Prediction of Autism at 3 Years from Behavioural and Developmental Measures in High-Risk Infants: A Longitudinal Cross-Domain Classifier Analysis.
G Bussu, E J H Jones, T Charman and 2 others
PMID 29453709WHAT IT FOUND
At the second visit, daily living scores gave a modest signal for later autism in high-risk infants, but early measures did not predict individual outcome.
The prediction model was much better at saying an infant would not develop autism than at identifying who would.
Key findings
01Infants who later developed ASD or atypical outcomes showed slower or decreasing developmental trajectories than low-risk and high-risk typical infants, with differentiation becoming significant around 14 months.
02For predicting later ASD in high-risk siblings, classifiers at 8 months were at chance level; at 14 months, daily living scores had the best performance (AUC = 71.3%), but positive predictive value was 28.5%.
03For predicting broader atypical development versus typical development, the best 14-month classifier combined Vineland adaptive scores and AOSI scores, but its advantage over other classifiers was not significant after correction.
STILL TO COME
How it was doneWhat they foundWhat it means for PTsWhat it means for OTsWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The sample was enriched for autism because all high-risk infants had an older sibling with autism, so the prediction results may not apply to infants in the general population. The low-risk group was not fully diagnosed: in Phase 1 low-risk infants did not receive an outcome evaluation, and no formal clinical diagnoses were assigned to low-risk infants. Missing data were handled with imputation, and some infants were excluded because they did not receive all evaluations or visits. Statistical power was limited; differences between the best classifier and other classifiers did not survive correction for multiple comparisons, and the authors called for replication in larger samples. The atypical outcome group was defined by test cut-offs and was highly variable, so it may not correspond to a single clinical condition. Experimenters knew infants' familial risk status, although assessments were blind to clinical outcome, so risk information could have influenced interpretation. The supplied text gives different AUC values for the best broader-atypical classifier (71.8% in the narrative and 70.8% in Table 2), so the exact performance is unclear.
Declared interests
The supplied text lists funding from the H2020 Marie Skłodowska-Curie Actions. No other conflict-of-interest declaration is given.
The easy way to misread this
Do not use these scores as a stand-alone test to predict autism in an individual infant. At 8 months the ASD prediction scores were at chance level, and of the infants flagged by the best 14-month ASD classifier, only 28.5% actually developed autism.