Figure 5
AUC assessed the risk of developing MACE within ten years of DM diagnosis. QRISK, QRISK2 and QRISK3 of developing a MACE within the ten years following the DM diagnosis. All the individual predictors used in the QRISK3 model were included in our analysis in addition to the predictors derived from our TDA analysis (such as distance to extremities, TDA phenotypes, TDA clusters and the distance to extremities). The list of other predictors includes the following: body mass index (BMI), migraine, systemic lupus erythematosus, corticosteroids, erectile dysfunction, HIV/AIDS, mental illness, rheumatoid arthritis, atypical antipsychotics, HDL cholesterol ratio, smoking status, chronic kidney disease, family history of coronary artery disease, ethnicity, index of deprivation, systemic blood pressure, atrial fibrillation, gender, treated hypertension. The AUC was calculated for these models and models using individual predictors of QRISK3. The results show that augmenting QRISK3 with TDA features significantly increased the AUC (to an extent larger than the one obtained when upgrading from QRISK2 to QRISK3). Moreover, TDA features score better than all individual predictors of QRISK3. Even when taken individually, they score better than most QRISK3 individual predictors. The predictive value of the QRISK3 model also improves by adding two features derived from the TDA graph: 1) the TDA-derived phenotype and 2) the distances to both extremities of the TDA graph (as defined by Dijkstra's algorithm).
