A cross-sectional study of undergraduate physical therapy students in China found that critical thinking disposition and deep learning do not move in a fixed, universal pattern. The association shifted by programme type and by academic year, suggesting that the educational setting can reshape how students approach reasoning, integration, and memorisation-heavy coursework.

The study drew on 791 students at Nanjing University of Chinese Medicine between April and October 2025. Investigators used sequential linear regression with HC3 robust inference and then explored dimension-level relationships with false discovery rate correction. The central result was straightforward but academically unsettling: overall, higher critical thinking disposition was inversely associated with deep learning scores, even though deep learning is usually treated as a desirable learning strategy in health professions education.

One result stood out immediately. The relationship was not uniform across the student body.

The association weakened or reversed in specific educational strata

Median deep learning score was 3.69 on a 1 to 5 scale, with an interquartile range of 3.47 to 4.02. Within the main model, programme type modified the relationship between critical thinking disposition and deep learning, with an interaction term of B = 0.171 and a 95% confidence interval of 0.060 to 0.282; P = 0.003. Academic year also mattered, with a significant second-year interaction of B = −0.174, 95% CI −0.335 to −0.014; P = 0.034.

Among second-year students, the adjusted simple slopes were especially striking. In the regular programme, the CTDI–DL slope was −0.369, with a 95% CI of −0.493 to −0.246. In the Sino-foreign cooperative programme, the slope was still negative but smaller in magnitude at −0.198, with a 95% CI of −0.303 to −0.094. In practical terms, the students most likely to be consolidating foundational habits appeared to show the sharpest inverse relationship between critical thinking disposition and deep learning, particularly in the regular programme.

The data suggest that an early curriculum can reward different cognitive habits depending on how it is structured. When coursework emphasizes rapid assimilation, reproduction, or assessment-driven recall, students who lean toward critical scrutiny may not score as highly on a standard deep-learning measure, even if their reasoning is more mature. In that setting, more reflective students may be wrestling with complexity rather than displaying the kind of surface academic behaviors captured by a self-report scale.

Dimension-level findings point to reasoning style, not just overall score

Exploratory analyses added another layer. Negative adjusted associations emerged for analytical ability, systematic ability, and cognitive maturity. Those subdimensions matter because they move the interpretation away from a simplistic “more critical thinking is better” frame. Analytical ability and systematic organization should, in theory, support higher-order learning, yet in this sample they tracked with lower deep-learning scores after adjustment.

One plausible explanation is that deep learning in this context may capture an orientation toward study strategy and academic engagement rather than pure intellectual rigor. Students with stronger critical thinking tendencies may question content, compare sources, and seek internal consistency, which can slow the pace of direct content absorption. In an assessment environment that rewards efficiency and conformity to expected answers, that style may appear less compatible with deep-learning self-ratings even when it is educationally valuable.

The nonlinear sensitivity analysis complicated the story without overturning it. Programme-related variation remained visible, but the difference in average marginal slopes was not significant, with P = 0.121. In other words, the primary linear model detected meaningful modification by programme type, while the nonlinear check did not provide equally strong confirmation on that specific average marginal metric. That leaves the main message intact but cautions against overextending it.

Curriculum structure may be shaping the measurement itself

The setting matters here. Undergraduate physical therapy education sits at the junction of biomedical knowledge, clinical reasoning, and hands-on application, so students are expected to integrate information rather than memorize it in isolation. Yet the way a programme is built can reward different learning behaviors. A cooperative international track may expose students to alternate pedagogic styles, broader language demands, or a different balance between independent study and guided synthesis, all of which could alter how critical thinking disposition translates into a deep-learning score.

This is where the study becomes most useful for educators. If critical thinking and deep learning are not aligned in a predictable way across all students, then a single curriculum design may be too blunt an instrument. Structured opportunities for knowledge integration, evidence-based reasoning, and reflection may help bridge the gap between cognitive disposition and academic approach, but the study does not prove that such interventions will work. It only shows where the mismatch appears most clearly.

The reliance on self-reported measures also shapes interpretation. Students were rating their own learning orientation and thinking disposition, not being observed in a skills lab or clinical simulation. Self-report tools can capture perception, but they may not fully reflect how students actually reason under pressure, solve patient problems, or retain clinically relevant material.

What this means for health professions education

For educators, the strongest implication is not that critical thinking is undesirable. It is that the pathway from critical thinking to deep learning is context dependent. A second-year student in one programme may be learning to question, organize, and integrate in a way that temporarily lowers a self-reported deep-learning score, while another student in a different programme may show a weaker version of the same pattern because the instructional environment is less demanding or less tightly scripted.

For curriculum planners, the result argues for closer attention to programme design, year of study, and the specific cognitive demands placed on students. A uniform approach to fostering deep learning may miss the students whose reasoning style is most likely to be influenced by educational context. The study also raises a quiet but important measurement issue: if a learning scale does not capture the productive tension between reflection and rapid assimilation, educators may misread strong thinkers as weaker learners.

The study remains limited by its single-centre design, cross-sectional structure, and dependence on self-report. Those constraints prevent causal claims and restrict generalisability beyond this university setting. The next step is not simply to repeat the survey elsewhere, but to observe whether the same programme-dependent pattern appears in longitudinal cohorts, clinical teaching environments, and objective assessments of reasoning and performance.


Reporting by Dr. A. Nikhil Krishnan, MBBS for Medicine World Foundation, with additional data provided by Frontiers in Immunology.