JMIR Publications

ABSTRACT

Background:

Prostate-specific antigen (PSA) screening involves complex trade-offs between early detection and risks of overdiagnosis. For older adults (aged $\geq$50 years), shared decision-making (SDM) is often hindered by health literacy limitations, sensory or cognitive impairments, and multimorbidity, which complicate risk comprehension. While traditional decision aids (DAs) provide foundational knowledge, they often lack personalization. Machine learning (ML) offers a potential solution through individualized recommendations, yet the psychological and behavioral consequences of ML-assisted SDM in geriatric populations remain poorly characterized.

Objective:

The study followed a two-stage design. First, a Model Establishment Group (MEG; N=507) was used to train and evaluate six ML algorithms based on clinical and values-clarification data. A Random Forest (RF) model was selected for its superior performance (mean AUC 0.933, 95% CI 0.902–0.963). Second, a randomized controlled trial (RCT) was conducted with 367 participants (mean age 61.2 years) assigned 1:1 to either the Machine-Learning Suggestion Group (MLSG; n=185) or the Control Group (CG; n=182). Both groups received video-based education, counseling, and values clarification; however, only the MLSG received an ML-generated "second opinion" recommendation. Primary and secondary outcomes were assessed using the Decisional Conflict Scale (DCS), State-Trait Anxiety Inventory (STAI), and Satisfaction with Decision (Sa) scale.

Methods:

The data used in our research were retrieved from Taiwan St. Joseph Hospital between October 2017 and April 2018, for middle-aged and older patients. A total of 507 participants constituted the Model Establishment Group (MEG) within six machine‑learning (ML) algorithms trained and evaluated. Subsequently, an additional 380 participants were recruited and randomized to either the Machine‑Learning Suggestion Group (MLSG) or the Control Group (CG) to evaluate effects on psychological outcomes (anxiety, satisfaction with information, decisional conflict).

Results:

Integrating personalized ML recommendations into SDM workflows provides significant emotional scaffolding for older men, reducing decisional distress and enhancing confidence without undermining autonomy. By addressing geriatric-specific vulnerabilities through a facilitated digital interface, this ML-driven approach effectively complements traditional clinical consultations. These findings support the scalable integration of AI-assisted decision support to foster truly patient-centered care in aging populations.

Conclusions:

These findings suggest that ML-based decision aids can complement, rather than replace, clinical consultations by embedding personalized, AI-driven recommendations into SDM workflows tailored to adults, thereby improving informed choice, enhancing decision self-efficacy, and sustaining reductions in decisional conflict over time. By explicitly addressing geriatric vulnerabilities and psychological determinants of choice, this approach extends the utility of traditional DAs, reduces uncertainty, and may ultimately lower decisional regret while supporting adherence and downstream quality-of-life outcomes in aging populations. Clinical Trial: Chinese Clinical Trial Registry: ChiCTR 2000034126; https://www.chictr.org.cn/showprojEN.html?proj=55682