Falls in Hospitals: Challenging Traditional Risk Assessments With New Insights Into Patient Mobility

ABSTRACT

Aims and Objectives

To explore the association between different aspects of patient functional mobility, specifically, mobility capability (i.e., what the patients could do) versus mobility performance (i.e., what the patients actually did) and hospital falls.

Background

Fall risk assessments are important strategies to mitigate inpatient falls, and mobility is a crucial factor in determining a patient's risk. However, different fall assessment tools vary in how they attribute risk based on mobility difficulties. Understanding how various aspects of mobility uniquely influence fall risk is essential for accurately capturing and assessing a patient's true fall risk.

Design

A retrospective analysis was conducted using routine electronic medical record data at three hospitals, encompassing 498 patients who experienced falls and 53,708 patients who did not fall.

Methods

We examined patient mobility in three distinct ways and their relationship with in‐hospital falls. Mobility was assessed within the first 48 h of admission using the mobility questions in the Johns Hopkins Fall Risk Assessment Tool (JHFRAT). Additionally, we evaluated other aspects of mobility using the AM‐PAC scale, which measures mobility capability, and the JH‐HLM scale, which assesses mobility performance.

Results

A negative linear/stepwise relationship was observed between both AM‐PAC scores and JHFRAT mobility scores with fall incidence, indicating that lower mobility capability is consistently linked to a higher risk of falls. In contrast, the relationship between JH‐HLM scores and falls followed an inverse U‐shaped curve, with a lower fall incidence in patients with the lowest mobility performance.

Conclusions

This exploratory study highlights that a one‐size‐fits‐all approach to assessing mobility may not accurately capture a patient's true fall risk, emphasising the importance of evaluating different aspects of patient mobility for a more precise assessment. By considering both functional mobility capacity and actual mobility performance, we can better understand and address the unique ways in which mobility impacts fall risk.

1. Introduction

Falls among hospitalised people pose a significant challenge in healthcare, particularly for older adults (Salari et al. 2022). Falls are difficult to completely prevent or predict and can lead to patient injury, prolonged hospitalisation and escalated costs (Costantinou and Spencer 2021; Strini et al. 2021). Patient mobility is fundamental to this problem, where mobility issues account for up to 30% of falls in hospitals (Capo‐Lugo et al. 2023; Growdon et al. 2017), so it is critical to understand how mobility relates to fall risk and prevention strategies (Fallen‐Bailey and Robinson 2021; Hoyer, Young et al. 2024; Kissane et al. 2023). While falls are recognised to result from a complex interplay of intrinsic and extrinsic factors—including cognitive status, medication profiles and environmental hazards—these have been widely studied and documented in the literature (Hendrich et al. 2020; Lindberg et al. 2020; McVey et al. 2024; Silva et al. 2023). Our study builds upon this foundational knowledge by focusing exclusively on mobility as a central, actionable factor in fall risk assessment. Functional mobility refers to an individual navigating their surroundings, influenced by various factors like range of motion, pain, strength, coordination, balance, motivation and endurance (Boonen and Maksymowych 2010). These factors collectively determine the extent of a person's movement capabilities and identify limitations that may require adjustments to their care plan to promote independence while ensuring safety. Traditional tools widely used for assessing fall risk, such as the Johns Hopkins Fall Risk Assessment Tool (JHFRAT), STRATIFY Risk Assessment Tool and Morse Fall Scale, include a mobility assessment to identify at‐risk patients (Morse 2006; Oliver et al. 1997; Poe et al. 2018; Strini et al. 2021). Common fall risk assessment tools often include only general questions about mobility limitations or focus narrowly on activities like walking or transferring. This approach can overlook how different aspects of mobility contribute to fall risk, leading to inconsistencies in how these tools attribute fall risk to mobility problems. For example, some tools might assign a high fall risk score to patients with increasing mobility issues, while others might assign low or no risk to those with severe mobility limitations. Given these varying approaches, our study aimed to provide a more comprehensive analysis of the contribution of mobility to fall risk by investigating the association between two distinct aspects of patient mobility—mobility capability (what patients could do if they tried) and mobility performance (what they actually did)—and hospital falls. This allowed us to explore how different dimensions of mobility influence fall risk in a hospital setting.

2. Methods

We conducted a retrospective analysis using historical data from the electronic health records at three hospitals in Maryland, United States of America, to examine if a broad association between mobility measures and falls can be observed. The three hospitals in this study represent both community and academic settings, chosen to enhance the generalisability of the findings across varied hospital environments. Data for this study were extracted from discrete fields within the electronic health record (EHR), ensuring consistency and accuracy. All demographic, clinical and fall event data were retrieved in a standardised format to facilitate accurate acquisition. Our health system follows the National Database for Nursing Quality Indicators (NDNQI) guidelines for fall and fall with injury data collection and reporting. A fall was defined in this study as an unplanned descent to the floor that may or may not result in injury, as recorded in the EHR by nursing staff at the time of occurrence and later reviewed by clinical nursing specialists for accuracy and also validated through submission to the NDNQI, a standardised database for tracking nursing‐sensitive measures (Staggs et al. 2014). The eligibility criteria were patients with a length of stay in hospital between 2 days and 62 days. The two‐day lower bound was to make sure we had assessment scores after admission. Exclusion criteria included patients with a length of stay longer than 62 days due to their lack of representation of our population and those from paediatrics, and psychiatry. We also excluded patients who remained physically located in the emergency department (ED) or the days they were treated in the ED. None of the excluded patients with a length of stay longer than 62 days experienced a fall during their hospital stays. We aim for an exploratory analysis and hence, therefore, a formal sample size calculation was not conducted. Instead, all eligible patients meeting the inclusion criteria were included, allowing a broad analysis across a generalised dataset. In these hospitals, falls are recorded by nursing staff in the electronic health records at the time of occurrence. They are then reviewed by clinical nursing specialists to ensure fidelity and are also submitted to the National Database of Nursing Quality Indicators. A descriptive analysis of the patients' demographic characteristics was conducted, comparing fallers and non‐fallers (Table 1). p‐values were calculated using t‐tests (Student 1908) for numerical variables and chi‐squared tests (Pearson 1900) for categorical variables between fallers and non‐fallers.

TABLE 1.

Population characteristics.

Characteristic	Non‐fallers	Fallers	p *
No. of patients	53,708	498
Mean age (SD)	61 (18)	63 (16)	0.003
Race group (%)
White	30,005 (55.9)	274 (55.1)	0.786
Black	17,696 (33.0)	171 (34.4)
Other	6007 (11.1)	53 (10.5)
Sex group (%)
Female	26,993 (50.3)	249 (50.0)	0.957
Male	26,707 (49.7)	249 (50.0)
Other	8 (0.0)	0 (0.0)
Payor category (%)
Commercial	15,128 (28.2)	121 (24.3)	0.083
Medicaid	6638 (12.4)	75 (15.1)
Medicare	25,820 (48.1)	251 (50.4)
Other	6122 (11.4)	51 (10.2)
Service category (%)
Medicine	30,300 (56.4)	284 (57.0)	< 0.001
Neurology	1685 (3.1)	43 (8.6)
Neurosurgery	3198 (6.0)	32 (6.4)
Oncology/Heme	4151 (7.7)	25 (5.0)
Orthopaedics	2105 (3.9)	4 (0.8)
Surgery	9876 (18.4)	89 (17.9)
Other	2393 (4.5)	21 (4.2)
Elixhauser comorbidity (%)
0–2	16,352 (30.4)	59 (11.8)	< 0.001
3	8638 (16.1)	52 (10.4)
4	8400 (15.6)	62 (12.4)
5–6	12,828 (23.9)	156 (31.3)
7+	7490 (13.9)	169 (33.9)
JHFRAT fall history (%)	116 (23.3)	6736 (12.5)	< 0.001
JHFRAT medication (%)
0 (no medication)	12,742 (23.7)	94 (18.8)	< 0.01
3 (one medication)	17,524 (32.6)	152 (30.5)
5 (two+ medication)	17,616 (32.8)	204 (40.9)
7 (sedated)	5826 (10.8)	48 (9.6)
JHFRAT cognition (%)
0	45,385 (84.5)	334 (67.1)	< 0.001
1	4076 (7.6)	69 (13.9)
2	796 (1.5)	16 (3.2)
3	278 (0.5)	15 (3.0)
4	1432 (2.7)	26 (5.2)
5	908 (1.7)	13 (2.6)
6	184 (0.3)	1 (0.2)
7	649 (1.2)	24 (4.8)
JHFRAT risk (%) a
High risk	9937 (18.5)	173 (34.7)	< 0.001
Low risk	14,872 (27.7)	68 (13.7)
Moderate risk	28,899 (53.8)	257 (51.6)

^a Calculated using the JHFRAT score upon 48 h of admission.

^*For numerical variables, p‐values are calculated using the t test, and for categorical variables, p‐values are calculated using the chi‐squared test.

We first identified the different mobility assessments completed by nurses within the first 48 h of admission. We began by examining the mobility components of the Johns Hopkins Fall Risk Assessment Tool (JHFRAT), which is a required field to be completed by nurses during every shift. These components include questions like ‘Requires assistance or supervision for mobility, transfer, or ambulation’ and ‘Unsteady gait’. Patients receive scores of 0, 2, 4 or 6 points, with higher scores indicating a greater difficulty with mobility. We then also examined two additional measures of mobility: (1) a measure of mobility capability, the Activity Measure for Post‐Acute Care Inpatient Mobility ‘6‐clicks’ Short Form (AM‐PAC), and (2) a measure of performance, the Johns Hopkins Highest Level of Mobility (JH‐HLM), which are both reliable and valid in the hospital setting (Hoyer et al. 2017). These two measures give different insights to understand and assess how patients move and function physically. The AM‐PAC evaluates basic mobility with six tasks, each scored on a 4‐point scale based on how much help from another person is required. The AM‐PAC is required to be recorded at least once daily by nursing staff and after every therapy session. An example of the AM‐PAC score would be a raw score from 6 to 24, or equivalently a t‐score from 17 to 58, with higher score indicates better mobility capability. The JH‐HLM is an 8‐point scale documented based on the patient's observed performed mobility (ranging from 1 = lying in bed to 8 = walking 250+ feet). It is also required to be recorded at least once per shift by nursing staff and after every therapy session. An example of the JH‐HLM score would be a score from 1 to 8, with higher score indicating better mobility performance. The selection of the AM‐PAC (for mobility capacity) and JH‐HLM (for mobility performance) was guided by their validation and effectiveness in hospital settings, as highlighted by the American Geriatrics Society White Paper on Mobility Assessment in Hospitalised Older Adults (Wald et al. 2019).

In this work, we aim for an exploratory analysis, and hence, examined the association between these mobility measures and the incidence of falls using a LOESS (Locally Estimated Scatterplot Smoothing) (Cleveland 1979) approach using R package loess with default parameters and visualised using R package ggplot2 (R Core Team 2023; Wickham 2016). Specifically, we plotted the fall rates against the JHFRAT Mobility scores, as well as the AM‐PAC and JH‐HLM scores and fitted a LOESS curve to model this relationship. To visualise the LOESS fit, we included a shaded region around the smoothed line. This shaded region represents the estimated mean of the fall rate plus or minus the standard error of the mean estimate. The width of this band provides an estimate of the variance in fall rate outcomes at different AM‐PAC and JH‐HLM scores, adjusted for the sample size.

To maintain consistency with the AM‐PAC and JH‐HLM plots, we reversed the x‐axis on the JHFRAT Mobility plot. This adjustment ensures that moving to the right on the x‐axis consistently indicates better mobility across all plots.

3. Results

The study population included a total of 54,206 patients who were admitted to the three hospitals between April 1, 2022 and October 1, 2023. There were 498 patients who experienced falls during their hospitalisation and 53,708 people who did not fall. Fallers had a mean age of 63 years, were 55% white, 50% female, 57% were admitted into a medicine service, and 35% were categorised as high fall risk according to the JHFRAT (Table 1).

In the mobility section of the JHFRAT, the association between fall risk and JHFRAT scores was approximately linear upon visual inspection. This means that for each one‐point increase in the JHFRAT score, the fall risk increased by a consistent amount across all score levels regardless of whether the scores are low, medium, or high. In other words, improved mobility scores were consistently associated with a lower likelihood of falls, suggesting a meaningful relationship between better mobility and reduced fall risk (Figure 1, Panel A).

FIGURE 1.

The LOESS smoothed trend of the fall Incidence against reversed JHFRAT mobility score (Panel A), AM‐PAC (Panel B), JH‐HLM (Panel C), with standard errors represented in light purple. The shaded region represents the estimated mean of the fall rate plus or minus the standard error of the mean estimate, based on the LOESS smoothing method. The width of the band around the mean provides an estimate of the variance in fall rate outcomes at different mobility scores, adjusted for the sample size. Black dots represent the observed fall incidence for each level of the mobility scoring tools. Data for 498 patients who experienced falls during their hospitalisation and 53,708 patients who did not fall.

However, the relationship between fall risk and mobility capacity and mobility performance differs. For mobility capability, we found a roughly linear negative relationship between AM‐PAC scores and fall incidence, like what we observed with the JHFRAT mobility scores. Specifically, as mobility capability increased, the incidence of falls consistently decreased (Figure 1, Panel B). This indicates that patients who required the most assistance to move also had the highest incidence of falls, while those with greater mobility capability experienced fewer falls.

Interestingly, the relationship between mobility performance using JH‐HLM and fall incidence followed an inverse U‐shaped curve (Figure 1, Panel C). Patients with JH‐HLM scores of 1 or 2, which indicate limited mobility performance at bed level, experienced fewer falls compared to those with moderate mobility performance, who move in the transition zone between bed level and walking. Conversely, patients with high mobility performance, as reflected by JH‐HLM scores of 6 or higher—which represent increasing walking distances—had the lowest incidence of falls.

4. Discussion

Our findings have several implications for fall prevention strategies in hospitals. First, the linear relationship between mobility capability (as measured by AM‐PAC) and fall incidence underscores the need for accurate assessment of a patient's functional mobility capacity. Our study highlights that identifying patients with more severely impaired mobility capacity should prompt early interventions because these patients are at the highest risk of falling. This finding challenges the traditional practice of categorising patients with significant mobility impairments as having low or no risk of falling (Morse 2006; Poe et al. 2018). These patients require special considerations, like the involvement of physical therapy and safe patient mobility equipment.

Second, the inverse U‐shaped relationship between actual performed mobility (as measured by JH‐HLM) and fall incidence illustrates a more complex relationship between mobility and falls (Capo‐Lugo et al. 2023). Patients require careful monitoring and support during mobility activity performance to safely navigate the transition from low levels of mobility (e.g., bed‐level) to higher ambulatory levels of mobility, such as ambulation (Lee et al. 2018). Indeed, our data suggest that patients with a combination of low AM‐PAC scores and moderate JH‐HLM levels are at highest risk for falling and may benefit from more tailored fall prevention interventions. Patients who stay in bed and perform extremely low levels of activity (e.g., bed‐level activity) have lower fall incidence, but importantly, their fall rates are still higher than those of patients who were ambulatory early in their hospitalisation. Patients who were walking early in their hospitalisation had the lowest incidence of falls, which supports strategies that promote safe patient mobility early in the hospital setting as a fall prevention strategy (Bainbridge et al. 2023; Fallen‐Bailey and Robinson 2021; Klein et al. 2022; Wyatt et al. 2020). Notably, another research team from China also notice the inverse‐U relationship between mobility performance and fall rate in their hospital, although using different assessment tools (Yan et al. 2024). Future studies are encouraged to validate these results in additional healthcare environments globally to ensure broader applicability and to understand how localised assessment practices may influence fall outcomes.

Our study also underscores the limitations of current fall risk assessment tools, which often fail to differentiate between the various dimensions of functional mobility. Given the critical role mobility problems play in fall events, there is a need for more advanced fall risk assessments that consider both a patient's mobility capability and their actual performed mobility as part of routine care. However, incorporating multiple redundant mobility assessments into nursing documentation is not ideal. For example, we recently demonstrated that the AM‐PAC score could replace the JHFRAT mobility scoring, which also focuses on mobility capability, thereby eliminating the need to score the JHFRAT mobility items separately (Stenum et al. 2024). This approach strikes a balance between achieving precision in identifying at‐risk patients and ensuring the feasibility of implementation for nursing staff. This approach could provide more targeted interventions, reducing the incidence of falls and improving patient outcomes. For example, by distinguishing between what patients are capable of versus what they actually perform, healthcare providers can design more tailored interventions. For instance, patients with high mobility capability but low performance may benefit from confidence‐building and encouragement to utilise their full mobility potential, thereby reducing their fall risk. Conversely, those with low mobility capability might require more supportive measures to enhance their overall mobility.

These mobility assessments may not only enhance fall risk evaluations but could also play a crucial role in assessing other patient risks, such as non‐home discharge (Young et al. 2023) and venous thromboembolism (Hoyer, Bhave, et al. 2024), adding substantial value to patient care. Additionally, improving the detection of high fall risk patients is increasingly important, especially considering the new Centers for Medicare & Medicaid Services Inpatient Prospective Payment System electronic clinical quality measure, Falls with Injury (FY 2025 Hospital Inpatient Prospective Payment System (IPPS) and Long‐Term Care Hospital Prospective Payment System (LTCH PPS) Final Rule—CMS‐1808‐F | CMS n.d.).

5. Conclusion

This study emphasises the crucial role of comprehensive mobility assessments in hospitalised patients' fall risk. Traditional fall risk assessment tools may miss key aspects of mobility, leading to misclassification, particularly in patients with severe or moderate mobility limitations. By incorporating both mobility capability and performance assessments into routine care, healthcare providers may improve the identification of mobility‐related at‐risk patients and could contribute to more customised interventions. This approach, combined with assessments of other important factors, such as cognitive impairment, would likely not only enhance fall prevention strategies but also improve overall patient care. Additionally, better detection of high fall risk patients has significant implications for hospitals under the new Centers for Medicare & Medicaid Services Inpatient Prospective Payment System electronic clinical quality measure, Falls with Injury (FY 2025 Hospital Inpatient Prospective Payment System (IPPS) and Long‐Term Care Hospital Prospective Payment System (LTCH PPS) Final Rule—CMS‐1808‐F | CMS n.d.). Moreover, it is essential to recognise that falls in hospitalised patients often result from a combination of both intrinsic factors, such as each individual patient's characteristics, including cognitive impairment, medication effects, comorbidities and extrinsic factors, including environmental hazards, staffing patterns and unit protocols (Hendrich et al. 2020; Lindberg et al. 2020; McVey et al. 2024; Silva et al. 2023). Future research could incorporate these additional factors alongside mobility assessments to develop a more holistic fall risk model. Such an approach would enable more targeted fall prevention strategies that address the full spectrum of contributing factors.

Author Contributions

E.H.H., D.L.Y., C.Z., E.C., K.G., Made substantial contributions to conception and design, or acquisition of data, or analysis and interpretation of data; Involved in drafting the manuscript or revising it critically for important intellectual content; Given final approval of the version to be published; Each author should have participated sufficiently in the work to take public responsibility for appropriate portions of the content; Agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Conflicts of Interest

The authors declare no conflicts of interest.

Peer Review

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer‐review/10.1111/jan.16866.

Data Availability Statement

The data that support the findings of this study might be available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.