Digital Cognitive Testing in Mitochondrial Disease: Validity and Challenges for Clinical Trial Use

ABSTRACT

Background

Primary mitochondrial disease is a group of genetic disorders caused by pathogenic variants in nuclear or mitochondrial DNA, often resulting in progressive neurodegeneration and cognitive decline. Current management is primarily supportive, though recent research offers hope for disease‐modifying treatments in the future. Selecting appropriate therapeutic outcomes for clinical trials in mitochondrial diseases is challenging due to limited sensitivity to changes, small sample sizes, and the burden of study related activities. This study aims to identify an efficient choice of cognitive endpoints for translational research.

Methods

This study compared digital cognitive assessments with traditional paper‐based tools. It included two cohorts: the Newcastle cohort of 45 patients recruited from the mitochondrial clinic Newcastle upon Tyne (UK) and the KHENERGYZE clinical trial cohort of 27 patients recruited from four European countries. Patients in the Newcastle cohort underwent two conventional cognitive assessments (Addenbrooke's Cognitive Examination and Montreal Cognitive Assessment), along with two computerized tests (Cogstate and Test of Attentional Performance). Potential confounding factors were also assessed.

Results

Both cohorts showed a high prevalence of moderate to severe perceived fatigue. Over 50% of patients showed reduced reaction times. Strong correlations were found between conventional and digital assessments. Several confounding factors such as education and employment were identified as influencing cognitive performance.

Conclusions

The findings support the understanding of mitochondrial disease as a slowly progressive condition, where impaired cognitive function is evident even in patients in the absence of devastating CNS manifestations such as stroke‐like episodes. Observed variability in cognitive performance may help detect meaningful changes over time.

1. Introduction

Primary mitochondrial disease (PMD) encompasses a group of genetic disorders caused by pathogenic variants in either nuclear or mitochondrial DNA, leading to impaired mitochondrial function and disrupted oxidative phosphorylation. Neurological and cognitive impairments are common features across the PMD spectrum [1]. Disease severity varies widely, ranging from milder forms characterised by slowly progressive myopathy and chronic progressive external ophthalmoplegia, maternally inherited diabetes and deafness, to severe forms such as mitochondrial encephalopathy with lactic acidosis and stroke‐like episodes (MELAS) [2]. Cognitive impairment in PMD mirrors this variability, ranging from subtle deficits to rapid deterioration due to neurodegeneration and brain atrophy. Therefore, prompt cognitive assessment should be considered when patients with PMD report cognitive difficulties [3].

Longitudinal studies have shown significant cognitive decline in patients with the m.3243A>G variant at 1 year, and more pronounced changes were observed, with statistically significant deterioration over 1 and more pronounced after 4 years [4]. Specific domains of cognitive function, such as working memory, processing speed and decision‐making, appear particularly susceptible to changes in adult patients with PMD. Monitoring these domains can help identify early deficits and guide compensatory strategies [5]. However, the diversity of neurocognitive tests used across studies and in clinical settings has introduced heterogeneity, limiting comparability and hampering meta‐analyses [6]. This underscores the need for a standardized and widely accepted cognitive assessment tool for use in clinical and research settings.

Currently, treatment options for PMD are largely limited to symptomatic management and supportive care. Nonetheless, advances in translational research have led to the emergence of clinical trials aimed at slowing disease progression [7] some of which directly target cognitive function [8]. Accurate and sensitive cognitive outcome measures are therefore essential for cognitive deficit identification and detecting clinically meaningful changes over time in these trials and guiding therapeutic decision‐making.

Digital technologies have driven the development of computerized cognitive assessments that offer interactive, efficient testing with improved sensitivity to subtle cognitive changes. A recent systematic review evaluated 10 digital cognition platforms and rated their psychometric, functional and technology aspects [9]. Cogstate has emerged as one of the most validated tools. Cogstate has shown strong correlations with traditional neuropsychological measures in conditions such as schizophrenia, depression, traumatic brain injury and AIDS dementia complex [10, 11]. Although Cogstate has not been previously validated in PMD, it has been selected as a primary cognitive outcome measure in the KHENERGYZE Phase 2b trial of sonlicromanol in patients with m.3243A>G‐associated phenotypes, including MELAS, MIDD and mixed forms [8].

Given the brain's high energy demands, mitochondrial dysfunction can result in impaired focus, alertness and cognitive symptoms that may be underrecognized in PMD [12]. Objective measures such as the Test of Attentional Performance (TAP) can help quantify these cognitive deficits. Incorporating such tools into clinical evaluations may enhance understanding of disease progression and inform therapeutic targets.

This study aims to evaluate the feasibility of computerized cognitive assessments for adult patients with mitochondrial disease in clinical research. It also explores changes and variability in cognition and alertness over a 1‐year period, alongside the influence of potential confounding factors on cognitive performance.

2. Materials and Methods

2.1. Study Population and Study Design

This study was approved by the London Queens Square Research Ethics Committee (22/PR/1052). The study comprises two patient cohorts: the Newcastle cohort (Ncl) and the screening cohort from the KHENERGYZE clinical trial (KH) [8]. The Ncl cohort was recruited at a single site in Newcastle upon Tyne, UK. The KH cohort was recruited across four different clinical sites in Europe (UK, Netherlands, Germany and Denmark), involving a multicentre research team. There were 45 adults (> 18 years) with PMD recruited prospectively to the Ncl cohort between December 2022 and March 2024. Each participant in the Ncl cohort underwent two assessments approximately 1 year apart (baseline and annual follow‐up). The KH cohort includes 27 adults (> 18 years) with PMD associated with the m.3243A>G pathogenic variant. For this study, screening data from the KHENERGYZE clinical trial were used. The study design is further illustrated in Figure 1.

FIGURE 1.

Study design. Patients recruited in the study (cohorts Ncl and KH) underwent the cognitive assessments; alongside, medical history and patient reported tests were collected and analysed. Patients recruited in the study (cohorts Ncl and KH) underwent the cognitive assessments; alongside, medical history and patient reported tests were collected and analysed. ACE, Addenbrooke's Cognitive Examination‐III; FSS, Fatigue Severity Scale; HADS, hospital anxiety and depression scale; MOCA, Montreal Cognitive Assessment; NMDAS, Newcastle Mitochondrial Disease Scale for Adults; TAP, test of attentional performance, alertness.

2.2. Study Assessments

In both cohorts, disease severity and symptom profiles were assessed using the Newcastle Mitochondrial Disease Scale for Adults (NMDAS), a semi‐quantitative clinical rating scale specifically developed for evaluating all forms of mitochondrial disease [13]. Briefly, the NMDAS comprises three sections: Section 1—Current Function (10 items), Section 2—System‐Specific Involvement (9 items) and Section 3—Current Clinical Assessment (10 items). Each item is scored on a scale from 0 (normal) to 5 (severely affected). Fatigue was evaluated using the Fatigue Severity Scale (FSS) in the Ncl cohort, and the Neuro‐QoL Fatigue Short Form in the KH cohort. Cognitive function and alertness were assessed in both cohorts using the Cogstate battery and the Test of Attentional Performance (TAP), respectively.

2.2.1. Computerized Cognitive Assessment 1. Cogstate

Cogstate computerized tests (Cogstate Software), study monitoring website (DataPoint) and deployment processes are compliant with the relevant clinical trial regulations and guidelines (http://www.cogstate.com/academic‐research). This assessment battery includes tests evaluating the executive domain, which encompasses functions attributable to the prefrontal cortex and its connections to the basal ganglia, also short‐term memory and learning. The tests administered in this study comprised the International Shopping List Task (ISL), Chase Test (practice test, excluded from statistical analyses), Groton Maze Learning Test (GML), Detection Task (DET), Identification Task (IDT), One Card Learning Task (OCL), One Back Test (ONB) and Social–Emotional Cognition Task (SEC). Test outcomes were reported both as raw scores (e.g., log₁₀‐transformed reaction times, number of correct responses) and as population‐based Z scores.

2.2.2. Computerized Cognitive Assessment 2. Test of Attentional Performance

The Test of Attentional Performance (TAP) Alertness subtest is a computerized assessment designed to measure reaction time under two distinct conditions. The first condition evaluates intrinsic alertness, capturing simple reaction time (in milliseconds) in response to a visual stimulus without any prior cue. The second condition assesses phasic alertness, wherein reaction time is measured in response to a critical stimulus that is preceded by an auditory warning tone. This condition reflects phasic arousal, or the temporal orientation of attentional focus [14]. Each test session generates descriptive statistics for individual runs as well as a summary of overall performance.

2.2.3. Other Assessments

The Addenbrooke's Cognitive Examination‐III (ACE‐III) [15] and Montreal Cognitive Assessment (MOCA) [16] were employed as validated cognitive screening tools and served as a comparator for the Cogstate and TAP test. Hospital anxiety and depression scale (HADS) [17] was used to evaluate symptoms of anxiety and depression.

Additional data were collected on factors that may influence cognitive test performance, such as limited familiarity with digital technology, level of education and clinical symptoms (e.g., ataxia). Furthermore, unstructured free‐text feedback was obtained from participants in the main cohort study to capture their subjective experiences with the Cogstate assessment.

2.3. Statistical Analysis

The sample size was determined based on clinically observed variability within a mitochondrial disease cohort and was consistent with sample sizes used in comparable Cogstate cohort based studies, for example, Alzheimer disease [18], HIV [19], bipolar disorders [20].

Descriptive statistics were used to characterise study cohorts and to analyse baseline data. T‐tests were implemented to investigate differences between cohorts and differences in test scores between baseline and follow‐up. Simple linear regression was employed to examine correlations between traditional paper‐based cognitive assessments (ACE‐III and MOCA) and computerized cognitive tests (Cogstate and TAP) and between baseline and follow‐up results. Additionally, the effects of potential confounding factors including clinical symptoms, aspects of social history, and medications were evaluated using linear regression or ANOVA, as appropriate.

2.3.1. Computerized Cognitive Assessment Data Analysis

2.3.1.1. Cogstate

For each test in the Cogstate battery, the most sensitive outcome measure was selected for analysis in accordance with the software's user guide. These measures included the number of correct responses, total errors and reaction time (ms). Raw scores were transformed into standardized Z scores, based on age‐ and sex‐adjusted normative data provided by the Cogstate developer. The Z‐scores allow fairer comparison between groups with different demographic profiles, better ensuring that observed differences or changes in scores reflect differences in cognitive ability rather than normal variation associated with age or sex.

2.3.1.2. TAP

The TAP Alertness test was divided into four blocks to control for potential fatigue effects. For both general reaction time (ms) and phasic alertness (temporal orientation of attention), the mean values, standard deviations (SD), and percentages of the predicted normative values were calculated.

3. Results

3.1. Cohort Description

This study included two independent cohorts comprising of patients with similar age distribution and gender proportions. Baseline demographics and clinical characteristics are detailed in Table 1. In the Ncl cohort, 13 distinct pathogenic variants were identified, with m.3243A>G being the most common (60%). The KH cohort consisted exclusively of individuals harbouring the m.3243A>G pathogenic variant. No significant differences in average disease burden were observed between the cohorts, as assessed by the NMDAS. This included each of the three section scores as well as the total NMDAS score (Table 4). Both cohorts demonstrated a high prevalence of moderate to severe fatigue, affecting over 60% of participants (Table S3). Additionally, in the Ncl cohort, moderate to severe symptoms of low mood and anxiety were reported by 20% and 26% of respondents, respectively (Table S1). Extended demographic data (e.g., age of disease onset, highest level of education and current employment status) and potential confounding factors for cognitive performance (e.g., body mass index, and use of antiseizure medications) were collected for the Ncl cohort only and are presented in Table 1.

TABLE 1.

Baseline demographic description for the main and KHENERGYZE cohorts.

Criteria	Newcastle study cohort (n = 45)		KHENERGYZE cohort (n = 27)
	Mean [Min–Max]	±SD	Mean [Min–Max]	±SD
Age at baseline (years)	49.3 [18.8–74.7]	14.8	46.0 [22–66]	10.8
Body mass index (kg/m2)	24.1 [14.1–37.3]	4.9	22.6 [16.9–35.7]	9.0
Duration of the symptoms (years since onset) Total n = 38	15.1 [1–42]	10.0

Sex	n (%)	n (%)
Male	13 (28.9)	7 (25.9)
Female	32 (71.1)	20 (74.1)

Ncl
Criteria	Category	n (%)
Highest level of education	School	8 (17.8)
	College	14 (31.1)
	University	23 (51.1)
Employment	Full time	14 (31.1)
	Part time	5 (11.1)
	Retired	15 (33.3)
	Unemployed	11 (24.4)
Technology experience	Some experience with digital technology	12 (26.7)
	Confident user of digital technology	33 (73.3)
Performance limiting factors (only if moderate or severe)	Motor/dexterity difficulty	7 (15.6)
	Visual impairment	2 (4.4)
	Speech disorders	1 (2.2)
Symptoms with the potential to impact computerized cognitive test performance (one patient can have more than one)	Ataxia	21 (46.7)
	Hearing Loss	31 (68.9)
	CPEO	13 (2.9)
	Eyelid ptosis	10 (22.2)
	Visual acuity impairment	4 (8.9)
	Dysarthria	1 (2.2)
	Dystonia	1 (2.2)
	Cognitive impairment	2 (4.4)
Current therapy with antiseizure medications	Yes	13 (28.9)
	No	32 (71.1)
Genotype	m.3243 A>G	27 (60.0) a
	Single large scale mtDNA deletions	3 (6.7)
	m.8344 A>G	3 (6.7)
	m.8993T T>C	2 (4.4)
	POLG1 (nuclear DNA)	3 (6.7)
	Other mtDNA variant (m.5690A>T, m.3761 C>A, m.3271T>C, m.14709T>C, m.5541C>T)	5 (11.1)
	Other nuclear DNA variant (NDUFB3, AFG3L2)	2 (4.4)

^a KHENERGYZE cohort 100%.

TABLE 4.

Mitochondrial disease assessment tool (NMDAS) scores for the Newcastle and KHENERGYZE cohorts.

NMDAS	Ncl		KH		Difference between cohorts two‐sided t‐test, p	Ncl		Annual difference, paired two‐sided t‐test, p
	Baseline					1 year follow up
	Mean [min–max]	SD	Mean [min–max]	SD		Mean [min–max]	SD
Total score (max 145)	22.9 [7–50]	10.5	21.6 [12–47]	9.3	0.633	24.9 [4–45]	10.3	0.172
Breakdown by sections
Section 1 Current function (max 50)	8.7 [0–24]	6.4	8.6 [3–22]	5.0	0.948	10.0 [0–23]	6.4	0.103
Section 2 System‐specific involvement (max 45)	7.6 [0–19]	4.4	8.2 [1–18]	3.9	0.577	7.6 [0–17]	4.6	0.112
Section 3 Current clinical assessment (max 50)	5.7 [0–13]	3.9	4.9 [1–14]	3.5	0.392	6.2 [0–16]	4.3	0.518

Note: Sample size is n = 25 for KH cohort (total and all sections). Baseline Ncl sample sizes are n = 43 (section 1), 36 (section 2), 38 (section 3), and 34 (total). Follow up Ncl sample sizes are n = 36 (section 1), 33 (section 2), 34 (section 3), 31 (total).

In the Ncl cohort, cognitive impairment, defined as a total ACE‐III score < 83 or MoCA score < 26, was identified in 26.2% and 32.6% of participants, respectively. The rest of the participants scored within the normal or borderline range on both assessments (Table S2). Within the Cogstate battery, tests primarily measuring reaction time such as Detection Task (DET), Identification Task (IDN) and One Back test (ONB)—demonstrated below‐average performance, with 56.8%–57.8% of test results yielding Z scores < −1. In contrast, a smaller proportion of participants (15.9%–26.7%) showed Z scores < −1 on tests assessing other cognitive domains including verbal learning (ISL test), executive function (GML test), visual learning (OCL test) and emotional recognition (SEC test). A similar pattern was observed in the KH cohort, suggesting that slower processing speed, rather than deficits in specific cognitive domains, may be a more prominent feature of cognitive impairment in patients with mitochondrial disease. Mean Z‐scores were not statistically significantly different between the two cohorts for all Cogstate tests excluding the DET test, for which Ncl scores were, on average, higher (Table 2).

TABLE 2.

Cogstate test scores at baseline for the Newcastle and KHENERGYZE cohorts.

Cogstate test	Outcome measure	Ncl				KH		Ncl and KH baseline comparison	Ncl
		Baseline							1 year follow up
		Raw score Mean [min–max]	SD	Z score Mean [min–max]	SD	Z score Mean [min–max]	SD	2‐sided T test, p	Raw score Mean [min–max]	SD	Z score Mean [min–max]	SD
ISL	Number of correct answers	22.2 [10–31]	5.0	−0.5 [−3.2 to 1.2]	1.0	−0.1 [−3.0 to 1.6]	1.0	0.122	23.7 [12–31]	4.3	−0.2 [−2.7 to 1.4]	0.9
GML	Number of errors	64.1 [24–145]	23.9	−0.4 [−3.1 to 1.1]	0.9	−0.2 [−2.5 to 1.2]	0.9	0.289	67.4 [30–169]	25.4	−0.5 [−4.0 to 1.1]	0.9
DET	Speed, mean of the log10‐transformed reaction time	2.6 [2.5–2.9]	0.1	−1.1 [−4.3 to 1.0]	1.2	−2.1 [−5.6 to 0.1]	1.7	0.015	2.7 [2.5–2.9]	0.1	−1.4 [−6.5 to 6.5]	1.9
IDN	Speed, mean of the log10‐transformed reaction time	2.8 [2.7–3.1]	0.1	−1.7 [−5.2 to 0.4]	1.4	−2.3 [−6.0 to 0.6]	1.3	0.099	2.8 [2.7–3.0]	0.1	−1.6 [−5.6 to 0.5]	1.4
OCL	Accuracy	0.9 [0.7–1.2]	0.1	−0.2 [−2.8 to 2.1]	1.1	0.1 [−1.7 to 1.6]	0.9	0.300	1.0 [0.8–1.2]	0.1	0.0 [−1.6 to 2.3]	1.0
ONB	Speed, mean of the log10‐transformed reaction time	3.0 [2.8–3.3]	0.1	−1.1 [−3.6 to 0.8]	1.1	−1.1 [−2.9 to 0.8]	0.9	0.825	3.0 [2.8–3.2]	0.1	−1.1 [−3.6 to 1.7]	1.3
SEC	Accuracy	1.1 [0.6–1.3]	0.2	−0.2 [−4.0 to 1.2]	1.1				1.1 [0.8–1.3]	0.1	0.1 [−3.0 to 1.2]	0.8

Statistical analysis of annual change (Ncl cohort)
Baseline and 1 year	Two‐sided paired t‐test, p	Linear regression between baseline and year 1 scores, F‐test p =, R 2	n (%) improved	n (%) no change	n (%) worsened	Δ Z score Mean	SD	95% CI
ISL	0.040	< 0.001, 0.271	25 (65.8)	1 (2.6)	12 (31.6)	−0.31	0.91	−0.62 to −0.01
GML	0.736	0.105, 0.071	19 (50.0)	0	19 (50.0)	0.06	1.12	−0.31 to 0.43
DET	0.607	0.206, 0.044	13 (34.2)	1 (2.6)	24 (63.2)	0.17	2.02	−0.49 to 0.83
IDN	0.788	0.001, 0.250	21 (55.3)	1 (2.6)	16 (42.1)	−0.06	1.40	−0.52 to 0.40
OCL	0.131	< 0.001, 0.417	23 (60.5)	2 (5.3)	13 (34.2)	−0.22	0.88	−0.51 to 0.07
ONB	0.743	< 0.001, 0.358	21 (56.8)	0	16 (43.2)	−0.06	1.09	−0.42 to 0.30
SEC	0.055	< 0.001, 0.365	21 (58.3)	1 (2.8)	14 (38.9)	−0.26	0.80	−0.53 to 0.01

Note: Sample size is n = 27 for all tests for KH cohort. Baseline sample size is n = 45 for Ncl for all tests excluding OCB (n = 44) and SEC (n = 44). Follow up sample size is n = 38 for Ncl for tests excluding SEC (n = 37).

Reaction time measured using the TAP Alertness test (no signal) ranged from 214 to 513 ms in the Ncl cohort and from 260 to 846 ms in the KH cohort. The normative threshold for healthy performance is < 250 ms. In the Ncl cohort, reaction time was significantly prolonged in 52.5% of participants, with values falling below 40% of the predicted norm. The standard deviation (SD) of reaction times in the TAP test was narrower in the Ncl cohort, indicating more stable performance across test blocks. In contrast, a wider SD was observed in the KH cohort. Additionally, KH cohort reaction times were typically (and statistically significantly) longer (Table 3, Figure 2). These differences potentially may reflect greater variability in testing environments across the recruitment sites as well as ascertainment bias coming from the clinical trial inclusion criteria.

TABLE 3.

TAP alertness test scores at baseline in Ncl and KH cohorts.

Alertness	Ncl		KH			Ncl				Ncl
	Baseline									1 year follow up
	Mean [min–max]	SD	Mean [min–max]	SD	2‐tailed test (Ncl vs KH cohort)	Mean (Index of phasic alertness) [min–max]	SD	Mean % of Norm [min–max]	SD	Mean [min–max]	SD	Mean (Index of phasic alertness) [min–max]	SD	Mean % of Norm [min–max]	SD
TAP no signal, mean (ms)	320.6 [214–513]	80.4	404.9 [260–846]	145.6	0.009	0.3 [−0.21–0.28]	0.12	43.7 [3–99]	33.3	321.8 [213–665]	97.1	0.00 [−0.24 to 0.24]	0.10	40.0 [2–98]	28.7
TAP with signal, mean (ms)	307.9 [209–546]	79.4	375.4 [233–712]	136.8	0.026					322.5 [209–695]	99.2

Statistical analysis of annual change (Ncl cohort)
Baseline and 1 year	Two‐sided paired T‐test, p	Linear regression, F‐test p =, R 2	n (%) improved	n (%) no change	n (%) worsened	Reaction time change, ms Mean	SD	95% CI
TAP no signal, mean	0.489	< 0.001, 0.571	19 (55.9)	1 (2.9)	14 (41.2)	7.9	66.0	−15.1 to 30.9
TAP with signal, mean	0.604	< 0.001, 0.454	16 (47.1)	1 (2.9)	17 (50.0)	−6.7	74.3	−32.6 to 19.3
Mean index of phasic alertness	0.279	0.058, 0.108	15 (44.1)	0 (3.8)	19 (55.9)	0.0	0.1	−0.02 to 0.07
Mean % of Norm	0.586	0.136, 0.098	10 (41.7)	2 (8.3)	12 (50.0)	3.8	33.7	−10.4 to 18.0

Note: Sample size is n = 27 for all scores for KH cohort. Baseline sample size for Ncl is n = 42 for Tap no signal, TAP with signal and Index of phasic alertness, and n = 40 for % of Norm. Follow up sample size is n = 36 for Tap no signal, TAP with signal and Index of phasic alertness, and n = 27 for % of Norm.

FIGURE 2.

TAP test score distribution in two independent cohorts. Distribution of reaction times (ms) measured by the Test of Attentional Performance under two conditions: With and without an audio signal, across two independent cohorts. Statistically significant differences in mean times were observed between cohorts.

3.2. Baseline Correlations

3.2.1. Correlation Between Digital and Paper‐Based Assessments

The primary outcomes of the TAP Alertness test—mean reaction times (ms) with and without an auditory warning signal—were significantly correlated with total scores on both the MoCA and ACE‐III assessments. Specifically, longer reaction times were associated with lower cognitive scores, suggesting a relationship between reduced processing speed and global cognitive performance (Figure S1).

Both ACE‐III and MoCA scores showed statistically significant positive correlations with Cogstate tasks assessing verbal memory and emotional recognition. In addition, significant negative correlations were observed with Cogstate tasks assessing reaction time, namely the IDN and ONB—meaning that longer reaction times were associated with lower total ACE‐III and MoCA scores.

TAP reaction time outcomes (both with and without a warning signal) were strongly correlated with Cogstate reaction time‐based tasks: (Det, IDN and ONB) (Table S6).

3.2.2. Confounding Factors

3.2.2.1. Demographic Factors

Overall, sex did not significantly influence cognitive test outcomes in this study, with one exception: female participants performed better on the Cogstate verbal learning task (International Shopping List), demonstrating an average Z‐score of 0.65 points higher than male participants.

A higher level of education was associated with better performance on both the ACE‐III and the TAP Alertness test. Similarly, employment status was a significant predictor of better cognitive outcomes.

Participants who were unemployed exhibited slower reaction times, averaging 79.4 ms longer (on the TAP Alertness test) compared to those in full‐time employment, or 87.6 ms longer compared to those employed full or part time. In addition, being employed (either full‐time or part‐time) was associated with better performance on the DET and IDN Cogstate tasks, which measure processing speed.

Technology experience had a limited impact on cognitive test outcomes. Self‐identified ‘confident technology users’ achieved Z‐scores 1.14 points higher on average on the IDN test than those reporting only ‘some experience with technology’. However, technology experience did not affect any other Cogstate tests or TAP Alertness test outcomes (Table S4, Figure S1). Notably, no participants in the Ncl cohort identified themselves as completely naïve to digital technologies.

3.2.2.2. Subjective Assessments

Symptoms of anxiety and depression, assessed using the HADS and the Beck Depression Inventory (BDI), did not show a statistically significant influence on cognitive performance as measured by the MoCA, ACE‐III, Cogstate battery, or TAP Alertness test in the Ncl and KH cohort. In the Ncl cohort, symptoms of fatigue, assessed using the Fatigue Severity Scale (FSS), showed statistically significant correlations with the ISL, GML and ONB tests. In each case, a higher fatigue level was associated with a reduced test performance. However, in the KH cohort, fatigue (assessed using Neuro‐QoL) did not show statistically significant correlations with any of the measures of cognitive performance (Table S4).

3.2.2.3. Clinical Factors

NMDAS, used as a marker of disease severity, demonstrated a significant correlation with TAP performance in the Ncl cohort. This included both the total NMDAS score and specific subsections—Section 1 (patient‐reported functional status) and Section 3 (clinical examination findings). Within the Cogstate battery, the ONB and SEC tasks also correlated with the NMDAS total score. However, no significant correlations with the NMDAS total score were observed in the KH cohort (Table S5).

Genotype effects were assessed by grouping cases into three categories: m.3243A>G (n = 27), other mtDNA variants (n = 13) and nuclear DNA variants (n = 5). No statistically significant effect of genotype was observed on Cogstate and TAP test outcomes.

Several indirect markers of disease severity such as disease duration (measured in years since onset), heteroplasmy levels, body mass index (BMI), and use of anti‐seizure medication were analysed for the Ncl cohort. A lower BMI (< 24) was associated with slower performance on the TAP Alertness test, by an average of 50.8 ms. A similar trend was seen in Cogstate tasks measuring processing speed, specifically ONB and DET. Additionally, use of anti‐seizure medication was associated with lower performance in the ONB and DET tests. Duration of disease correlated with performance on several Cogstate tests, including ONB, ISL, DET, IDN and SEC, but showed no significant relationship with TAP outcomes. Lower heteroplasmy levels (< 40%) in blood were associated with better performance on the ONB and DET tasks, with average Z‐scores 0.88 and 0.85 points higher, respectively.

3.3. Annual Change in Performance

3.3.1. Cognitive Assessment: Cogstate, TAP, ACE‐III and MOCA

Apart from the ISL (verbal learning) test, there were no statistically significant changes in mean outcome between the baseline and 1‐year follow‐up for the Cogstate test Z‐scores (Table 2, Figure 3), Tap performance with and without signal (Table 3, Figure S2), or ACE and MOCA overall scores (Table S2). The ISL test demonstrated a spontaneous statistically significant improvement in Z‐score (from −0.5 to −0.2); however, this change is not considered clinically meaningful. Score trajectories were approximately evenly split between minor improvement and decline, and score variability (standard deviation and 95% confidence intervals) remained comparable to baseline.

FIGURE 3.

Cogstate battery test population‐based Z‐score distribution in two independent cohorts. Distribution of Z scores from the seven Cogstate battery tests across two independent cohorts shows that most participants exhibit cognitive deficits, particularly in DET, IDN, and ONB tests, which assess reaction time to visual stimuli.

3.3.2. Confounding Factors: Fatigue, Anxiety and Depression

Fatigue levels, as measured by the FSS, did not change significantly over 1 year. Participants were approximately equally likely to report higher or lower fatigue scores compared to baseline, with a 95% CI of −4 to +5 for the change in points (Table S3). A similar pattern was observed for symptoms of anxiety and depression measured by the HADS. Mean changes in total scores were < 1 point, with 95% CI of −0.7 to +0.9 for anxiety and −1.6 to +0.5 for depression, indicating no statistically significant shifts (Table S1).

3.3.3. Burden of the Condition: NMDAS

NMDAS showed a mean increase of approximately 2 points over 1 year, suggesting some worsening of disease burden. This change was primarily driven by Section 1, which reflects the patient's subjective experience. However, no statistically significant change was observed in the total NMDAS score or in any individual section (Table 4).

3.4. User Feedback

Qualitative analysis of participant feedback was based on recurring keywords and the overall context of the comments. The majority of participants (n = 26; 56%) provided neutral or overall positive feedback regarding the computerized assessments. Six participants (15%) offered unprompted comments indicating a preference for digital testing over traditional paper‐based assessments, whereas three participants (7.5%) expressed a preference for paper‐based tests. Seven participants (17.5%) reported that the Cogstate assessment felt overly long and repetitive, which contributed to fatigue, reduced concentration, and a need for comfort breaks. Additionally, the SEC task within the Cogstate battery was described as ‘strange’ by four participants (10%), primarily due to the stylized nature of the facial expression drawings used in the task.

4. Discussion

PMD are currently in focus with several ongoing preclinical studies and clinical trials being conducted. Historically, translational efforts have focused on seizures, myopathy and visual impairment. However, mitochondrial dysfunction has been implicated in multiple biological mechanisms [21] and also linked to neurodegeneration and cognitive decline [6], which have not been as extensively researched, despite having devastating impacts on patients and their caregivers. Recently, fatigue and cognitive performance have been studied with digital cognitive assessments, which have been used as inclusion criteria and outcome measures in several mitochondrial‐related clinical trials [8].

This study assessed cognitive performance at baseline and annually using computerized tools, analysing their correlation with traditional assessments and the impact of potential confounders to inform future trial design. Two independent, yet demographically and clinically comparable, cohorts enabled analysis of variability and reproducibility.

Across both cohorts, computerized tests that measure reaction time identified processing speed deficits in over 50% of participants. This exceeds the approximately 30% of patients with cognitive deficits identified by ACE‐III, MoCA and other Cogstate tests. Correlations between TAP and Cogstate reaction time tasks suggest a consistent slowing of processing speed, regardless of test type. Together, these findings indicate that processing speed may be more profoundly affected than discrete cognitive functions such as memory or learning in individuals with mitochondrial disease.

Annual progression and trajectories across all cognitive tests were stable, with no statistically significant changes observed, particularly in the absence of major medical events, for example, stroke‐like episodes or metabolic decompensation. The agreement between tests supports the reliability and internal consistency of the assessments. Additionally, correlations between baseline and annual scores across multiple cognitive measures suggest consistent patient performance, minimal assessor bias and overall good reproducibility.

Notably, most participants in both cohorts reported moderate to severe fatigue, consistent with the known burden of PMD [22]. This subjective symptom significantly affected cognitive performance in 3 tests but remained stable over the 1‐year period. Subjective assessments of anxiety and depression remained stable and did not significantly affect cognitive performance. These findings highlight the importance of incorporating both subjective and objective outcome measures in clinical trials to fully capture disease burden.

The KH and Ncl cohorts did not show consistent correlations between digital cognitive assessments and the NMDAS. Although this may reflect an incidental finding, it may also reflect the differences in recruitment strategies, clinical settings and assessment practices across countries. This underscores that variability in cognitive outcomes can arise not only from patient‐related factors but also from differences in clinical environments and examiner consistency. NMDAS scores worsened by 2 points, which is not statistically significant; however, it reflects worsening of the condition and is in line with the previous findings [22].

This study also indicates that several confounding factors may influence cognitive performance and should be considered in study design and analysis. Although some assessments already adjust for age and education, we found that unemployment, lower BMI, use of anticonvulsants, and higher heteroplasmy levels were associated with poorer cognitive outcomes in some tests. These variables may also be considered as indirect markers of disease severity. In contrast, common mitochondrial symptoms such as chronic progressive external ophthalmoplegia, eyelid ptosis, ataxia and muscle weakness did not significantly affect cognitive performance.

Participant feedback indicated that completing all seven Cogstate tests in succession was lengthy (up to 40 min) and fatiguing. Since several tasks assess reaction time and produce comparable results, a streamlined battery may be more suitable for future studies. We recommend focusing on ISL, ONB and SEC—targeting learning, memory and social–emotional recognition—along with a single reaction time task, such as TAP or IDN, to efficiently evaluate processing speed.

In summary, digital cognitive assessments are effective for detecting mild to moderate cognitive impairment in individuals with mitochondrial disease. Their correlation with paper‐based methods supports their validity, whereas reduced floor and ceiling effects enhance their sensitivity and provide more granularity across a wider performance range. Findings also suggest that processing speed, rather than isolated cognitive domains, may be more significantly affected in many patients with mitochondrial disease. Longitudinal data show minimal variability over 1 year, indicating that cognitive performance remains largely stable in the absence of major clinical events.

Although participants noted that lengthy test batteries could be fatiguing, computerized assessments were generally well received, requiring minimal digital literacy. They also overcome some limitations of traditional tools, such as reliance on fine motor skills, and offer potential for remote administration, reducing participant burden and improving accessibility for future clinical trials.

Finally, several factors, such as education, BMI, or use of anti‐seizure medication, should be considered in study design to ensure accurate interpretation and effective case–control matching.

Author Contributions

Oksana Pogoryelova: study design, data acquisition, data analysis, interpretation, manuscript writing. Jan Smeitink: data acquisition, data interpretation, manuscript revision. Herma Renkema: data acquisition, data interpretation, manuscript revision. Rosabeth White: statistical data analysis, data interpretation, manuscript revision. Frankie Barton: data acquisition, manuscript revision. Rachel Lyon: data acquisition, manuscript revision. Jane Newman: scientific advisor, critical review of the study proposal, data acquisition, manuscript revision. Aye Moe: data acquisition, manuscript revision. Yi Shiau Ng: scientific advisor, critical review of the study proposal, manuscript revision. Gráinne Siobhan Gorman: scientific advisor, critical review of the study proposal, manuscript revision.

Funding

The software licence and patient travel arrangements were funded by the charity organisation Mito Foundation, Australia.

Ethics Statement

The study was approved by the London Queens Square Research Ethics Committee (Reference number: 22/PR/1052).

Consent

Written informed consent was obtained from all participants prior to their inclusion in the study.

Conflicts of Interest

Jan Smeitink (Founder and CEO) and Herma Renkema (CSO) are employees of Khondrion, sponsor of the KHENERGYZE trial.

Supporting information

Figure S1: Heatmap illustrating the correlations among cognitive test scores, fatigue levels and anxiety and depression scales for Ncl and KH cohorts.

Figure S2: Annual change in processing speed for each participant in the Kh cohort, measured using the TAP test under two conditions: with signal and without signal.

Table S1: Hospital anxiety and depression assessment at baseline and 1 year follow up.

Table S2: Baseline description of the cognitive performance in the Ncl cohort over 1 year as assessed by Addenbrookes (ACE) and Montreal (MOCA) scales and at baseline in KH cohort assessed by ACE. There was no statistically significant change of scores in the cohort, and baseline scores correlated with the scores at 1 year.

Table S3: Fatigue scores at baseline assessed by fatigue severity score and Neuro‐QoL short fatigue form in the Newcastle and KHENERGYZE cohorts, respectively.

Table S4: Correlation analysis between paper based cognitive assessment (MOCA, ACE and CFQ questionnaire) and computerized tests.

Table S5: Correlation analysis 2. Correlation analysis between digital cognitive assessments and clinical disease severity assessment tool.

Table S6: Correlation analysis 3. Correlation analysis between two digital cognitive assessments which measure reaction time to a stimulus.

Funding Statement

This work was funded by Mito Foundation grant G190.

Data Availability Statement

Anonymised, aggregated data underlying the findings of this study may be made available upon reasonable request to the corresponding author. Any request for data access will be reviewed by the relevant local institutional authorities to ensure compliance with participant consent and applicable local data protection policies. Data will only be shared in accordance with ethical guidelines and where appropriate approvals are in place.