Neuropsychiatric lupus (NPSLE) is characterized by inflammation in the brain with common symptoms of depression and anxiety. Early detection is crucial as it may change the treatment regimen; however, current approaches are costly and resource intensive. Therefore, we propose that leveraging current work using linguistics in NLP detection of mental health symptoms can be advantageous in early detection of NPSLE. This study is a proof-of-concept using 20 interviews from N=20 adolescents (10-17 years) diagnosed with Lupus. Our results suggest that linguistic feature-based models supported by Word2Vec embeddings offer an interpretable output compared with BERT models, while maintaining competitiveness in depression, and improvement over BERT in anxiety detection. This work may transform early screening methods in paediatric contexts and can be adapted to other clinical populations.