Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2023 Oct 21;13(10):1513.
doi: 10.3390/jpm13101513.

Predicting Problematic Behavior in Autism Spectrum Disorder Using Medical History and Environmental Data

Jennifer Ferina  1   2 , Melanie Kruger  2   3 , Uwe Kruger  1 , Daniel Ryan  4 , Conor Anderson  4 , Jenny Foster  4 , Theresa Hamlin  4 , Juergen Hahn  1   2   5
Affiliations

Predicting Problematic Behavior in Autism Spectrum Disorder Using Medical History and Environmental Data

Jennifer Ferina et al. J Pers Med. .

Abstract

Autism spectrum disorder (ASD), characterized by social, communication, and behavioral abnormalities, affects 1 in 36 children according to the CDC. Several co-occurring conditions are often associated with ASD, including sleep and immune disorders and gastrointestinal (GI) problems. ASD is also associated with sensory sensitivities. Some individuals with ASD exhibit episodes of challenging behaviors that can endanger themselves or others, including aggression and self-injurious behavior (SIB). In this work, we explored the use of artificial intelligence models to predict behavior episodes based on past data of co-occurring conditions and environmental factors for 80 individuals in a residential setting. We found that our models predict occurrences of behavior and non-behavior with accuracies as high as 90% for some individuals, and that environmental, as well as gastrointestinal, factors are notable predictors across the population examined. While more work is needed to examine the underlying connections between the factors and the behaviors, having reasonably accurate predictions for behaviors has the potential to improve the quality of life of some individuals with ASD.

Keywords: autism spectrum disorder; challenging behavior; machine learning.

PubMed Disclaimer

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Figure 1
Figure 1
Inclusion criteria for each behavior cohort of the study (aggressive behavior—AGG; self-injurious behavior—SIB; BOTH refers to AGG and SIB). A formal ASD diagnosis upon entry to the center excluded the most individuals, while time periods with little data also excluded a high number of individuals. Allowing either behavior provided more data for individuals, so fewer participants were eliminated in the BOTH cohort, while the smallest cohort was the SIB group. However, individuals with one behavior that was much more frequent than the other were excluded from the BOTH cohort when there were few examples of no behavior.
Figure 2
Figure 2
The filtered behavior patterns over time for each individual and for all behaviors examined. Behavior patterns differ largely; while some people typically have a behavior on most days, others rarely exhibit behaviors. Thus, class imbalance is of concern when predicting individuals’ behaviors. Subfigure (a) refers to individuals with aggressive behavior, (b) refers to individuals with SIB, and (c) refers to individuals with either aggressive behavior or SIB.
Figure 2
Figure 2
The filtered behavior patterns over time for each individual and for all behaviors examined. Behavior patterns differ largely; while some people typically have a behavior on most days, others rarely exhibit behaviors. Thus, class imbalance is of concern when predicting individuals’ behaviors. Subfigure (a) refers to individuals with aggressive behavior, (b) refers to individuals with SIB, and (c) refers to individuals with either aggressive behavior or SIB.
Figure 3
Figure 3
Times of data collection from Day 0 to Day 1, starting on 07/01/2015 as an example. The behavior day starts at 7 a.m., while the weather and BM data are both defined from midnight to midnight. While the moon data also has midnight recorded as a day start, the moon data were taken at 7 a.m. to align with the behavior data collection start. The behavior data from 07/01 7 a.m.–07/02 7 a.m., sleep data starting the evening of 07/01 to morning of 07/02, BM data starting at midnight 07/01 to midnight 07/02, moon data midnight 07/01 to midnight 07/02 collected at 7 a.m. 07/01, and weather data midnight 07/01 to midnight 07/02 are used to predict behavior 07/02–07/03 starting at 7 a.m.
Figure 4
Figure 4
Linear results followed by direct kernel results for all individuals and cohorts: AGG at the top, SIB in the middle, and BOTH at the bottom. For all cohorts, the median linear model performance is close to 50%, indicating that other methods must be explored to achieve useful predictions. In almost all cases, the median BCR for the direct kernel approach is above 50%, and for many individuals the predictions are substantially better.
Figure 4
Figure 4
Linear results followed by direct kernel results for all individuals and cohorts: AGG at the top, SIB in the middle, and BOTH at the bottom. For all cohorts, the median linear model performance is close to 50%, indicating that other methods must be explored to achieve useful predictions. In almost all cases, the median BCR for the direct kernel approach is above 50%, and for many individuals the predictions are substantially better.
Figure 4
Figure 4
Linear results followed by direct kernel results for all individuals and cohorts: AGG at the top, SIB in the middle, and BOTH at the bottom. For all cohorts, the median linear model performance is close to 50%, indicating that other methods must be explored to achieve useful predictions. In almost all cases, the median BCR for the direct kernel approach is above 50%, and for many individuals the predictions are substantially better.
Figure 5
Figure 5
The sensitivity (left), specificity (middle), and BCR (right) for each model compared to the percentage of behavior occurring in the case of sensitivity and specificity, and the maximum class imbalance for BCR. Individuals with an average BCR above 80% are shown in black. All three cohorts (AGG on top, SIB in the middle, and BOTH at the bottom) follow similar trends for sensitivity, specificity, and balanced accuracy. Individuals with a smaller amount of class imbalance, i.e., they either exhibit a somewhat similar number of days with as well as without behaviors and show a higher BCR relative to their average class imbalance, while individuals with a higher amount of class imbalance are typically predicted poorly.
Figure 6
Figure 6
Plots of feature importance across all individuals show that BM, moon, sleep, and weather features are represented in the sensitivity analysis for the most individuals, while the allergen and menses features are less represented, except in the case of menses for the SIB cohort. Note that the results for menses, marked with an asterisk, are only calculated for the number of female individuals in each cohort.
Figure 6
Figure 6
Plots of feature importance across all individuals show that BM, moon, sleep, and weather features are represented in the sensitivity analysis for the most individuals, while the allergen and menses features are less represented, except in the case of menses for the SIB cohort. Note that the results for menses, marked with an asterisk, are only calculated for the number of female individuals in each cohort.
Figure 7
Figure 7
Hierarchical clustering of the data for the individuals where the balanced accuracy exceeds 80%. The clustering results for the entire population are very similar to the one for the individuals where high prediction accuracy was possible (results not shown).
See this image and copyright information in PMC

Similar articles

See all similar articles

References

    1. Maenner M.J., Warren Z., Robinson Williams A., Amoakohene E., Bakian A.V., Bilder D.A., Durkin M.S., Fitzgerald R.T., Furnier S.M., Hughes M.M., et al. Prevalence and Characteristics of Autism Spectrum Disorder Among Children Aged 8 Years—Autism and Developmental Disabilities Monitoring Network, 11 Sites, United States, 2020. MMWR Surveill. Summ. 2023;72:1–14. doi: 10.15585/mmwr.ss7202a1. - DOI - PMC - PubMed
    1. Faras H., Al Ateeqi N., Tidmarsh L. Autism spectrum disorders. Ann. Saudi Med. 2010;30:295–300. doi: 10.4103/0256-4947.65261. - DOI - PMC - PubMed
    1. Dominick K.C., Davis N.O., Lainhart J., Tager-Flusberg H., Folstein S. Atypical behaviors in children with autism and children with a history of language impairment. Res. Dev. Disabil. 2007;28:145–162. doi: 10.1016/j.ridd.2006.02.003. - DOI - PubMed
    1. Tye C., Runicles A.K., Whitehouse A.J.O., Alvares G.A. Characterizing the Interplay Between Autism Spectrum Disorder and Comorbid Medical Conditions: An Integrative Review. Front. Psychiatry. 2018;9:751. doi: 10.3389/fpsyt.2018.00751. - DOI - PMC - PubMed
    1. Vargason T., Grivas G., Hollowood-Jones K.L., Hahn J. Towards a Multivariate Biomarker-Based Diagnosis of Autism Spectrum Disorder: Review and Discussion of Recent Advancements. Semin. Pediatr. Neurol. 2020;34:100803. doi: 10.1016/j.spen.2020.100803. - DOI - PMC - PubMed

Grants and funding

This research received no external funding.

LinkOut - more resources