ABSTRACT
Artificial intelligence (AI) applied to the genome sciences has the potential to revolutionize healthcare. Yet to fully harness the predictive power of AI, at least in the fields of oncology and infectious disease, evolutionary theory must be brought to bear. In oncology, AI is uniquely suited to analyze the complex latticework of correlations among the many genomic and environmental influences that constitute cancer risk. It also makes possible the evolutionary theory-inspired concepts of next-generation cancer treatment, such as evolutionary traps, adaptive therapy, and treatment vaccination. In infectious disease, AI promises the rapid diagnosis of a pathogen's current drug resistance profile, as well as the prediction of its potential to develop resistance. Using anticipatory diagnostics, drug regimens can be tailored to probabilistically channel pathogens toward less resistance-prone genotypes to avoid the emergence of resistance. Advanced computational methods are also used in antimicrobial drug design and to anticipate outbreaks of infectious disease and the evolution of epidemics, such as the SARS-CoV-2 pandemic. In detailing these advances, we discuss illustrative examples of the productive collaboration of data scientists, evolutionary theorists, epidemiologists, and clinicians. In addition, we briefly note the dangers of overreliance on advanced computational tools that involve "black box” algorithms and question whether they undermine the synthesis of Mendelian genetics and Darwinian theory. Yet, insofar as evolutionary theory is used for hypothesis algorithm development and AI for data creation and analysis, this problem may be avoided, and the potential of both will be realized. © Springer Nature Switzerland AG 2022.
ABSTRACT
Introduction: Arterial and venous thromboembolism are common complications in COVID-19. Micro-macro thrombosis-related organ dysfunction can confer an increased risk for mortality. The optimal dosage of anticoagulation (AC) in COVID-19 patients remains unclear. Interim data from adaptive randomized control trials (ATTACC, REMAP-CAP, and ACTIV-4a) showed divergent results of therapeutic AC (TAC) versus usual care AC for the primary outcome of organ support free days in hospitalized COVID-19 patients. Components of CHA 2DS 2-VASc, a model originally built for predicting ischemic stroke in atrial fibrillation, are consistent with independent risk factors for COVID-19 severity and mortality. Herein, we analyzed the performance of the CHA 2DS 2-VASc model in hospitalized COVID-19 patients for predicting arterial and venous thromboembolic events, which could potentially aid in risk stratification of hospitalized patients and guide AC dosing. Methods: This is a large, retrospective, multicenter cohort study that included all adult patients from one tertiary care and five community hospitals with PCR-proven SARS-CoV-2 infection between 3/1/2020 and 12/1/2020. The primary composite outcome was acute arterial thromboembolism (ATE) and venous thromboembolism (VTE). We identified patients with ATE [cerebrovascular accident (CVA), myocardial infarction (MI) including both ST-segment elevation MI and non-ST-segment elevation MI], and VTE [deep vein thrombosis (DVT) and pulmonary embolism (PE)] using ICD -10 codes. Mean and standard deviation were reported for continuous variables;proportions were reported for categorical variables. To compare the groups, the Chi-square test was used for categorical variables, and the t-test was used for continuous variables. CHA 2DS 2-VASc scores were calculated on admission and were used as a measure of the predictive accuracy of the scoring system. Sensitivity and specificity with different cut-offs of CHA 2DS 2-VASc scores were calculated. All statistical tests were 2-sided with an α (significance) level of 0.05. All data were analyzed using R version 4.0.5. Results: Among 3526 patients, a total of 619 patients had thromboembolic events: 383 had ATE and 236 had VTE. Of 383 patients who had ATE, 350 patients were found to have acute MI, 48 had CVA, and 15 had both MI and CVA. In patients with VTE, 134 had DVT, 168 had PE, and 66 had both DVT and PE (Figure 1). We analyzed the primary composite outcome of ATE and VTE (group 1) vs no ATE and VTE (group 2). Baseline characteristics are included in Table 1. The in-patient all-cause mortality rate was 28.4% in group 1 vs 12.6% in group 2 (p<0.001). The mean hospital length of stay was 12.3 days in group 1 vs 8.8 days in group 2 (p<0.001). Group 1 had a mean CHA 2DS 2-VASc score of 3.3 ±1.6. vs 2.7±1.7 in group 2 (p<0.001) (Figure 2). At CHA 2DS 2-VASc scores of 3 and 4, the model had a specificity of 46% and 67% and sensitivity of 68% and 42% respectively for predicting ATE/VTE. The CHA 2DS 2-VASc score of 5 had a specificity of 86% and sensitivity of 25%. The score of 7 had 98% specificity but 3% sensitivity (Table 2). Conclusion: Our results suggest that the CHA 2DS 2-VASc model for arterial and venous thromboembolism has a moderate performance. The CHA 2DS 2-VASc score of 5 has a high specificity, though low sensitivity, for predicting thromboembolism. The CHA 2DS 2-VASc score can be used as an adjunct risk stratification tool to initiate TAC. [Formula presented] Disclosures: No relevant conflicts of interest to declare.