This cross-sectional study was conducted via anonymous online surveys using Wenjuanxing, a widely-used survey platform in China, among university students in Guangdong Province from February 1 to March 15, 2025. The study protocol was approved by the Institutional Review Board of the Third Affiliated Hospital of Southern Medical University (Approval no.: 2025-ER-008) and strictly adhered to the principles of the Declaration of Helsinki. Electronic informed consent was voluntarily provided by all participants prior to participation in the study. Given the non-sensitive nature of the survey content and absence of complex legal implications, according to national legislation and institutional requirements, university students aged 16–17 years were deemed competent to provide informed consent independently; therefore, additional written informed consent from their legal guardians or next of kin was not required.

The questionnaire structure was developed based on the framework of the “Residents' Medication Behavior Risk KAP Questionnaire” from the Chinese Pharmaceutical Association (33), which underwent revisions through the Delphi expert consultation method with three pharmaceutical professors to achieve localized adaptation for university students' medication scenarios (see Supplementary Material 1 for details). The finalized questionnaire comprised two sections: demographic characteristics (10 items), serving as independent variables for subsequent analysis; and KAP dimensions, including medication knowledge (28 items), medication attitudes (13 items), and medication practices (25 items).

Based on data released by the Guangdong Provincial Department of Education (21), the total college student population in Guangdong Province is approximately 2.6 million. The required sample size was calculated using the following formula:

where Z = 1.96 (95% confidence level), p = 0.5, and E = 0.05. This yielded a minimum sample size of 385. Accounting for a 10% invalid responses and a design effect (Deff = 1.2), the adjusted sample size was: n′ = 385 × (1 + 0.1) × 1.2 = 512.

Statistical analysis was performed using SPSS software (version 26.0; IBM Corp., Armonk, NY, USA). Demographic characteristics were described using frequencies and percentages [n (%)]. As the KAP scores were not normally distributed, they were described using median and interquartile range (IQR). Inter-group comparisons of KAP scores were performed using the Mann–Whitney U-test (for two groups) or the Kruskal–Wallis H-test (for three or more groups), based on the number of groups. To further explore the factors influencing scores in each dimension of the KAP, multiple linear regression models were constructed with the scores of each dimension as the dependent variable. All variables significant in the univariate analyses were included in these models. The unstandardized regression coefficients (B) and standardized regression coefficients (β) were calculated. Multicollinearity within the models was assessed using the variance inflation factor (VIF) and tolerance. p-value < 0.05 was considered statistically significant for all analyses.

A cross-sectional survey design was employed in this study. A total of 880 questionnaires were distributed. After excluding invalid responses, 816 valid questionnaires remained, yielding a valid response rate of 92.7%. The sample comprised 430 males (52.7%) and 386 females (47.3%). The predominant age group was 18–20 years (70.2%). Most respondents were enrolled in diploma programs (71.1%) or undergraduate programs (23.4%). The majority were non-medical/non-pharmaceutical majors (72.5%), enrolled in full-time programs (95.5%), and attended vocational colleges (66.5%). Participants' urban and rural residences were nearly evenly distributed (urban: 50.9%; rural: 49.1%). The most frequently reported monthly household income per capita was 3,000 to 5,000 RMB, accounting for 29.5% of respondents.

Table 1 presents the analysis of relationships between demographic variables and KAP dimension scores. The results indicated that gender showed no statistically significant differences across any KAP dimensions (p > 0.05). However, significant differences (p < 0.05) emerged across various dimensions for age, current education level, major, study mode, institution type, institution region, urban/rural residence, family income, and sources of medication knowledge in some dimensions.

The KAP scores of the respondents concerning self-medication are presented in Table 2. Overall, students showed the highest qualification rate in the knowledge dimension (93.50%). Within this dimension, 43.38% achieved a “Good” level, and 50.12% were at a “Fair” level.

The qualification rate in the attitude dimension was 75.25%, with 23.04% achieving a “Good” level and 52.21% at a “Fair” level. However, 24.75% of students failed to meet the qualification criteria in this dimension.

Regarding the practice dimension, while the qualification rate (75.74%) was comparable to that of the attitude dimension, the proportion achieving a “Good” level was significantly lower (only 6.50%). Students at a “Fair” level accounted for 69.24%, and the proportion of students classified as “Unqualified” was 24.26%.

Multiple linear regression models were used to analyze the influence of demographic variables on self-medication KAP scores. The detailed results are presented separately for knowledge, attitude, and practice scores.

As shown in Table 3, compared with students aged 16–17 years, significantly lower knowledge scores were observed in students aged 18–20 years (B = −14.990, 95% CI: −20.475 to −7.833, p = 0.018), 21–24-year-old (B = −12.825, 95% CI: −17.579 to −4.633, p = 0.048), and 25–28-year-old (B = −23.370, 95% CI: −28.324 to −11.946, p = 0.004). Undergraduate students (B = −7.589, 95% CI: −10.062 to −3.002, p = 0.032) and master's degree students (B = −11.076, 95% CI: −14.902 to −3.668, p = 0.049) demonstrated significantly lower knowledge scores than diploma students. Part-time students had significantly higher knowledge scores than full-time students (B = 12.100, 95% CI: 9.110–16.378, p = 0.001). Furthermore, students primarily obtaining medication knowledge from friends or peers had significantly lower scores than those whose primary source was family (B = −14.453, 95% CI: −17.176 to −6.878, p = 0.005).

As indicated in Table 4, no significant associations were identified between demographic variables and attitude scores.

As presented in Table 5, practice scores were significantly lower in the 18–20-year-old (B = −12.973, 95% CI: −22.042 to −3.904, p = 0.005) and 21–24-year-old (B = −10.980, 95% CI: −20.266 to −1.694, p = 0.021) groups compared with the 16–17-year-old group. Moreover, part-time students demonstrated significantly higher practice scores compared to full-time students (B = 7.102, 95% CI: 1.888–12.316, p = 0.008), while those citing friends or peers as their primary source had significantly lower practice scores (B = −11.522, 95% CI: −18.909 to −4.135, p = 0.002).

The regression models accounted for limited variance, with adjusted R² values of 0.057 for the knowledge score model and 0.021 for the practice score model, suggesting that additional influential factors not included in these models might exist.

The results of this study indicate the presence of a significant gap between KAP regarding self-medication among the university student population. Although participants demonstrated overall high levels of medication knowledge, notable deficiencies were observed in their medication-related attitudes, with a considerable proportion holding negative or ambiguous perceptions toward safe medication practices. Issues related to actual medication behaviors were even more pronounced. Students who possessed adequate knowledge frequently failed to implement safe medication practices effectively.

Comparisons with previous studies reveal both similarities and differences. A KAP survey involving 7,557 residents in Western China reported that participants' overall medication knowledge was “good” (72.77 ± 22.91), whereas their attitudes (32.89 ± 10.64) and practices (71.27 ± 19.09) were rated as “average,” according to established scoring criteria (33). Similarly, a survey of 471 residents in Haikou, Hainan Province, found that participants had “good” overall knowledge levels (52.2 ± 13.08) but only “average” medication-related attitudes (27.34 ± 8.14) and practices (51.54 ± 9.22). These findings highlight an increased risk of irrational medication use, particularly among individuals with lower educational attainment (34). However, a study of 3,272 adult residents in Shanxi Province yielded divergent results, indicating “fair” overall medication knowledge, but “good” attitudes and practices. This study also emphasized higher medication risks among male respondents and individuals from low-income and lower-education backgrounds (35). Such discrepancies suggest that the relationship and translation among KAP components may exhibit varying patterns influenced by demographic characteristics and regional variations.

Furthermore, our results align with studies conducted among university students in other countries and regions, including Saudi Arabia, Thailand, and Colombia, which similarly identified significant challenges in translating medication-related knowledge into safe practices within this population (36–38). This common global challenge underscores the notion that the knowledge-to-practice gap arises not only from individual factors, such as insufficient personal experience, but is also strongly associated with systemic issues. A lack of sufficient emphasis on safe medication practices within educational frameworks and healthcare systems at national and regional levels appears to be a critical contributing factor (39–42). Therefore, the subsequent sections of this paper will draw upon international experiences with interventions to propose targeted strategies designed to effectively bridge this gap and promote the translation of medication knowledge into consistently safe practices among university students.

The observed disconnect between knowledge and practice in university students' self-medication is not an isolated phenomenon but exemplifies the pervasive “knowledge-practice gap” within public health. Simply enhancing health knowledge does not automatically lead to corresponding improvements in health behaviors, a persistent challenge across various domains of public health, including self-medication.

This gap is especially pronounced in the context of self-medication. For instance, a survey of pharmacy students in Uganda found that while 87.38% possessed good self-medication knowledge and 96.03% held positive attitudes toward medication, only 27.34% actually practiced appropriate medication use, highlighting a significant “knowledge-practice gap” (82). Similarly, a study by Ali et al. involving 1,630 adults in Egypt reported that although 55.0% of respondents had good medication knowledge, this knowledge level did not significantly predict actual medication practices (aOR = 1.15, 95% CI: 0.90–1.48, p = 0.268). In contrast, medication attitudes significantly influenced behavior (aOR = 0.44, 95% CI: 0.36–0.55, p < 0.001) (83). These findings indicate that individuals, despite being fully aware of the risks associated with self-medication, may still engage in inappropriate practices due to cognitive biases, limited access to healthcare, economic constraints, or other practical barriers.

Similar phenomena are widespread across other areas of public health. Yeh et al. in their study on chronic disease management, found that diabetic patients exhibited high levels of foot care knowledge (86.2% correct responses); however, the correlation between knowledge and actual foot care practices was only moderate (r = 0.31), indicating substantial barriers to effectively translating knowledge into behavior (84). Additionally, an Australian study of adolescents aged 15–25 revealed a significant gap between their awareness of healthy eating and physical activity and their actual adherence to related public health recommendations (85). Similarly, a multinational survey by Eltewacy et al. involving 12,606 university students' blood donation behaviors, further corroborated this phenomenon: although health science students possessed significantly higher blood donation knowledge than non-health science students (34.7 vs. 15.7%, p < 0.001), actual blood donation rates did not differ significantly between the groups. This emphasizes that actual health behaviors are strongly influenced by factors beyond knowledge alone, including attitudes, healthcare access, and risk perceptions (86).

Based on key findings from this study and relevant public health literature, targeted intervention strategies are recommended to comprehensively enhance the appropriateness of university students' self-medication across KAP dimensions.

This study has several limitations. First, the cross-sectional design inherently limits the control over potential confounding variables. Thus, unmeasured factors may influence the results, complicating their interpretation. Additionally, the online survey method used in this study, while ensuring participant anonymity and convenience, introduces risks of selection bias and response inaccuracies. Such risks include potential underrepresentation of specific populations and the possibility of careless or rushed responses, all of which may compromise data reliability. Furthermore, data collected exclusively via self-report are susceptible to subjective biases, such as socially desirable responding or idealized self-representation by participants.

Second, the study's limited geographical scope and focus on specific universities and academic majors may restrict sample representativeness. Thus, caution is required when generalizing the findings to broader populations or other regions.

In terms of statistical analyses, the relatively low coefficient of determination (R²) obtained from multiple regression models indicates that these models accounted for only a modest proportion of the variance in KAP dimension scores. This suggests that other influential variables not included in the current analysis might exist, potentially restricting the comprehensiveness and generalizability of the conclusions. Moreover, the measurement instruments demonstrated limited sensitivity, particularly evident within the medication knowledge dimension, where a high pass rate of 93.50% suggested a pronounced “ceiling effect” (109). This ceiling effect implies that high overall knowledge levels could obscure subtle but meaningful differences among variables, making statistical detection challenging.

Therefore, future research should broaden the geographical and demographic scope of the survey, ensure more representative sampling methods, and integrate objective measurement techniques to enhance validation and extend the current findings.