Exploring clinical applications for a novel multi-task functional assessment: matching appropriate technology to clinical need

Woodgate, Samuel 2021. Exploring clinical applications for a novel multi-task functional assessment: matching appropriate technology to clinical need. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Download (8MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (98kB)

Abstract

Huntington’s Disease (HD) is an autosomal-dominant, progressive, and ultimately fatal neurodegenerative disorder which results in complex array of motor, cognitive, behavioural, and functional symptoms. At present no disease modifying therapies are available however numerous potential therapies are under active development. Due to the progressive nature of the disease, there is a particular focus on therapies that target the earliest stages of HD, with a view to slowing progression before too much damage has occurred. To help prove the efficacy of such potential therapies, as well as for facilitating effective clinical management, sensitive assessments of HD disease state are required. It has however been repeatedly shown throughout the literature that existing assessments methods used in HD are unsuitable for measuring subtle changes in disease symptom progression. This presents a clear problem for the development of potential treatments as well as clinical management and as such there is an ongoing drive to develop new assessment strategies for HD. In response to this need for new HD assessment strategies (specifically regarding functional symptoms) the Clinch Token Transfer Test (C3t), a timed upper-body dexterity test, was developed. The C3t has been shown in previous work to be sensitive to various gold-standard HD assessments and an instrumented variant shown to related to general upper-body motor function. This thesis expands on this previous work with the goal simplifying uptake of the C3t, providing further evidence of the C3ts utility in HD assessment and exploring its relationship with chorea, a common early-stage HD motor symptom, using data from wrist-mounted accelerometers worn during the test. Additionally, this thesis details and critiques the development and deployment of a remote data collection platform (RDCP) designed for the C3t which facilitated the collection of much of the data used in this study. First, understanding of the C3t and the scores it contains was developed using C3t and clinical data from one-hundred and five HD gene-positive participants of varying disease stages (pre-manifest to TFC Stage 3) of which thirty-three had 1-month and 12-month follow-up data. Four clinical measures were included in the study – the UHDRS-TMS, the Composite Unified Huntington’s Disease Rating Scale (CUHDRS), the Prognostic Index Normalised for HD (PINHD), and a chorea score from the summed chorea components of the UHDRS-TMS. C3t scores were available for all visits, clinical measures were only available for the baseline and 12-month visits. Analysis of the C3t scores distribution within the cohort showed six of the fourteen scores were mostly invariant and so could be removed from the C3t protocol and further analysis. Six additional scores were also ultimately recommended for removal –two as they were solely derived from the invariant scores, two which showed no relationship with any of the studied clinical measures, and two which extremely high correlations with the C3t time scores but require extra work to produce making them effectively redundant. The two remaining C3t scores, both time-taken scores, were highly correlated with each clinical measure (Spearman’s R, UHDRS-TMS r=0.69; CUHDRS r=-0.69; PINHD r=0.83) and could be used as independent variables in regression models to estimate the CUHDRS and UHDRS-TMS with a low degree of error (normalised mean absolute error (N-MAE), UHDRS-TMS=9.4%; CUHDRS=11.0%). No relationship was found between any C3t score and the summed chorea score. Effect sizes calculated for the C3t scores and each clinical measure between the baseline and 1-month visits (C3t scores only) and baseline and 12-month visits (C3t scores and clinical measures) were inconclusive. Finally, it was found that study site and test version did not impact regression models produced using the C3t time scores to estimate the clinical measures. As no relationship was observed between any non-instrumented C3t score and the summed chorea score signal features thought to be sensitive to chorea were decided upon and extracted from instrumented C3t data. Data were drawn from fifty-five HD gene-positive participants who wore two GeneActiv tri-axis accelerometers, one on each wrist, whilst taking the C3t. In keeping with recommendations from reviewed literature and expert clinician advice, features were chosen whose hypothesised relationship with chorea would be simple to explain clinically. Two time-domain features were ultimately generated – the number of peaks in a signal and the width between the peaks. To study the impact of different methods of feature generation variations of these features were produced. Variations included generating the features from acceleration and jerk signals, using different high-pass filters prior to feature generation, and combining features generated from different mixes of axes and C3t tasks. Strong correlations were found between the generated features and whole-body chorea (r=0.81), upper-body chorea (r=0.79), and the UHDRS-TMS (r=0.85). These features could also estimate each clinical measure with a low degree of error (N-MAE = 15.3%, 14.8%, and 12.2% respectively). Filter frequency had a large impact on feature quality, with the best performing feature using a bandpass filter of 7.5Hz-0.3Hz, suggesting this may be a good frequency band to use for generating features sensitive to chorea. Axes and task makeup had minimal impact on feature quality. Features generated from jerk tended to outperform those generated from acceleration, however the difference was marginal. Both sets of analysis relied heavily on data collected using the developed RDCP. The developed system facilitated the synchronisation of timestamps between the C3t task times and sensor recordings. It also facilitated the transmission of C3t data from remote study sites. Although the system was by large successful several design flaws along with issues involving the GeneActiv accelerometers reduced the amount of data ultimately available. By assessing the issues encountered by study sites six recommendations for future similar research and software platforms were developed. First, sensors should be chosen based on both technical suitability and usability. In this project sensors were chosen primarily based on technical suitability and availability. In practice however usability of the sensors was found to be poor, with many sites and clinicians reporting significant issues properly using the sensors. Future projects should trial sensors with the clinicians who are using them prior to be selected. Second, development of the RDCP took place whilst the C3t was still being developed. As such part the way through the platform’s development the second version of the C3t was released. This necessitated re-working some of the underlying software, increasing development time. Whilst this can be unavoidable, future work may wish to properly take subsequent test versions into account when designing software systems and building them in a more generic manner such that modifications are as simple as possible to make. Third, the Waterfall software development methodology was used in this study despite Agile being the methodology typically preferred in industry. The rationale was that academic projects typically have their requirements set far ahead of the project starting and, in such cases, Waterfall can provide a quicker more streamlined development cycle. However, the requirements of the software changed throughout the project making waterfall unsuitable for use. Regardless, future projects should still consider Waterfall as a viable methodology when development software systems for research projects in cases where those projects are fully defined before they are started. Fourth, whilst training was given to clinicians using the RDCP testing of that training was not conducted. As such although the system appeared straightforward to operate and was found easy to use by local clinical teams, some other teams found the system hard to operate. Future work should conduct at least some testing of any training provided and provide access to an online ‘how to use’ resource. Fifth, projects which include data collected from sensor devices, particularly those that include multiple study sites, should implement automatic monitoring of data quality. In this study sensor data was not reviewed until after it had been fully collected. As such, data quality issues were not detected until it was too late to do anything about. Additionally, reviewing sensor data quality is a specialist operation most study managers will have little to no expertise in. As such, engineers working on such projects should develop systems to review data automatically and send reports to specialist personal capable of reviewing sensor data as it comes in. Finally, any clinician-facing software systems should include a function for reporting issues. In addition to enhanced training clinicians should be able to send reports from within the software itself when aspects of it are not functioning correctly. This would allow the development team to isolate ‘pain points’ within the software and apply fixes proactively.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Engineering
Uncontrolled Keywords:	Huntington’s Disease, Movement Disorders, Chorea , Accelerometers,Machine Learning
Date of First Compliant Deposit:	31 May 2022
Last Modified:	06 Jul 2023 01:58
URI:	https://orca.cardiff.ac.uk/id/eprint/150165

Actions (repository staff only)

Edit Item