Comparing Clinician and Machine Learning Predictions for Self-Injury Treatment
Author Information
Author(s): Moa Pontén, Oskar Flygare, Martin Bellander, Moa Karemyr, Jannike Nilbrink, Clara Hellner, Olivia Ojala, Johan Bjureberg
Primary Institution: Karolinska Institutet
Conclusion
Both clinicians and a machine-learning algorithm showed similar prediction accuracy for adolescents' self-injury treatment outcomes, but further research is needed to improve these predictions.
Supporting Evidence
- Clinician predictions had an accuracy of 63%, while the machine learning model had an accuracy of 67%.
- Emotion dysregulation was identified as the most important predictor of treatment outcome.
- Both prediction methods performed better than a baseline model that predicted all patients would reach NSSI absence.
Takeaway
This study looked at how well doctors and computers can guess if teens will stop hurting themselves after treatment. They found that both did about the same job, but there's still room to get better.
Methodology
The study compared clinician predictions with a machine learning model using data from a randomized controlled trial on internet-delivered therapy for adolescents with nonsuicidal self-injury.
Potential Biases
The study may have biases due to the reliance on clinician predictions and the potential for overfitting in the machine learning model.
Limitations
The sample size was limited, which may affect the generalizability of the results, and the outcome measures differed between groups.
Participant Demographics
{"gender_distribution":{"female":126,"male":7,"non-binary":5},"age_mean":15.03,"age_sd":1.24}
Statistical Information
P-Value
0.205
Confidence Interval
95% CI 0.41 to 0.58
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website