Algorithmic Assessments in Deciding on Voluntary, Assisted or Involuntary Psychiatric Treatment.

Abstract

The challenges in assessing whether psychiatric treatment should be provided on voluntary, assisted or involuntary legal bases prompted the development of an assessment algorithm that may aid clinicians. It comprises a part that assesses the incapacity to provide informed consent to treatment, care or rehabilitation. It also captures the patient’s willingness to receive these treatments, the risk posed to the patient’s health or safety, financial interests or reputation and risks of serious harm to self or others. By following various decision paths, the algorithm yields one of four legal states: a voluntary, assisted, or involuntary state or that the proposed intervention should be declined. This study examined the predictive validity and the reliability of this algorithm. It was applied 4052 times to 135 clinical case narratives by 294 research participants. The legal states yielded by the algorithm had high statistical significance when matched with the gold standard (Chi-squared = 6963; df = 12; p < 0.001). It was accurate in yielding the correct legal state for the voluntary, assisted, involuntary and decline categories in 94%, 92%, 88% and 86% of the clinical case narratives, respectively. For internal reliability, a correspondence model accounted for 99.8% of the variance by which the decision paths clustered together fittingly with each of the legal states. Inter-rater reliability testing showed a moderate degree of agreement among participants on the suitable legal state (Krippendorff’s alpha = 0.66). These results suggest the algorithm is valid and reliable, which warrant a subsequent randomised controlled study to investigate whether it is more effective in clinical practice than standard assessments.