This study is a randomized, double-blind, concurrent, sham-controlled trial designed to evaluate the safety and efficacy of Cognilum, an investigational medical device, for the treatment of symptoms of ASD in two- to six-year-old children.

The study protocol was approved by the institutional review board (WCG IRB approval # 1280247) and is registered at ClinicalTrials.gov (Identifier: NCT04660552).

The US Food and Drug Administration (FDA) is likely to make a decision regarding a minimal clinically important difference (MCID) based on a paper by Jurek et al. (76) which suggested that a clinically significant difference (CSD) is a 4.5-point reduction in CARS scores after treatment in the active group (see further discussion in the Limitations Section). We expected at least a 1- to 2-point reduction in CARS scores in the Sham group (see Results Section for the mean reductions in both groups).

This was the first human trial evaluating the Cognilum^™ (JelikaLite Corp, New York United States) device. Without previous testing in humans to assess its safety and potential efficacy (powered to detect a CSD), we relied on published research and studies which used 11 participants (77) and 40 participants (74, 78), and estimated that a sample size of 30 participants would be adequate for statistical analysis. The sample size for this trial was set at 30 for feasibility.

The study enrolled children of both sexes aged two to six years, who had previously been diagnosed with ASD by a licensed professional according to the Diagnostic Statistical Manual of Mental Disorders, Fifth Edition (DSM V). Most participants received their diagnosis at 24 months during an evaluation provided by the NYS Department of Health Early Intervention program, which typically includes a series of tests such as ADOS, CARS, TABS, and ADI-R. Approximately half of the participants in each group received one or more types of behavioral therapy (i.e., Applied Behavioral Analysis [ABA], speech therapy, occupational therapy, or physical therapy), which did not prevent them from participating in the trial, see Table 1. Exclusion criteria for the study included regular use of medications or a history of seizures. More details on the inclusion and exclusion criteria can be found in Table 2. There were no changes in the trial methods after the trial began.

Participants were randomly assigned to the Active or Sham Control groups using a random sequence generation program at www.random.org. The random sequence was created beforehand, and assignment to each group was conducted by a research assistant who was not involved in the evaluation or enrollment of the participants. The enrollment was conducted by licensed clinicians, the first and last authors in the study. No restrictions were applied to the randomization process. Before enrollment, each participant signed an informed consent form acknowledging that they had the same chance of being assigned to the Active or Sham condition.

The study was conducted in two private medical offices (one location was in Brooklyn NY and another one was in Manhattan, NY). Both locations were approved by the Institutional Review Board. Both locations are easily accessible by public transportation (subway and bus) and by car (with garages available nearby). New York City has a diverse population, both racially and economically, with autism prevalence rates similar to those of the rest of the country, approximately 3% (Autism and Developmental Disabilities Monitoring Network, 2018).

Treatment with tPBM was administered using Cognilum^™, an investigational medical device specifically designed to treat ASD in young children. The manufacturer is Sterling Medical Devices, Moonachie, NJ 07074. Upon reviewing JelikaLite LLC’s application for risk review classification, on October 14, 2020, the FDA concluded that the proposed clinical investigation involving Cognilum and autistic children aged two to six was a nonsignificant risk (NSR) device study. This classification was received because the Cognilum device does not fulfill the criteria for a significant risk (SR) device as defined under 21 CFR 812.3(m) of the investigational device exemptions (IDE) regulation (21 CFR 812). Subsequently, on December 29, 2021, based on the results of this study, the FDA awarded Cognilum a “Breakthrough Device Designation.”

The device weighed approximately 500 grams, was wireless, comfortable for children to wear due to the soft materials used in its construction and did not impede children’s mobility during treatment. The device was adjustable using Velcro to fit children aged two to six years. The current version of the Cognilum device cannot be used with any other device simultaneously (i.e., EEG or pacemakers). The battery used was rechargeable and the maximum use of the battery was several hours. The device specifications included 6 LEDs emitting pulsed light (40 Hz) at a wavelength of 850 nm, with a total maximum power less than 300 mW. The internal design of the device included a control panel, wires, and LEDs, as shown in Figure 1.

All LEDs in the investigational device could simultaneously deliver pulses of NIR light at 40 Hz, 300 mW maximum power to the patients’ scalps, hypothetically modulating mitochondrial function and inducing functional brain connectivity, across several selected brain areas. Treatment sessions were administered twice a week for 8 weeks. To avoid inducing hyperactive behavior, the duration of treatment gradually increased over the course of each subsequent session, up to 6 minutes, and then it was maintained constant.

For the Sham condition, the device performed identically to the active device in all observable behaviors, with the exception that it did not emit any NIR light. The research assistants set the device as active or sham, based on the randomization sequence that was available to them. The experience of the participants during both active and simulated conditions was indistinguishable. The NIR wavelength used by Cognilum (850 nm) is not visible to the human eye and did not cause tissue heating at the administered dose. Caregivers, participants, and assessors were blinded to whether light stimulation was administered.

The principal investigator who is the first author (a board-certified pediatric psychiatrist) and the last author, a licensed clinical psychologist who frequently diagnoses children using CARS, conducted the initial evaluation in person at the evaluation site. The evaluation included administering initial and final CARS. The initial CARS evaluation was performed immediately before the first treatment session and lasted about an hour. The treatment session was administered twice a week for 8 weeks. Each treatment session lasted approximately 30 minutes. Each treatment session involved collecting EEG data for approximately 10 minutes, depending on the tolerance of each child, tPBM treatment for five to six minutes, followed by a second EEG data collection for another 10 minutes, in as many as participants possible. During tPBM treatment sessions, participants were allowed to walk around the office, watch cartoons or play with toys. Participants were encouraged to sit quietly during EEG collection, sitting on their parents’ lap and held by their parents to minimize movement.

Childhood Autism Rating Scales, Second Edition (CARS, 2^nd Edition), were evaluated before (baseline) and after the course of treatment. CARS is a validated clinical rating scale that can be used by a trained clinician to rate items indicative of ASD after direct observation of the child (79). European Medicines Agency (EMA) guidelines recommend it as an outcome measure for the clinical development of treatments for ASD. Furthermore, a consensus study has been conducted on the minimal clinically important difference in core symptoms resulting from therapeutic interventions (76). The scale consists of fifteen items that correspond to the different core domains (e.g., verbal communication, emotional response, and relationships with people) that can be affected by ASD. Total scores can range from a low of 15 to a high of 60; scores below 30 indicate that the individual is in the nonautistic range, scores between 30 and 36.5 indicate mild to moderate autism, and scores from 37 to 60 indicate severe autism (76).

Weekly interviews were conducted with each parent about changes in child behavior and functioning by a blinded research assistant by phone weekly. Parents were asked to maintain a diary documenting their child’s behavior, including newly produced words, changes in comprehension of instructions, consistency of eye contact, sleep patterns, tantrum severity and frequency, anxiety severity, social interaction, and eating behavior. Furthermore, qualitative data was collected through brief weekly interviews with blinded research assistants who were unaware of the experimental condition of the participant. These qualitative data provided further information.

The parents were asked the following questions:

Although tPBM technology is not known to cause any hair or skin disorders, and the same applies to EEG caps, the last open question would have grasped these data.

EEG signals were collected in children before and after each treatment session. Data collection was performed using a Brain Scientific-produced clinical grade device, called NeuroEEG. The device was a thirty-two-channel dry electrode pediatric EEG softcap. The company has since started to manufacture disposable EEG caps, and this particular device has since been discontinued because it is reusable. The EEG was collected only for those children who tolerated wearing the EEG cap and could sit quietly for a period of time.

The last author, a licensed psychologist experienced in the use of CARS in clinical practice, conducted blinded Before and After CARS ratings for all participants. The final CARS evaluations were performed immediately after the final treatment session. CARS scores were analyzed using the independent sample t-test. We compared the differences from the baseline between both groups, calculating both the average difference and 95% CI within each group and between groups.

Responses to the eight weekly questions were scored as either: a positive change representing an improvement with a value of +1, a negative change representing a regression with a value of −1, or a third option without noticeable change with a value of 0. Any appearance of a side effect (e.g., hyperactivity, transient tics) was given a score of 1 if the subject presented with that symptom or a score of 0 if that symptom was not reported. Each subject received two final cumulative scores, representing behavioral changes addressed by the above questionnaire and the appearance of side effects. The scoring of the behavioral data was performed by an analyst (fifth author of the paper) who was fully blinded to the experimental hypothesis and the treatment conditions of the participants. Cumulative scores were compared using a Wilcoxon–Mann–Whitney two-sample rank sum test.

The analytical team, fully blinded to both the experimental hypothesis and the condition of the participants, removed artifacts due to the movement of the participants manually. The frequency power was measured for each participant during the trial time (for those who tolerated wearing the EEG cap). The average power of each wavelength (alpha, beta, theta, and delta) was analyzed. Data analysis was performed using SciPy¹ and Statsmodels.² Matplotlib³ was used to plot the data.

The study employed Mixed Linear Models to analyze the effects of a treatment versus sham on two dependent variables, Net_theta (the average power of theta waves and Net_delta) (the average power of delta waves), across both physical time and a scaled time period.

The scaling of data across was applied to all data, uniformly, for all waves, and it is unlikely that the reported results are achieved solely due to scaling. Net_delta and Net_theta were used as measures, which lead to 48 total number of observations.

After conducting the statistical analysis over actual time, the following scaling was performed: the time scale was standardized so that the first observation for each participant was at point 0 and the last observation was at point 1. The time scaling accounted for individual timelines, with 0 representing a patient’s first day and 1 their last day in the study: Time_scaled = (Time – Time_start) / (Time_end – Time_start).

Additionally, the power of each brain wave (alpha, beta, theta, and delta), measured in percentages of 100, was analyzed over time. EEG power represented the amount of activity in certain frequency bands of the signal (80).

We used linear correlation (Pearson’s R) to test the relationship between the power of theta and delta waves. We also used linear correlation (Pearson’s R) to test the relationship between changes in the power of theta and delta waves and changes in the CARS scores of participants.

There were no predetermined stopping guidelines for the interim analysis of data. In case of adverse effects, the Institutional Review Board would have been notified and, if necessary, the study stopped. There were no changes in the trial results after the start of the study.

Thirty participants completed this trial, according to the original Institutional Review Board approval; 16 participants were in the Active group.

Participants were recruited through social networks, local schools, and centers that provide behavioral therapy to children diagnosed with ASD. “Trialfacts”⁴ assisted with the recruitment. The study began in March 2021. The study ended in October 2021. The recruitment period was approximately 6 months. The trial was completed once the Institutional Review Board-approved 30 participants completed the entire course of the trial. The day the first participant started was 17 March 2021 and the day the last participant completed the study was 28 October 2021.

The study initially recruited 34 participants, but four were lost and replaced throughout the trial. A participant withdrew immediately after the first session, as it appeared that he had an absence seizure. A seizure was never confirmed; this child was randomized to a sham condition. The parents of another participant decided not to continue after one session. One participant was replaced after one session due to their parents’ refusal to cut the child’s long hair. Another participant was replaced after the first session because his mother found it difficult to travel to the experimental site. Figure 2 shows the flow chart of participant selection and recruitment, as well as their attrition during the trial.

Initially, there were 15 participants randomized to the Active and Sham groups. However, due to an experimental error, treatment was mistakenly administered to a replacement participant who was randomized to the Sham group, in the final analysis there were 16 children in the Active group and 14 children in the Sham group.

The experience of the participants under the Sham and Active conditions was identical: The device was placed on their heads and activated by the research assistant. Children were allowed to watch cartoons or YouTube videos or play with their parents or research assistants. EEG was collected from participants in both Active and Sham groups.

At the beginning of the trial, an initial evaluation was performed to assess the severity of ASD in participants using the CARS score. Table 1 displays the mean, median, and range of the participant scores. Most participants had moderate to severe ASD, with a CARS score higher than 35. There were no significant differences in age, gender, ethnicity, language status, treatment status, or CARS scores between the Active and Sham groups.

Every participant who completed the trial was evaluated with CARS post-treatment. There were no missing CARS posttreatment evaluations (see Figure 2).

In the Active group, the before treatment CARS scores had a mean score value of 43.5 (n = 16) with a standard deviation (SD) of 5.7, the after-treatment scores had a mean score of 33.7 with an SD of 5.0. In the Sham group, the before treatment CARS scores had a mean score value of 40.6 (n = 14) with SD of 7.2, the after-treatment scores had a mean score of 38.0 with SD of 8.4. Table 3 shows the mean (M) and SD of the CARS scores for the Active and Sham groups before and after treatment. The difference in the change in CARS between the two groups was 7.23 (95% CI 2.357 to 12.107, p = 0.01).

In addition to analyzing the changes in CARS scores based on averages, we also conducted respondents-based analyses, quantifying the number of participants in Active and Sham groups, whose change in CARS was greater than 4.5 points (as it has been suggested by Jurek et al. (76) to be the minimal clinically important difference). The result was significant X² = 8.48, p = 0.004. By the end of the trial, 87% of the participants in the Active group (14) had achieved an equal or greater than 4.5 point reduction in CARS scores, only 35% of the participants of Sham group (5) achieved an equal or greater than 4.5 point reduction in CARS scores. The number of subjects that achieved this minimal clinically important difference in the Active group was statistically significantly higher compared to the Sham group. Therefore, the treatment was considered to have clinically meaningful efficacy.

Statistical analysis indicates a significant difference between parental ratings of the above eight categories (Secondary Outcomes) of behavior between the Active and Sham groups. The median cumulative changes in behavior [quartile] of the Active and Sham groups, respectively, were 20 [19, 21] and 16 [15.1, 16.9] (Wilcoxon rank sum W = 163.5, nA = 16 nS = 14, p = 0.05 two-tailed, rpb = 0.41).

During the trial, some participants experienced side effects, which were expected based on previous studies on tPBM (40, 81), and had been reported to the Institutional Review Board as a possible side effect before the study. A potential reason for these side effects is the increase in cerebral blood flow because of tPBM treatment. There were no serious adverse events during the study.

Four participants in the Active group exhibited overexcitement during the first 3 weeks of the trial, as noted by their therapists and teachers, who were unaware of their participation in the study. Overexcitement included running around excessively and pounding on the chest in one participant. Two children in the Active group also reported mild headaches, which resolved after several sessions and could have been due to increased cerebral blood flow. A mother gave her child Tylenol to alleviate the discomfort, but the headaches resolved on their own by session 6. One participant in the active condition experienced an increase in transient tics, which were present prior to the study and resolved after completion of the trial. Consistent with previous studies, several participants exhibited increased irritability; however, this research did not measure irritability on a separate scale. CARS score includes dimensions related to irritability (i.e., emotional response). Despite these side effects, all parents of the affected participants chose to continue with the trial as they observed visible improvements in their children’s eye-contact, concentration, and receptive and expressive language.

None of the participants with active or sham conditions reported experiencing skin or hair disorders during or after the trial. The difference in the appearance of side effects between the Active and Sham groups was not significantly different. The median (quartile) of the Active and Sham groups, respectively, were 2.5 [2, 4] and 1.5 [1, 3.8] (Wilcoxon rank-sum W = 207, nA = 16, nS = 14, p = 0.69 two-tailed, rpb = 0.08). All side effects were mild and resolved quickly (Table 4).

Seventeen out of thirty participants tolerated wearing an EEG cap in at least two sessions, and two observations were lost due to experimental error. After data cleaning (removal of motion artifacts), forty-eight observations were available. A mixed linear model was used to analyze changes in delta and theta over scaled time.

We observed trends in delta and theta bandwidths and used Mixed Linear Models to evaluate the effect of a treatment versus sham on two health markers, Net_theta and Net_delta across time. If there were two observations per participant within 1 day, the average power of these two observations was taken. The model includes fixed effects for treatment, scaled time, and their interaction, as well as random effects to capture patient-specific variability. The analysis aims to provide information on both the overall efficacy of the treatment and its interaction with time, considering the differences between individual patients.

We noticed a decrease in the power of delta waves over time in the active group, and the increase in the power of theta waves over time in the active group, prompting us to examine the correlation between CARS and EEG. Our objective was to explore the association between behavioral data (measured by CARS scores) and EEG (measured by delta and theta power over time). This is an exploratory study, no corrections were made for multiple comparisons, and crude p-values were reported. Tables 5, 6 present the EEG analysis of Net_delta values and Net_theta using mixed linear models through physical time.

While changes over time were observed in Net_delta (Coefficient = −0.141, 95% CI -0.258 to −0.023, p = 0.02), no trend in differences between the Active group and the Sham group could be identified.

To ensure that data from all participants had equal weight in the analysis, the time scale was standardized so that the first observation for each participant was at point 0 and the last observation was at point 1. The time scaling accounted for individual timelines, with 0 representing a patient’s first day and 1 their last day in the study: Time_scaled = (Time – Time_start) / (Time_end – Time_start).

Table 7 presents the EEG analysis of Net_delta values across scaled time using mixed linear models.

No significant difference was observed between conditions (Active vs. Sham) for Net_delta (Coefficient = −4.261, 95% CI -12.542 to 4.020, p = 0.313), while we observed a significant effect with Scaled Time (Coefficient = −6.713, 95% CI -12.520 to −0.907, p = 0.023). An interaction between Active vs. Sham and Scaled Time was noted (Coefficient = 7.521, 95% CI -0.517 to 15.559, p = 0.067), suggesting a potentially greater decrease in delta power in the Active group over time compared to the Sham group (Figures 3A,B).

Table 8 presents the EEG analysis of Net_theta values across scaled time using mixed linear models.

There was a significant difference observed between conditions (Active vs. Sham) for Net_theta (Coefficient = 9.547, 95% CI 0.027 to 19.067, p = 0.049), and a significant effect with Scaled Time (Coefficient = 6.429, 95% CI -0.020 to 12.877, p = 0.051). A marginal interaction between Active vs. Sham and Scaled Time was noted (Coefficient = −8.287, 95% CI –17.199 to 0.26, p = 0.068), suggesting a potentially greater increase in theta power in the Active group over time compared to the Sham group (Figures 4A,B).

The analysis revealed a significant positive correlation between the change in CARS scores and the change in delta waves, indicating that participants who had a reduction in CARS scores also had a reduction in their delta waves (r = 0.62, p = 0.008, n = 17) (see Figure 5A). We also found a significant negative correlation between the power of theta waves and the change in the CARS scores (r = −0.66, p = 0.004, n = 17) (Figure 5B); this indicates that participants who had a decrease in CARS scores experienced an increase in the power of their theta waves (n = 17). These findings suggest a possible relationship between behavioral symptoms (measured by CARS) and the distribution of brainwaves. We also found a negative correlation between the change in the power of delta waves and the change in the power of theta waves (r = −0.52, p = 0.03, n = 17) (Figure 5C).

In addition, we analyzed the changes in the power of the ratios of brainwaves over the course of the trial. There were no significant changes.