Abstract:
Chronic kidney disease (CKD) progression in patients with diabetes has been identified by the
National Kidney Foundation as a major health issue in New Zealand for almost a decade. To date,
most studies have used a single outcome measure for kidney function as an end- point to drive
patient management decisions. However, it has been recently identified that disease progression,
i.e., trajectories, are more important for the management of patients with CKD rather than a
single estimate. The individual changes over time mean that groups of individuals fall in different
groups, or clusters. Yet, most methods that aim to find such clusters rely on the researcher to
choose the number of clusters before carrying out parame- ter estimation. In this thesis, we
identified methods that identify groups of individuals that arise from different distributions and
outlined the limitations of such models, namely that these models utilize exploratory decisions
rather than confirmatory. To reduce this bias and to let the data determine the number of clusters,
we extended a trans-dimensional Bayesian algorithm to capture the true heterogeneity in a
longitudinal dataset that was representa- tive of routine clinical practice in New Zealand.
We found that this Bayesian algorithm could successfully estimate the number of clusters
using the data itself. We also showed that that, given sufficient visits of the patient to
clinical practices, their disease trajectory can be estimated. Using our model, the trajectories
of CKD in patients with diabetes that were identified can be applied in primary health care to
drive patient management decisions to
help reduce the clinical and epidemiological impact of CKD.