The PainChek® Pain Assessment Tool: Harnessing AI and Reducing Subjectivity to Assess Pain in People with Dementia

TL;DR

PainChek® is a new AI-based tool that objectively assesses pain in people with dementia, showing high agreement and reliability compared to a traditional pain scale.

Contribution

PainChek® introduces an objective AI-based method for pain assessment in dementia patients, reducing subjectivity.

Findings

PainChek® showed high agreement with the Abbey Pain Scale, especially for no pain, moderate, and severe pain categories.

PainChek® outperformed the Abbey Pain Scale in test-retest reliability with an intraclass correlation of .73.

Predictive values for PainChek® were high for no pain and moderate pain but lower for severe pain due to fewer severe cases.

Abstract

Pain assessment in people with dementia is challenging and can result in inadequate treatment. Most pain assessment tools require subjective quantification of pain/discomfort. PainChek® is a new assessment tool that harnesses a) artificial intelligence to provide objective pain ratings of facial expression and b) dichotomous ratings of observable non-facial pain indicators. This study aimed to validate PainChek® by examining the agreement, reliability, and predictive validity between PainChek® and the Abbey Pain Scale (APS), a more subjective pain scale commonly used in Australia. Participants (N = 103) were US nursing home residents with moderate-to-severe cognitive impairment. Participants were assessed for pain during rest and post-movement by two blinded raters, each administering one pain assessment tool. Agreement between raters was 92.7% (no pain), 69.1% (mild), 75% (moderate), and 83.3% (severe), suggesting high levels of agreement between PainChek® and APS assessments. Agreement rates, except in the mild condition, exceeded a priori hypotheses and were similar across rest/post-movement conditions. Intraclass correlations show that PainChek® (.73, 95% CI: .62-.81) outperformed APS (.59, 95% CI: .44-.71) on test-retest reliability. Bootstrapping methods yielded predictive values that were 75.5%, 89%, 82.8%, and 38.3% for no pain, mild, moderate, and severe. Low predictive value for severe pain was likely due to few occurrences of severe pain assessments. Results suggest that PainChek® performs on par with the APS, while providing a more objective format that enables greater accessibility for a variety of staff to administer reliably. Decreasing subjectivity from the pain assessment process may facilitate accuracy and lead to more appropriate treatment.