A new study from Shanghai Jiao Tong University and Tongji University found that DeepSeek R1 and Google's Gemini Pro 2.5 bring different analytical strengths to healthcare research, with neither model universally superior but each capturing distinct insights when combined with human analysis. Researchers conducted a hybrid analysis comparing three approaches to understanding why cancer patients in China hesitate to participate in clinical trials, revealing how multiple AI models can complement traditional qualitative research methods.

The study analyzed data from 11 in-person interviews with cancer patients and 219 comments from two Chinese online health communities, Zhihu and Yuaigongwu. Three parallel analytical methods were used: traditional human-led thematic analysis, Gemini Pro 2.5-assisted analysis, and DeepSeek R1-assisted analysis. All three methods received identical, structured prompts to ensure fair comparison.

What Themes Did Each AI Model Identify Differently?

The results revealed distinct analytical patterns rather than a clear winner. DeepSeek R1 and human researchers identified eight shared themes, including family-involved decisions and service-related factors. Gemini Pro 2.5 and human researchers identified three shared themes, including regional disparities and autonomous decision-making. DeepSeek R1 uniquely identified the theme of "insufficient clinical data," while Gemini Pro 2.5 uniquely identified "lack of information resource," and human researchers alone identified "recognition of medical value".

All three analytical methods jointly identified seven core themes shaping patient decisions about trial participation:

• Treatment Selection: Patients' preferences for specific therapeutic approaches and concerns about which treatment options would be offered in trials.
• Financial Burden Relief: Economic considerations and whether trial participation would reduce out-of-pocket costs for cancer care.
• Uncertain Therapeutic Efficacy: Patient doubts about whether experimental treatments would actually work for their condition.
• Uncertainty Regarding Control Groups: Concerns about being assigned to a placebo or standard-care control arm instead of the experimental treatment.
• Lack of Cognition: Limited understanding of how clinical trials work and what participation entails.
• Misconceptions: False beliefs about trial procedures, risks, or benefits that discourage participation.
• Physician Trust: The critical role of trust in doctors' recommendations when deciding whether to enroll.

How Can Researchers Use Multiple AI Models Effectively?

The study demonstrated that hybrid analytical frameworks combining human expertise with multiple large language models (LLMs), which are AI systems trained to recognize patterns and extract themes from text, can produce more comprehensive insights than any single approach alone. Human analysis provided contextual depth and cultural sensitivity, while LLMs offered efficiency and identified additional thematic dimensions that researchers might have overlooked.

The researchers noted that the complementary value of this approach addresses a key limitation in previous health research. Most prior studies relied on either exclusively qualitative or exclusively quantitative methods, which often fail to capture the full picture of patient perspectives. By combining online discourse analysis with in-depth interviews and multiple AI analytical tools, researchers achieved both representativeness and depth.

Why Does Clinical Trial Enrollment Matter So Much?

The challenge of patient enrollment in clinical trials is substantial and has real consequences for medical progress. More than 80% of trials fail to complete required sample enrollment within target timeframes, and as many as 19% of clinical trials are terminated due to insufficient enrollment. Adult participation rates in trials remain extremely low, ranging from 2% to 8% over the long term.

In China specifically, the burden is significant. According to 2022 data, China recorded 4.825 million new cancer cases and 2.574 million cancer-related deaths annually, accounting for 24% and 26% of global totals respectively. Clinical trials are essential for advancing cancer treatment and introducing innovative therapies, yet low enrollment rates delay progress and increase costs.

By using AI to better understand patient decision-making, researchers can develop more effective recruitment strategies tailored to specific populations. Understanding that cancer patients in China weigh family involvement, financial burden, and physician trust equally suggests that trial recruitment materials must address these specific concerns rather than focusing solely on medical efficacy.

What Are the Practical Implications for Healthcare Organizations?

This research suggests several actionable insights for improving clinical trial recruitment. First, organizations should recognize that patient decisions involve complex, multidimensional factors encompassing treatment expectations, economic considerations, risk perceptions, cognitive factors, and relational dynamics. Second, the study indicates that different AI models excel at different analytical tasks, implying that healthcare organizations should evaluate multiple AI tools rather than assuming one model will perform best across all contexts.

The researchers concluded that future recruitment strategies should prioritize patient-centered communication, transparent trial information, and culturally tailored approaches. The hybrid analytical framework demonstrated that combining human judgment with AI analysis can uncover insights that either approach alone might miss. As healthcare systems worldwide grapple with low clinical trial enrollment, this research suggests that thoughtfully integrating multiple AI tools into research workflows may play an increasingly important role in understanding patient barriers and designing effective recruitment strategies.