Proposal for Multimorbidities in Diabetic Patients Essay

Proposal for Multimorbidities in Diabetic Patients Essay

Type 2 diabetes is one of the most common chronic conditions and also occurs with many morbidities. Consequences of the multimorbidities in diabetic patients negatively affect both the patients and the healthcare providers as they are associated with high mortality, decreased quality of life and increased consumption of healthcare resources.

Multimorbidity in diabetic patients is not well understood and existing research provides a limited description of its causes and consequences. Many research articles are often based on cross-sectional data and do not consider the temporal evolution of multimorbidity patterns and disease progression.Proposal for Multimorbidities in Diabetic Patients Essay.

While there is some research that focuses on disease progression, they neglect irregular patient visits and do not adequately model interventions. Existing research usually ignores the critical dependence of historical illness and intervention on future illnesses and care. Proposal for Multimorbidities in Diabetic Patients Essay. Research in this area is growing but most investigations have focused on predicting morbidities in isolation from one another and not in the presence of other morbidities.


To address these issues, we introduce a neural network for disease progression modeling, intervention recommendation and risk prediction.

Our neural network will read all Electronic Medical Records (EMR) of diabetic patients, store previous medical history including all morbidities and all implemented intervention, deduce current disease states and predict future medical outcomes.

Our model is built on Long Short-Term Memory (LSTM) which is a recurrent neural network with memory cells to store medical history. At each time step, the LSTM accepts an input, updates its memory cells and returns an output. Memory is maintained through a forget gate that regulates the passing of memory from one time step to another and is updated only when new input is received at each time step.

The output is determined by the memory and regulated by an output gate. In our framework, the LSTM models the disease trajectory together with the medical procedures of a patient which are recorded in a time-stamped sequence of medical admissions. The input to the LSTM is EMR information extracted from medical admissions while the output represents illness states at the time of medical admission.

The memory cells in the LSTM neural network are used to store, update and forget disease history over time-stamped episodes. The inferred history is then combined to deduce current disease states and future prognosis. Similar to human memory, our model has a recency effect i.e. more recent events have greater impact on future medical risks.

Our framework has three aims. The first aim is to model disease progression along with all available morbidities and to investigate the progression from one stage to another of the same disease, the recurrence of a disease or the transition to a new disease. The second aim is to provide intervention recommendations i.e. predicting a subset of treatment procedures for the current diagnoses. The final aim is to provide prognosis and assess the future risk of re-admission or mortality within a defined period after discharge from EMRs. Proposal for Multimorbidities in Diabetic Patients Essay.