Abstract

Exaggerating or faking the symptoms of a mental health disorder can confound structured diagnostic interviews and hinder clinical psychiatric assessments (1, 2). We introduce an artificial intelligence (AI) framework for detecting symptom fabrication in mental health assessments, illustrated here for Post-Traumatic Stress Disorder (PTSD) diagnoses, for which malingering is a known problem (3, 4), partially ascribable to the potential for secondary financial gain from positive diagnoses. Algorithm VeRITAS employs novel generative AI to infer statistical dependencies inherent in true response patterns, and flags responses which violate these subtle constraints. With a study sample of \(n=651\) patients, VeRITAS is estimated to have an Area Under the Curve (AUC) of \(\geq 0.95\pm 0.02\), with sensitivity \(> 95\%\), specificity \(>88\%\) respectively, and positive likelihood ratios between \(9.9 - 19.77\) achievable based on the population prevalence of malingering in the context of PTSD diagnosis. We show that in our methodology having training in forensic psychiatry, or other relevant mental health experience, is not helpful in deceiving the algorithm. Our tool offers an objective, disease-specific, fast (average time \(\leq 4\) min) approach to detect fake PTSD, and if adopted, can ensure that healthcare resources and disability concessions reach those genuinely in need, while helping to maintain integrity of clinical data. Moreover, the ability to identify and help patients who might be malingering due to other mental health conditions, poverty or socio-economic compulsions can improve general health outcomes in disadvantaged communities.

Software

truthnet: Python package for analysis of arbitrary response data using our approach