AI-generated research brief — verify at source
ML-Driven PSA Screening Recommendations Reduce Decisional Conflict in Older Men
Men who received personalized machine learning–generated PSA screening recommendations reported meaningfully lower decisional conflict than controls, with a mean score difference of -3.77 points on the Decisional Conflict Scale (95% CI -5.55 to -1.99; Cohen d = -0.44; P < .001). Decision satisfaction was also significantly higher in the ML recommendation group, and screening acceptance rates tracked closely with the direction of the ML suggestion.
What Was Studied
This study examined whether integrating a personalized, machine learning–generated recommendation into a shared decision-making workflow could reduce the decisional conflict experienced by middle-aged and older men when considering PSA screening. The primary outcome was decisional conflict, assessed using the Decisional Conflict Scale, with secondary outcomes including state anxiety and satisfaction with the screening decision. PSA screening is an especially suitable target for this approach because it involves nuanced trade-offs between the benefits of early cancer detection and the risks of overdiagnosis, which are particularly difficult for older adults with limited health literacy, sensory or cognitive impairments, or multiple comorbidities to navigate.
How It Was Studied
The research used a two-stage design. In the first stage, data from 507 men were used to train and evaluate six machine learning algorithms on clinical and values-clarification data; a random forest model was selected based on its superior discriminatory performance (AUC 0.933, 95% CI 0.902–0.963). In the second stage, a randomized controlled trial enrolled 367 men aged 50 years and older (mean age 64.34, SD 10.30 years), allocated 1:1 to the ML suggestion group (MLSG; n = 185) or a control group (CG; n = 182). Both groups received video-based education, counseling, and values clarification; only the MLSG additionally received a personalized ML-generated “second opinion” recommendation regarding whether to accept or defer PSA screening at this time.
What Was Observed
- Decisional conflict was significantly lower in the MLSG, with the total Decisional Conflict Scale score approximately 3.8 points lower than in the control group (MD -3.77, 95% CI -5.55 to -1.99; Cohen d = -0.44; P < .001). Men in the MLSG also reported greater perceived support, more adequate advice, and higher decision confidence on individual DCS items.
- Decision satisfaction was substantially higher in the ML group across all measured items, with a total Satisfaction with Decision score difference of -7.38 points in favor of the MLSG (95% CI -8.54 to -6.18; P < .001), indicating a clinically meaningful improvement in how men felt about their final screening choice.
- Targeted emotional benefits were observed, even though overall anxiety scores did not differ between groups. MLSG participants reported meaningfully reduced worry (STAI item 6: MD -0.98, 95% CI -1.20 to -0.76; Cohen d = -0.89; adjusted P < .001) and modestly increased calmness (STAI item 1: MD 0.30, 95% CI 0.06–0.54; Cohen d = 0.25; adjusted P = .01).
- Behavioral choices aligned strongly with the ML recommendation: among MLSG participants receiving an “accept” recommendation, 50.7% (34/67) chose to accept screening, compared with 24.2% (44/182) in the control group (P < .001). Conversely, when the system recommended “not now,” only 17.8% (21/118) of MLSG participants chose to accept, demonstrating that the tool influenced but did not dictate decision-making.
Why This Matters
PSA screening decisions are inherently complex, and older adults face specific barriers — including cognitive and sensory limitations and multimorbidity — that make standard decision aids insufficient. This study demonstrates that embedding a machine learning recommendation as a personalized “second opinion” within an existing shared decision-making framework can reduce both decisional distress and emotional burden without overriding patient autonomy, as participants were not simply deferring entirely to the algorithm. The findings suggest that AI-assisted decision support could be scaled in clinical settings serving aging populations who require more individualized guidance in cancer screening contexts.
How to Read This Result
This well-designed RCT provides encouraging evidence that ML-enhanced shared decision-making meaningfully improves decisional and emotional outcomes for older men considering PSA screening; however, because the abstract does not report study limitations, it remains unclear how well these findings generalize across different clinical settings, cultural contexts, or healthcare systems, and whether the observed decision-making benefits translate into improved long-term clinical or behavioral outcomes.
Limitations
The abstract does not explicitly report study limitations.