Polygenic risk for hearing abilities and longitudinal cognitive outcomes in older adults

1 Introduction

Alzheimer's disease (AD) and other dementias affect 57 million people worldwide (World Health Organization, 2025). AD and other dementias typically have a preclinical stage in which no behavioral symptoms are reported, but biological changes in the brain are observed (Livingston et al., 2024). The Lancet Commission on Dementia Prevention, Intervention and Care has reported that addressing hearing loss during mid-life contributes to the highest percentage of the total modifiable risk for dementia due to its high prevalence among older adults (Livingston et al., 2024). However, an “auditory biomarker” of cognitive decline has yet to be elucidated, although the potential of central auditory processing tests for serving as a biomarker for cognitive decline has been supported by previous studies (Gates et al., 2011; Häggström et al., 2020; Levett et al., 2025).

Central auditory processing generally refers to the wide array of cognitive, attentional, linguistic, and network analyses necessary for processing auditory signals in order to facilitate complex listening. The hypothesis linking central auditory processing and cognitive decline suggests that cognitive demands are high for performing central auditory processing tasks which interact with pathological synaptic mechanisms, resulting in neuronal degeneration (Griffiths et al., 2020). In support of this hypothesis, accumulation of molecular biomarkers of AD pathology have been observed in the medial temporal lobe, which is thought to have a role in auditory processing (Griffiths et al., 2020; Khan et al., 2014). A recent scoping review by de Azevedo Picinini et al. (2025) reported dichotic tests, which assess binaural central auditory processing, are sensitive in detecting early signs of cognitive impairment (Mohammed et al., 2022; Tuwaig et al., 2017). Similarly, Gates et al. (2008) reported that dichotic sentence identification was more sensitive to memory impairments than dichotic digits among adults with mild cognitive impairment and dementia. However, it is yet to be determined which central auditory processing tests are suitable for and tolerated by the general older adult population (Levett et al., 2025).

A previous cross-sectional examination of community-dwelling older adults in five cohorts of the Rush Alzheimer's Disease Center (RADC) found that complex tests of auditory processing such as dynamic and static spectral-pattern discrimination contributed significantly to model predictions of global cognition (Sheft et al., 2015). Although central auditory dysfunction is common in individuals with hearing loss and cognitive impairment, few studies have examined the longitudinal relationship between central auditory tests and cognitive outcomes (Tarawneh et al., 2022; Häggström et al., 2020). The current study reports on the longitudinal cognitive outcomes of the RADC cohorts to determine whether auditory performance on complex tests of auditory processing at the time of evaluation were predictive of global cognition at follow-up.

Additionally, years of research examining the link between hearing loss and cognitive decline has highlighted the role of genetic predisposition for hearing loss and AD with suggestions of shared genetic vulnerabilities (Abidin et al., 2021) and pleiotropy (Brenowitz et al., 2020) as a common factor (Levett et al., 2025). However, studies have reported mixed findings on whether genetic risk for AD predicts hearing difficulty (Mitchell et al., 2020; Abidin et al., 2021; Brenowitz et al., 2020) and whether genetic risk for HL predicts cognitive outcomes (Jiang et al., 2024; Wang et al., 2022), with most studies examining self-reported hearing. The current study addresses the gaps in the literature on the association between genetic predisposition to AD and HL, central auditory processing, and cognitive decline. Therefore, the objective for this study was to examine the relationship between simple and complex hearing abilities and longitudinal cognitive outcomes after controlling for genetic risk for AD and HL.

2 Methods

2.1 Participants and data source

5,247 adults (mean = 77.1 ± 7.9 years) enrolled into one of five longitudinal cohorts of the RADC, including the Religious Orders Study (ROS), Memory and Aging Project (MAP), Minority Aging Research Project (MARS), the Clinical Core (CC), and the Latino Core (LATC) participated in this study. The ROS cohort started recruitment in 1994 and consists of clergy men and women recruited from across the United States, whereas the MAP, MARS, CC, and LATC consist of community dwellers recruited from the Chicagoland area (Arvanitakis et al., 2018; Barnes et al., 2012; Bennett et al., 2012a,b). Participants from all cohorts underwent annual cognitive testing, which provided the longitudinal outcome data for the present analyses. In 2012–2013, participants in the CC were invited to participate in a study examining the relationship between measured hearing abilities and cognitive outcomes in a cross-sectional examination previously published in Sheft et al. (2015) (see flowchart in Figure 1 for enrollment, screening, and inclusion for both the original cohort dataset and the subset with measured hearing abilities). Only adults with no prior diagnosis of dementia at baseline were included in the study. Only the Mini-Mental State Examination was administered to participants in the measured-hearing group as described in Sheft et al. (2015). Informed consent was obtained from all individuals. All procedures were reviewed by the Institutional Review Board at Rush University Medical Center. Additional details of the cohorts and data source are available at: https://www.radc.rush.edu/docs/parentStudyDesigns.htm.

2.2 Outcome

Global cognition was computed as the standardized score across 19 tests in five domains of cognition, including perceptual speed, semantic memory, episodic memory, working memory, and visuospatial ability. Details of these tests are described elsewhere (Bennett et al., 2005). Briefly, the seven tests of episodic memory included the immediate and delayed recall of East Boston Story and Story A of the Wechsler Memory Scale-Revised, Recall and Recognition, and Word List Memory. The three tests of semantic memory were Verbal Fluency, Boston Naming, and National Adult Reading Test, while the three tests of working memory included Digit Span Forward, Digit Span Backward, and Digit Ordering. The four tests of perceptual speed were Symbol Digit Modalities Test, Number Comparison, and two measures of a modified Stroop Test. Finally, the two tests of visuospatial ability were Judgement Line Orientation and Standard Progressive Matrices. The individual scores for each test were converted to z scores using the combined cohort baseline means and standard deviations for each test. Z scores were then averaged across domains to get the global cognition composite score.

2.3 Exposures

2.3.1 Self-reported hearing ability

Self-reported hearing was obtained during annual interviews conducted by a research assistant. The following questions were asked: (1) “(with/without a hearing aid) can you usually hear and understand what a person says without seeing his face if that person talks to you in a normal voice in a quiet room?” and (2) “(with/without a hearing aid) how often do you have difficulty understanding people over the telephone?”. Both self-reported hearing measures were analyzed as binary indicators of hearing difficulty. Hearing trouble in a quiet environment was coded as 0 = no trouble understanding and 1 = trouble understanding. Difficulty understanding speech on the phone was dichotomized as 0 = no difficulty (“Rarely or Never”) and 1 = any difficulty (“Almost Always” or “Occasionally”).

2.3.2 Pure-tone audiometry

Hearing abilities were measured in the participants' homes using a laptop and Sennheiser HD 280 Pro headphones with sound levels calibrated for an average adult ear canal using Knowles Electronic Manikin for Acoustic Research. Pure tone thresholds were obtained for 500, 1,000, 2,000, and 4,000 Hz and averaged to provide a speech-frequency pure-tone average in the better-hearing ear only. Otoscopy, tympanometry, and bone-conduction audiometry were not performed.

2.3.3 Quick speech in noise (QuickSIN)

The QuickSIN test consisted of sentences presented with four-talker babble with a starting signal-to-noise ratio (SNR) of 25 dB, decreasing in 5 dB steps to 0 dB. The sentences were presented at 70 decibel (dB) hearing level (HL) or the most comfortable level using a diotic presentation under headphones. The SNR loss averaged across two test lists was the SNR needed for 50% of the keywords in each sentence to be repeated correctly.

2.3.4 Static and dynamic spectral ripple discrimination

Starting Phase of a Spectral Ripple Discrimination: the two tests of frequency pattern discrimination are described in Sheft et al. (2015). Briefly, a cued two-interval forced-choice procedure was used, and participants were asked to report which stimulus differed from the cue. In the first task, the stimuli consisted of 500-ms sinusoidal ripples presented at 80 dB sound pressure level (SPL). The ripples were passed through a speech-shaped noise filter with energy between 200 and 500 Hz. The starting phase of the ripples was initially set at 2.3 radians and varied in 0.54-radian steps until the participant was no longer able to detect a difference between the cue and test stimuli. The ending phase difference was termed the delta ripple phase, with a smaller delta indicating better frequency pattern discrimination ability.

Frequency-Modulated (FM) Pattern Discrimination in Noise: In the second task, the stimuli consisted of 1,000 Hz stochastically frequency-modulated sinusoids using 5-Hz noise modulators and a fixed maximum frequency excursion of 400 Hz. The 500-ms modulated sinusoids were also centered in 1,000-ms speech-shaped noise maskers with a fixed level of 80 dB SPL. The level of the FM sinusoids varied in 6-dB steps to estimate the SNR needed to discriminate the frequency variation. Similar to the first task, participants indicated whether the test stimuli differed from the cue in six SNR values, ranging from −18 dB to 12 dB. The ending SNR was the level at which the participant was able to discriminate between the cue and the test stimuli. Thresholds for both tests were based on 42 trials presented in a single block and estimated using a psychometric function for the two-interval, forced choice (2IFC) testing paradigm described in Sheft et al. (2015).

2.3.5 Global hearing score

All hearing measures were standardized using z-scores based on means and standard deviations of the entire sample. In addition to standardizing measured hearing abilities, a global hearing score was also computed as the average of standardized scores on (1) better ear average of pure-tone thresholds for speech frequencies; (2) dB SNR loss on QuickSIN; (3) just-detectable change in radians for the starting phase of a spectral ripple in a signal filtered at 200–500 Hz; and (4) dB SNR for pattern discrimination in noise for a frequency modulated stimulus. The use of the global hearing score was motivated by the vastly different hearing tasks utilized to encapsulate these various measures into a single metric.

2.4 Covariates

Polygenic risk scores (PRS) were constructed using PRSice2 across different p-value thresholds for including a particular single-nucleotide polymorphism (SNP) in the PRS. PRS for AD was based on data from the International Genomics of Alzheimer's Project (Bellenguez et al., 2022), whereas PRS for HL was computed based on the genome-wide association study (GWAS) of hearing difficulties in the United Kingdom (UK) Biobank (Cherny et al., 2020; Abidin et al., 2021). In short, the candidate SNPs from GWAS are used to identify risk variants in the genotype samples and the association between SNPs predicting the target phenotype in the target sample is evaluated using linear models and a p-value for the association is obtained. PRS computed from SNPs that met two p-value thresholds of 1 and 5 × 10⁻⁸ were used in separate linear mixed models. However, only model outputs with p-value threshold of 1 are reported. We chose this threshold because it is more inclusive of regions of interest in the genomic code than stricter p-value thresholds. While p = 1 has previously shown to explain the highest proportion of variance in the self-reported hearing data within the TwinsUK dataset using the UK Biobank GWAS (Cherny et al., 2020), using a stricter p-value threshold did not change the relationship between hearing and cognitive outcomes (Supplementary Table 3). Details related to genotyping can be found elsewhere (Tasaki et al., 2018).

Additional covariates included vascular disease risk, age, sex, education, and race. A composite measure of vascular disease risk was used based on self-reported history of hypertension, diabetes, and smoking, where responses ranged from 0 to 3, with three indicating a higher risk of vascular disease. Age was calculated by subtracting the date of birth from the baseline assessment date and dividing by 365.25. Sex was a self-reported response to male/female. Years of education were based on self-reported numbers of years of school at baseline. Race was self-reported as follows: (a) White, (b) Black or African American, (c) American Indian or Alaska Native, (d) Native Hawaiian or Other Pacific Islander, (e) Asian, (f) Other, and (g) Unknown. When racial/ethnic groups were compared, only non-Latino Whites and non-Latino Blacks were considered.

2.5 Statistical analysis

Separate linear mixed models were constructed to examine associations of self-reported and measured hearing abilities with global cognition. For all models, time was modeled as years since the analytic baseline, and at least two cognitive assessments at or after the analytic baseline were required. The analytic baseline was defined as the first visit with available data for the relevant hearing exposure and global cognition outcome; for the measured-hearing subset, this corresponded to the visit at which objective auditory testing was performed. Models included random intercepts and random slopes for time, allowing participants to vary in cognitive level at the analytic baseline and in the rate of cognitive change during follow-up. Hearing-by-time interaction terms were used to test whether hearing ability was associated with the rate of subsequent cognitive change.

Models were adjusted for age, sex, education, vascular disease risk, and relevant interactions with time. For the global measured-hearing analysis, we first fit a model adjusted for demographic and vascular covariates and then additionally adjusted for PRS for AD and HL. For the analyses of self-reported hearing and individually measured hearing tasks, the final adjusted models included the covariates specified in the corresponding tables. All fixed effects were entered simultaneously within each fitted model.

Missing data were handled using available-case analysis without imputation. Linear mixed-effects models included all available person-visit observations with non-missing outcome, exposure, and covariate data for the corresponding model. In the primary longitudinal analyses, global cognition and self-reported difficulty understanding speech on the phone had no missing observations, while self-reported hearing trouble in a quiet environment had 37 missing observations among 24,930 person-visit observations (0.15%). For the measured-hearing subset, analyses included participants with valid data for the relevant auditory measure and cognitive outcome.

Spearman's correlational analyses were performed to examine the relationship between PRS for AD and age-related HL. For analyses other than determining inclusion of SNPs in PRS scores, statistical significance was determined at p < 0.05. All statistical analyses were performed using SAS (Version 9.4, SAS Institute Inc., Cary, NC, USA) and R version 4.4.3 (https://www.R-project.org/).

3 Results

3.1 Participant characteristics

Of the 5,247 adults who had completed baseline cognitive assessment in one of the five cohorts of the RADC, 3,506 had self-reported hearing data as well as global cognition as of when these data were extracted in October 2023. Of these, 274 were excluded due to only one follow-up, leaving 3,232 participants in the primary analysis (Table 1). The participants' race was coded as White, Black/African American, Asian, and Other Races, with proportions of 55%, 34%, 10% and 0.6%, respectively. The average follow-up for this group was 6.9 ± 4.6 years. In a subset (n = 113), additional measures of auditory abilities were collected at a single time point as described in Sheft et al. (2015) with mean follow-up time of 8.6 ± 2.3 years. This group included similar proportions of Black/African American or Other Races (n = 50; 44%) and White/Caucasians (n = 58; 51%), and was predominantly female (n = 99; 88%), with a mean age of 76 ± 6 years and mean education of 15.4 ± 3.17 years, respectively (Table 2).

Table 1. Basic characteristics of the participants in the original cohort with self-reported hearing data.

Table 2. Basic characteristics of the participants in the subset with measured hearing data.

3.2 Self-reported hearing and cognitive outcomes

The output of the mixed-effects modeling of self-reported hearing and global cognition with the full set of covariates showed that after controlling for genetic predisposition, there was no statistically significant association between self-reported/perceived hearing and cognitive ability. This was true for both models with subjective difficulty in quiet (β = −0.067, p = 0.107) and subjective difficulty on the telephone (β = −0.019, p = 0.481) (Supplementary Table 1).

3.3 Measured hearing and cognitive outcomes

Global hearing, as a composite score of standardized scores on all four auditory tests, was measured at the analytic baseline. In this model, global hearing was associated with global cognition at the analytic baseline (β = −0.243, p < 0.001). However, the global hearing-by-time interaction was not significant (β = −0.012, p = 0.535), indicating no evidence that baseline global hearing was associated with the rate of subsequent cognitive change during the cognitive follow-up (Table 3; Supplementary Figures 1, 2). Additional linear mixed models (see Supplementary Table 2) were evaluated with individual auditory tests as primary exposures and personal, demographic, and full set of covariates (age, sex, education, genetic risk for HL and AD, vascular disease risk, and years in study). Race was not included in these models due to the disproportionate racial/ethnic makeup and the relatively small size. Therefore, findings from the measured hearing analyses should be interpreted with caution and may not fully generalize to the larger cohorts or to more racially diverse older adult populations.

Table 3. Coefficients for linear mixed effects regression for global hearing and global cognition with and without AD and HL genetic predisposition.

Model 1 is based on 990 observations from 113 participants, whereas Model 2 is based on 907 observations from 103 participants.

^aTime represents time in years since baseline.

^bAge represents the mean-adjusted age at analytical baseline at which first hearing measure was available.

^cEducation is based on years of school reported at baseline.

^dVascular Disease Risk represents a sum of responses to self-reported hypertension, diabetes and smoking history.

^ePolygenic Risk for AD was based on the data from the International Genomics of Alzheimer's Project (Bellenguez et al., 2022).

^fPolygenic Risk for ARHL was derived from a genomic-wide association study of hearing difficulties in the UK Biobank (Cherny et al., 2020; Abidin et al., 2021).

^gHearingGlobal represents the composite of standardized scores on all 4 hearing measures (pure tone average (B4PTA), QuickSIN, SpectralRipple, and FMinNoise).

Bolded values indicate statistical significance with p < 0.05

After controlling for genetic risk and other covariates, hearing sensitivity measured using pure-tone audiometry at baseline was not associated with global cognition (β = −0.008, p = 0.056). Additionally, Quick-SIN at baseline was not associated with global cognition at baseline (β = −0.019, p = 0.052) nor with the rate of subsequent cognitive change (β = −0.003, p = 0.339). On the other hand, the two measures of complex auditory abilities (FM pattern discrimination in noise and spectral ripple discrimination) were significantly associated with global cognition at the analytic baseline (SpectralRipple: β = −0.257, p < 0.001; FMinNoise: β = −0.024, p < 0.001; Supplementary Table 2). However, their interactions with time were not significant, indicating no evidence that these baseline auditory measures were associated with the rate of subsequent cognitive change (SpectralRipple: β = 0.002, p = 0.946; FMinNoise: β = −0.001, p = 0.583; Supplementary Figure 1).

4 Discussion

The current study investigated the relationship between self-reported and measured hearing abilities on simple and complex auditory tasks and longitudinal cognitive outcomes after controlling for genetic risk factors for AD and HL and other covariates. In a large sample of older adults with self-reported hearing abilities (n = 3,232) and follow-up cognitive testing, there was no association between self-reported hearing abilities in quiet or on the telephone, after controlling for genetic predisposition for AD and HL and other covariates. While self-reported hearing difficulty has been shown to have a positive predictive value of 0.89–0.96 for detecting hearing loss at 4,000, 6,000, and 8,000 Hz—frequencies which are affected first by age-related HL (Hannula et al., 2011), self-report is thought to underestimate hearing abilities in older adults with cognitive decline (Fu et al., 2022; Levett et al., 2025). This discrepancy between self-report and measured hearing ability in older adults with cognitive decline likely explains our non-significant findings; however, other explanations such as slow and gradual onset of age-related HL, the potential use of coping or compensatory strategies utilized by older adults, and stigma/denial related to hearing loss and ageism, all of which contribute to the discrepant self-reports by older adults, cannot be ruled out (Pichora-Fuller et al., 2015). Alternatively, the questions used in this study to evaluate self-report hearing abilities were not sensitive to the difficulty experienced by older adults. Future studies using self-report measures of hearing should consider complementing subjective reports with objective measures such as those implemented in this study.

Using objective hearing measures collected at a single analytic baseline visit in a subset of the participants, we found that global hearing ability was associated with cognitive level at that visit but not with subsequent cognitive change during follow-up. The beta coefficient for global hearing was −0.243 after controlling for standard covariates as well as genetic covariates. Loughrey et al. (2018) performed a meta-analysis of nine cohort studies and found a significant negative association between hearing loss as measured by pure-tone audiometry and global cognition (β = −0.14 [−0.19−0.09]). The estimates from the current study are slightly higher than previously reported estimates likely because the assessment extended beyond the traditional gold standard pure-tone audiometry to include central auditory processing tasks. Additionally, genetic predispositions to both HL and AD were included as covariates, which may have improved model predictions. Neverthless, future studies should examine the mechanisms by which genetic risk for AD interacts with risk for HL to explain performance on central auditory processing tests and association with cognitive outcomes. Given the limited sample size and scope of the current study, only correlations were observed - causal genetic links and confounders were not examined, but would be worthwhile future directions using epidemiological approaches such as Mendelian randomization (Jiang et al., 2024).

Currently, there is controversy regarding the use of behavioral tests of auditory processing in older adults with mild cognitive impairment (MCI) and AD because neuropathology may cause underlying central auditory dysfunction (Merten et al., 2026; Humes, 2005). However, proponents of including behavioral tests of auditory function, such as central auditory processing tests, argue that these tests increase the sensitivity for detecting auditory-cognitive interactions based on reports of poorer central auditory function even in adults with subjective memory complaints who are not yet clinically diagnosed with MCI or AD (de Azevedo Picinini et al., 2025; Mohammed et al., 2022; Jayakody et al., 2020; Tuwaig et al., 2017; Gates et al., 2008). In parallel, others have suggested that the sensitivity of central auditory processing tests may differ by the type of dementia. For example, Häggström et al. (2020) reported differences between dichotic digits test in adults with MCI that converted to other types of dementia, such as frontotemporal dementia but not AD. In the current study, it was not possible to determine the differential sensitivity of our spectral pattern discrimination ability on dementia types due to the small sample size and low conversion rate to AD and MCI, which was 12.4% (n = 1) and 27.5% (n = 28), respectively.

The current study suggests that spectral ripple starting phase discrimination might be more sensitive to global cognition, given the larger effect size compared to frequency-modulated signal detection in noise. However, it should be noted that even an effect size of 0.257 is considered small (Nieminen, 2022) and thus, clinical significance of this test as a standalone measure may be limited at this time. Previous studies of central auditory processing abilities and cognitive function in older adults (see review by Tarawneh et al., 2022) have largely considered speech perception in noise, which involves multiple cognitive processes including linguistic abilities, pattern analysis, and working memory (Griffiths et al., 2020). Sheft et al. (2012, 2015) proposed non-speech dependent auditory processing tasks to examine interactions between hearing and cognitive ability. Using these non-speech dependent tasks, we found a significant association between spectrotemporal pattern discrimination ability and global cognition that persisted even after controlling for genetic predisposition to AD and HL. While the current findings did not find a longitudinal relationship between complex hearing abilities and cognition, likely due to the small sample size, findings suggest the possibility of using pattern discrimination tests as proxies for assessing cognition in older adults. Specifically, static and dynamic pattern discrimination tests can be added to the list of possible screening tools to be evaluated at earlier time points in adulthood, particularly for detecting neural changes in the medial temporal lobe (Griffiths et al., 2020; Khan et al., 2014).

4.1 Limitations

While our study has many strengths, including multiple measures of hearing spanning from simple pure-tone detection to speech-in-noise ability, an intensive cognitive test battery measured longitudinally, and a large group of self-reported hearing data combined with genetics, there are also a few limitations. One limitation of our work is that only behavioral measures of hearing were assessed. Incorporating neuro/electrophysiological measures in future work may address the limitations of the current literature, which largely focuses on behavioral measures of hearing. Another limitation of this study is the inherent selection bias for all participants returning for follow-up testing in that the participants agreed to donate or consider donating for brain autopsy.

Additionally, many participants in the original cohort were evaluated only once, leading to many missing follow-ups and thereby introducing another source of selection bias. Additionally, many of our covariates (e.g., vascular disease risk) were based on self-report rather than clinical data. Future studies could incorporate more standardized, clinical measures as well as additional common-cause factors including diet, which has recently been shown to reduce the risk of hearing loss and its association with longitudinal cognitive outcomes in older adults (Liu et al., 2025). Lastly, the polygenic risk scores for HL in the current study were based on the UK Biobank, which is largely composed of data from participants of Northern European (White/British) descent. PRS in the current study may be less predictive of global hearing in non-European than in European participants as previously reported by Cherny et al. (2020). Thus, future studies with larger, more diverse biobanks are needed. Moreover, because race was not included in the measured-hearing models due to the small and racially biased sample with over-representation of Black/African American group and under-representation of Asian or other minority racial groups, the findings from these analyses may not fully generalize to the general U.S. population.

5 Conclusions

While self-reported hearing abilities were not associated with global cognition in the larger sample, complex auditory pattern discrimination tests that engage central auditory processing were associated with baseline global cognition, even after controlling for genetic risk for AD and HL and other covariates in a subset of the original sample. However, prediction of longitudinal cognitive outcomes was likely hindered by the small sample size. The current findings highlight the need for future large-scale studies using peripheral, central and neurophysiological measures of auditory function to better understand the association and directionality between hearing loss and cognitive decline/dementia.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fauot.2026.1872446/full#supplementary-material

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Footnotes

Edited by: Justin Yao, Rutgers, The State University of New Jersey - Busch Campus, United States

Reviewed by: David Toloza-Ramirez, Andres Bello University, Chile

Shraddha Shende, Illinois State University, United States