Feasibility of large-scale development-focused EEG research in a hospital-based primary care setting

Abstract

Background:

The age of diagnosis of autism spectrum disorder varies by sociodemographic factors, elevating the need for earlier, unbiased screening and diagnostic measures. Electroencephalography (EEG) is a promising biomarker for early autism detection, but feasibility and validation in primary care settings and in underserved populations remains underexplored. This study evaluates the feasibility of embedding longitudinal EEG and survey data collection into the primary care setting to address these care differences.

Methods:

Baby Steps is a longitudinal study being conducted in a hospital-based urban primary care clinic serving predominantly low-income families. Infants were recruited prior to their 4-month well-child visits, with EEG and parent surveys collected at 4-, 9-, and 12-month visits. Data collection was integrated into the clinic workflow to minimize disruption. Feasibility was assessed based on the first 250 participants enrolled through metrics of recruitment, retention, and data quality, with bivariate analyses examining predictors of retention.

Results:

To reach 250 enrolled participants, 856 eligible families were approached for recruitment. Enrollment reflected clinic demographics (53% Hispanic/Latina/e, 36% Black), and 64% were publicly insured. 94% of participants were retained through 12 months. EEG data were collected at 2 or more time points for 87% of participants. Full survey completion ranged from 65–78% at each time point. No sociodemographic factors were significantly associated with completion of surveys.

Conclusions:

Embedding EEG research in primary care is feasible and facilitates the inclusion of underserved populations. This approach enhances research applicability and may serve as a model for similar efforts in pediatric research.

Introduction

Autism spectrum disorder affects more than 2% of children in the United States, with diagnosed children having a broad range of functional outcomes^1,2. There is considerable evidence that the earlier a child receives developmental interventions, the better their later outcomes^3–7. The American Academy of Pediatrics (AAP) and others have created guidelines for autism diagnosis and treatment that emphasize early connection to care⁸. However, the average age of autism diagnosis in the United States remains approximately four-years of age, indicating that many children are not receiving critical early services¹.

Current best practices for screening for autism rely on parent-completed screening questionnaires, such as the MCHAT R/F, administered during primary care well-child visits in the second year of life⁸. While over 75% of pediatricians report screening universally for autism using questionnaires^9,10, the average age at diagnosis is four years, and there are disparities in age of diagnosis of children with autism, with Black and Hispanic children diagnosed later than White children^11–14. While some of this diagnostic disparity is due to differential access to diagnostic evaluation after detection of autism risk, other factors, including language barriers, trust in the health care system, and perceptions of typical development, likely also reduce the sensitivity of questionnaire-based approaches to screening in some populations^15–19. Most recent prevalence estimates from the Autism and Developmental Disabilities Monitoring Network confirmed higher rates of autism for Black, Hispanic, and Asian children compared to non-Hispanic White children; in addition, lower socioeconomic status (SES) was associated with autism prevalence ^1,20. While these trends suggest improvements in screening and access to services among groups that have historically been underserved, these disparities highlight the importance of promoting detection of those with increased likelihood of autism and connection to services in lower income children of color, a group historically not well represented in autism research.

There are several candidate biomarkers that could provide an alternative to screening questionnaires, and aid in earlier identification of children at increased likelihood for autism or developmental delays. One promising method for earlier risk detection is electroencephalography (EEG), a non-invasive and relatively low-cost method to measure brain activity. Longitudinal, lab-based studies of infants with elevated likelihood of autism (based on having an autistic sibling) have observed differences in EEG measures between infants with and without an autistic sibling and found high sensitivity and specificity of predicting later autism diagnosis^21–23. However, these studies have focused on infants with autistic siblings, not a general population, and have taken place in a more controlled lab setting. Therefore, it is unclear whether collecting high quality EEG within the busy clinic setting is feasible and whether prior lab-based findings will translate to the clinical setting.

It is critical that biomarkers for autism and associated developmental delays are developed and validated across diverse populations and in clinically relevant settings to ensure the effective translation of research findings to patient care. This paper describes our work exploring the feasibility of conducting a longitudinal, EEG-based research study (Baby Steps) aimed at identifying measures associated with later autism and/or developmental delay in a large, racially diverse, urban primary care setting largely serving families with low income. We describe the development and design of the Baby Steps Study, enrollment and participant retention metrics, survey and EEG completion, and assessment of EEG data quality.

Methods:

Brief overview of study design

The Baby Steps Study is a single-site longitudinal, prospective cohort study taking place at the Boston Children’s Hospital Primary Care Clinic (CHPCC). The study follows infants from 4-months to 2-years of age. As of February 2025, the study had enrolled 454 infants. Infants seen at CHPCC are recruited and enrolled prior to or during their 4-month well child visit. Longitudinal data including 5-minute resting state EEG, parent surveys, and review of the medical record are completed in the clinic at 4-, 9-, and 12-month well-child visits. After 24-months of age, participants complete a research developmental evaluation at the Labs of Cognitive Neuroscience at Boston Children’s Hospital to determine autism and developmental outcomes. This feasibility analysis focuses on enrollment and data collection within the well-child clinic, but not the 24-month developmental assessments as enrolled participants have just begun to age into the final assessment time point. Here we present data from the first 250 participants enrolled, all of whom have passed their first birthday. All study procedures were approved by the Boston Children’s Hospital Institutional Review Board.

Description of Study Development

The study design was carefully considered in collaboration with primary care clinic leadership and staff in order to optimize enrollment and data collection within this particular clinic population and to minimize disruption to clinic flow. CHPCC is a large, academic primary care clinic located at a tertiary children’s hospital, serving approximately 16,000 patients. The clinic’s population identifies as primarily Hispanic/Latina/e (33%), African American/Black (34%), and multiracial (19%); 74% of the population is publicly insured, and 8% of the clinic population has complex, chronic disease, which is higher than the national average in primary care populations. The practice cares for 1500 patients with autism. The clinic has dedicated care coordination and social work resources. On average, 100 well child-care visits are completed per day on site by faculty, fellow, and resident providers. The clinic is staffed by 100 providers and 20 nurses.

In addition to vetting the study with nursing and provider leadership, a provider champion was chosen to act as a co-investigator on the study and helped to provide orientation to the clinic and integration with clinical workflows. In-clinic space was identified to store the EEG machine and research supplies. Research staff were introduced to providers and nurses at staff meetings and shadowed clinical visits to identify potential opportunities to approach families for enrollment and collect data that would not interfere with completion of medical visits. To reduce interference with clinic flow, the study team also met regularly with nurses and pediatricians to elicit feedback on current processes.

To reduce burden for families, both recruitment and data collection occurred in the primary care clinic either before or after the scheduled 4-, 9-, and 12-month well-child visit. The research study team was responsible for identifying eligible or already enrolled participants scheduled for well-child visits. Enrolled families were sent phone, text, and email reminders about upcoming visits. Families received $25 for each research visit completed in the first year, as well as $10 for travel compensation. Families received $75 with completion of the lab-based developmental assessment at 24 months, as well as a report describing their child’s developmental skills.

Enrollment

Potential participants were first identified using medical record queries identifying those attending a 2-month-old well-child visits at the clinic; chart review was used to determine eligibility. Research assistants approached families of eligible patients during the two-month well-child care visit, provided information on the study, and offered an opportunity to enroll in the study at four months. Parents were informed that the longitudinal study was aimed at understanding brain development and that at 2 years of age, the research team would complete developmental assessments, including screening and assessment for autism, and that the parents would receive report about their child’s development. Those families who expressed interest were contacted by email or phone to go over the consent form in advance. If they were not able to be contacted, the consent form would be discussed and signed when they were re-approached at the four-month visit and enrolled in the study.

Participants:

Participant demographics are shown in Table 1. Exclusion criteria included neurologic, genetic, or ophthalmologic disorders, birth at <32 weeks gestational age, intrauterine growth restriction, or born small for gestational age, intrauterine opioid exposure, and major NICU complications. Infants also needed to be from either predominantly English or Spanish speaking households, with 75% of an infant’s awake hours in the presence of caregivers who speak either English or Spanish.

Table 1.

Demographic information

	N = 250
Sex	129 M, 121 F
Race, n (%)
African American or Black	89 (36)
Asian	1 (0.4)
Native American or Alaska Native	5 (2)
Native Hawaiian or Pacific Islander	1 (0.4)
White	26 (10)
More than one	41 (16)
Not answered	87 (35)*
Ethnicity, n (%)
Hispanic or Latino/a	133 (53)
Not Hispanic or Latino/a	105 (42)
Not answered	12 (5)
Primary Language, n (%)
English	145 (58)
Spanish	43 (17)
English and Spanish	62 (25)
Insurance type, n (%)
Medicaid	160 (64)
Private	77 (31)
Uninsured	1 (0.4)
Not answered	12 (5)
Monthly household income, n (%)
<$2,100	54 (22)
$2,101–$3,700	60 (24)
$3,701–$7,400	49 (20)
$7,401–$12,700	37 (15)
>$12,700	36 (14)
Not answered	14 (6)
200% Poverty line (2021), n (%)
Below the poverty line	125 (50)
Above the poverty line	106 (42)
Not answered	19 (8)
Marital Status, n (%)
Single	117 (47)
Cohabitating	43 (17)
Married	71 (28)
Separated, divorced, widowed	7 (3)
Not answered	12 (5)
Maternal Education, n (%)
< High School	22 (9)
High school/GED	60 (24)
Associate’s degree/some college	84 (34)
Bachelor’s degree	50 (20)
> Bachelor’s degree	23 (9)
Not answered	11 (4)

^*80% of participants who did not respond to this question about race identified as Hispanic when asked about ethnicity.

Data Collection:

Parent Surveys:

Parents are asked to complete several surveys at each visit (Table 2). Surveys cover parent reports of development (Ages and Stages Questionnaire), temperament (Infant Behavior Questionnaire), and parental perceived stress (Perceived Stress Scale), along with demographic information. In addition, several variables were collected via medical record abstraction. All surveys were available in English and Spanish, and provided in whichever language was preferred by the parent.

Table 2.

Survey Responses

Income Sufficiency, N= 250
Never, n (%)	30 (12)
Rarely, n (%)	26 (10)
Sometimes, n (%)	83 (33)
Usually, n (%)	55 (22)
Always, n (%)	44 (18)
Not Answered, n (%)	12 (5)
Perceived Stress
	4-months (n = 196)	9-months (n = 167)	12-months (n = 151)
mean (SD)	11 (7.62)	13 (7.57)	13 (7.50)
Low, n (%)	121 (62)	95 (57)	84 (56)
Moderate, n (%)	70 (36)	65 (39)	61 (40)
High, n (%)	5 (3)	7 (4)	6 (4)
Depression Measures
	4-months (n = 204)	9-months (n = 173)	12-months (n = 159)
mean (SD)	5.1 (5.28)	6.9 (10.45)	6 (9.62)
Above clinical cutoff, n (%)	19 (9)	25 (14)	18 (11)

Outcome Measures at 2-years of age:

Participants are invited to come to the lab (under 1 mile from the primary care clinic) to complete developmental assessments. This includes the Mullen Scales of Early Learning (MSEL⁸⁶), a level 2 autism screener – the Rapid Interactive Screening Test for Autism in Toddlers (RITA-T²⁴), and the Autism Diagnostic Observation Schedule-2²⁵, for those toddlers with elevated scores on the RITA-T.

EEG Acquisition

Resting EEG data is collected using a 128-channel HydroCel Geodesic Sensor Net (HGSN) connected to a NetAmps 400 amplifier (Electrical Geodesic Inc.) while infants watch a video of infant toys for 5 mins. The EEG equipment is secured on a portable cart, and collection occurs in a quiet area of the clinic with curtains providing additional privacy (Figure 1).

Figure 1:

Portable EEG Cart

The EEG recording is referenced online to a single vertex electrode during acquisition and channel impedance is kept below 50 kΩ; signals are sampled at 1000 Hz. EEG data are processed using the Batch Automated Processing Platform (BEAPP: ²⁶) with integrated Harvard Automated Preprocessing Pipeline for EEG (HAPPE: ²⁷) using the parameters described in ²⁸. EEG recordings are rejected using the following HAPPE data quality measures: Fewer than 20 segments (40 s of total EEG), percent good channels <80%, percent independent components rejected >80%, mean artifact probability of components kept >0.3, and percent variance retained <25%.

Feasibility Analysis

In order to assess study feasibility, we assessed metrics related to patient enrollment and retention in the study, time burden for families engaged in the study, and relationships between patient characteristics and retention in the study. Surveys were considered partially complete if the caregiver started the survey but did not complete all sections. Surveys were considered fully completed if all questions were answered, including if parent chose “Prefer not to answer.”

Results

From February 2022 to July 2023 (16 months), 250 participants were successfully enrolled in the study (Figure 2). A total of 1364 patients were screened for eligibility based on available medical records, and 982 families were approached in the clinic for recruitment. Of those 856 were eligible for enrollment. At the time of recruitment 394 families (46% of 856) expressed interest in being enrolled in the study, and 63% of those interested families enrolled (250 of 394). Reasons for not enrolling included (1) child aging out between time of recruitment and first study visit, (2) families not having enough time to complete the EEG at their 4-month well-child visit, and (3) families deciding they no longer wanted to participate.

Figure 2:

Consort diagram of participant recruitment and enrollment.

Demographic information of enrolled participants is shown in Table 1. Demographics of enrolled participants match the overall clinic population with families primarily identifying as Latina/e (53%) or Black (36%) or more than one race (16%) and 64% reporting being publicly insured. The majority of infants (66%) had an older sibling, and 16 (6%) had an older sibling with autism.

Survey responses (Table 2), indicate moderate to high levels of stress for many families in the study, with 22% reporting “never” or “rarely” having income sufficient to meet their family’s needs, and 39% reporting moderate to high levels of stress on the Perceived Stress Scale.

Data Collection

During the first two years, the study team was comprised of four full time research assistants dedicated to recruitment and data collection in the primary care clinic. At 4-, 9-, and 12-month well child visits, enrolled families were asked to complete a set of parent surveys (Table 4), and 5-minutes of resting state EEG data were collected. Families were sent links to surveys one week ahead of the clinic visit via REDCap email, and received calls reminding them about the co-occurring research study at their well-child visit. EEG was collected in a secluded area of the waiting room with mobile privacy screens. In the majority of cases, participants completed their well-child appointment prior to EEG; the average time required to complete EEG data collection was 20 minutes. Despite receiving survey links in advance, the majority of families who completed completed surveys did so on the same day as their appointment (70% of those who completed surveys).

Table 4.

Bivariate associations between participant characteristics and study data

	Percent with complete survey data at 12 months n (%)	p-value (chi-square or t-test of means)	Mean (SD), complete survey data at 12 months	Mean (SD) no compete survey data at 12 months
Language(s) spoken at home, n=235		0.25
English, n=137	92 (67)
English and Spanish, n=59	35 (59)
Spanish, n=39	21 (54)
Income Sufficiency, n=227		0.13
Always/usually sufficient, n=96	68 (71)
Sometimes, Rarely, or Never Sufficient, n=131	80 (61)
Insurance type, n=226		0.12
Public, n=153	95 (62)
Private, n=73	53 (73)
Perceived stress, measured at 9 months n=166		0.07	13.4 (7.4)	10.9 (8.1)
CES-Depression scale (measured at 9 months), N=172		0.80	6.8 (10.2)	7.3 (11.4)

Table 3 shows completion rates of parent surveys and EEG data for the 250 participants enrolled at 4 months. As completion was tied to whether participants came to their well-child visits, the completion rates for follow up visits was impacted by clinic no shows, cancellations, and rescheduling. In addition, infants who presented with cold symptoms were quarantined and tested for COVID, precluding research assessment. 89% of enrolled participants provided some amount of longitudinal data, and 72% provided data at all three time points. EEG was collected at all three time points for 68% of enrolled participants. However, many participants did not complete all the surveys provided (Table 3). While there is missing longitudinal data due to missed appointments or partial survey completion, 94% of participants were retained in the study at the 12-month time point.

Table 3.

Data Collection

	4-months	9-months	12-months
Clinic appointment completed	250	215	211
No clinic appointment	0 (0)	26 (12)	24 (11)
Dropped out or transferred facilities	0 (0)	9 (4)	15 (7)
Survey Data Collected, n (%) *Percent calculated with clinic appointment completed
Surveys fully complete	190 (76)	142 (66)	150 (71)
Surveys partially/fully complete	235 (94)	177 (82)	175 (82)
EEG Data Collected, n (%)
EEG’s Collected	246	197	192
EEG’s with good quality	203 (83)	190 (96)	178 (93)
Excluded for sleeping	38 (15)	3 (2)	2 (1)
Any data collected, n (%)	249 (99)	205 (95)	197 (93)
	N = 250
Longitudinal Data, n (%)	Across three available time points
Any data from two timepoints	223 (89)
Any data from all three timepoints	179 (72)
At least two EEGs collected	218 (87)
All three EEGs collected	169 (68)

Predictors of research study engagement

Given our intention to enroll a socially vulnerable population into the study, we sought to understand whether primary language, socio-economic status, and mental health impacted engagement in the study. Missing data in the study occurred due to participants not coming to their well-child visits or not fully completing surveys, which was typically the last portion of the study visit. Therefore, we utilized 12-month survey data completion as our outcome of interest, including in our analysis those who did not come to clinic for a well child visit. We performed bivariate analysis with chi-square or t-tests of means to examine associations between complete survey data at 12 months and demographic and survey measures. We found no significant relationships between perceived income sufficiency, insurance type, depression, perceived stress, or language spoken at home and complete survey data at 12 months (Table 4). However, we did see trends such that participants with higher perceived income sufficiency had higher rates of survey completion compared to those with lower income sufficiency (71% versus 61%, p=0.13), and similar findings for those with private versus public insurance (73% versus 62%, p=0.12). Though not significant, those with complete surveys at 12 months had higher mean perceived stress scores, which would not be the expected trend (13.4 versus 10.9, p=0.07).

Discussion

This study demonstrates the feasibility of collecting longitudinal EEG and parent survey data within a busy primary care setting. Further, this work demonstrates an innovative approach to recruiting and retaining a diverse sample of participants, including populations historically underrepresented in pediatric research, especially autism research. Our results underscore the importance of primary care-based research in reaching diverse, lower-income populations.

Through this study, we identified several unique benefits to conducting research in the primary care setting. First, by embedding research in the primary care setting, studies can reduce barriers to participation for communities that may face logistical, systemic, or trust-based obstacles to accessing traditional, lab-based studies. Basing enrollment and data collection in primary care may support higher enrollment and retention rates, especially across longitudinal studies. This is especially important when considering goals in ensuring that research includes racially and socioeconomically diverse samples in the field of developmental neuroscience. That our research population was well matched to the racial, ethnic, and insurance characteristics of our clinic population suggests that we may have successfully reduced barriers to research participants and overcome some of the inherent sampling biases seen in lab-based research^29,30.

Second, successful data collection within the clinic setting provides a road map for efficient and effective translation to clinical implementation. For example, research EEG data collected in the clinic is more likely to reflect real-world clinical conditions where the ambient noises and sounds cannot be controlled, allowing for more accurate assessment of the effectiveness of EEG as a potential clinical screening tool. Thus, early partnerships within primary care can improve the applicability of findings to primary care practices. When important screening tests can be located within the site of usual care, adherence to screening increases^31,32. In this study, we successfully collected usable and high-quality EEG data, supporting the feasibility of collection in the primary care setting which is inherently less controlled than a lab-based setting. Third, by working within the primary care setting, the research team was able to leverage medical record data to screen participants for eligibility and reduce survey burden by extracting some data directly from the medical record.

A key component of our success was the collaborative framework established between research and clinical staff. Prior to initiation of data collection, we integrated our study team into the clinic flow, partnering closely with nursing and provider teams to minimize disruptions to patient care and establish familiarity and trust. Both in-the-moment check ins with the clinical team through the workday, as well as regular meetings with clinic staff and leadership allowed us to receive ongoing feedback and adjust protocols to align with changes in clinical operations.

Our analysis of data collection and study engagement also highlight areas for improvement. While the vast majority of enrolled participants provided longitudinal data and few participants dropped out of the study, a substantial number of families did not fully complete surveys. Only 30% of participants completed surveys prior to their well-child visit despite receiving these surveys in advance; thus survey completion was often tied to whether the family had sufficient time during or after their visit to complete the surveys. Reducing the number of surveys or separating out survey completion from other research related activities (e.g. EEG) may increase completion rates. Future analyses in this study may provide key insights regarding which surveys provide the most useful information alongside EEG, and may allow for fewer surveys in subsequent studies. Further analysis of parent measures indicated that English and Spanish speaking families had similar rates of survey completion. Parents reporting Income Sufficiency in the usual-always range and those with private health insurance had higher rates of survey completion but differences were not significant (p = 0.13). In larger studies, however, these trends could translate to significant differences in who is contributing data. Interestingly, those with compete surveys at 12 months also had higher reported perceived stress at 9 months, which may suggest that moderate stress levels promote study engagement in some way.

There are several limitations that should be considered. First, this study was completed at a single site with a large patient population. Multi-site studies embedded in smaller practices will likely have additional logistical barriers to consider. Second, this study occurred at an academic medical center and while the primary care practice is not specific to children receiving specialty care at the academic center, it is likely that they are more familiar with research in primary care and elsewhere at the medical center, and thus more accepting of participating in research. We did not collect information from families on why they did or did not enroll in the study. However, many families and clinicians did report anecdotally that the formal developmental testing at age 2 was a draw to participation in the study; it is possible that our enrollment rates may have been lower without this built-in benefit. Finally, significant resources are needed to successfully run this type of study given that once families are enrolled, the research team is responsible to be available whenever future appointments are scheduled. Thus, the research team had to have broad availability and redundancy in order to minimize missing opportunities for data collection.

While our study has demonstrated the feasibility of research-based EEG acquisition in the academic primary care setting, successful translation to universal screening in a wide variety of practice settings requires additional advances in EEG acquisition and analysis. Specifically, an FDA approved EEG device for screening would need to be a reasonable cost, easy to use, and easy to interpret. Advances in technology and new devices for screening have been implanted into primary care settings in the past – for example vision screening devices are now commonly used in primary care to identify amblyopia and refractive errors. We also note that while high levels of staffing were needed for research recruitment, consenting, and data collection, EEG data collection in infants can be completed by a single trained person, similar to vaccine administration or a blood draw.

In summary, this study highlights the feasibility and benefits of conducting EEG research within a primary care setting and underscores the potential of primary care-based studies to advance research equity by increasing access to underrepresented populations. Future work will focus on the analysis of the EEG and survey data to examine the relationships between early neural activity and later developmental outcomes in this representative, real-world sample. This approach can serve as a model for similar efforts aiming to expand inclusivity and applicability of developmental research.

What’s New:

This study demonstrates the feasibility of conducting EEG research within a primary care setting and highlights the potential of primary care-based studies in improving enrollment and data collection across the general population, including those historically underrepresented in research.