Preliminary Evaluation of the Engagement and Effectiveness of a Mental Health Chatbot

Key Findings

• — Analyzed real-world outcome data from 3,629 Vitalk users across three program tracks (anxiety, low mood, stress)
• — Found reductions in anxiety, depression, and stress symptom scores associated with program engagement, using Tobit regression to account for pre-intervention severity
• — Documents a built-in risk-detection module: when a user's free-text input contains language directly or indirectly linked to suicidality (e.g. 'life isn't worth living'), the module triggers and the chatbot delivers crisis information, including national suicide-line details and, where appropriate, a follow-up conversation with a Vitalk healthcare professional
• — Vitalk is built with a preventative mental-health design (not solely crisis-response), delivered as short daily conversational sessions over a 90-day, three-phase program

Methodology Notes

Observational real-world data analysis (not a randomized controlled trial), 3,629 users, symptom questionnaires (anxiety/depression/stress) at baseline and end of each 30-day phase, Tobit regression modeling. Developed and evaluated by a Brazil-based team (TNH Health, São Paulo).

Sources