Logistic regression beats transformers in long-term depression prediction
No Transformer Beats Six Covariates: Long-Horizon Prediction of Depressive Symptoms from Childhood Essays
Computation and LanguageArtificial Intelligence
Summary
This paper looks at whether advanced AI models called transformers can predict depression from childhood essays written 12 years earlier. It turns out a simple logistic regression model using just six childhood factors predicts adult depressive symptoms better than any of the tested text-based transformers. Adding the essay information from transformers doesn't improve the prediction. The authors show that, for predicting depression far in the future, simple models with key personal data can outperform complex AI text models.
What this means in practice
- •For clinical data analysts: Use childhood demographic and health variables to better predict adult depression risk than current text-based AI models.
- •For mental health screening teams: Prioritize simple structured data over complex language models for long-term depression risk identification in youth cohorts.
Tested on one dataset.
Authors
Daniel Kua, Emrul Hasan, John-Jose Nunez, Frances Chen
Abstract
Natural language processing (NLP) models can detect depression-related language in text written near the time symptoms are measured, but whether pretrained transformers can predict depressive symptoms from text written twelve years earlier is largely untested. In the National Child Development Study, a British birth cohort, we predict probable depressive symptoms at age 23 from essays the same people wrote at age 11. Our baseline, a logistic regression on six childhood covariates, outperforms every text model that sees only the essay: seven fine-tuned transformers, a bag-of-words model, frozen embeddings and four zero-shot large language models. Its area under the receiver operating characteristic curve (AUC-ROC) is 0.737 against 0.670 for the best transformer on the primary seed, and no added text score detectably raises the baseline's AUC-ROC. None of the five domain-pretrained transformers detectably beats its general-domain control after Bonferroni correction. For long-horizon prediction, the baseline remains the model to beat.