A Multidisciplinary Approach
Mouse moves and Keyboard actions are recorded by monitoring software so we can track all user interface operations which people carry out, e.g. selecting menu options, selecting icons, etc. SAMS tracks use of email and Word documents so we can capture text for analysis. The software also tracks use of the Internet; however, we only capture a limited set of data- Microsoft Windows user interface operations, and text created by the volunteer user. Everything else is blanked out. We also take care to encrypt and anonymise data so no personal details or sensitive data (e.g. passwords or using on line banking) are recorded.
The recorded data is stored on computers in the Universities of Lancaster and Manchester with strict security. We analyse the data with two techniques:
Data Mining: this looks for patterns in user behavior i.e. keyboard actions, mouse movements and user interface actions such as opening and closing Windows. Data mining techniques can discover trends or changes from a normal pattern. We compare each person with themselves so look for changes over time, as well as comparing each person against population wide standards.
Text mining: this analyses email messages and word document text for normal patterns using known standards for frequencies of words used, grammar, etc to look for and exceptions and mistakes which may indicate cognitive problems. As with data mining we compare people with themselves and population wide standards.
The results from data and text mining are linked to potential 'cognitive indicators', which are measures that medical experts use in diagnostic tests for dementia. Some examples are:
Our hypothesis is that we can detect subtle changes in behavior that may indicate cognitive problems associated with dementia. If we manage to detect possible problems then the SAMS system will urge people to take follow up tests which already exist of the web, or to contact their doctor for further tests.
Contextual Inquiry - Expert Reference Group
A key challenge in analysing computer-use is the extensive range of behaviours that could be captured. Therefore, we convened a reference group of experts from clinical and cognitive neurosciences to gain consensus on a) which computer use behaviours are most likely to be sensitive and specific to detecting early cognitive decline, and b) which domains of cognitive function each of these behaviours would be dependent on. The expert reference group first participated in two workshops focussed on identifying which computer-use behaviours (e.g. mistakes or slowed functioning) are most likely to indicate early cognitive decline. They then completed a follow-up survey to indicate the extent to which each of the twenty-two shortlisted computer-use behaviours could indicate impairment in six cognitive domains.
Cross-sectional study - cognitive assessments
Participants completed a battery of assessments to determine memory and thinking abilities prior to conducting a semi-directed computer-use session. These assessments are designed to measure functioning over a range of cognitive domains including memory, attention, inhibition, processing speed and executive functioning. These cognitive measures will be analysed alongside the computer use-behaviours defined by our expert reference group. The aim is to determine whether increasing levels of cognitive dysfunction are related to poorer computer-use behaviour at the individual level, and thus provides increased confidence that we are able to capture poorer memory and thinking abilities via day-to-day computer use behaviours.
Longitudinal study - cognitive assessments
A similar cognitive testing battery to the cross-sectional study will be implemented at three time points throughout the 9 month period of continuous data capture of computer-use behaviour; start point (0 months), mid-point (4 months) and end-point (8-9 months). Any change in normal computer-use patterns throughout this period will be compared against the cognitive battery to determine whether these changes may be due to cognitive decline.