Effective Diagnosis and Treatment of Age-related Disease Through Time-varying Modelling

Within healthcare, there is a tradition of using measurements taken at a single point in time, or at a small number of contact points to infer diagnoses and treatment plans. In the healthcare of today we have many measurements taken that are dense in observations across time yet the traditional summary of this information in a single number (e.g., max, mean, most current) is prevalent. By applying non-statioonary time series analysis we hope to improve the decisions made in healthcare by taking all data into account instead of single values. Dr. Killick is an expert in non-stationary time series analysis and is seeking to expand knowledge in healthcare technologies in order to drive further statistical research motivated by challenges in healthcare technologies.
This discipline hopping proposal identifies two initial areas of healthcare, related to ageing, where utilising this information across time will provide a novel perspective on patients and their care. This discipline hop will propel Dr Killick into two areas of healthcare; orthopaedics and colorectal surgery, in order to 1) learn the required underlying science of the measurements taken; and 2) identify clinical needs to inform modelling. We provide further detail on the motivation behind the two identified areas below.

Osteoporosis (OP) is a debilitating condition caused by a reduction in bone mineral density (BMD) associated with age. It primarily affects post-menopausal women with 1 in 3 affected at 80 years of age. The current NHS approach to assessing bone density is to take a DXA scan of the hips and lumbar spine (L1-4). From the DXA scan, measurements of density are taken at 6 points, although typically only 1 or 2 points are used clinically. In practice, bone density varies across the DXA image and there is a different fracture risk depending on whether loss occurs in one specific region or uniformly. Current treatment improves the bone density at the measured locations but at a high risk of fractures in other locations which are not monitored, these are more challenging to treat and manage. Thus using statistical techniques to create a more accurate assessment of how bone density varies both across bones and across time will not only aid diagnosis of patients, but also spark new drug development that treats the whole bone rather than specific areas.

Colorectal cancer affects over 41,000 people every year, is the third most common malignancy in the UK, and the only curative treatment is surgery. This is however associated with significant risks. The incidence is strongly related to age with the highest rates in the 85-89 age group (44% of new cases are people aged 75 and over). An audit within the NHS suggested that colorectal cancer patients with higher level of fitness have better outcomes after surgery and longer overall survival. Cardiopulmonary Exercise Testing (CPET) is a method used to assess fitness for surgery. Standard use of CPET output is to take the maximal/peak oxygen uptake (VO2 max/peak) and use this as a measure of cardiorespiratory fitness. Generally, CPET can more effectively identify high risk patients before surgery than other clinical risk factors and is therefore a critical component within the decision tree for whether a patient undergoes surgery. A common feature when using CPET on elderly patients is that their VO2 peak values alone are very similar providing little predictive power of surgical outcomes. In contrast, the entire time series of breath by breath measurements produces a marked difference between patients. This motivates us to provide a classification of patients utilising the full time series structure of their CPET progression. This will identify high risk patients and, following further investigation of the biological science, may indicate new pre-operative regimes to reduce post-surgery outcomes.

In addition to the academic beneficiaries, the work proposed in this project will have great impact on others. The methodology created in this project will have short-term impact through the specific applications in colorectal surgery and orthopaedics through collaborations with our project partners. Longer term impact will come from encouraging these methods to be adopted more widely in the NHS and other disciplines. A secondary, but still important, aspect to impact will be the personal development of Dr Killick and the wider team in this important translational research area.
The main routes to impact will be:
1. Making computer code to implement the methodology for a general situation available as a free-to-use R and Python packages. The PI has a track record of making code freely available and ensuring impact around software.
2. Direct work with industrial and scientific collaborators to ensure adoption following the project. In addition the PI will make new collaborations through the interdisciplinary conferences and interacting with the NHS project partners that will facilitate impact within a broader long-term research career.
Research will also be disseminated widely, both within Statistics, in related disciplines (Machine Learning and Signal Processing), and in application areas of colorectal surgery and orthopaedics. The career development of the PI will be aided by the wide-range of training opportunities planned, which address areas such as scientific knowledge, public engagement and impact.