learning methods can be leveraged to identify top predictors of non-calcified coronary burden in patients with psoriasis. These were confirmed by unadjusted linear regression models. Known traditional risk factors for cardiovascular disease provide a risk assessment at the population-level but often fall short when precisely assessing an individual’s risk.(14, 15) For instance, one study demonstrated that the Atherosclerotic Cardiovascular Disease score, which is commonly used in clinical practice, to determine whether someone needs to be on a statin, often overestimates risk in both males and females in multiple ethnic groups in the modern American primary prevention cohort.(16) Machine learning is promising in that it can be potentially applied to provide a more personalized risk assessment of a patient’s subclinical disease and future CV event risk, given the patient’s clinical characteristics (e.g. psoriasis). In fact, machine learning has been applied previously to offer cardiovascular disease risk prediction and evaluation in multiple studies.(17–19) However, none have applied machine learning to identify top predictors of early rupture prone plaque as assessed by non-calcified coronary burden.
Patients with chronic systemic inflammatory diseases, such as psoriasis, have accelerated atherosclerosis with increased cardiovascular disease risk and events(2, 20, 21), in part due to the impact of chronic inflammation on subclinical cardiovascular disease (elevated non-calcified coronary burden by CCTA).(3) It is also important to note that obesity and metabolic syndrome which are associated with these inflammatory states promote a pro-inflammatory profile with increased inflammatory cytokines such as IFN-gamma, IL-1beta, IL-6, TNF-alpha, CRP and reduced anti-inflammatory mediators like adiponectin.(22) The interplay between inflammation and metabolic syndrome contributes to the pathogenesis of coronary artery disease. Thus, it is interesting that the top predictors of non-calcified coronary burden by machine learning were related to obesity, dyslipidemia, and inflammation.
CCTA has long been utilized for the characterization of coronary plaque burden and has been extensively compared to and validated against the gold standard, intravascular ultrasound.(23) CCTA provides characterization of not only lumen stenosis and arterial remodeling, but also plaque subcomponents, including calcified, non-calcified, and high-risk features.(5) Studies have shown that there is an increase in high-risk non-calcified coronary burden in acute coronary syndrome patients, which in turn are prone to rupture and consequently cause future cardiovascular events.(5, 24) Our machine learning algorithm demonstrated that the top predictors of non-calcified coronary burden in psoriasis patients were factors related to markers of obesity (body mass index, adiposity, waist-to-hip ratio), dyslipidemia (apolipoprotein A1, lipoprotein particles, cholesterol efflux capacity, apolipoprotein B), and inflammation (erythrocyte sedimentation rate, high sensitivity CRP, absolute immature granulocyte count, absolute monocyte count, white blood cells), which is consistent with our understanding of the pathophysiology of atherosclerosis.(25) Each of these top predictors determined by the machine learning algorithm was evaluated via simple linear regressions. Apolipoprotein A1, HDL and cholesterol efflux capacity were found to be negatively associated with non-calcified coronary burden, consistent with our understanding of atherogenesis in psoriasis patients.(10) The importance of the machine learning algorithm is that it takes into consideration multiple variables, which permits ranking of each variable in predicting non-calcified coronary burden, which is not provided by linear regression models.
A major limitation of this study is that the analysis only included baseline values from the patients’ first visit. Additional follow-up, along with recorded CV events in the future will be required to provide better insight into whether the machine learning algorithm can correctly stratify a patient into the appropriate risk category based on their estimated non-calcified coronary burden. Finally, given that machine learning was applied to a specific population, namely patients in the Psoriasis Atherosclerosis Cardiometabolic Initiative, we were unable to provide any prognostic accuracy and validation on an external cohort with non-calcified coronary burden and these clinical values to further confirm our results.
In conclusion, our 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 CCTA. 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 cardiovascular disease risk.
ACKNOWLEDGEMENTS:
We would like to acknowledge and thank the NIH Clinical Center outpatient clinic-7 nurses for their invaluable contribution to our study.
FUNDING:
This study was supported by the National Heart, Lung and Blood Institute (NHLBI) Intramural Research Program (HL006193-05). This research was also made possible through the NIH Medical Research Scholars Program, a public-private partnership supported jointly by the NIH and generous contributions to the Foundation for the NIH from the Doris Duke Charitable Foundation (DDCF Grant #2014194), the American Association for Dental Research, the Colgate-Palmolive Company, Genentech, Elsevier, and other private donors. The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
DISCLOSURES:
Dr. Mehta is a full-time US government employee and has served as a consultant for Amgen, Eli Lilly, and Leo Pharma receiving grants/other payments; as a principal investigator and/or investigator for AbbVie, Celgene, Janssen Pharmaceuticals, Inc, and Novartis receiving grants and/or research funding; and as a principal investigator for the National Institute of Health receiving grants and/or research funding.
Dr. Gelfand in the past 12 months has served as a consultant for Coherus (DSMB), Dermira, Janssen Biologics, Merck (DSMB), Novartis Corp, Regeneron, Dr. Reddy’s labs, Sanofi and Pfizer Inc., receiving honoraria; and receives research grants (to the Trustees of the University of Pennsylvania) from Abbvie, Janssen, Novartis Corp, Regeneron, Sanofi, Celgene, and Pfizer Inc.; and received payment for continuing medical education work related to psoriasis that was supported indirectly by Lilly and Abbvie. Dr. Gelfand is a co-patent holder of resiquimod for treatment of cutaneous T cell lymphoma.
ABBREVIATION:
- CCTA
Coronary computed tomography angiography
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
IRB APPROVAL: Yes.
SUBMISSION DECLARATION and VERIFICATION:
The work described has not been published previously (except in the form of an abstract, a published lecture or academic thesis), is not under consideration for publication elsewhere, its publication is approved by all authors and tacitly or explicitly by the responsible authorities where the work was carried out, and, if accepted, it will not be published elsewhere in the same form, in English or in any other language, including electronically without the written consent of the copyright-holder.
All other authors declare no conflicts of interests in relation to the work presented in this manuscript.
Supplemental material available at: http://dx.doi.org/10.17632/9r4bbws2th.2
REFERENCES:
- 1.Mehta NN, Yu Y, Pinnelas R, Krishnamoorthy P, Shin DB, Troxel AB, et al. Attributable risk estimate of severe psoriasis on major cardiovascular events. Am J Med. 2011;124(8):775 e1–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Gelfand JM, Dommasch ED, Shin DB, Azfar RS, Kurd SK, Wang X, et al. The risk of stroke in patients with psoriasis. The Journal of investigative dermatology. 2009;129(10):2411–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Lerman JB, Joshi AA, Chaturvedi A, Aberra TM, Dey AK, Rodante JA, et al. Coronary Plaque Characterization in Psoriasis Reveals High-Risk Features That Improve After Treatment in a Prospective Observational Study. Circulation. 2017;136(3):263–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Gelfand JM, Neimann AL, Shin DB, Wang X, Margolis DJ, Troxel AB. Risk of myocardial infarction in patients with psoriasis. JAMA. 2006;296(14):1735–41. [DOI] [PubMed] [Google Scholar]
- 5.Motoyama S, Ito H, Sarai M, Kondo T, Kawai H, Nagahara Y, et al. Plaque Characterization by Coronary Computed Tomography Angiography and the Likelihood of Acute Coronary Events in Mid-Term Follow-Up. J Am Coll Cardiol. 2015;66(4):337–46. [DOI] [PubMed] [Google Scholar]
- 6.Motwani M, Dey D, Berman DS, Germano G, Achenbach S, Al-Mallah MH, et al. Machine learning for prediction of all-cause mortality in patients with suspected coronary artery disease: a 5-year multicentre prospective registry analysis. Eur Heart J. 2017;38(7):500–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Av Rosendael, Maliakal G, Kolli K, Beecy A, Al’Aref S, Dwivedi A, et al. Maximization of the usage of coronary CTA derived plaque information using a machine learning based algorithm to improve risk stratification; insights from the CONFIRM registry. Journal of Cardiovascular Computed Tomography. 2018;12(3):204–9. [DOI] [PubMed] [Google Scholar]
- 8.Harrington CL, Dey AK, Yunus R, Joshi AA, Mehta NN. Psoriasis as a human model of disease to study inflammatory atherogenesis. Am J Physiol Heart Circ Physiol. 2017;312(5):H867–H73. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Kwan AC, May HT, Cater G, Sibley CT, Rosen BD, Lima JAC, et al. Coronary Artery Plaque Volume and Obesity in Patients with Diabetes: The Factor-64 Study. Radiology. 2014;272(3):690–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Salahuddin T, Natarajan B, Playford MP, Joshi AA, Teague H, Masmoudi Y, et al. Cholesterol efflux capacity in humans with psoriasis is inversely related to non-calcified burden of coronary atherosclerosis. European heart journal. 2015. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Boulesteix AL, Janitza S, Kruppa J, Konig IR. Overview of random forest methodology and practical guidance with emphasis on computational biology and bioinformatics. Wires Data Min Knowl. 2012;2(6):493–507. [Google Scholar]
- 12.Westreich D, Lessler J, Funk MJ. Propensity score estimation: neural networks, support vector machines, decision trees (CART), and meta-classifiers as alternatives to logistic regression. J Clin Epidemiol. 2010;63(8):826–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Breiman L. Random Forests. Mach Learn. 2001;45(1):5–32. [Google Scholar]
- 14.Franco M, Cooper RS, Bilal U, Fuster V. Challenges and opportunities for cardiovascular disease prevention. Am J Med. 2011;124(2):95–102. [DOI] [PubMed] [Google Scholar]
- 15.Greenland P, Knoll MD, Stamler J, Neaton JD, Dyer AR, Garside DB, et al. Major risk factors as antecedents of fatal and nonfatal coronary heart disease events. Jama-J Am Med Assoc. 2003;290(7):891–7. [DOI] [PubMed] [Google Scholar]
- 16.DeFilippis AP, Young R, McEvoy JW, Michos ED, Sandfort V, Kronmal RA, et al. Risk score overestimation: the impact of individual cardiovascular risk factors and preventive therapies on the performance of the American Heart Association-American College of Cardiology-Atherosclerotic Cardiovascular Disease risk score in a modern multi-ethnic cohort. Eur Heart J. 2017;38(8):598–608. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Kalscheur MM, Kipp RT, Tattersall MC, Mei C, Buhr KA, DeMets DL, et al. Machine Learning Algorithm Predicts Cardiac Resynchronization Therapy Outcomes: Lessons From the COMPANION Trial. Circ Arrhythm Electrophysiol. 2018;11(1):e005499. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Colombet I, Ruelland A, Chatellier G, Gueyffier F, Degoulet P, Jaulent MC. Models to predict cardiovascular risk: comparison of CART, multilayer perceptron and logistic regression. Proc AMIA Symp. 2000:156–60. [PMC free article] [PubMed] [Google Scholar]
- 19.Ambale-Venkatesh B, Yang X, Wu CO, Liu K, Hundley WG, McClelland R, et al. Cardiovascular Event Prediction by Machine Learning: The Multi-Ethnic Study of Atherosclerosis. Circ Res. 2017;121(9):1092–101. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Naik HB, Natarajan B, Stansky E, Ahlman MA, Teague H, Salahuddin T, et al. Severity of Psoriasis Associates With Aortic Vascular Inflammation Detected by FDG PET/CT and Neutrophil Activation in a Prospective Observational Study. Arterioscler Thromb Vasc Biol. 2015;35(12):2667–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Mehta NN, Azfar RS, Shin DB, Neimanns AL, Troxel AB, Gelfand JM. Patients with severe psoriasis are at increased risk of cardiovascular mortality: cohort study using the General Practice Research Database. European Heart Journal. 2010;31(8):1000–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Monteiro R, Azevedo I. Chronic inflammation in obesity and the metabolic syndrome. Mediators Inflamm. 2010;2010. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Fischer C, Hulten E, Belur P, Smith R, Voros S, Villines TC. Coronary CT angiography versus intravascular ultrasound for estimation of coronary stenosis and atherosclerotic plaque burden: a meta-analysis. J Cardiovasc Comput Tomogr. 2013;7(4):256–66. [DOI] [PubMed] [Google Scholar]
- 24.Dey D, Achenbach S, Schuhbaeck A, Pflederer T, Nakazato R, Slomka PJ, et al. Comparison of quantitative atherosclerotic plaque burden from coronary CT angiography in patients with first acute coronary syndrome and stable coronary artery disease. J Cardiovasc Comput Tomogr. 2014;8(5):368–74. [DOI] [PubMed] [Google Scholar]
- 25.Linton MF, Yancey PG, Davies SS, Jerome WGJ, Linton EF, Vickers KC. The Role of Lipids and Lipoproteins in Atherosclerosis In: MDText.com I, editor. Endotext2000. [Google Scholar]