Abstract Body: Objective
Multimorbidity is a growing but understudied global challenge in an aging world. Cardiometabolic disease (CMD) and cancer are the 2 most common chronic diseases, yet their transition patterns remain unclear, and reliable tools for early prediction of disease trajectories across the lifespan are scarce.

Methods
This study was conducted on 429,909 UK Biobank participants without baseline CMD and cancer. CMD included type 2 diabetes, coronary disease, heart failure, and stroke. A multistate analysis was used to investigate transition patterns and identify multiomics signatures for transitions from baseline to single morbidity and to multimorbidity. Machine learning prediction models for disease trajectories were constructed using genomics, metabolomics, and proteomics data, and predictive performance was assessed.

Results
During a median follow-up of 15 years, 105,855 participants developed morbidity and 15,044 developed multimorbidity of CMD and cancer. Participants with multimorbidity had a 9%-30% higher mortality risk and 2.4-5.3 years of shorter survival time than those healthy or with single morbidity. Development of CMD before cancer presented a poorer prognosis than the reverse order. Distinct and shared multiomics signatures underlying disease trajectories were identified. Top 10 multiomics signatures presented a dose-response characteristic from health to single morbidity and to multimorbidity of CMD and cancer. Proteomics scores showed superior prediction performance than metabolomics and genomics scores. For 10-year outcome prediction, proteomics scores showed varying performance across disease trajectories, with area under receiver-operating characteristic curves ranging from 0.64 for health-to-only cancer trajectory to 0.90 for health-to-CMD-to-death trajectory, significantly better than the base model and traditional clinical model.

Conclusions
This study revealed disease trajectories of CMD and cancer with varying prognostic implications that were predicted with differing accuracy through multiomics approaches.