Lightweight Language Models are Prone to Reasoning Errors for Complex Computational Phenotyping Tasks · Activ.news