Abstract
Hip fractures significantly affect patients' health and quality of life. Despite therapeutic treatments, many patients continue to experience poor prognoses that include recurrent fractures and mortality, especially the older ones. Therefore, understanding important factors associated with post-fracture prognoses is critical. This study focuses on patients' multimorbidity trajectories and examines how the trajectory's time span, number of coexisting chronic diseases, and sequential disease patterns relate to distinct prognostic outcomes. From the National Health Insurance Research Database in Taiwan, we obtain a sample of 128,822 patients who suffered an initial hip fracture between 1996 and 2011. We use this sample to analyze the relationships between multimorbidity trajectories and prognostic outcomes after an initial hip fracture. The results reveal that a patient's multimorbidity trajectory's time span and number of chronic diseases significantly associate with his or her post-fracture prognosis. In addition, essential sequential patterns of chronic diseases relate to post-fracture prognoses too. We then leverage the discovered relationships to develop a cross-attention neural network method for estimating patients' post-fracture prognoses and demonstrate its predictive utilities relative to several prevalent machine leaning methods. This study underscores the importance of leveraging the time span, number of chronic diseases, and sequential disease patterns in patients' multimorbidity trajectory profile to estimate their prognoses within three years of an initial hip fracture, which can support physicians' clinical decisions and patient management.