fNIRS Studies of Individuals with Speech and Language Impairment Underreport Sociodemographics: A Systematic Review

Abstract

Introduction:

Functional near-infrared spectroscopy (fNIRS) is a promising tool for scientific discovery and clinical application. However, its utility depends upon replicable reporting. We evaluate reporting of sociodemographics in fNIRS studies of speech and language impairment and asked: (1) Do refereed fNIRS publications report participant sociodemographics? (2) For what reasons are participants excluded from analysis?

Method:

This systematic review was preregistered with PROSPERO (CRD42022342959) and followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses protocol. Searches in August 2022 included the terms: (a) fNIRS or functional near-infrared spectroscopy or NIRS or near-infrared spectroscopy; (b) speech or language, and (c) disorder or impairment or delay.

Results:

Searches yielded 38 qualifying studies from 1997 to present. Eight studies (5%) reported at least partial information on race or ethnicity. Few studies reported SES (26%) or language background (47%). Most studies reported geographic location (100%) and gender/sex (89%).

Discussion:

Underreporting of sociodemographics in fNIRS studies of speech and language impairment hinders the generalizability of findings. Replicable reporting is imperative for advancing the utility of fNIRS.

Language impairment is common, with nearly 8% (one in 12) children ages 3 to 17 years in the United States in 2012 having a communication disorder (Black et al., 2015). Of this 8%, speech problems (5%) and language problems (3.3%) were the most prevalent types of communication disorder, and over one-third of children with communication disorders had co-occurring communication disorders (34%; Black et al., 2015). In recent years, the prevalence estimates of common specific communication disorders, developmental language disorder and autism, are one in 14 and one in 36 (Maenner et al., 2023; National Institute on Deafness and Communication Disorders [NIDCD], 2023). These disorders are characterized by wide heterogeneity in language and communicative outcomes (Norbury et al., 2016; Schaeffer et al., 2023), with unknown etiology (NIDCD, 2023). Behavioral assessment fails to capture abilities in all individuals with communication disorders and can yield inconsistent outcomes (e.g., Plesa Skwerer et al., 2016). Hence, our efforts to understand the underlying mechanisms of language, provide accurate diagnosis, and develop supports are stymied (Butler et al., 2023). Studying the functional neural correlates of language in children with language impairment as a more direct probe relative to behavioral assessment can advance these efforts.

Functional near-infrared spectroscopy (fNIRS) offers a way to sensitively assess neural responses in individuals with speech and language impairment who show little variability on behavioral assessments (Butler et al., 2020; Pinti et al., 2020). fNIRS is a noninvasive, portable neuroimaging tool that uses the absorption of near-infrared light to measure hemodynamic oxyhemoglobin and deoxyhemoglobin concentrations in the cortex as a proxy for direct neural responses, similar to the BOLD signal of functional magnetic resonance imaging (fMRI; Ferrari & Quaresima, 2012; Jöbsis, 1977; Scholkmann et al., 2014). The cap and optodes of fNIRS are similar to the cap and electrodes of electroencephalography (EEG; Pinti et al., 2020). Though fNIRS does not use gel like many EEG systems, a good fNIRS signal requires adequate contact between optodes and the scalp (Parker & Ricard, 2022).

fNIRS has several advantages for assessment of neural responses in individuals with speech and language impairment. First, fNIRS is more robust to head movement than EEG or MRI (Crosson et al., 2010), making it more accessible to individuals who cannot remain motionless, including those with speech and language impairments (Cosgrove et al., 2022; Pinti et al., 2020). Second, fNIRS does not require individuals to be constrained inside an enclosed space and is portable, allowing for inclusion of participants with disabilities who cannot be measured in a scanner environment or who are unable to travel to a lab (Pinti et al., 2020). MRI bores are small and noisy, which present challenges to individuals with sensory issues such as noise sensitivity or claustrophobia (Cosgrove et al., 2022). Further, MRI bores cannot accommodate individuals with large, afro-textured hair or hair extensions with metal, excluding some Black participants (Parker & Ricard, 2022; Webb et al., 2022). These factors contribute to eagerness about use of fNIRS with individuals with speech and language impairment.

However, the utility of fNIRS is limited because it is insufficiently inclusive for all racial groups. As with EEG (Choy et al., 2021), current fNIRS systems cannot achieve adequate contact between optodes and the scalp for individuals with coarse or curly hair (Webb et al., 2022; Yücel, 2023). Though inclusive EEG electrodes exist (Etienne et al., 2020), to our knowledge, no such optodes for fNIRS exist. Further, fNIRS may not work as well on individuals with higher levels of skin pigmentation, because light passes differently through darker skin (Webb et al., 2022). Preliminary evidence from 69 primary-school children with shaved heads from the Ivory Coast suggests fNIRS may achieve adequate signal in Black neurotypical individuals with high levels of skin pigmentation (Jasińska et al., 2023). Yet, other studies from thousands of individuals show light (including that used in fNIRS systems), does not pass as well through the skin of individuals with darker skin (Fawzy et al., 2022; Gottlieb et al., 2022; Henry et al., 2022; Wong et al., 2021; Yücel, 2023). These factors may serve to exclude BIPOC from fNIRS studies and limit the generalizability of findings (Girolamo et al., 2022b).

An important step to address potential exclusion from fNIRS studies of speech and language impairment is to properly report race and exclusion criteria. A systematic review by Butler et al. (2020) examined studies using fNIRS to understand the neural correlates of speech and language in individuals with or at a high likelihood of having speech and language impairment. While valuable, that review did not consider participant sociodemographics, a crucial aspect of replicable and ethical research (Sabik et al., 2021). Performance on speech and language tasks varies by language community, which can coincide with race and ethnicity. For example, Black individuals who use African American English may use language forms specific to that variant of English (Washington et al., 2018). Together with other sociodemographic variables (e.g., gender/sex, SES; Cosgrove et al., 2022), examining race and ethnicity is key to understanding the broad applicability of fNIRS. The present systematic review asked: Do refereed fNIRS publications examining speech and language impairment report participant sociodemographics and the reasons for exclusion of participants?

Intersection of Race & Disability in Neuroscience

To frame sociodemographics in fNIRS studies of speech and language impairment, we look to Dis/ability Studies and Critical Race Theory (DisCrit), a theory from special education that is relevant to clinical research (Annamma et al., 2013). DisCrit posits that dis/ability and race are social constructs primarily involving the reaction of others to individual differences and not individual differences themselves; these constructs can reinforce one another and exacerbate experiences of marginalization (Annamma et al., 2013). As social constructs, race and dis/ability have no fixed definition but change depending on sociocultural context. As Gilpin and Taffe (2021) note, the 1790 U.S. Census listed “slave” as a racial/ethnic category. DisCrit draws from intersectionality theory, which posits that BIPOC may have multiple identities each associated with experiences of marginalization that intersect with one another and give rise to nuanced marginalization (Crenshaw, 1989, 1991). Thus, experiences of marginalization associated with individual identities are not simply additive; rather, they depend on how identities function within a social-structural context (Bauer, 2014). For example, while 62% of states underrepresented Black students in the primary disability category of speech and language impairment, 14% overrepresented them (Robinson & Norton, 2019). A Black student may not have a disability but be misinterpreted as having one due to inadequate cultural sensitivity (Heath, 1983). Conversely, a student may have language impairment but be misinterpreted as having typical language due to inadequate linguistic sensitivity (Oetting et al., 2019). Each type of disproportionate representation arises from being racialized as Black and perceptions about dis/ability, but the exact experiences of marginalization differ.

In human neuroscience, DisCrit and intersectionality point to potential systematic exclusion of BIPOC with speech and language impairment. Even without intent to exclude, research engagement methods and fNIRS methods may be insufficiently inclusive of BIPOC with speech and language impairment (Girolamo et al., 2022b). Data from fMRI and EEG studies support this possibility. Though a comprehensive review is beyond the scope of this report, we provide two examples. In the ABCD study, adolescents with low noise resting state fMRI data were more likely to be female, white and non-Hispanic, of higher SES, and to have higher neurocognitive skills and lower levels of neurodevelopmental disorder traits than participants with high noise data (Cosgrove et al., 2022). Thus, fMRI may fail to equitably obtain usable data by race, gender/sex, SES, and dis/ability. In turn, Choy et al. (2021) reviewed EEG studies in neurotypical minority adults. There, just 6% (5 of 81) of studies reported including Black participants but did not specify whether they excluded their data; only 5% (4 of 81) cited literature on engaging BIPOC in research (Choy et al., 2021). Together with underreporting of race in EEG and MRI studies (Goldfarb & Brown, 2022), these findings motivate an evaluation of sociodemographics as reported in fNIRS studies of speech and language impairment.

Reporting Sociodemographic Variables

One step in ensuring fNIRS studies of speech and language impairment are generalizable is using reporting practices that reflect rich within-group heterogeneity versus reifying white as a neutral norm and interpreting findings from white participants from Western, educated, industrialized, rich, and democratic (WEIRD) countries as broadly generalizable (Heinrich et al., 2010; Roberts & Mortenson, 2022). For example, allowing participants to write in their race and ethnicity as Korean or Naxi, rather than requiring them to select “Asian” (which refers to an entire continent), is more inclusive and specific (Buchanan et al., 2021). Such a perspective aligns to both DisCrit (Annamma et al., 2013) and Diversity Science, which recognizes interindividual differences as reflections of sociocultural realities (Plaut, 2010). It is unclear to what extent common reporting guidelines in neuroscience embody these perspectives.

Reporting Guidelines: APA, and AMA, and fNIRS Studies per Yücel et al. (2021)

Guidance for reporting and evaluating participant sociodemographics in fNIRS studies of speech and language impairment come from the American Psychological Association (APA; 2020, 2021) and American Medical Association (AMA; Flanagin et al., 2021; JAMA Network Editors, 2020). Specific guidance for reporting fNIRS studies comes from the Society for Functional Near-Infrared Spectroscopy (Yücel et al., 2021). As is common practice in reporting guidelines for various study types (Girolamo et al., 2022a), these guidance sources instruct authors to report participant flow but not to provide reasons for attrition. These guidelines also show inconsistency in what and how to report sociodemographic variables; see Table 1. These variables of race/ethnicity, gender/sex, SES, and language come from prior work documenting the relevance of these factors in neuroimaging studies and speech and language research.

Table 1.

Reporting Guidelines for Sociodemographics

Topic	American Medical Association	American Psychological Association	Best practices for fNIRS studies
variables to report	race and ethnicity; age; sex/gender; SES using specific categories & reporting all categories (e.g., not “white”, “male” as default)	specific information on age, disability status, sex, gender identity, racial and ethnic identity, SES, intersectionality of these attributes	mean age & variation, inclusion and exclusion criteria (e.g., pathologies, native language), gender distribution, other relevant features: handedness, ethnicity, SES
where to report	Methods: demographic measures used, who identified demographics and how (e.g., self-report), reason for collecting race/ethnicity (e.g., funding requirements); Results: demographics	Methods: sample demographics, justification of samples & overview of sample inclusion efforts; Results: participant flow; Discussion: constraints on generalizability (e.g., sampling validity, ecological validity)	not specified
race and ethnicity	race: group of people connected by common descent or origin; ethnicity: group membership regarded as of common descent, or having a common national or cultural tradition	race: social construction & categorization of people based on perceived shared physical traits that result in the maintenance of a sociopolitical hierarchy; ethnicity: a particular type of culture (e.g., language) related to common ancestry & shared history	not defined
gender and sex	gender: cultural indicator of a person’s personal & social identity; sex: biological characteristics of males and females; report all categories	gender: attitudes, feelings & behaviors that a culture associates with a person’s biological sex; gender identity: person’s psychological sense of their gender sex: biological sex assignment; report all categories	not defined
SES	not defined	encompasses quality of life attributes & opportunities afforded to people within society: income, educational attainment, occupational prestige & perceptions of social status & class	not defined

Note. American Medical Association guidance from Jama Network Editors (2020) and Flanagin et al. (2021). American Psychological Association (APA) guidance from APA (2020, 2021). Best practices for fNIRS studies from Yücel et al. (2021). SES = socioeconomic status.

Race and Ethnicity.

The AMA (Flanagin et al., 2021) and APA (2020, 2021) define race and ethnicity as distinct social constructs arising from shared culture or history, with each underlining the importance of providing specific information (e.g., on race category versus “multiracial”). While Yücel et al. (2021) do not mention race or define ethnicity, they suggest reporting ethnicity, as the ethnicity of participants can differ from the ethnicity of the population at data collection sites.

Gender and Sex.

The AMA (Flanagin et al., 2021; JAMA Network Editors, 2020) and APA (2020, 2021), but not fNIRS reporting guidelines (Yücel et al., 2021), define gender and sex. The APA (2020, 2021) additionally defines gender identity. Following these guidelines, sex involves biological assignment at birth, while gender and gender identity involve how a person perceives themselves and how society perceives an individual (American Psychological Association, 2020, 2021; Flanagin et al., 2021; JAMA Network Editors, 2020). Thus, study reporting should treat sex, gender, and gender identity as distinct constructs and specify what they report. Only reporting males or reporting males and females without specification of gender or sex minimize these important sociodemographic dimensions.

Socioeconomic Status.

The APA and AMA recommendations offer inconsistent guidance on SES. The AMA (Flanagin et al., 2021; JAMA Network Editors, 2020) and Yücel et al. (2021) suggest reporting SES but provide no specific guidance; the AMA also advocates the use of “person first” language (e.g., “people of low resource backgrounds” versus “the poor”). In contrast, the APA (2020, 2021) defines SES as a multivariate construct comprised of indicators such as income, educational attainment, occupational prestige, and subjective perceptions of social status and class. Assessing SES with one indicator is insufficient; for example, a doctoral student may have a low income but high educational attainment and access to family wealth and occupational opportunities. Amid varying guidance, adopting the multivariate definition of the APA (2020, 2021) may provide the most insight on SES.

Geographic Location.

No guidelines detail how authors should report where data collection takes place. The APA (2020, 2021) recommends that authors describe the setting (i.e., a university lab) and location (i.e., a specific city) of data collection. In conducting research with BIPOC from clinical populations, especially if they are from a population of low incidence, authors must balance reporting information relevant for reproducibility and generalizability with protecting participant privacy and confidentiality (Girolamo et al., 2022a). For example, it could be possible to identify an individual person based on specific demographics, geographic and sample characteristics. In either case, reporting the geographic location of data collection sites or explaining why authors elect not to report this information should be the norm.

Languages Spoken.

Reporting guidelines for fNIRS studies include reporting participant language(s) as part of the study selection criteria (Yücel et al., 2021). Other reporting guidelines do not specify reporting language(s) spoken; however, they do indicate reporting relevant characteristics alongside basic sociodemographic variables. Thus, language background or language of research activities would be a relevant characteristic to report in studies of speech and language impairment. As with other sociodemographic variables, however, reporting language in studies of speech and language requires specificity. For instance, reporting that participants speak Chinese is insufficient for understanding characteristics. Chinese languages and variants of Chinese languages, such as Putonghua (standardized Mandarin) and Cantonese, can differ enough to be mutually unintelligible or to require different language norms.

Summary

Overall, the methodological limitations of neurotechnology and evidence of exclusion of BIPOC, including those from clinical populations, in EEG and fMRI studies, suggest there may also be exclusion in fNIRS studies of speech and language impairment. Though reporting guidelines relevant to fNIRS studies appear to partly align to relevant theories pertaining to sociodemographic diversity (Annamma et al., 2013; Plaut, 2010), inconsistency in these guidelines may serve to limit the generalizability of studies.

The Current Study

Given what is known about the limitations of fNIRS and underrepresentation in human neuroscience, the present systematic review extends the findings of Butler et al. (2020) and examined reporting of participant sociodemographics in fNIRS studies of speech and language impairment. Research questions were:

1. To what extent do refereed fNIRS publications examining speech and language impairment report participant sociodemographics including race and ethnicity, gender and sex, socioeconomic status, geographic location, languages spoken?

2. Why do fNIRS studies examining speech and language impairment exclude participants, and how is this exclusion documented?

METHOD

This systematic review included empirical studies using fNIRS and a speech or language task for individuals with or at elevated likelihood of having speech and language impairment. The review protocol was preregistered with PROSPERO (CRD42022342959) and followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses protocol (Page et al., 2021). As an extension of a previous systematic review, the methods in this study were similar to those of Butler et al. (2020), which evaluated empirical studies using fNIRS to investigate the neural basis of speech and language disorders. Key differences were the literature included and the focus of this report on participant sociodemographic reporting and inclusion. This review included all articles from Butler et al. (2020) save Chaudhary et al. (2017), which was retracted (The PLOS Biology Editors, 2019). It also included articles published after Butler et al. (2020): Gilmore et al. (2021), Marks et al. (2022), Mushtaq et al. (2020), and Pecukonis et al. (2021).

Search Procedures

Prior to conducting database searches, the first author searched for other published reviews and protocols in these databases in August 2022: Cochrane Database of Systematic Reviews, Campbell Systematic Reviews, JBI Evidence Synthesis, and PsycINFO; this search did not yield addition relevant reviews or protocols and ensured that the present review was not redundant with other reviews or literature search protocols. Next, the authors conducted a database search for relevant reports (versus reviews or protocols), also in August 2022: MEDLINE via the PubMed interface and PsycINFO. Searches included terms related to three essential concepts: (a) fNIRS or functional near-infrared spectroscopy or NIRS or near-infrared spectroscopy, (b) speech or language, and (c) disorder or impairment or delay. In addition, the term “fNIRS or functional near-infrared spectroscopy or NIRS or near-infrared spectroscopy” was employed to search several journals: Journal of Speech, Language, and Hearing Research; American Journal of Speech-Language Pathology; and Language, Speech, and Hearing Services in Schools. The search strategy also included hand-searching citation lists in these results for additional qualifying studies. Search results were uploaded into Covidence (Veritas Health Innovation, 2021), which automatically removed duplicates and housed all template data collection forms, extracted data, and data used for all analyses. The first, second, and third authors screened titles and abstracts, followed by complete texts for eligibility. The authors were blind to others’ decisions, discussing any disagreements until they reached consensus.

Selection Criteria

This review process included empirical studies published in refereed journals in English with no restriction on year of publication. Inclusion criteria consisted of studies of individuals with speech or language disorders; infants who have older siblings with autism spectrum disorder (ASD), as they have an elevated likelihood of ASD and language impairment (Butler et al., 2020); and mild cognitive impairment (MCI), because it is an early indicator of dementia. This review also included studies that used speaking or listening tasks, with a minimum of three or more participants. It excluded single participant studies using fNIRS, many of which focus on brain and computer interfaces for individuals with locked-in syndrome and bedside monitoring, which are beyond the scope of this review.

Exclusion Criteria

Given our focus on speech and language impairment, we excluded studies beyond the scope of this report. First, this review excluded studies that did not involve a speaking or listening task, such as nonverbal working memory, motor, visual, executive function, and attention tasks that did not have a communicative component. Second, this review also excluded studies of autistic traits in neurotypical individuals from the general population, because they do not involve speech or language impairment or a communication disorder (Sasson & Bottema-Beutel, 2022). Third, this review excluded etiologies that do not have speech and language or communication as a core characteristic, such as attention-deficit/hyperactivity disorder, bipolar disorder, depressive disorder, public speaking anxiety, social anxiety disorder, and schizophrenia (American Speech-Language-Hearing Association, 2016).

Risk of Bias Analysis

This review appraised the risk of bias of qualifying studies. Cochrane (2022) has a risk of bias tool for observational studies, which is for the quality assessment of diagnostic studies (QUADAS-II; Whiting et al., 2011). However, its applicability is not immediately obvious to the present study. Therefore, the authors considered completeness of reporting in terms of sociodemographics (Viswanathan et al., 2012); see criteria (a) to (e). These dimensions contribute to the interpretation of the findings (i.e., internal validity of the study) and to the generalizability of the findings (i.e., external validity; Washington et al., 2018). The authors made a quality rating of high or low for each study for each of the following participant criteria:

1. Race and ethnicity: studies providing information on race and ethnicity received a high-quality rating;

2. Gender and sex: studies providing information on participant gender and sex received a high-quality rating;

3. SES: studies providing information on participant SES received a high-quality rating;

4. Geographic location: studies providing information on participant geographic location (i.e., where data collection took place) received a high-quality rating;

5. Language background: studies providing information on the language(s) that participants used to complete research activities, including variants of a given language, received a high-quality rating.

In addition, the authors evaluated each study for outcomes reporting, which involved providing information about screening, attrition, and characteristics of the final sample:

1. Screening reporting: studies providing information on screening procedures and results received a high-quality rating;

2. Attrition reporting: studies providing information on and explanations for attrition received a high-quality rating;

3. Selective outcome reporting: studies providing information on all participant outcomes or stating that participants were not excluded for reasons unrelated to performance that could be attributed to selective reporting (e.g., excluding outliers that may have impacted results) received a high-quality rating.

Data Extraction and Synthesis

The authors analyzed studies in terms of reporting of sociodemographics and reasons for exclusion. The first, second, and third authors independently extracted and synthesized data in Covidence (Veritas Health Innovation, 2021), discussing disagreements until they reached consensus. Though prior work indicates that original study authors generally do not respond to requests for information (Gabelica et al., 2022), the authors contacted the corresponding author of articles which received a low-quality rating to ask for race and ethnicity. If the corresponding author did not initially respond, the authors emailed a second time. If the corresponding author no longer had a working email address, the authors used publicly available information (e.g., work website) to find current contact information or contacted a coauthor from a recent publication. The study characteristics in this review include: (a) age, (b) population, (c) sample size, (d) race and ethnicity, (e) gender and sex, (f) SES, (g) geographic location, (h) language(s) spoken or used by participants, and (i) reasons for exclusion; see Tables 2 to 7.

Table 2.

Risk of Bias Analysis of Sociodemographic Reporting in fNIRS Studies of Speech and Language Impairment

Reference	Race/Ethnicity	Gender/Sex	SES	Location	Lang	Attrition	Screen	Select	Total
	1	2	3	4	5	6	7	8
Autism Spectrum Disorder
Pecukonis et al. (2021)	low	high	high	high	high	high	low	low	23456
Yeung et al. (2019)	low	high	low	high	low	low	low	low	24
Edwards et al. (2017)	low	high	high	high	high	high	high	high	2345678
Lloyd-Fox et al. (2018)	low	high	low	high	low	low	low	high	248
Keehn et al. (2013)	low	high	low	high	low	low	low	high	248
Funabiki et al. (2012)	low	low	low	low	low	low	low	low
Iwanami et al. (2011)	low	low	low	high	high	low	low	low	45
Kawakubo et al. (2009)	low	high	low	high	low	high	low	high	2468
Minagawa-Kawai et al. (2009)	low	low	low	high	high	low	low	low	45
Kuwabara et al. (2006)	low	high	low	high	low	low	low	low	24
Developmental Speech and Language Disorders
Marks et al. (2022)	high	low	low	high	high	low	low	low	145
Cutini et al. (2016)	low	low	low	low	high	high	low	high	568
Fu et al. (2016)	low	low	low	low	high	low	low	high	58
Sela et al. (2014)	low	high	low	high	high	low	low	low	245
Song et al. (2014)	low	high	low	high	high	low	low	low	245
Deafness & Cochlear Implants
Mushtaq et al. (2020)	low	low	low	high	high	low	low	high	458
Anderson et al. (2017)	low	low	low	high	high	low	low	low	45
Bisconti et al. (2016)	low	high	low	low	high	high	low	high	2568
Olds et al. (2016)	low	low	low	low	high	low	low	low	5
Van de Rijt et al. (2016)	low	high	low	high	high	low	low	high	2458
Dewey & Hartley (2015)	low	low	low	high	high	low	low	high	458
Sevy et al. (2010)	low	low	low	low	high	high	high	high	5678
Dementia, Alzheimer’s Disease & Mild Cognitive Impairment
Katzorke et al. (2018)	low	high	low	high	high	low	high	high	24578
Yeung et al. (2016)	low	high	low	high	low	low	low	low	24
Metzger et al. (2015)	low	low	low	high	low	low	low	low	4
Arai et al. (2006)	low	high	low	low	low	low	low	low	2
Fallgatter et al. (1997)	low	high	low	low	low	low	low	high	28
Hock et al. (1997)	low	high	low	low	low	low	low	high	28
Locked-in Syndrome
Naito et al. (2007)	low	low	low	low	low	low	low	low
Neurologic Speech Disorders/Dysarthria
Caliandro et al. (2013)	low	high	low	high	high	low	low	high	2458
Stroke/Aphasia
Gilmore et al. (2021)	low	low	low	high	high	high	low	high	4568
Hara et al. (2017)	low	high	low	high	low	low	low	low	24
Sakatani et al. (1998)	low	low	low	low	low	low	low	low
Stuttering
Jackson et al. (2019)	low	low	low	high	high	low	low	high	458
Hosseini et al. (2018)	low	low	low	low	low	low	low	high	8
Walsh et al. (2017)	low	high	low	low	high	low	low	high	258
Sato et al. (2011)	low	low	low	high	high	low	low	low	45
Traumatic Brain Injury
Rodriguez Merzagora et al. (2014)	low	low	low	low	low	low	low	high	8

Note. SES = socioeconomic status. Location = geographic location of data collection. Lang = language background of participants. Screen = reported screening outcomes. Select = selective outcome reporting.

Table 7.

Geographic Location & Languages in fNIRS Studies of Speech and Language Impairment

Age in	Population	N	Language	Reference
United States
5-11	RI	24, 73 TD	English	Marks et al. (2022)
20-72	PWA	6, 41 TD	English*	Gilmore et al. (2021)
16-28 months	EL-ASD	14, 18 LL-ASD	English	Pecukonis et al. (2021)
PWS: 27 (8), TD: 26 (3)	PWS	15, 15 TD	English	Jackson et al. (2019)
8-16	CWS	14, 16 PWS, 16 TD	English**	Hosseini et al. (2018)
4-6 months	EL-ASD	5, 16 LL-ASD	English*	Lloyd-Fox et al. (2018)
3 months	EL-ASD	21, 17 LL-ASD	English	Edwards et al. (2017)
7-11	PWA	16, 16 TD	English	Walsh et al. (2017)
21-74	CI users	14, 13 NH	English/ASL & English	Bisconti et al. (2016)
8-12	SLI	15, 15 TD	English	Fu et al. (2016)
23-86	CI users	32, 35 NH	English	Olds et al. (2016)
32-52	TBI	6, 11 TD	English*	Rodriguez Merzagora et al. (2014)
3, 6, 9 & 12 months	EL-ASD	27, 3 LL-ASD	English	Keehn et al. (2013)
3-12 & 22-28	CI users	36, 22 NH	English	Sevy et al. (2010)
Japan
42-75	PWA	8	Japanese*	Hara et al. (2017)
10-22 & 20-32	ASD	11, 12 TD		Funabiki et al. (2012)
18-35	ASD	20, 18 TD	Japanese	Iwanami et al. (2011)
3-12 & 18-44	PWS	23	Japanese	Sato et al. (2011)
9-16 & adults	HFA, PDD-NOS	27, 54 TD		Kawakubo et al. (2009)
18-37	PDD	10, 10 TD	Japanese*	Kuwabara et al. (2006)
6-11	ASD	9, 9 TD	Japanese*	Minagawa-Kawai et al. (2009)
22-80	ALS-TLS	17 ALS-TLS, 23 ALS		Naito et al. (2007)
MCI: 63 (6.4), AD: 59.2 (3.9)	MCI, AD	15 MCI, 15 AD, 32 TD	Japanese*	Arai et al. (2006)
Germany
70-77	MCI	55, 55 TD	German	Katzorke et al. (2018)
50-90	probable AD	16		Metzger et al. (2015)
67.3 (10.6)	DAT	10, 10 TD	German	Fallgatter et al. (1997)
71 (10)	AD	19, 19 TD		Hock et al. (1997)
UK
6-12	CI users	19, 20 NH	English*	Mushtaq et al. (2020)
36-78	Deaf, CI users	15, 17 NH	English	Anderson et al. (2017)
M = 12	dyslexia	15, 15 TD	English	Cutini et al. (2016)
18-60	Deaf	30, 30 NH	English, BSL, Portuguese	Dewey & Hartley (2015)
China
Grades 3-5	dyslexia	20, 20 TD	Chinese: Putonghua	Song et al. (2014)
48-59	PWA	10, 6 poststroke nonaphasia, 13 TD		Sakatani et al. (1998)
Hong Kong
11-18	ASD	22	Chinese*	Yeung et al. (2019)
60-91	MCI	26, 26 TD	Chinese*	Yeung et al. (2016)
Israel
M=25.7 (2.7)	dyslexia	17, 34 TD	Hebrew*	Sela et al. (2014)
Italy
44.5 (14.2)	DM1	29, 30 TD	Italian*	Caliandro et al. (2013)
Netherlands
55-59	CI	5, 33 NH	Dutch	Van de Rijt et al. (2016)

Note. EL-ASD = elevated likelihood of autism spectrum disorder (ASD). LL-ASD = low likelihood of ASD. PWA = persons with aphasia. CI users = users of cochlear implants. NH = normal hearing. MCI = mild cognitive impairment. TD = typically developing. PWS = persons who stutter. CWS = children who recovered from stuttering. AD = Alzheimer’s dementia. PDD = pervasive developmental disorder. RI = reading impairment. SLI = specific language impairment. DM1 = myotonic dystrophy Type 1. HFA = high-functioning autism, PDD-NOS = PDD not otherwise specified. DAT = dementia of the Alzheimer type. AD = Alzheimer’s dementia. TBI = traumatic brain injury. ALS = amyotrophic lateral sclerosis. ALS-TLS = ALS total locked-in syndrome. BSL = British Sign Language. Yeung et al. (2016, 2019) reported use of Chinese language assessments but did not specify language (e.g., Mandarin, Cantonese). Fallgatter et al. (1997) and Dewey & Hartley (2015) offered information on participants’ language background in their response to a request for race/ethnicity information.

^*= provided information on language of assessments but not on participants’ language background.

^**= information was inferred, as a subset of participants came from another report with this information, Walsh et al. (2017).

RESULTS

Study Flowchart

The search yielded 319 qualifying records; see Figure 1. After removing 36 duplicates, screening 283 records by title and abstract led to exclusion of 230 records. Assessment of 53 full texts led to exclusion of 15 articles. The final review included 38 reports, with a total of 709 participants with and 736 participants without speech and language impairment.

Figure 1.

PRISMA flowchart (Page et al., 2021)

Risk of Bias Analysis: Reporting of Participant Sociodemographics

Overall, few studies consistently followed reporting guidelines; see Table 2. Per the total column of Table 2, which shows the criteria that received a high-quality rating, no study received high quality ratings for all criteria. One study (3%) received high quality ratings on seven of eight criteria (Edwards et al., 2017), and one study (3%) received a high-quality rating for reporting of race/ethnicity (Marks et al., 2022). No study used specific labels or reported data intersectionally (e.g., race by gender by SES). Few studies received high quality ratings for reporting of SES (n = 2, or 5%), screening outcomes (n = 3, or 8%), or attrition (n = 7, or 16%).

In contrast, studies were likely to receive high quality ratings for reporting of gender/sex (n = 19, or 50%), geographic location of data collection (n = 24, or 63%), language of study activities (n = 22, or 58%), and complete versus selective outcome reporting (n = 20, or 53%). In all, these reporting patterns demonstrate inconsistency with reporting guidelines, limiting generalizability of findings and replicability of methods. From a theoretical perspective, such reporting also serves to mask issues about inclusion in research, as underreporting works against transparency of potential systemic issues in research bias.

Reporting of Participant Sociodemographics

Race and Ethnicity

Most studies failed to report or make available information on participants race and ethnicity (n = 1196 of 1577 participants, or 76%); see Table 3. Two studies partly reported race and ethnicity (Marks et al., 2022; Pecukonis et al., 2021), and seven studies had authors who provided post hoc information on race and ethnicity upon request (n = 7, or 18%; Caliandro et al., 2013; Fallgatter et al., 1997; Iwanami et al., 2011; Lloyd-Fox et al., 2018; Keehn et al., 2013; Metzger et al., 2015; Minagawa-Kawai et al., 2009). In these nine studies, 280 of 383 (73%) of participants were white. In contrast, 15% of participants were Japanese, 9% were multiracial, 1% were other, and just 1% were Black. In addition, 2% of participants were Hispanic or Latino. One consideration is that race and ethnicity differ by location. Therefore, in studies not funded by NIH (which mandates using certain categories), these categories may not be meaningful. Yet, without complete reporting, it is impossible to know. Reporting of race and ethnicity showed minimal improvement over time; see Figure 1 in Supplemental Materials 1 (SM1).

Table 3.

Race and Ethnicity of Participants in fNIRS Studies of Speech and Language Impairment

	All Studies		Studies Reporting Race/Ethnicity
Category	N = 1579		n = 383
Race	n	%	n	%
Asian: Japanese	56	4%	56	15%
Black	4	0%	4	1%
multiracial	34	2%	34	9%
American Indian	0	0%	0	0%
white	280	18%	280	73%
Other	4	0%	4	1%
Missing	5	0%	5	1%
Not reported	1196	76%	0	0%
Ethnicity
Hispanic/Latinx	8	1%	8	2%

Note. Only some studies reported Hispanic/Latinx separately from race. Pecukonis et al. (2021) reported race as “white” (n = 28) and “some other race or more than one race,” (n = 4) under “Other.” Iwanami et al. (2011) and Minagawa-Kawai et al. (2009) responded all participants were Japanese. Lloyd-Fox et al. (2018) responded 13 participants were multiracial using the following terminology: white/white British + Black/Caribbean (n = 2), white/white British and Jewish (n = 2), white British and Arabic/Middle Eastern (n = 1), white/white British and Asian (n = 2), British and Chinese (n = 1), white Irish and Norwegian/US (n = 1), white mixed US/British (n = 2), mixed Asian (n = 1), mixed Eastern European and African (n = 1).

In studies with available information on race and ethnicity (n = 9), participants were mostly white (59%-100%); see Table 4. Reporting collapsed racial and ethnic variability. For example, Marks et al. (2022) described 20% of participants as multiracial or multiethnic with no further specification. Similarly, Pecukonis et al. (2021) reported participants as “white” and “some other race or more than one race,” which effectively reifies white as the norm. With the caveat that relevant racial and ethnic categories depend on data collection site and may differ from NIH categories, these data indicate that, when information is reported or available, white participants comprise most of the evidence base on fNIRS studies of speech and language impairment. In responding to queries, some authors shared that information on race and ethnicity was unavailable (n = 4, or 11%; Edwards et al., 2017; Hosseini et al., 2018; Jackson et al., 2019; Walsh et al., 2017) or that they did not collect this information (n = 6, or 16%; Anderson et al., 2017; Dewey & Hartley, 2015; Katzorke et al., 2018; Mushtaq et al., 2020; Olds et al., 2016; van de Rijt et al., 2016). All other authors (n = 19, or 50%) did not respond to information requests.

Table 4.

Participants by Race and Ethnicity in Studies with Available Information on Race and Ethnicity (n = 383)

Race														Ethnicity		Reference
Asian: Japanese		Black		multiracial		Indigenous		white		other		Missing		Hispanic/Latinx
n	%	n	%	n	%	n	%	n	%	n	%	n	%	n	%
0	0%	3	3%	19	20%	0	0%	76	78%	0	0%	0	0%	0	0%	Marks et al. (2022)
0	0%	2	2%	0	0%	2	2%	28	88%	4	6%	0	0%	2	6%	Pecukonis et al. (2021)
0	0%	0	0%	13	33%	0	0%	23	59%	0	0%	3	8%	0	0%	Lloyd-Fox et al. (2018)
0	0%	0	0%	0	0%	0	0%	16	100%	0	0%	0	0%	0	0%	Metzger et al. (2015)
0	0%	0	0%	0	0%	0	0%	59	100%	0	0%	0	0%	0	0%	Caliandro et al. (2013)
0	0%	1	2%	3	5%	0	0%	58	91%	0	0%	2	3%	6	9%	Keehn et al. (2013)
38	100%	0	0%	0	0%	0	0%	0	0%	0	0%	0	0%	0	0%	Iwanami et al. (2011)
18	100%	0	0%	0	0%	0	0%	0	0%	0	0%	0	0%	0	0%	Minagawa-Kawai et al. (2009)
0	0%	0	0%	0	0%	0	0%	20	100%	0	0%	0	0%	3	0%	Fallgatter et al. (1997)

Note. Marks et al. (2022) reported multiracial or multiethnic as “multiracial.” Pecukonis et al. (2021) reported race as “white” (n = 28) and “some other race or more than one race,” (n = 4) captured under “Other.” Two participants had missing race/ethnicity in Keehn et al. (2013). Iwanami et al. (2011) and Minagawa-Kawai et al. (2009) responded that all participants were Japanese. Lloyd-Fox et al. (2018) reported by email 13 participants were multiracial using the following terminology: white/white British + Black/Caribbean (n = 2), white/white British and Jewish (n = 2), white British and Arabic/Middle Eastern (n = 1), white/white British and Asian (n = 2), British and Chinese (n = 1), white Irish and Norwegian/US (n = 1), white mixed US/British (n = 2), mixed Asian (n = 1), mixed Eastern European and African (n = 1).

Gender and Sex

In studies reporting any information on gender or sex, 57% (n = 696 of 1231) of participants were male and 43% (n = 535 of 1231) were female; see Table 5. Information was unavailable for 22% of participants (n = 348 of 1579). Thirty-four studies (89%) reported at least some information on gender/sex. Of the four studies (11%) that did not report any information, one reported comparing the gender distribution of participants with and without a traumatic brain injury and found they did not differ (Rodriguez Merzagora et al., 2014). Another study reported having “40 male and female” participants but did not specify the proportions (Naito et al., 2007). Two studies did not specify whether they collected or considered gender or sex in data collection (Anderson et al., 2017; Sevy et al., 2010).

Table 5.

Gender/Sex in fNIRS Studies of Speech and Language Impairment

Age	Population	total N	Categories				Reference
			male		female
			n	%	n	%
Reported Gender
3 months	EL-ASD, LL-ASD	38	23	61%	15	39%	Edwards et al. (2017)
42-75	PWA	8	6	75%	2	25%	Hara et al. (2017)
6-12	CI users, NH	39	19	49%	7		Mushtaq et al. (2020)
60-91	MCI, TD	52	13	25%	39	63%	Yeung et al. (2016)
Reported Sex
16-28 months	EL-ASD, LL-ASD	32	16	50%	16	50%	Pecukonis et al. (2021)
70-77	MCI, TD	110	51	46%	59	54%	Katzorke et al. (2018)
7-11	PWS, TD	32	24	75%	8	25%	Walsh et al. (2017)
21-74	CI users, NH	27	5		5		Bisconti et al. (2016)*
grades 3-5	dyslexia, TD	40	24	60%	16	40%	Song et al. (2014)
11-18	ASD	44	36	82%	8	18%	Yeung et al. (2019)
4-6 months	EL-ASD, LL-ASD	36	20	56%	16	44%	Lloyd-Fox et al. (2018)
3, 6, 9 & 12 months	EL-ASD, LL-ASD	64					Keehn et al. (2013)*
MCI: 63 (6.4), AD: 59.2 (3.9)	MCI, AD, TD	62	28	45%	34	55%	Arai et al. (2006)
18-37	PDD, TD	20	15	75%	5	25%	Kuwabara et al. (2006)
48-59	PWA, poststroke non-aphasia, TD	29	11		5		Sakatani et al. (1998)
Did Not Specify Whether Gender or Sex Reported
5-11	RI, TD	97	48	49%	49	51%	Marks et al. (2022)
20-72	PWA, TD	47	25	61%			Gilmore et al. (2021)
PWS: 27 (8), TD: 26 (3)	PWS, TD	30			7	23%	Jackson et al. (2019)
8-16	CWS, TD	46	34	74%			Hosseini et al. (2018)
M: 12	dyslexia, TD	30	22	73%			Cutini et al. (2016)
8-12	SLI, TD	30	16	53%			Fu et al. (2016)
23-86	CI users, NH	67			35	52%	Olds et al. (2016)
55-59	CI users, NH	33	15	45%	18	55%	van de Rijt et al. (2016)
18-60	Deaf, NH	60	24	40%	36	60%	Dewey & Hartley (2015)
50-90	probable AD	16			10	63%	Metzger et al. (2015)
25.7 (2.7)	dyslexia, TD	51	25	49%	26	51%	Sela et al. (2014)
44.5 (14.2)	DM1, TD	59	32	54%	27	46%	Caliandro et al. (2013)
10-22 & 20-32	ASD, TD	23	20	87%	3	13%	Funabiki et al. (2012)
18-35	ASD, TD	38	26	68%	12	32%	Iwanami et al. (2011)
3-12 & 18-44	PWS	23	20	87%	3	13%	Sato et al. (2011)
9-16 & adults	HFA, PDD-NOS	78	53	68%	25	32%	Kawakubo et al. (2009)
6-11	ASD, TD	18	14	78%	4	22%	Minagawa-Kawai et al. (2009)
67.3 (10.6)	DAT, TD	20	9	45%	11	55%	Fallgatter et al. (1997)
71 (10)	AD, TD	56	22	39%	34	61%	Hock et al. (1997)
Gender/Sex Not Reported
36-78	Deaf, CI users, NH	32					Anderson et al. (2017)
32-52	TBI, TD	17					Rodriguez Merzagora et al. (2014)
3-12 & 22-28	CI users, NH	58					Sevy et al. (2010)
22-80	ALS locked-in, ALS	40					Naito et al. (2007)

Note. Mushtaq et al. (2020) reported males and females for cochlear implant users but only reported males for comparison group. Bisconti et al. (2016) and Sakatani et al. (1998) did not report sex for comparison group. Keehn et al. (2013) reported males and females per timepoint but not the total sample. Rodriguez Merzagora et al. (2014) only noted that gender distribution did not differ between TBI and control group. Naito et al. (2007) reported “40 male and female” participants but did not specify further. EL-ASD = elevated likelihood of autism spectrum disorder (ASD). LL-ASD = low likelihood of ASD. PWA = persons with aphasia. CI users = users of cochlear implants. NH = normal hearing. MCI = mild cognitive impairment. TD = typically developing. PWS = persons who stutter. CWS = children who recovered from stuttering. AD = Alzheimer’s dementia. PDD = pervasive developmental disorder. RI = reading impairment. SLI = specific language impairment. DM1 = myotonic dystrophy Type 1. HFA = high-functioning autism, PDD-NOS = PDD not otherwise specified. DAT = dementia of the Alzheimer type. AD = Alzheimer’s dementia. TBI = traumatic brain injury. ALS = amyotrophic lateral sclerosis.

Although most studies at least partly reported gender or sex, reporting practices were inconsistent yet improved over time; see Figures 2 and 3 in SM1. Nineteen of 34 studies (56%) did not distinguish between gender and sex and simply reported “male” and “female.” Eleven studies (32%) reported sex, and four studies (12%) reported gender. In addition, reporting was sometimes incomplete. Four studies (11%) only reported male participants (Cutini et al., 2016; Fu et al., 2016; Gilmore et al., 2021; Hosseini et al., 2018), and three studies (8%) only reported female participants (Jackson et al., 2019; Metzger et al., 2015; Olds et al., 2016). Two studies (6%) reported sex for the clinical group but not for the comparison group (Bisconti et al., 2016; Sakatani et al., 1998), while one study (3%) reported the gender distribution for the clinical group but only reported males for the comparison group (Mushtaq et al., 2020). Still further, reporting was incomplete in other ways. For instance, one study (3%) reported the distribution at each timepoint of data collection versus the distribution of the entire sample, with some (but not all) participants completing data collection at multiple timepoints (Keehn et al., 2013). Given the relevance of gender diversity in clinical populations, such as autism (Walsh et al., 2018; Warrier et al., 2020), these data limit the generalizability of findings.

Socioeconomic Status

Of 38 studies, 28 (74%) did not report any information on SES; this reporting trend remained consistent over time; see Figure 4 in SM1. Of the 10 studies that at least partly reported SES, two (20%) reported multiple SES indicators; see Table 6. Both reported household income plus maternal level of education alone or level of education for both mothers and fathers. Pecukonis et al. (2021) reported household income and maternal level of education using a scale with four points, while Edwards et al. (2017) reported household income on a Likert scale with eight points and level of education for mothers and fathers on a Likert scale with nine points. Thus, there was inconsistency in measurement scale.

Table 6.

Socioeconomic Status in fNIRS Studies of Speech and Language Impairment

Age in Years	Population	N	Distribution	Reference
Reported Household Income (HHI) & Maternal Level of Education (MLE) or Father Level of Education (FLE)
3 months	EL-ASD	21, 17 LL-ASD	HHI on 8-point scale, low likelihood of ASD = 7.29 (2.02), high likelihood of ASD: 7.6 (1.35), with 7 = $65-$75,000 & 8 = $>75,000; MLE: LL-ASD 6.35 (1.32), EL-ASD 5.75 (1.48); FLE: LL-ASD 5.35 (2.40), EL-ASD 5.5 (1.50) on a 9-point scale, with 1 = some high school, 5 = 4-year degree, 6 = some graduate school, 9 = professional degree	Edwards et al. (2017)
16-28 months	EL-ASD	14, 18 LL-ASD	HHI on 4-point scale, LL-ASD: 3.9 (3-4), EL-ASD: 3.6 (1-4), with 1 = $25-$35,000 and 4 = >$75,000; MLE on 4-point scale, LL-ASD: 2.9 (1-4), EL-ASD: 3.2 (1-4) with 1 = community college, some college, or professional degree and 4 = master’s or doctorate	Pecukonis et al. (2021)
Reported Parental Level of Education (PLE)
5-11	RI	24 RI, 73 TD	M = 8.94 (5-11) on an 11-point scale; 8.94 = some post-baccalaureate or master’s level schooling, 5 = some associate’s level or certificate training, 11 = doctorate	Marks et al. (2022)
7-11	PWS	16, 16 TD	stuttering M = 6.2 (0.8), 4-7 on a 7-point scale (1 = < 7th grade education, 7 = completion of a graduate or professional degree & comparison group matched to stuttering group on PLE	Walsh et al. (2017)
Reported Participants’ Educational Experience or Level of Education
70-77	MCI	55, 55 TD	MCI: M = 9.4 (2.9) years, Comparison: M = 9.8 (2.7) years	Katzorke et al. (2018)
60-91	MCI	26, 26 TD	MCI: 7.94 (4.41) years, Comparison: 9.58 (3.26) years	Yeung et al. (2016)
57-87	probable AD	16	M = 10.21 (3.33) years	Metzger et al. (2015)
44.5 (14.2)	DM1	29, 30 TD	DM1: M = 10.29 years (3.3), TD = 9.1 years (3.1)	Caliandro et al. (2013)
67.3 (10.6)	DAT	10, 10 TD	9 (90%) secondary school, 1 (10%) university degree	Fallgatter et al. (1997)
Reported Non-Specific Information
25.7 (2.7)	dyslexia	17, TD	participants recruited from middle-class secondary schools	Sela et al. (2014)

Note. EL-ASD = elevated likelihood of autism spectrum disorder (ASD). LL-ASD = low likelihood of ASD. HHI = household income. Self-Ed. = participant’s level of education. EL-ASD = elevated likelihood of ASD. LL-ASD = low likelihood of ASD. RI = reading impairment. TD + typically developing. PWS = persons who stutter. MCI = mild cognitive impairment. AD = Alzheimer’s Disease. DM1 = Myotonic dystrophy - Type 1. DAT = dementia of the Alzheimer-type.

Some studies used singular or unspecific indicators of SES. Two of 10 studies (20%) reported parental level of education (Marks et al., 2022; Walsh et al., 2017). Five studies (50%) reported participant years of education (Caliandro et al., 2013; Katzorke et al., 2018; Metzger et al., 2015; Yeung et al., 2016) or level of educational attainment, without specifying the scale used (Fallgatter et al., 1997). Sela et al. (2014) only reported that participants were recruited from middle class secondary schools. In addition to only being descriptive, this information pertains to schools and not participants. Overall, amid broad underreporting of any information on SES, these data show inconsistency in how SES is measured and reported.

Geographic Location, Income Level, and Language

All institutions of the 38 qualifying studies were in WEIRD countries (n = 24, or 63%) or industrialized countries (n = 14, or 37%) that qualified as high-income countries per the World Bank (2022); see Table 7. Fourteen studies (37%) took place in the United States, 10 (26%) in Japan, four (11%) in Germany, four (11%) in the United Kingdom, two in China (5%), two in Hong Kong (5%), one in Israel (3%) and one in Italy (3%). Reporting any information on where data collection took place improved over time; see Figure 5 in SI1.

Nearly half the studies (n = 18, or 47%) provided information on participant language background, with one additional study (n = 1, or 3%) spontaneously offering this information post hoc in their response to a request for information about race and ethnicity (Dewey & Hartley, 2015). Some studies did not report language background but indicated the language used for participant research activities (n = 13, or 34%). However, it was not always clear which language was used. For instance, Yeung et al. (2016, 2019) reported use of Chinese language assessments but not a specific variant of Mandarin (e.g., Putonghua) or Cantonese. Similarly, few studies with participants who spoke English specified a variant of English (e.g., General American English). Nevertheless, reporting any information on participant language improved over time; see Figure 6 in SI1. Together with geographic location, these data suggest global discrepancy in who conducts fNIRS studies pertaining to speech and language impairment and how they administer language materials and record language use by the participants.

Reporting of Reasons for Attrition from Studies

Screening

A minority of studies (n = 15 of 38, or 39%) reported exclusion of participants. No study reported examining sociodemographics in sharing information on attrition or exclusion; see Table 8. Four of these 15 studies (73%) reported exclusion due to selection criteria during screening. Most of the listed reasons for attrition were unrelated to sociodemographics.

Table 8.

Reported Reasons for Attrition from fNIRS Studies of Speech and Language Impairment

Age	Population	N	Reasons	Reference
Screening: Selection Criteria
16-28 months	EL-ASD	14, 18 LL-ASD	family history of ASD (n = 4 LL-ASD), too old (<8 months; n = 1 LL-ASD, 1 EL-ASD)	Pecukonis et al. (2021)
70-77	MCI	55, 55 TD	history of CNS sickness (n = 54 TD, 2 MCI), not right-handed (n = 28 TD, 4 MCI), not native German speaker (n = 14 TD, 3 MCI)	Katzorke et al. (2018)
3-12 & 18-44	PWS	23	not right-handed (n = 2 adults, 4 children, 1 preschooler)	Sato et al. (2011)
9-16 & adults	HFA, PDD-NOS	27, 54 TD	not right-handed (n = 1 HFA and 1 sibling, but no change in statistical outcomes)	Kawakubo et al. (2009)
Attrition: Dropped from Study
16-28 months	EL-ASD	14, 18 LL-ASD	no 24- or 36-month visits (n = 9 LL-ASD, 1 EL-ASD)	Pecukonis et al. (2021)
36-78	Deaf, CI users	15, 17 NH	unrelated medical reasons (n = 1 CI user)	Anderson et al. (2017)
55-59	CI users	5, 33 NH	unexplained attrition (n = 33 NH in methods, n = 21 in results)	Van de Rijt et al. (2016)
3, 6, 9 & 12 months	EL-ASD	27, 3 LL-ASD	unexplained total attrition (26%)	Keehn et al. (2013)
Attrition: Insufficient Data
16-28 months	EL-ASD	14, 18 LL-ASD	insufficient trials (<16; n = 2 LL-ASD, n = 4 EL-ASD); experimenter error (n = 1 EL-ASD)	Pecukonis et al. (2021)
6-12	CI users	19, 20 NH	did not complete fNIRS task (n = 1 CI user)	Mushtaq et al. (2020)
70-77	MCI	55, 55 TD	missing genetic data (n = 4 TD, 2 MCI), no fNIRS data available (n = 6 TD, 2 MCI)	Katzorke et al. (2018)
3 months	EL-ASD	21, 17 LL-ASD	participants who heard <16 trials (8/each stimulus type)	Edwards et al. (2017)
60-91	MCI	26, 26 TD	unable to complete fNIRS tests (n = 17)	Yeung et al. (2016)
Attrition: Poor fNIRS Data Quality
6-12	CI users	19, 20 NH	channels with poor optode-scalp contact	Mushtaq et al. (2020)
16-28 months	EL-ASD	14, 18 LL-ASD	fNIRS data unusable (n = 1 LL-ASD, 2 EL-ASD)	Pecukonis et al. (2021)
70-77	MCI	55, 55 TD	poor fNIRS signals (n = 44 TD, 9 MCI)	Katzorke et al. (2018)
36-78	Deaf, CI users	15, 17 NH	excessive motion & poor contact between fNIRS optodes and the scalp (n = 1 CI user)	Anderson et al. (2017)
3 months	EL-ASD	21, 17 LL-ASD	insufficient total measurement channels; did not report n but groups did not differ in number of channels that they contributed to analysis	Edwards et al. (2017)
21-74	CI users	14, 13 NH	poor channel availability, poor data quality, or difficulties in placing the fNIRS headband (n = 4 CI users, 3 NH)	Bisconti et al. (2016)
M = 12	dyslexia	15, 15 TD	data too noisy for HbO analysis (n = 3 dyslexia, 3 TD)	Cutini et al. (2016)
6-11	ASD	9, 9 TD	motion artifacts (n = 9 ASD)	Minagawa-Kawai et al. (2009)
22-80	ALS-TLS	17, 23 ALS	poor signal separation (≤75%): n = 7/23 cases TLS, 10 non-TLS)	Naito et al. (2007)
Selective Outcome Reporting: Incomplete Reporting of Results
16-28 months	EL-ASD	14, 18 LL-ASD	reported HbO only	Pecukonis et al. (2021)
23-86	CI users	32, 35 NH	selected hemodynamic response function “by eye”	Olds et al. (2016)
60-91	MCI	26, 26 TD	reported HbO only	Yeung et al. (2016)
Grades 3-5	dyslexia	20, 20 TD	partial reporting of linear regression results	Song et al. (2014)
10-22 & 20-32	ASD	11, 12 TD	reported HbO only	Funabiki et al. (2012)
18-35	ASD	20, 18 TD	reported HbO only	Iwanami et al. (2011)
MCI: 63 (6.4), AD: 59.2 (3.9)	MCI, AD	15 MCI, 15 AD, 32 TD	excluded negative HbO changes	Arai et al. (2006)
18-37	PDD	10, 10 TD	unclear whether they used HbO or Hbr	Kuwabara et al. (2006)
48-59	PWA	10, 6 poststroke nonaphasia, 13 TD	no quantitative reporting of hemodynamic response function	Sakatani et al. (1998)

Note. EL-ASD = elevated likelihood of autism spectrum disorder (ASD). LL-ASD = low likelihood of ASD. MCI = mild cognitive impairment. TD = typically developing. CI users = users of cochlear implants. NH = normal hearing. PWS = persons who stutter. HFA = high-functioning autism, PDD-NOS = PDD not otherwise specified. ALS = amyotrophic lateral sclerosis. ALS-TLS = ALS total locked-in syndrome. PWA = persons with aphasia.

Attrition

Four of 38 studies (27%) reported exclusion due to participant attrition. However, two of these studies only provided attrition rates or non-specific reasons, such as stating that attrition was due to unknown reasons. Five of 38 studies (33%) reported attrition due to insufficient data, with three studies stating that participants were unable to complete the task and one stating that no fNIRS data was available (i.e., it was missing). As with dropping out of a study, no specific reasons were provided for why there was insufficient data.

A minority of studies (n = 9 of 38, or 24%) reported reasons for poor fNIRS data quality, with most reasons being non-specific: having unusable data (Pecukonis et al., 2021), poor signal quality (Katzorke et al., 2018), insufficient number of measurement channels (Edwards et al., 2017), or due to multiple reasons (Anderson et al., 2017; Bisconti et al., 2016). Other studies listed more specific reasons: poor contact between optodes and the scalp (Mushtaq et al., 2020), poor signal separation (Naito et al., 2007), and motion artifacts (Minagawa-Kawai et al., 2009). No study reported exclusion by sociodemographic variables. In all, amid underreporting of exclusion rates, studies also underreported reasons for exclusion.

Summary

Analysis of 38 studies using fNIRS to investigate speech and language impairment revealed broad underreporting of race and ethnicity, gender and sex, and participant SES. When reported, exclusion data did not consider sociodemographic variables. These methodological issues hinder the potential of fNIRS as a neuroimaging tool that is feasible for use with diverse individuals with speech and language impairments.

DISCUSSION

Underreporting Limits Generalizability and Reproducibility

Consistent with DisCrit (Annamma et al., 2013), data from this review indicate that findings from fNIRS studies of speech and language impairment represent a biased sample on at least two dimensions: race/ethnicity and SES. In studies with available information on race or ethnicity, all non-white groups comprised less than 10% of participants except for Asian participants, who were all Japanese, and comprised 15%. Referring to nationality versus specific information does not provide information relevant to appreciating the rich heterogeneity in sociocultural background, hair types, and skin pigmentation relevant to speech, language, and fNIRS. As data were not reported intersectionally, there are further concerns about bias; being BIPOC and of low SES are each tied to experiences of marginalization that could make research participation less accessible (Crenshaw, 1989). More consistent reporting using intersectional indicators is needed, so that systematic exclusion is duly noted, and pressure can be exerted on researchers as a community to correct these biases. We call for careful consideration of representation, with participant sociodemographics representative of the context in which studies were conducted (e.g., in Norway, 2.6% of individuals are African or of African descent) (Office of the High Commissioner for Human Rights, 2022).

Though not a primary aim of this study, 50% of studies had corresponding authors or coauthors who did not respond to multiple requests for information, and 26% of study authors reported they did not collect or have access to this information. Only 18% provided information. In comparison, Gabelica et al. (2022) found 79% of 1792 authors of BioMed Central journal articles that indicated they would share data upon reasonable request did not reply to emails; 7% shared data. In the present case, insufficient information on participant sociodemographics limits the weight of the evidence from study findings. Further, authors do not commonly mention sociodemographics as a limitation. This dual conundrum of underreporting and failure to acknowledge limitations is not endemic to fNIRS. For example, 38% of 301 autism studies spanning 100,245 autistic participants did not report information on intellectual ability, though it is known that about 50% of the autistic population has intellectual disability, and 69% of studies did not report lack of generalizability as a limitation (Russell et al., 2019). One simple corrective action would be to implement journal reporting standards that are applied consistently, so that the burden is not on readership to seek out sociodemographic information from authors.

While we have direct information that authors exclude relevant information on race and ethnicity, the failure to include race or ethnicity in the main report or supplemental materials effectively treats these constructs as irrelevant (Webb et al., 2022). Omission of this information also does not align to current reporting guidelines and minimizes the impact of known methodological issues that disproportionately increase the likelihood of excluding BIPOC from clinical neuroscience studies (Girolamo et al., 2022b). At a detailed level, not including this information makes it impossible to pinpoint where potential breakdowns in representation occurs in fNIRS speech and language impairment studies: recruitment, screening, or analysis.

Intersectional Reporting of SES in fNIRS Studies

To arrive at a more realistic understanding of the variability that exists in a broad representative sample of the population, scientists must identify ways to implement theoretically sound practices per Diversity Science and explore heterogeneity within a group without ascribing to stereotyping practices, such as only studying BIPOC of low SES, which can suggest a singular correspondence between being BIPOC and of low SES (Plaut, 2010). One way to address this concern is to develop meaningful sociodemographic variables, ideally yielding quantitative information plus interpretative text. We recognize that researchers must balance reproducible science via transparent reporting with protecting participant rights to confidentiality and privacy, especially if providing such information could potentially risk inadvertent identification within that local community (Girolamo et al., 2022).

Nevertheless, the need for meaningful sociodemographic reporting remains. Children of lower SES are less likely to have usable neuroimaging data with low noise (Cosgrove et al., 2022). In this review, 10 of 38 studies (26%) reported SES yet varied widely in their SES indicators and the scales they reported. Such inconsistency is to be expected, as the scale and relevant indicators of SES vary by location, and there are currently no recommended scales or guidelines. A commonly used measure, the Hollingshead scale, was published nearly 50 years ago (Hollingshead, 1975). Further, Edwards et al. (2017) and Pecukonis et al. (2021) reported household income of over $75,000 USD as the upper end of the scale. However, there is no information with which to interpret that scale, such as how that amount aligns to the cost of living. In contrast, household income may be a less relevant indicator of SES in countries with higher levels of social support. Clearly, there is no singular solution to reporting SES, but there is a clear need to be more transparent in selection and explanation of SES indicators.

Limitations of Our Review

This review had several limitations. First, it only included studies in English. Exclusion of studies published in other languages prevents fully understanding the state of the literature, especially considering overrepresentation of individuals from WEIRD countries in science (Henrich et al., 2010). Our finding that all studies included in this review took place in WEIRD or industrialized countries may be an artifact of this criterion. The cost of fNIRS systems means it is only accessible to researchers with the means to purchase, build, and maintain such a system. A second limitation arises from underreporting. An absence of information about sociodemographics and reasons for exclusion does not constitute systematic exclusion per se. A third limitation is the scope of the review itself, as individuals with speech and language impairment and fNIRS studies using speech or language tasks comprise only a subset of all BIPOC with dis/abilities and all fNIRS studies. It is unknown to what extent the findings of this review are representative of fNIRS studies more broadly.

Finally, underreporting of participant sociodemographics hinders understanding of the true nature of systematic exclusion in fNIRS studies of speech/language impairment. Regardless, convenience sampling is extremely common, leading to several large-scale initiatives in at least the United States to mitigate systematic exclusion of BIPOC in clinical research: All of Us, an National Institutes of Health (NIH)-wide research program (National Institutes of Health, 2023) with at least $2.16 billion in funding through 2026, with community-based consultation of minoritized populations (e.g., Indigenous nations); advancing minority health as an NIH-wide strategic goal (Jones et al., 2019; National Institutes of Health, 2021), and; promoting health equity through use of community-based methods as a strategic goal of NIDCD (2023b).

Future Directions

The findings of this review highlight the urgent need for further examination of reporting biases and an action plan to address these deficiencies. Clearly, studies do not fully align their reporting guidelines for sociodemographics in fNIRS research, as well as clinical research overall (Choy et al., 2021; Goldfarb & Brown, 2022). As a starting point, we call upon fNIRS scientists, together with other community members (e.g., BIPOC participants from clinical populations, other neuroimaging scientists), to engage in critical conversations about current reporting guidelines (APA, 2020, 2021; Flanagin et al., 2021; JAMA Network Editors, 2020; Yücel et al., 2021), the APA Equity, Diversity, and Inclusion Toolkit of (APA Journals, 2021), and journal reporting standards. Though no singular set of criteria is perfect, discussing and acting upon these issues and professional guidelines will generate more inclusive policies that we envision manifesting in two specific ways.

First, as a noninvasive neuroimaging tool that may be more widely accessible to clinical populations (Butler et al., 2020), fNIRS is only as useful as its ability to work equitably on all in the population. Recent reports commissioned by the Society for fNIRS underline best practices for fNIRS publications (Yücel et al., 2021), as well as developments of and advances in neurophotonic tools (Ayaz et al., 2022). We recognize that individual labs have developed innovative solutions, such as preprocessing pipelines and capping tips, to use fNIRS on a range of skin tones and hair types. However, these solutions are not publicly available. We urge development of a collaborative report centered on sociodemographic reporting in fNIRS studies and inclusive innovations. Some variables, such as hair type and skin tone, would benefit from measures that do not rely on subjective judgements, such as the Fitzpatrick scale (Goon et al., 2021; Okoji et al., 2021); alternatives are offered by an ongoing equity study of fNIRS (Yücel, 2023). At the same time, providing highly specific measures with a constellation of other data, can make participants identifiable; there is a clear need to develop a minimum set of criteria for reporting that upholds participant privacy and confidentiality. Such a report by the fNIRS community is an ideal place to foster these conversations.

Second, journal editors and conference committees need to develop policies that value sociodemographic reporting. In addition to requiring authors to report some minimum set of sociodemographic variables, editorial boards and committees must decide more nuanced aspects of reporting. For instance, authors are often constrained by word and page limits, especially if following best practices for reporting fNIRS (Yücel et al., 2021). Possible solutions might be to provide this information in the supplemental materials rather than the main document, though doing so risks downplaying the importance of such information, or creating a dedicated reporting section not included in the word limits. Although increasing page or word limits might incur a financial cost, these costs could be absorbed by professional societies or the publishing industry. Increasing page or word limits in the context of conference submissions may be less likely to incur additional cost. As with developing sociodemographic reporting guidelines for fNIRS studies, this issue of how to design and implement policies is complex and requires systemic, sustained action. These are just two of many future directions, amid an extant body of work and calls centered on increasing BIPOC participant representation in neuroscience studies at large.

Conclusion

This systematic review documents sociodemographic underreporting and underrepresentation in fNIRS studies of speech and language impairment over the past 25 years. Findings support systematic exclusion at the intersection of race and disability in research, though the weight of the evidence is limited by underreporting. To ensure research is generalizable to all in the population, efforts must be made to develop and implement reporting guidelines, policies, and practices for sociodemographic data that are specific to fNIRS.

Supplementary Material

Supplementary Material

Supplemental Materials 1. Figures showing the incidence of fNIRS studies of speech and language impairment reporting participant sociodemographics

Availability of Data and Materials:

The authors confirm that all data analyzed during this study are included in this published article.