Pharmacoutilization data-driven artificial intelligence-assisted diagnosis algorithm to improve the pharmacological treatment of pain and agitation in patients suffering from severe dementia.

The number of diagnoses and drug prescriptions for dementia patients is poorly available. Delay in the diagnosis of Alzheimer's disease (AD), for which missed diagnoses amount to over half cases, and undertreatment of chronic and neuropathic pain mirroring excessive use of harmful antipsychotics and antidepressants is reported. Our study aimed at diagnosing AD through the most advanced artificial intelligence (AI) methodologies even in patients who escaped clinical observation. To this end, pharmacoepidemiology data collected as part of the first retrospective community study in a wide sample of 298,000 individuals, 84,235 aged over 60 years, were used to set up an AI algorithm for the rescue of missed diagnoses of AD. The core of the algorithm consisted in the management of time series represented by pharmacological therapies through a distance matrix and in the use of autoencoders. Patients without a diagnosis of AD based on pharmacotherapy were 114.920, while diagnosed patients were 1.150, mainly aged between 75 and 84 years, pointing at late start of treatment. Increased use of antidepressants, neuroleptics, and mood stabilizers is found in patients treated with acetylcholinesterase inhibitors (AChEIs) and memantine, while nonsteroidal anti-inflammatory drugs, paracetamol-codeine and opioids are mostly prescribed to patients not receiving AChEIs and memantine. The classification model demonstrated good global accuracy at the end of training, equal to 79.12%. Further studies and longitudinal monitoring of patients are needed to improve disease detection and management. The deep learning-based pharmacoutilization algorithm generated in the present study will aid the diagnosis of AD and the understanding of neuropsychiatric symptoms treatment.