Identifying trends in patients with chronic kidney disease (CKD) using routinely collected data

This 4-week project extended some prior work in data analytics to improve outcomes for renal patients. Many individuals with the end-stage renal disease receive haemodialysis either at home or in a centre in order to sustain life. Intradialytic hypotension (IDH) is a common and challenging complication of haemodialysis. An IDH event can lead to patient discomfort, dialysis interruption, hospitalization, increased morbidity, increased mortality, cerebral ischemia, vascular access thrombosis, and cardiovascular events.

Besides these medical consequences, IDH also leads to an increase in service costs. Early intervention of such a dangerous clinical event is highly desirable for prevention, yet no tool for the management and decision-making of IDH has been available until now. An accurate prediction of IDH would help healthcare professionals and patients with planning, decision-making, and building management strategies. Such predictive capability may improve patient care, save healthcare costs, and reduce the burden on healthcare professionals. We have conducted an initial study showing some predictive ability, but more work is
needed to produce a more reliable model. Following this, a short contract (4 weeks) was provided to Lars Jacobson (Research Assistant) to investigate the data further and improve the possibility of using machine learning (ML) techniques in predicting IDH by analysing various standard clinical variables measured by clinicians.

We carried out these activities:
1) Data cleaning
2) Data and Feature analysis
3) Investigate two ML models which were built previously
4) Compare the model outcomes with all patent data vs. personalised data.

The expected outputs of this project were to:
1) Clean the Data
2) Conduct data and feature analysis
3) Investigate two ML models which were built previously
4) Compare the model outcomes with all patent data vs. personalised data.

The actual work carried out was cleaned, and prepared the data accordingly. Following this, data and feature analysis were performed, which explored the importance of variables in predicting IDH. Preliminary prediction models using machine learning (ML) algorithms were investigated. The prediction models were also used to investigate the total data set and the data grouped properties per patient. The outcomes of comparing the total data set and grouped properties per patient data show that the feature importance increases for grouped properties per patient data.