A Mechanistic Insight into Sources of Error of Visual Working Memory in Multiple Sclerosis

Reviewer #1 (Public Review):

This study compares visuospatial working memory performance between patients with MS and healthy controls, assessed using analog report tasks that provide continuous measures of recall error. The aim is to advance on previous studies of VWM in MS that have used binary (correct/incorrect) measures of recall, such as from change detection tasks, that are not sensitive to the resolution with which features can be recalled, and to use mixture modelling to potentially disentangle different contributions to overall performance. This aim is met in part, but there are some problems with the authors' interpretation of their findings:

- How can the authors be confident the performance deficits in the patient groups are impairments of working memory and not visual or motor in nature? I appreciate there was some kind of clinical screening, but it seems like there should have been a control condition matched to the experimental tasks with only the memory components removed.
- The participant groups are large, which is definitely a strength, but not particularly well-matched in terms of demographics, with notable differences in age (mean and spread), years of education and gender. These could potentially contribute to differences in performance between groups and tasks.
- The authors interpret the mixture model parameter described as "misbinding error" as reflecting failures of feature binding, and propose a link to hippocampus on that basis, however there is now quite strong evidence that these errors (often called swaps) are explained mostly or entirely by imprecision in memory for the cue feature (bar color in this case), e.g. McMaster et al. (2022), already cited in the ms.

The methodology of the ROC analyses should be described in more detail: it is not clear what measures are being used to classify participants or how.

There are a number of unusual choices of terminology that could potentially confuse or mislead the reader:
- The tasks are not "n-Back" tasks by the usual meaning: they are analog report tasks with sequential presentation.
- The terms recall "error","variability", "precision" and "fidelity" are used idiosyncratically. Variability and precision usually refer to the same thing: they describe the dispersion or spread of errors. The measure described as recall error in the sequential tasks is presumably absolute (or unsigned) error. For the mixture model parameters I suggest describing them more explicitly in terms of the mixture attributes, e.g. "Von Mises SD", "Target proportion", "Non-target proportion" "Uniform proportion".

Reviewer #2 (Public Review):

The authors applied two visual working memory tasks, a memory-guided localization (MGL), examining short-term memory of the location of an item over a brief interval, and an N-back task, examining orientation of a centrally presented item, in order to test working memory performance in patients with multiple sclerosis (including a subgroup with relapsing-remitting and one with secondary progressive MS), compared with healthy control subjects. The authors used an approach in testing and statistically modelling visual working memory paradigm previously developed by Paul Bays, Masud Husain and colleagues. Such continuous measure approaches make it possible to quantify the precision, or resolution, of working memory, as opposed to measuring working memory using discretised, all-or-none measures.

The authors of the present study found that both MS subgroups performed worse than controls on the N-back task and that only the secondary progressive MS subgroup was significantly impaired on the MGL task. The underlying sources of error including incorrect association of an object's identity with its location or serial order, were also examined.

The application of more precise psychophysiological methods to test visual working memory in multiple sclerosis should be applauded. It has the potential to lead to more sensitive and specific tests which could potentially be used as useful outcome measures in clinical trials of disease modifying drugs, for example.

However, there are some significant limitations which severely affect the scientific validity and interpretability of the study:

a) There is a striking lack of key clinical information. The inclusion and exclusion criteria are unclear and a recruitment flowchart has not been provided. Therefore it is unclear what proportion of MS patients were ineligible due to, for example, visual impairment. Basic clinical data such as EDSS scores, disease duration, treatment history, and performance on standard cognitive testing were not provided. Basic clinical and demographic data for each subgroup were not provided in a clear format. This severely limits the interpretability of the study and its significance for this clinical population. For example, might it be that the SPMS patients performed worse on the MGL task because they were more cognitively impaired than RRMS patients? That question might be easily answered, but the answer is unclear based on the data provided.

b) The study is completely agnostic to the underlying pathophysiology. There is no neuroimaging available, therefore it is unclear how the specific working memory impairments observed might relate to lesioned underlying brain networks which are crucial for specific aspects of working memory. This severely limits the scientific impact of the results. This limitation is acknowledged by the authors, but the authors did not put forward any hypotheses on how their results may be underpinned by the underlying disease processes.

c) The present study does not compare the continuous-report testing with a discrete measure task so it is unclear if the former is more sensitive, or more feasible in this patient group, although this may not have been the purpose of the study.