**Background:** Heart failure with preserved ejection fraction (HFpEF) is a growing public health concern with limited treatment options, partly due to its complex, comorbidity-driven pathophysiology. Unlike HF with reduced ejection fraction (HFrEF), HFpEF is associated with a wide range of comorbidities that may be causal factors. Systems medicine approaches, such as network medicine, can model phenome-wide comorbidity profiles to improve understanding of HFpEF and infer associated genetic profiles. This study aimed to apply a network medicine approach to describe comorbidity patterns in HFpEF and investigate shared genetic backgrounds.
**Methods:** The study retrospectively analyzed 569 comorbidities in 29,047 HF patients (8,062 HFpEF, 6,585 HFrEF, 3,018 HFmrEF) from a German university hospital (2008-2021). HF was defined using ICD-10 codes and clinical criteria (e.g., NT-proBNP >120 ng/ml, LVEF <50%). Multiple correspondence analysis (MCA) assessed variance in comorbidity profiles between HF subtypes. Machine learning classifiers (random forest and elastic net) identified distinctive comorbidity profiles for HFpEF and HFrEF. A comorbidity network (HFnet) was built using Fisher's exact test and phi correlations, and disease clusters were identified via network clustering. A multilayer heterogeneous network (HFhetnet) integrated the HFnet with gene networks (Omnipath, PPI, GO) and disease-gene associations (DisGeNET). Gene prediction for HFpEF and HFrEF was performed using random walk with restart on multiplex heterogeneous networks (RWR-MH), with seed nodes from the comorbidity profiles. Predicted genes were prioritized by ranking differences between HFpEF and HFrEF. Transcriptomic data from a murine HFpEF model (high-fat diet + L-NAME) were used for corroboration.
**Key Results:** MCA showed high variance between HFpEF and HFrEF comorbidity profiles (39.5% explained by subtype). The elastic net classifier achieved an AUROC of 0.777, identifying 71 comorbidities for HFpEF (including hypertensive heart disease, sleep apnea, osteopenia, breast cancer, rheumatoid arthritis) and 29 for HFrEF (including myocardial infarction, ischemic heart disease, tobacco abuse, chronic kidney disease). The HFnet contained 569 nodes and 19,347 edges, with 9 disease clusters (DCs) capturing important comorbidity groups (e.g., DC1: cardiac/endocrine/respiratory; DC6: rheumatoid/osteologic/psychiatric; DC8: neoplastic). HFpEF patients showed higher similarity to DC1, DC2, DC6, and DC8, while HFrEF patients were more similar to DC3, DC4, and DC5. Gene prediction from the HFpEF comorbidity profile identified candidates involved in fibrosis (COL3A1, LOX, SMAD9, PTHL), hypertrophy (GATA5, MYH7), oxidative stress (NOS1, GSTT1, XDH), and ER stress (ATF6). The top 50-100 predicted HFpEF genes were significantly enriched in overexpressed genes in the murine HFpEF transcriptome (p<0.05), while HFrEF genes were not.
**Clinical Implications:** This study demonstrates that comorbidity profiles can distinguish HFpEF from HFrEF and provide insights into HFpEF pathophysiology. The identification of diverse comorbidities (neoplastic, osteologic, rheumatoid) supports the view of HFpEF as a systemic syndrome. The predicted gene candidates, particularly those related to fibrosis and oxidative stress, may represent novel therapeutic targets. The network medicine approach offers a framework for integrating clinical data with molecular networks to understand complex diseases like HFpEF. However, limitations include the retrospective design, use of ICD-10 codes, potential selection bias from a tertiary care center, and lack of temporal data. Future studies should validate these findings in other populations and explore disease trajectories.