Falls among older adults constitute a critical global public health crisis, imposing substantial burden on individuals, healthcare systems, and societies worldwide. Globally, approximately 684,000 fatal falls occur annually, with over 80% occurring in low- and middle-income countries, while an estimated 37.3 million non-fatal falls require medical attention each year (1, 2). Falls represent the second leading cause of unintentional injury death globally and result in more years lived with disability than transport injuries, drowning, burns, and poisoning combined (3, 4). The economic impact is equally staggering, with fall-related healthcare expenditures in high-income countries reaching billions of dollars annually (4). Beyond immediate physical injuries, falls precipitate cascading consequences including reduced functional independence, institutionalization, fear of falling, social isolation, and premature mortality, fundamentally compromising quality of life among affected individuals (5, 6).

Evidence-based fall prevention strategies have been extensively documented in international literature, with multifactorial interventions demonstrating effectiveness in reducing fall incidence among community-dwelling and hospitalized older adults (7, 8). The 2022 World Falls Guidelines provide a comprehensive framework for evidence-based practice, recommending individualized risk assessment, exercise programs, environmental modifications, medication review, and patient education as core components of effective prevention (1). Contemporary risk assessment tools, such as the Fall Risk Assessment Scale developed and validated specifically for rehabilitation hospital patients, provide standardized methods for identifying high-risk individuals and targeting interventions appropriately in acute inpatient settings (9, 10). Systematic reviews have consistently demonstrated that multifactorial interventions can reduce fall rates by 20–30% when properly implemented (11, 12). However, the translation of these evidence-based recommendations into routine clinical practice remains inconsistent across settings and populations, highlighting a persistent research-practice gap (13, 14).

Central to successful fall prevention is the recognition that patients and caregivers play active roles in implementing preventive strategies (15). Knowledge, attitudes, and behaviors (KAB) regarding fall prevention emerge as crucial modifiable factors influencing engagement with evidence-based interventions (16, 17, 89, 90). Studies have revealed that inadequate knowledge about fall risk factors, negative attitudes toward prevention strategies, and suboptimal adoption of protective behaviors significantly impede effective fall prevention (18, 19). Research demonstrates that healthcare providers’ beliefs and attitudes toward fall prevention directly influence their screening practices and patient education delivery, underscoring the importance of comprehensive assessment of KAB across all stakeholders (16, 20, 91). Furthermore, patient education interventions have shown promise in improving fall risk awareness, knowledge levels, and adoption of preventive behaviors among older adults (21).

Health literacy—defined as the capacity to obtain, process, and understand basic health information needed for appropriate health decisions—has emerged as a critical determinant of successful fall prevention engagement (22, 23). Lower health literacy correlates with poor understanding of fall risks, reduced adherence to prevention recommendations, and increased fall rates among older populations (23). Recent scoping reviews highlight the growing recognition of health literacy’s role in fall prevention strategies, yet substantial knowledge gaps persist regarding optimal approaches to deliver education that accommodates varying literacy levels (22, 24). The quality of patient education delivery, including clarity, cultural appropriateness, and utilization of teach-back methods to verify comprehension, significantly influences knowledge retention and behavioral outcomes (25). Despite this recognized importance, systematic assessments of patient education quality and its relationship to fall prevention outcomes remain limited.

China, home to the world’s largest aging population with over 280 million adults aged 60 years and older, faces unprecedented challenges in fall prevention (26). The burden of falls among Chinese older adults has escalated dramatically, with incidence rates increasing substantially from 1990 to 2019, particularly among the oldest-old population (27, 28). Studies report fall prevalence rates ranging from 8.6 to 26.4% among community-dwelling Chinese older adults, varying by geographic region and population characteristics (29, 30). In mainland China, falls constitute the leading cause of injury-related mortality among older adults, yet fall-related disability-adjusted life years (DALYs) have remained relatively stable, suggesting improvements in acute care but persistent prevention gaps (31, 32). The rapid pace of population aging, projected to reach 400 million older adults by 2035, coupled with evolving family structures and increasing empty-nest households, intensifies the urgency for effective fall prevention strategies (33).

Unique contextual factors shape fall prevention challenges in China. Cultural beliefs regarding aging, family caregiving traditions, and stigma associated with acknowledging vulnerability to falls may influence help-seeking behaviors and acceptance of preventive interventions (34, 35). China’s healthcare system, characterized by hospital-centric care delivery and variable integration of preventive services, presents distinct implementation challenges for evidence-based fall prevention programs (36). Furthermore, substantial urban–rural disparities exist in healthcare access, health literacy levels, and fall prevention resource availability, necessitating context-specific approaches (37). Despite increasing recognition of falls as a preventable health issue, systematic evaluations of fall prevention knowledge, attitudes, and behaviors among Chinese hospitalized older adults populations and their caregivers remain scarce (37, 38).

Critical research gaps persist both globally and within the Chinese context regarding the interplay between patient education quality, health literacy, and fall prevention outcomes. While international evidence supports patient education as a fall prevention component, limited research has systematically examined the quality dimensions of education delivery—including clarity, usefulness, confidence-building, and teach-back verification—and their associations with knowledge, attitudes, behaviors, and actual fall outcomes (39, 40). The role of caregivers in fall prevention education has received insufficient attention, despite caregivers’ crucial position in supporting older adults’ daily safety (41, 42). Additionally, communication barriers stemming from limited health literacy have been identified as obstacles to effective fall prevention education, yet few studies have quantitatively assessed these barriers’ relationship to fall risk in hospital settings (40, 43). In China specifically, no studies have comprehensively evaluated the cascade of fall prevention education delivery—from initial risk awareness communication through teach-back assessment—among hospitalized older adults and their family caregivers, nor examined how education quality and health literacy levels relate to fall prevention knowledge, attitudes, behaviors, and fall events.

This single-center cross-sectional study conducted at a tertiary hospital in China aimed to: (1) assess fall prevention knowledge, attitudes, and behaviors among hospitalized older adult patients and their family caregivers; (2) evaluate the quality of fall prevention patient education delivery, health literacy levels, and communication barriers; (3) examine the relationships between education quality, health literacy, KAB domains, and fall or near-fall events; and (4) identify modifiable factors associated with fall risk to inform targeted intervention development for Chinese hospital settings.

Falls and near-falls among hospitalized older adults remain a clinically consequential and system-relevant safety problem, particularly in aging societies where hospital-based care is increasingly tasked with both acute management and prevention. In this large single-center cross-sectional study of 3,223 respondents (1,644 hospitalized patients aged ≥65 years and 1,579 family caregivers), several findings were notable. First, despite pronounced age differences between respondent types, patients and caregivers displayed broadly comparable fall-prevention knowledge and attitudes, whereas caregivers reported substantially higher self-reported preventive behaviors. Second, the fall-prevention education process exhibited marked attrition at the stage of comprehension verification, with only approximately half of respondents reporting teach-back, and fewer than half achieving fully correct teach-back responses. Third, health literacy was frequently limited, communication barriers were moderate, and—counter to prevailing assumptions—knowledge, attitudes, behaviors, health literacy, and barriers were essentially uncorrelated.

A central observation was that overall knowledge about fall risk factors and prevention strategies was modest and highly heterogeneous across specific items, with substantial gaps in core determinants that are typically emphasized in guideline-based risk communication. Although aggregate knowledge scores were similar between patients and caregivers (approximately 58–59% accuracy), recognition of prior falls as a strong predictor of future falls and understanding of home modification benefits were each below one-third of respondents. These deficits are clinically consequential, because prior falls represent a cornerstone of risk stratification and trigger multifactorial assessment in international and national guidance, including the World Guidelines for Falls Prevention and Management, which highlight prior falls and environmental hazards as actionable determinants of subsequent risk (1). Notably, the pattern of uneven knowledge aligns with scoping and systematic reviews showing that older adults frequently recognize generic intrinsic risk factors (e.g., weakness, dizziness) but under-recognize the prognostic value of previous falls and the effectiveness of specific preventive strategies (e.g., home hazard modification, vitamin D where indicated) (18, 21). In contrast, the similarity between patients and caregivers in overall knowledge challenges a frequent assumption that caregivers inherently possess superior fall-prevention literacy; rather, caregivers may be exposed to the same episodic, non-tailored health communication as patients during hospitalization, yielding parallel knowledge profiles.

Attitudinal profiles in the present study were uniformly moderate-to-positive, with little between-group variation and high endorsement that falls are serious, that prevention matters, and that clinicians should discuss fall risk. This configuration is consistent with broader literature showing that positive general attitudes toward fall prevention can coexist with incomplete understanding of risk determinants and limited uptake of specific recommendations, particularly when educational content is generic or when perceived feasibility is low (16, 18). The comparatively low endorsement of environmental modification as worthwhile is also concordant with cultural and practical barriers reported among older Chinese adults and families, including beliefs about aging, concerns about household disruption, and constrained resources for home adaptations (34, 37). In settings where hospital discharge planning and community support are variable, environmental modification may be viewed as outside the immediate locus of control, thereby weakening attitudinal commitment even when falls are perceived as serious (37).

The most pronounced between-group difference was behavioral engagement, with caregivers reporting higher preventive behaviors than patients (total behavior score 48.80 ± 4.31 vs. 45.10 ± 4.08; p < 0.001; Cohen’s d = 0.88). In principle, this finding is plausible because caregivers often function as the operational agents of prevention—managing environments, supervising mobility, and facilitating care navigation—whereas hospitalized older adults may have restricted mobility, acute illness, or functional limitations that reduce autonomous engagement. Qualitative studies from hospital contexts similarly emphasize that caregivers often assume responsibility for safety behaviors, although they frequently report insufficient tailored education and limited opportunity to clarify uncertainties with health professionals (15, 55). However, the present results also reveal a critical implementation discordance: adherence to provider recommendations was markedly low (mean ≈2.1/5 = 42% adoption) in both groups despite positive attitudes toward clinician involvement and despite caregivers reporting high engagement in nearly all other preventive behaviors (mean scores 3.5–3.6/5 = 70–72% adoption). This juxtaposition suggests that “agreement” with prevention principles does not translate to adoption of clinician-directed actions, potentially because recommendations are perceived as impractical, insufficiently explained, inconsistently reinforced, or competing with acute care priorities (56, 57). This “performance-practice gap”—wherein respondents perceive themselves as actively engaged in fall prevention yet simultaneously report poor adherence to the specific recommendations provided by healthcare professionals—has several plausible explanations. First, the measured behaviors (e.g., wearing appropriate footwear, ensuring adequate lighting, using assistive devices) may represent autonomous safety actions that caregivers and patients implement based on common sense or prior experience, whereas “following provider recommendations” refers specifically to clinician-directed advice that may be perceived as impractical, poorly explained, or misaligned with the patient’s functional status or home environment. Second, social desirability bias may inflate self-reported engagement in generic preventive behaviors while respondents feel more comfortable honestly disclosing non-adherence to provider advice. Third, provider recommendations may be delivered as generic, non-tailored instructions (e.g., “be careful getting up,” “remove hazards at home”) that lack actionable specificity, reducing perceived feasibility and motivation for adoption. Finally, the low adherence may reflect insufficient shared decision-making during education delivery—recommendations that are not co-developed with patients and families may be viewed as externally imposed mandates rather than collaboratively designed strategies, undermining ownership and follow-through. These mechanisms converge on the conclusion that high self-reported engagement in fall prevention behaviors does not guarantee alignment with evidence-based professional guidance, underscoring the need for patient-centered education approaches that prioritize actionable, context-specific recommendations delivered through collaborative dialogue rather than unidirectional information transmission. Such an implementation gap aligns with evidence from process evaluations and qualitative work indicating that fall-prevention messages are often delivered briefly, with limited shared decision-making, and without structured reinforcement or verification of comprehension (14, 55).

A major contribution of this study is the operationalization of fall-prevention education as a cascade, enabling identification of the principal attrition points. Approximately 70% of respondents reported being told about fall risk, two-thirds reported receiving education, but only about half reported that teach-back was used. Moreover, at the whole-sample level, 46.6% reported fully correct teach-back responses; among those who were asked to teach back (n = 1,671), 89.9% reported fully correct responses. This pattern is consistent with prior observations that educational delivery is commonly documented as “provided” without systematic assessment of understanding, and that teach-back—despite its established role in improving comprehension across health contexts—is not routinely implemented due to time constraints, workflow limitations, and variable staff training (25, 49, 58). Recent qualitative work underscores that even when education is delivered, patients and caregivers may perceive it as generic and insufficiently personalized to their functional status or environment, which may reduce salience and retention (59, 60). Importantly, the education quality ratings (clarity, usefulness, confidence) were moderate and similar across respondent types, indicating that perceived “quality” alone may not capture whether critical content was delivered, tailored, or retained (61).

The findings also intersect with emerging literature emphasizing that hospital fall prevention is not solely dependent on patient education but on a multi-layered safety system that includes staff practices, environmental design, and organizational culture. Education is a necessary component of multifactorial interventions in many frameworks, but its effect may be conditional on concurrent structural supports, such as safe staffing, standardized risk assessment tools, mobility protocols, medication review processes, and environmental hazard controls (1, 7, 39). Therefore, the observed cascade attrition likely reflects system-level implementation constraints rather than individual-level unwillingness. These results suggest that institutional efforts should prioritize routinized comprehension verification and actionable discharge-oriented counseling, rather than assuming that risk notification and information provision are sufficient.

Health literacy is increasingly recognized as a determinant of preventive engagement in older adults, and prior work has linked limited health literacy to poorer understanding of fall risk and reduced adherence to preventive recommendations (22, 23). In this study, however, mean health literacy screening scores hovered near the inadequacy threshold, and approximately two-fifths of both groups fell into the limited-literacy range. Furthermore, perceived communication barriers were moderate. Notably, despite these gradients, correlations among knowledge, attitudes, behaviors, health literacy, and barriers were essentially null (|r| ≤ 0.02). This finding represents a theory-challenging observation that contradicts the foundational assumptions underlying many fall prevention education interventions. Conventional KAB models implicitly assume that knowledge gains lead to attitude changes, which in turn motivate behavioral adoption—a sequential pathway that should generate positive intercorrelations among domains (62, 63). The complete absence of such associations in our inpatient sample suggests that KAB constructs, while validated in community-based settings where individuals have autonomy and time to implement changes, do not function as an integrated framework in acute hospital environments. This has critical implications: educational interventions targeting knowledge alone are unlikely to modify behaviors or attitudes in hospitalized older adults, and effective inpatient fall prevention requires fundamentally different theoretical models that account for institutional constraints, acute clinical factors, and the limited agency patients have over their immediate environment during hospitalization.

Several scientific explanations may account for this apparent disconnect. First, measurement non-equivalence across respondent types and across constructs may attenuate observed correlations. Knowledge items were scored dichotomously, whereas attitudes and behaviors were Likert-based, and these different scaling properties can reduce linear associations, especially when distributions are narrow (e.g., attitudes clustering around similar values) (64, 65). Second, inpatient environments constrain behavioral expression; even when knowledge or attitudes are present, institutional routines and patient functional status may dominate behavior, obscuring relationships that might be more evident in community settings. Third, caregivers’ reported behaviors may reflect intention or perceived responsibility rather than directly observed actions, introducing social desirability and recall bias that can weaken associations with knowledge and attitudes (66, 67). Fourth, the content delivered through education may not map directly onto the specific knowledge items assessed, limiting the extent to which education quality relates to knowledge or behavior outcomes. Finally, contemporary fall risk is multifactorial and may be more strongly influenced by acute illness severity, delirium, medication changes, mobility limitations, and ward environment than by KAB constructs, particularly over short time horizons (1, 68).

Perhaps the most clinically provocative result was the positive association between health literacy, communication comfort, and reported fall/near-fall outcomes. Unlike standard clinical models where fall risks are often inversely related to health literacy, the adjusted analysis (Table 7) showed that individuals with higher health literacy (aOR = 1.11, 95% CI 1.03–1.19; p = 0.006) and greater comfort asking questions (aOR = 1.17, 95% CI 1.07–1.27; p < 0.001) were more likely to report events. This finding almost certainly reflects reporting and detection bias rather than any causal relationship between literacy and fall risk. Specifically, patients with higher health literacy and communication comfort are more capable of recognizing ambiguous events (e.g., brief loss of balance, stumbling, grabbing for support) as reportable “near-falls,” whereas individuals with limited literacy may dismiss identical events as non-incidents or lack the vocabulary to articulate them when questioned. This interpretation is further supported by the inclusion of near-falls in the outcome definition—events that are inherently subjective and require cognitive recognition and linguistic articulation to be captured in self-report assessments. Importantly, this mechanism implies that the observed association represents differential event detection and reporting, not differential event occurrence, and causal interpretations would be inappropriate. For readers unfamiliar with inpatient fall dynamics, it is critical to recognize that higher literacy facilitates accurate reporting of events that actually occurred, rather than increasing actual fall risk—a distinction with important implications for interpreting fall surveillance data and designing literacy-appropriate assessment protocols. This may reflect a “reporting bias” where more literate and engaged individuals are better at recognizing and articulating near-miss events, whereas those with lower literacy may under-report. Alternatively, highly engaged patients may be more mobile and independent, paradoxically increasing their exposure to risk scenarios. Moreover, discrimination remained poor even in the full model (AUC = 0.577), and pseudo-R² values indicated that the models explained only a small fraction of outcome variance. This low predictive utility highlights a theoretical mismatch: while KAB constructs are distal determinants suitable for predicting long-term community trends, they lack explanatory power for acute inpatient falls driven by proximal, dynamic triggers. These findings contrast with some community-based studies in which lower health literacy, weaker prevention knowledge, and poorer adoption of safety behaviors correlate with higher fall risk (21, 69). However, they align with the broader evidence base that, in hospital settings, falls are heavily influenced by clinical acuity, medications (particularly psychoactive drugs), gait instability, delirium, toileting urgency, staffing and supervision, environmental layout, and timing factors—domains not fully captured by the present KAB-centered framework (70, 71). Critically, the consistently low model discrimination (maximum AUC = 0.577) reflects not merely statistical insufficiency but a fundamental conceptual mismatch: the distal determinants measured in this study (knowledge, attitudes, behaviors, education exposure, perceived quality) are unlikely to exert short-term causal effects on fall events that are predominantly driven by acute clinical instability, rapidly changing medication regimens, delirium, severe mobility impairment, and immediate environmental hazards. This theoretical limitation underscores that KAB frameworks, while relevant for community-based prevention where individuals have autonomy over long-term behavioral adoption, have limited explanatory power in acute inpatient rehabilitation settings where proximal clinical and environmental factors dominate risk pathways over hours to days rather than weeks to months.

The counterintuitive direction of the univariate health literacy signal warrants careful interpretation. The most parsimonious explanation is detection and reporting bias: individuals with higher literacy possess superior capacity to recognize ambiguous balance disturbances as “near-falls,” articulate these events using appropriate medical terminology, and feel comfortable disclosing them during structured interviews—whereas individuals with limited literacy may experience identical physiological events but dismiss them as inconsequential or lack the language to describe them accurately. This mechanism is particularly relevant given the subjective nature of near-fall definitions and the reliance on self-report outcome ascertainment in this study. A plausible explanation is reverse causality: patients who experience a fall or near-fall may have more contact with clinicians, receive more explanations, and consequently report higher confidence in interacting with healthcare systems, artificially inflating health literacy scores (72, 73). Critically, these interpretations converge on the conclusion that the observed positive association does not reflect a causal effect of literacy on fall risk, but rather differential event detection, reporting, or post-event healthcare engagement. Readers should interpret these findings as evidence that literacy influences how fall events are recognized and communicated, not whether they occur. Alternatively, higher literacy may correlate with greater mobility and independence, which could increase exposure to fall opportunities in a hospital environment. Another explanation is detection or reporting bias: individuals with higher literacy may better recognize and report near-fall events, whereas those with lower literacy may under-report. Finally, residual confounding by unmeasured clinical factors (e.g., gait impairment, sedative exposure, delirium) could generate a spurious positive association in unadjusted analysis (74, 75). These interpretations are consistent with the well-recognized limitations of cross-sectional designs for causal inference and underscore the need for prospective designs with standardized outcome ascertainment (76).

While age, comorbidity count, and time spent alone showed no association, male sex and urban residence emerged as strong predictors of fall events in this cohort (p < 0.001). This contrasts with global literature often citing female preponderance in falls but may reflect risk-taking behaviors or mobility patterns among male inpatients in this setting (72, 77). Moreover, prior fall history was assessed among patients, but the models’ restriction to patients for higher-order models may limit power to detect modest effects when predictor distributions are constrained (78, 79). Additionally, the narrow age range among patients (mean 73.1 ± 6.4 years) reduces variance, diminishing the detectable age effect compared with studies spanning wider adult age ranges.

Although the present study found limited evidence that KAB domains and education quality metrics directly predict fall/near-fall events, the data nonetheless identify actionable targets for improving patient-centered prevention in Chinese hospital settings. First, the education cascade highlights a clear, modifiable system failure: comprehension verification via teach-back is inconsistently implemented (used with only ~52% of respondents), and among the total sample, only 46.6% achieved fully correct teach-back responses. Hospitals should implement the following system-level interventions to address this gap: (1) Embed mandatory teach-back verification into electronic medical record (EMR) workflows by creating a structured documentation field that requires nurses to record whether teach-back was attempted, the patient’s response accuracy (correct/partial/incorrect), and any re-education provided before the fall risk assessment can be electronically signed and closed. This EMR-based forcing function ensures teach-back becomes a non-optional component of fall prevention education rather than a discretionary practice dependent on individual clinician initiative. (2) Implement structured nurse training protocols that include didactic instruction on teach-back methodology, simulated practice sessions with standardized patients, competency assessment through direct observation of at least three teach-back interactions, and ongoing quality monitoring through monthly audits of EMR documentation with feedback to nursing units. Training should emphasize that teach-back is not “testing” the patient but rather verifying that the clinician’s explanation was clear and understandable. (3) Develop literacy-stratified education materials tailored to the 41.5% of patients with limited health literacy identified in this study, including visual aids depicting fall risk factors and prevention strategies (e.g., pictorial guides showing proper use of assistive devices, bathroom safety equipment, appropriate footwear), simplified language scripts for verbal education (avoiding medical jargon and using action-oriented phrasing), and availability of interpreter support for patients with language barriers beyond Mandarin proficiency. These materials should be pilot-tested with low-literacy patient advisory groups to ensure comprehensibility. Embedding teach-back into routine workflows—particularly for high-risk patients—could improve knowledge retention, align expectations, and facilitate safer transitions to home. Embedding teach-back into routine workflows—particularly for high-risk patients—could improve knowledge retention, align expectations, and facilitate safer transitions to home. Teach-back implementation is feasible within existing nursing and allied health roles and has demonstrated effectiveness in improving patient understanding and reducing readmissions in other clinical contexts (80, 81). Importantly, successful implementation likely requires structured training and organizational support, consistent with staff perspectives emphasizing that brief but frequent education, supported by institutional culture, is essential (82).

Second, the pattern of knowledge deficits suggests that education content should be more strategically targeted. Emphasis on the prognostic importance of prior falls, the benefits and practicalities of environmental modification, and the role of vitamin D and strength/balance interventions (where clinically appropriate) may address the largest knowledge gaps. Notably, these topics align with international guidelines and high-quality evidence syntheses supporting multifactorial interventions and exercise-based approaches for fall prevention (7, 83). Third, the discordance between positive attitudes and low adherence to provider recommendations implies that education should prioritize actionable, context-specific behaviors and shared decision-making, rather than general messaging. In practice, this may involve bedside demonstrations, mobility planning, toileting strategies, medication counseling, and family-inclusive safety planning tailored to the patient’s functional status and home environment (84, 85).

Fourth, the high prevalence of limited health literacy underscores the need for literacy-sensitive communication. Health literacy-informed approaches—including simplified language, pictorial aids, repetition, and opportunities for questions—may improve comprehension and engagement, even if health literacy was not an independent predictor of events in this cross-sectional design. Furthermore, addressing perceived communication barriers may require clinician training in patient-centered communication and time allocation for education, consistent with literature documenting limited dialogue between professionals, older adults, and caregivers about fall prevention during and after hospitalization (85, 86). Finally, the weak predictive performance of KAB-centered models suggests that future risk stratification and prevention efforts should integrate richer clinical and environmental data. Incorporation of validated inpatient fall risk tools, objective mobility measures (e.g., gait speed, Timed Up and Go), delirium screening, medication burden metrics, and ward-level environmental and staffing variables may yield improved prediction and more actionable intervention targets (71, 87). In the Chinese context, where rural–urban differences in health outcomes and healthcare utilization persist despite broad insurance coverage, multilevel approaches that incorporate social determinants and post-discharge environment may be particularly relevant (36, 88).

This study has several strengths, including a large sample size with dyadic patient–caregiver participation, comprehensive measurement of KAB domains alongside education quality, health literacy, and communication barriers, and the use of a cascade framework to identify specific implementation points where education fails. The progressive modeling strategy also provides transparent evidence that the examined domains, as operationalized here, have limited explanatory value for fall/near-fall events in this setting. Several limitations should temper interpretation. Most fundamentally, this study operationalized a KAB-centered conceptual model that, while appropriate for community-based fall prevention research, exhibits inherent theoretical limitations when applied to acute inpatient fall events. The consistently poor predictive performance (AUC = 0.577, pseudo-R² = 0.013) is not merely a statistical issue but reflects that the measured distal determinants (knowledge, attitudes, behaviors, education quality, health literacy) are conceptually mismatched to the proximal, rapidly evolving clinical factors that dominate inpatient fall risk over short hospitalization periods. Effective inpatient fall prediction and prevention require integration of acute clinical variables (delirium status, medication burden, objective mobility measures, staffing adequacy, environmental hazards) with patient engagement constructs, rather than relying on education-centric frameworks alone. The cross-sectional design precludes causal inference and is susceptible to reverse causality, particularly regarding health literacy and education exposure. Outcomes were based on self-report and included near-falls, which may be variably interpreted, potentially introducing misclassification; future studies should triangulate events using incident reports and clinical documentation. Several key inpatient fall determinants were not measured, including delirium, medication classes (e.g., sedatives), objective functional status, staffing ratios, ward environmental hazards, and timing of events, all of which may explain the low model discrimination. Self-reported behaviors may be influenced by social desirability, particularly among caregivers. Finally, the single-center design may limit generalizability across China’s heterogeneous hospital systems, although the large sample and inclusion of both patients and caregivers increase the relevance of findings for similar tertiary settings.

This large hospital-based study revealed that while fall prevention knowledge and attitudes were comparable between older adult patients and caregivers, critical gaps persisted regarding high-risk factors such as prior fall history and environmental modifications. Despite caregivers demonstrating higher preventive behaviors, the study identified systemic implementation failures, including inconsistent teach-back verification and low adherence to provider recommendations. Crucially, while traditional knowledge and education quality metrics failed to predict outcomes, health literacy and communication comfort emerged as significant independent predictors of fall reporting, suggesting that patient engagement capacity fundamentally influences event recognition. These results challenge the sufficiency of education-centric models in acute care, indicating that inpatient falls are driven by multifactorial clinical and environmental determinants rather than knowledge deficits alone. Consequently, prevention strategies must shift from passive information delivery to system-embedded, literacy-sensitive interventions that enforce comprehension verification and integrate active patient engagement with objective clinical risk protocols.