1 April 2019
Verbal paired associates: Effects of age and education in automated remote testing
New data presented at AD/PD 2019 demonstrated the feasibility and validity of conducting automated, verbal recall tasks on participants’ own devices. This research has implications for recruitment and monitoring in Alzheimer’s disease trials.
The automated assessment of memory enables (i) efficient recruitment in Alzheimer’s disease (AD) trials, and (ii) at home monitoring of cognition. Whilst, this has been achieved for computerised visual memory tests, automated verbal recall tasks has remained problematic, due to dependence on trained raters.
Here, we report data from an automatically administered and scored test of Verbal Paired Associates learning (aVPA), collected in a large sample of remotely tested participants.
We were interested in examining:
- The feasibility of large-scale automated verbal testing on participants’ own devices
- The validity of this testing modality, examining the ability to detect previously observed performance differences by task condition and participant demographics
2588 participants were tested at home, on their own devices, using the Cambridge Cognition Neurovocalix platform for automated verbal testing.
Participants recalled four easy (semantically related) and four hard (unrelated) pairs. The task terminated when participants recalled all items correctly, or after three unsuccessful attempts.
We extracted the following measures from the aVPA task:
- Immediate Memory (number correct on the first trial)
- Total Recall (Total correct across all trials)
- Learning (Trial 3 score – Trial 1 Score)
- Difficulty Ratio (total correct easy trials – total correct hard trials)
Consistent with other versions of the task, there was a significant reduction in errors across the three learning trials (1A).
We also observed fewer errors for the semantically related than semantically unrelated trials (p<0.001).
These results are in line with our expectations (1B) based on models of verbal episodic memory and previous data from pencil-and-paper versions of this task.
Figure 1: Performance of participants across different task conditions showing improvement with repetition (A), and distribution of error scores by word pair difficulty (B).
These findings suggest that the aVPA is assessing verbal episodic memory and learning.
Relationship to Participant Demographics
Verbal memory is known to show age-related decline. Therefore, we would expect performance on this task to be lower in older participants. The relationship between participant characteristics and performance on the automated VPA was examined by ordinary least squares regression:
- aVPA Total Errors was significantly associated with Age, Gender and Level of Education
- aVPA Immediate Recall was associated with Age and Education
These data replicate well-established findings from conventionally administered VPA tasks, supporting the validity of remote verbal testing using ASR technology.
We observed expected differences in performance depending on task condition, and associations between age, gender, education and task outcome measures.
In future work, we aim to extend these findings in patients with neurodegenerative disease.
Dr Francesca Cormack, Director of Research & Innovation