This study is an aggregated registry comprising of a retrospective medical record review
of individuals from multiple sites. The general approach is to create a large,
consolidated, global registry of existing cohorts of patients referred for noninvasive
testing using computed tomography (CT). Anonymized images and structured data, including
demographics, risk factors, outcomes and CT results will be obtained from multiple sites.
The types of images to be analyzed and quantified are non-contrast CT (NCCT) scans and
coronary CT angiography (CCTA). CAC score will use Agatston's method while CAD will be
assessed using registry data of CCTA reads.
The data collected will include risk factors and demographics such as age, sex,
ethnicity, hypertension, smoking, diabetes, dyslipidemia and family history of CAD.
Outcomes such as death and myocardial infarctions will be included in the dataset if
available. All data received will be anonymized and de-identified. Study team members
will check through the study data to ensure that all study data is accurately collected
and complete.
The data elements of different cohorts may not harmonize or match with each other. There
could be missing data elements or different data inputs. As such, omission or imputation
may be used to perform analyses. To minimize data heterogeneity in format, sites will be
provided with a standard template and data dictionary. This will complete the initial
data harmonization and expected data elements. The collected dataset would then be
harmonized by the biostatistics team prior to analysis.
The approximate total study size n = 200,000. Assuming an area under the receiver
operating curve (AUC) of 0.70 for existing PTP and CAC methods, this proposal is
adequately powered to detect an increase of 0.05 in AUC using a two-sided z-test at a
significance level of 0.05. Continuous variables will be expressed as mean and standard
deviation. Categorical variables will be expressed as absolute numbers and percentages.
Distributions will be tested for normality using Shapiro-Wilk statistics. Non-normally
distributed variables will be represented as median with 25th to 75th interquartile
range. Comparison of normally distributed continuous variables will be performed using
Student's t test for paired and unpaired data. Non-normally distributed variables will be
compared using Mann-Whitney Rank Sum tests and Kruskal-Wallis tests. Comparison of
categorical data will be performed using Chi-square and Fisher's Exact Tests where
appropriate.
Differences in outcomes over time will be analyzed by the Kaplan-Meier analysis with
log-rank test for each outcome. Using Cox regressions analysis univariate and
multivariate regression analyses will be performed. Univariate analysis will include
pre-event variables with p values <0.10. Variables that showed a significant (p<0.05)
correlation with the endpoints, after univariate analysis, will be considered in the
multivariate models. Odds ratios and 95% confidence interval will be calculated.
Statistical significance was established as p<0.05. Advanced machine learning techniques
(e.g. neural networks) may be applied.