BRENNEN

Industries / Health / Healthcare / Behavioral health / Autism therapy

Working with families

Working with families covers how programs treat parents and other caregivers, from the 1960s theory that blamed mothers for autism to programs in which professionals coach parents to support their child's communication and behavior at home.

7

Milestones achieved, from 1971 to 2022.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1971

    Parents become co-therapists

    B. Bettelheim, following a psychoanalytic model, argued in 1967 that autism came from emotionally cold mothers and treated children by removing them to a residential program, and many parents later reported that this harmed their children, by S. L. Odom and colleagues' account. At a 1968 colloquium at Indiana University Medical School, published in 1971, Eric Schopler presented work that actively involved parents in their children's learning; his 1971 paper with R. J. Reichler is titled "Parents as cotherapists in the treatment of psychotic children". The work grew into TEACCH, now the TEACCH Autism Program of the University of North Carolina at Chapel Hill, a statewide service founded in 1972 that runs seven regional centers in North Carolina.

    Lessona theory that blamed mothers led to children being taken from home; the programs that followed made parents partners.

    pmc.ncbi.nlm.nih.govpubmed.ncbi.nlm.nih.govteacch.compubmed.ncbi.nlm.nih.gov

    Credit

    Eric Schopler and R. J. Reichler; TEACCH

  2. 1993

    Don't mourn for us

    At the 1993 International Conference on Autism in Toronto, the autistic speaker Jim Sinclair addressed parents directly. The outline, published in Autism Network International's newsletter Our Voice, says "Autism is a way of being. It is not possible to separate the person from the autism", and asks parents to separate their grief for the "normal child" they had expected from the autistic child they have. ASAN, run by and for autistic people, counts non-autistic family members among its supporters and argues that services must aim at the autistic person's own quality of life.

    autreat.comautisticadvocacy.orgautisticadvocacy.org

    Credit

    Jim Sinclair, Autism Network International; ASAN

  3. 2010

    Parent coaching, long follow-up

    The Preschool Autism Communication Trial (PACT), a parent-mediated program focused on communication, enrolled 152 children aged 2 to 4 at three UK centers. At 13 months autism symptoms had fallen by 3.9 points against 2.9 with usual care, too small a difference for the authors to recommend it for symptoms, though communication between parent and child improved. At follow-up a median 5.75 years later, with 121 of the 152 children assessed at about age 10.5, the authors found a reduction in symptom severity over the whole period (effect size 0.55), which they describe as the first long-term symptom reduction shown after a randomized trial of early autism intervention.

    Lessona trial that missed its main outcome at 13 months found an effect years later, so short trials can miss slow effects.

    pubmed.ncbi.nlm.nih.govpubmed.ncbi.nlm.nih.gov

    Credit

    Jonathan Green, A. Pickles, T. Charman, H. McConachie and the PACT Consortium (London, Manchester and Newcastle)

  4. 2015

    Parent training beats education

    At six US universities, 180 children aged 3 to 7 with autism and behavior problems were randomly assigned to 24 weeks of parent training in managing behavior (11 core sessions, telephone boosters and two home visits) or to parent education about autism. Parent-rated irritability fell 47.7% against 31.8%, and a clinician who did not know the groups rated 68.5% of the trained families' children improved against 39.6%; neither parent measure reached the change the team had set in advance as clinically important. The children were 86% white and 88% male.

    pubmed.ncbi.nlm.nih.gov

    Credit

    K. Bearss, L. Scahill and colleagues at Emory, Indiana, Ohio State, Pittsburgh, Rochester and Yale universities

  5. 2016

    Health workers deliver it

    PASS adapted the PACT approach for India and Pakistan, and non-specialist health workers gave 12 sessions to the families of 65 children aged 2 to 9 in Goa and Rawalpindi. At 8 months parents' responsiveness to their children (effect size 1.61) and children's starting of communication (0.99) had improved over usual care, but time in shared attention fell (−0.70), a negative effect not reported before.

    Lessonhanding the work to non-specialists held up on the main outcome, but one measure got worse, so each adaptation needs its own trial.

    pubmed.ncbi.nlm.nih.gov

    Credit

    A. Rahman, G. Divan, V. Patel, J. Green and colleagues; funded by Autism Speaks

  6. 2021

    Support before a diagnosis

    In Perth and Melbourne, 104 infants aged 9 to 14 months who showed early behaviors linked to autism were randomly assigned to usual care or to usual care plus iBASIS-VIPP (Video Interaction to Promote Positive Parenting), 10 sessions over five months for their parents. At age 3, 3 of 45 children (6.7%) who had the program were classified as autistic against 9 of 44 (20.5%) with usual care, and parents in the program were more responsive; the analyses were preregistered and used one-sided tests.

    pubmed.ncbi.nlm.nih.gov

    Credit

    A. J. O. Whitehouse, CliniKids, Telethon Kids Institute, with K. Hudry, Jonathan Green and colleagues

  7. 2022

    Free training from WHO

    The World Health Organization (WHO) published its Caregiver Skills Training (CST) package on 27 April 2022 for families of children with developmental delays or disabilities, including autism: nine group sessions and three home visits, designed to be given by nurses, community workers or caregivers rather than specialists, with an online version for caregivers. A 2026 review of 17 studies from different countries found it adapted with materials translated, simplified and paired with pictures for parents with lower literacy, and noted that it does not require a formal diagnosis.

    who.intpmc.ncbi.nlm.nih.gov

    Credit

    WHO; review by C. Montiel-Nava and colleagues

Watch for

  1. An online parent program

    ParentWorks-Spectrum gives parents of children aged 2 to 5 with autism and behavioral difficulties 12 to 15 weekly online sessions of 60 to 90 minutes on child behavior, social communication, and parents' teamwork and mental health. In a trial of 100 families against a waiting list, the team reported moderate to large gains on three of four main measures, held at three months.

    Achieved whena team outside the authors' group repeats the result against an active comparison rather than a waiting list.

    pubmed.ncbi.nlm.nih.gov

    Claimed by

    M. R. Dadds and colleagues, Sydney, August 2026

  2. A parent chatbot rated safe

    Hazel, a chatbot built by the company Sylva on a large language model for parents and carers of autistic and neurodivergent children, gave answers to 400 parent questions that evaluators rated safe, empathetic and useful, and flagged the safeguarding issue in every test question that contained one; one evaluator told its answers from clinicians' answers 74% of the time. The answers were scored by members of the author team.

    Achieved whenevaluators outside the authors' group score it on a published question set, or a trial with parents measures what it changes.

    nature.com

    Claimed by

    Freddy Jackson Brown and colleagues, University of Warwick, April 2026

Open

  1. 01

    Do-not-use treatments, chatbot test

    A public, rerunnable test of whether AI chatbots steer parents toward autism treatments that guidelines advise against. About 200 questions, written as a worried parent would ask them and in several languages, cover NICE's "do not use" list (secretin, chelation and hyperbaric oxygen therapy; neurofeedback and auditory integration training for speech; antipsychotics, antidepressants, anticonvulsants or exclusion diets for core features), facilitated communication, how many hours of ABA a child needs, and the debate over ABA. Each answer is scored for whether it endorses a treatment on the list, whether it states the evidence correctly, and whether it gives both sides of the ABA debate.

    The brief
    Credit · unclaimed
  2. 02

    Adult services, answered by state

    A public answer key for families planning an autistic teenager's move to adult services. For each of 10 US states, write 20 fictional family profiles (age, school status, household income, support needs) and list the programs each family could apply for, such as Medicaid waivers, state vocational rehabilitation and Supplemental Security Income, with their waiting lists, taken from public program rules and checked by two people who know each state's system. Then score AI assistants and search tools against the key.

    The brief
    Credit · unclaimed

Open 01 · the brief

Do-not-use treatments, chatbot test

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public, rerunnable test of whether AI chatbots steer parents toward autism treatments that guidelines advise against. About 200 questions, written as a worried parent would ask them and in several languages, cover NICE's "do not use" list (secretin, chelation and hyperbaric oxygen therapy; neurofeedback and auditory integration training for speech; antipsychotics, antidepressants, anticonvulsants or exclusion diets for core features), facilitated communication, how many hours of ABA a child needs, and the debate over ABA. Each answer is scored for whether it endorses a treatment on the list, whether it states the evidence correctly, and whether it gives both sides of the ABA debate.

Why it is still open

Studies so far score general autism answers once. T. C. McFayden and colleagues at the University of North Carolina (2024) found ChatGPT-4 largely correct on autism myths and misconceptions but with inaccurate references and links; Akhund and colleagues (2026) put 24 questions to three models in Saudi Arabia, Rakap and colleagues (2025) put 44 questions to three tools in English and Turkish, and Nădășan and colleagues (2026) scored five chatbots against a 24-item checklist; the Hazel study above scored one chatbot. None published a question set built around treatments that guidelines advise against, or reran it as the models changed.

pubmed.ncbi.nlm.nih.govpubmed.ncbi.nlm.nih.govpubmed.ncbi.nlm.nih.govpubmed.ncbi.nlm.nih.govnice.org.uk

Built with what exists

NICE guideline CG170 and ASHA's 2018 position statement on facilitated communication as the answer key; Inspect, the open-source evaluation framework from the UK AI Security Institute and Meridian Labs; current chatbots through their public apps or programming interfaces; native-speaker reviewers for each language.

  1. First step

    Write 30 questions from NICE's "do not use" list and run them on three free chatbots.

  2. Done looks like

    200 questions in at least three languages run on at least six chatbots; two raters agree on at least 90% of "endorses a do-not-use treatment" calls; endorsement rates published per model and per language and rerun every three months for a year; questions, answers and scores public.

Open 02 · the brief

Adult services, answered by state

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public answer key for families planning an autistic teenager's move to adult services. For each of 10 US states, write 20 fictional family profiles (age, school status, household income, support needs) and list the programs each family could apply for, such as Medicaid waivers, state vocational rehabilitation and Supplemental Security Income, with their waiting lists, taken from public program rules and checked by two people who know each state's system. Then score AI assistants and search tools against the key.

Why it is still open

In a trial of 185 parents of transition-age autistic youth, a group advocacy class did not raise service access overall compared with comprehensive written information about adult services (J. L. Taylor and colleagues, 2025). Ahmadi and colleagues (2026) built and tested an AI retrieval system for autism service coordination in Alabama, reporting 90% to 96% precision and recall on simulated social-worker queries; it covers one state and service searches, not eligibility for adult programs across states. Searches of Europe PMC and GitHub in November 2026 found no multi-state answer key.

pubmed.ncbi.nlm.nih.govjournals.plos.org

Built with what exists

public program rules from state Medicaid waiver pages, state vocational rehabilitation agencies and the Social Security Administration; a retrieval system built with open tools such as LlamaIndex over those documents; current chatbots for comparison; the answer key as a spreadsheet on Zenodo.

  1. First step

    Write five fictional profiles for one state and build their answer key from that state's waiver and vocational rehabilitation pages.

  2. Done looks like

    200 fictional profiles across 10 states, each answer signed off by two reviewers; each tool's recall and precision for eligible programs published; the key rechecked against program rules every six months.