Typical progression patterns-sequences and timing of the conditions that patients progress through from a healthy state to a complication of diabetes or hypertension- can represent distinct disease mechanisms, knowledge that would be tremendously useful in optimizing care and in understanding the etiology of diabetes and hypertension. Patient coverage and follow-up times of EHR data available to most institutes do not allow for observing patients from the onset of the disease to the complications.
We aim to reconstruct the progression patterns from a unique combinations of two data sets: the University of Minnesota clinical data repository of 2,000,000 patients with relatively short follow-up times and the Mayo Clinic's exceptionally clean and complete Rochester Epidemiology Project (REP) data set covering 100,000 patients with long follow-up. First, we extract the individual patients' trajectories. From these trajectories, we extract all progression pairs, sequences of two directly or indirectly subsequent conditions. We also estimate and the risks of complications this progression confers upon the patient, as well as the progression time distribution between the pair of conditions. We represent these pairs as a typical progression and a catalog of exceptions (atypical pairs between the same two conditions that differ in history, medication or other details and have significantly different outcomes). Finally, using the Mayo Clinic data with its long follow- up times as scaffolding, we reconstruct the progression patterns from the progression pairs.
Typical progression patterns?sequences and timing of the conditions that patients progress through from a healthy state to a complication of diabetes or hypertension? can represent distinct disease mechanisms, knowledge that would be tremendously useful in optimizing care and in understanding the etiology of diabetes and hypertension. Patient coverage and follow-up times of EHR data available to most institutes do not allow for observing patients from the onset of the disease to the complications. We aim to reconstruct the progression patterns from statistically significant partial progression patterns discovered from a large clinical data repository at the UMN with short follow-up using the Rochester Epidemiology Project (REP) data set at Mayo Clinic with its long follow-up times as scaffolding.
Kim, Era; Pieczkiewicz, David S; Castro, M Regina et al. (2018) Multi-Task Learning to Identify Outcome-Specific Risk Factors that Distinguish Individual Micro and Macrovascular Complications of Type 2 Diabetes. AMIA Jt Summits Transl Sci Proc 2017:122-131 |
Oh, Wonsuk; Yadav, Pranjul; Kumar, Vipin et al. (2017) Estimating Disease Onset Time by Modeling Lab Result Trajectories via Bayes Networks. IEEE Int Conf Healthc Inform 2017:374-379 |
Hu, Zhen; Melton, Genevieve B; Moeller, Nathan D et al. (2016) Accelerating Chart Review Using Automated Methods on Electronic Health Record Data for Postoperative Complications. AMIA Annu Symp Proc 2016:1822-1831 |
Oh, Wonsuk; Kim, Era; Castro, M Regina et al. (2016) Type 2 Diabetes Mellitus Trajectories and Associated Risks. Big Data 4:25-30 |