A META-ANALYSIS OF FALL RISK IN OLDER ADULTS WITH ALZHEIMER’S DISEASE

Abstract

Objectives:

Falls are the leading cause of injuries in older adults. Although it is well recognized that Alzheimer’s disease (AD) increases the fall risk of older adults, the reported fall risk in people with AD varies drastically. The principal purpose of this study was to summarize and synthesize previous studies reporting fall risk-related metrics in people with AD.

Design:

This was a meta-analysis.

Setting and Participants:

Thirty-one studies reporting relevant fall data among 4,654 older adults with AD were included.

Methods:

The fall prevalence, average number of falls, rate of recurrent fallers, and rate of injured fallers of included studies were meta-analyzed using random-effects models with inverse variance weights.

Results:

The pooled annual fall prevalence in older people with AD is 44.27% with an average annual number of falls of 1.30/person and a yearly rate of recurrent fallers of 42.08%. The reported rate of injured fallers was 45.0%.

Conclusions and Implications:

Our results reinforce that people with AD experience a higher fall risk than their cognitively healthy counterparts. The pooled fall metrics in this meta-analysis extend our understanding of the fall risk in people with AD. In addition, standardized approaches are needed to report fall-related data for people with AD.

SUMMARY

Our meta-analyses revealed a high annual fall prevalence (44.27%) and number of falls (1.30 falls/person) in older adults with Alzheimer’s disease. The methods to report fall risk should be standardized.

INTRODUCTION

Falls present a serious health threat in older adults globally ¹. A recent meta-analysis of 104 studies including 36 million older adults without cognitive impairment reported that 26.5% of people aged 65 and over fall at least once annually ² with costly consequences ^3–5. Dementia increases the risk of falling and sustaining injuries by twice-thrice among older adults ^{6, 7}, and up to 80% of people with dementia fall annually ⁸. Alzheimer’s disease (AD) accounts for approximately 70% of dementia cases ⁹. AD denotes a specific initiation and progression of cognitive and functional deterioration linked to aging ¹⁰. People with AD experience a gradual decline in memory, reasoning abilities, and other aspects of cognitive function ¹¹.

The AD prevalence in older adults is about 10–30% ¹². Given the large and fast-increasing size of people with AD, it is crucial to study the fall risk in this population to develop fall prevention programs. As the population ages and life expectancy extends, the risks of falls and AD upsurge ¹³. The commingling effect of the aging-induced fall risk and AD makes falls a severe public health concern. To develop effective fall prevention treatments, it is vital to soundly understand the magnitude of falls for robust estimates of the healthcare costs related to fall prevention and treatments for people with AD.

Although most previous studies suggested that AD increases the fall risk in older adults ^14–16, some studies reported that the fall prevalence in this population is not significantly different from older adults without cognitive impairment ^{17, 18}. The reported annual fall prevalence in people with AD varied considerably between 25% ¹⁹ and 73.96% ²⁰. There is a lack of consensus on the fall prevalence in people with AD, which could cause barriers to identifying potential fallers and developing fall prevention interventions targeting this population. Such wide fall prevalence could be due to the small sample sizes (as small as 14 ²¹) in the previous studies. Therefore, it is imperative to determine the fall prevalence using large sample sizes in this population. Over the past decades, many articles with various sample sizes documented the fall prevalence in this population ^{16, 21, 22}. It is possible to perform a meta-analysis summarizing fall prevalence in this population. Unfortunately, no such meta-analysis has been conducted.

Besides the fall prevalence, additional fall-related metrics deserve consideration, such as the average number of falls, the rate of recurrent fallers, and the rate of injured fallers. These extra measurements could provide more information about the fall risk of people with AD, which will be helpful for fall education, prediction, and prevention in this population. For example, the information regarding the proportion of recurrent fallers may furnish beneficial insight into how the fall history is related to future falls. This could enhance the accuracy of predicting future falls. The number of falls and the rate of injured fallers may provide important knowledge for making relevant policies and developing effective fall prevention programs targeting people with AD.

As indicated above, the literature has reported a wide range of fall prevalence in people with AD and lacks exploration of other fall risk metrics for this population. Therefore, the primary purpose of this meta-analysis was to summarize and synthesize previous studies reporting fall risk in older adults with AD and to obtain the fall risk over a large sample size. Four outcome measures were used to quantify the fall risk: fall prevalence, the average number of falls, the rate of recurrent fallers, and the rate of injured fallers (Fig. 1). The findings of this meta-analysis could contribute valuable insights into the fall risk among individuals with AD. A comprehensive understanding of fall risk in this population could be beneficial for our efforts to tackle fall prevention for this population.

Figure 1.

Definitions of fall metrics, including fall prevalence, the average number of falls, the rate of recurrent fallers, and the rate of injured fallers.

METHODS

2.1. Search strategy and selection criteria

Our literature search was conducted within APA PSYCINFO, CINAHL, EBSCO, Google Scholar, MEDLINE, and PUBMED databases between September and December 2022. The search terms and keywords were developed in consultation with a librarian or based on the Medical Subject Headings (MeSH) terms (Appendix 1). No time range or region restrictions were applied. The reference lists of all included articles were manually reviewed by two authors. The meta-analysis adhered rigorously to the methodological guidelines established by the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) and followed the well-recognized Cochrane Collaboration protocol.

A study was included if it 1) was conducted in older adults aged 65 years or over with AD; 2) reported at least one of the four fall risk metrics; and 3) was published in peer-reviewed English journals. The exclusion criteria encompassed the following items: 1) the study reported falls in people with dementia not caused by AD (e.g., Parkinson’s disease, Lewy body dementia, mild cognitive impairment, etc.) and 2) the study was published as an abstract to safeguard the rigor of our meta-analysis. Articles with any study design that reported at least one of the four targeted fall risk measurements, either retrospective or prospective, were considered. Studies only reporting fall-related measurements after interventions were excluded to ensure that the fall risk measurements unbiasedly reflect people with AD.

2.2. Selection of studies

After duplicate removal from the initially searched articles, the full texts of the remaining articles were checked according to the inclusion and exclusion criteria. To avert selection bias, two authors (S.K. and C.S) independently assessed the article’s eligibility, and any disagreement was resolved by the third author (F.Y.). Thirty-one eligible articles were included in this meta-analysis.

2.3. Study quality assessment

To assess the quality of the included studies, a checklist appropriate to the study type was used. The quality of randomized controlled trials (RCT) or non-randomized controlled studies (NRS) was evaluated using the Physiotherapy Evidence Database (PEDro) scale ²³. PEDro scores range from 0–10 (a greater score indicates higher quality), with 6 as the cutoff score for high-quality studies ²⁴. PEDro scores were extracted from the official PEDro database when available ²³. The authors assessed the quality of studies absent from the PEDro database. For observational studies, the STROBE (Strengthening The Reporting of Observational Studies in Epidemiology) checklist was used ²⁵. The STROBE checklist consists of 32 items within six general sections with a possible maximum score of 32. A study with a STROBE score of 16 or higher was deemed to have good methodological quality ². A funnel plot analysis assessed the publication bias.

2.4. Data retrieval and outcomes determination

Three categories of information were extracted: 1) publication characteristics (author(s), year, country/region), 2) sample characteristics (sample size, age, dementia type), and 3) study characteristics (fall tracking duration and approach, fall prevalence, number of fallers if the fall prevalence was not directly given, mean and standard deviation (SD) of the average number of falls, number of recurrent fallers, and number of injured fallers).

A person who had experienced multiple falls was considered a recurrent faller during the respective fall tracking duration. The extracted data were used to calculate our primary (fall prevalence) and secondary (number of falls, rate of recurrent fallers, and rate of injured fallers) outcome measures, if not directly reported by the respective study. If data necessary to determine the fall metrics were not reported, the authors were contacted.

As the primary outcome, fall prevalence was calculated as the ratio of the number of fallers to the total number of participants in each study (Fig. 1). The secondary outcomes included the following variables. The average and SD of the number of falls were determined as the total number of falls divided by the sample size (Fig. 1). The rate of recurrent fallers (or injured fallers) was computed by dividing the number of recurrent fallers (or injured fallers) by the number of fallers. If an included article was an interventional study, we utilized fall risk-related data reported prior to the intervention. This approach ensures that the fall risk information used in our meta-analyses was not affected by any intervention. For an interventional study that involved multiple groups, the pre-intervention fall risk information was aggregated for the control/placebo and training/intervention groups.

2.5. Meta-analysis

Meta-analyses were conducted using the random-effects model with inverse variance weights within Review Manager (RevMan) 5.3 (Nordic Cochrane Centre, Denmark). The 95% confidence intervals (CI) for the outcome measures were estimated. The meta-analysis results were presented using forest plots. For the fall prevalence, separate meta-analyses were performed for three fall tracking durations (4, 6, and 12 months). The number of fallers and the sample size of each study were fed into RevMan to calculate the pooled fall prevalence over each tracking duration. The meta-analysis for the average number of falls was performed for studies with a duration of 12 months. The mean and SD of the number of falls for each study were entered into RevMan. Two meta-analyses summarized the rate of recurrent fallers in two tracking durations: 6 and 12 months. The respective number of recurrent fallers and the number of fallers were entered into RevMan. Since the rate of injured fallers was reported over 4 and 12 months by only one study for each, no meta-analysis was conducted for this measurement.

RESULTS

3.1. Study selection

The initial literature search yielded a total of 5,831 studies (Fig. 2). After removing the qualitative studies, duplicates, and titles or abstracts that were irrelevant to the research purpose (5,026), the remaining studies (805) were reviewed based on the inclusion and exclusion criteria. The exclusion criteria led to the removal of 770 studies due to their inclusion of individuals with MCI. Thirty-five studies met the criteria, and further review led to the elimination of four more studies ^26–29. These four studies were removed from our meta-analysis because they used the same participants as other included studies ^{26, 27}, reported fall risk information collected after intervention ²⁸, or did not specify the dementia type ²⁹. Therefore, 31 studies were qualified for the meta-analyses.

Figure 2.

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram showing the study selection process. The four studies removed at the last stage were due to different reasons: duplicate samples, post-intervention fall risk measurements, or unspecified dementia type.

3.2. Study characteristics

The 31 eligible studies were published between 1987 ^{30, 31} and 2022 ¹⁹ and included five RCT ^{20, 32–35}, four NRS ^{17, 18, 22, 36}, and 22 observational studies ^{14–16, 19, 21, 30, 31, 37–51}. The mean PEDro score for the RCT and NRS was 4.44±2.06 (Appendix 2), and the mean STROBE score for the observational studies was 18.14±3.22 (Appendix 3). According to the PEDro and STROBE cutoff points, the quality of the included studies was deemed fair to good.

Study populations included older adults with AD with an average age spanning between 65 ⁴² and 86.6 ²⁰ years (Table 1). The sample sizes varied significantly among studies: 14 ²¹ to 2,490 ²². The total number of people with AD in the 31 studies was 4,654. The studies were conducted in several countries: Australia ^{21, 35, 36, 46, 52}, Brazil ^{15, 16, 32, 38, 41, 43, 53}, Canada ^{40, 51}, Finland ³⁴, France ²⁰, Japan ^{22, 44, 45, 47, 48}, Pakistan ⁴², Spain ³³, Turkey ⁵⁰, United Kingdom ^{14, 37, 39}, and the U.S. ^{19, 30, 31, 49}. The living environments of the participants for the included studies also varied significantly: community ^{15, 20, 21, 31, 32, 34–36, 38, 41, 43, 45, 46, 49, 52}, nursing homes ⁴⁷, and AD caring centers ⁴⁰. The information on the living conditions was not reported for the other studies ^{18, 22, 30, 48, 50} (Table 1).

Table 1.

Summary of study and participant characteristics of all included studies.

Study	Publication year	Fall tracking duration (months) *	Country	Sample size	Study design	Participant mean age (years)	Fall data collection approach	Living environment	Fall definition provided
Allan 14	2009	−12	U.K.	38	Observational	79	NA	Care clinics	Yes
Allan 37	2005	−12	U.K.	40	Observational	78.6	NA	Care clinics	Yes
Ansai 15	2017	−12	Brazil	38	Observational	76.7	NA	Community-dwelling	No
Ansai 38	2018	6	Brazil	37	Observational	76.6	Monthly telephone calls	Community-dwelling	Yes
Ballard 39	1999	−6	U.K.	35	Observational	80	Daily fall diaries	NA	No
Biju 19	2022	12	U.S.	48	Observational	74.9	Monthly calendars	NA	Yes
Borges 16	2015	−12	Brazil	26	Observational	75	Face-to-face interview	Community-dwelling	No
Buchner 30	1987	36	U.S.	117	Observational	79	NA	NA	No
Camicioli 40	2004	12	Canada	42	Observational	82.6	Bimonthly chart reviews	Alzheimer care units	Yes
Cezar 32	2021	−12	Brazil	35	RCT	79.3	NA	Community-dwelling	No
Coelho 41	2012	−4	Brazil	23	Observational	77.8	Questionnaire	Community-dwelling	No
Dev 42	2021	12	Pakistan	140	Observational	65	Monthly telephone calls	NA	No
Goncalves 43	2018	6	Brazil	38	Observational	76.6	Monthly telephone calls	Community-dwelling	Yes
Horikawa 44	2005	12	Japan	104	Observational	74.1	Caregiver reports	Home	Yes
Kato-Narita 18	2009	−12	Brazil	45	NRS	80	Caregiver reports	NA	Yes
Kudo 45	2009	4	Japan	51	Observational	74.1	Face-to-face interview	Community-dwelling	Yes
Lorbach 46	2007	−12	Australia	21	Observational	79.3	NA	Community-dwelling	No
Morris 31	1987	48	U.S.	44	Observational	71.4	NA	Community-dwelling	No
Nakamura 47	1996	24	Japan	97	Observational	75.2	Nursing staff reports	Nursing home	No
Oki 48	2021	−12	Japan	47	Observational	80.9	Caregiver reports	NA	Yes
Puente-Gonzalez 33	2021	−12	Spain	72	RCT	76.7	Monthly telephone calls	Center of AD	Yes
Roitto 34	2018	−12	Finland	179	RCT	77.6	NA	Community-dwelling	No
Ryan 49	2011	−6	U.S.	43	Observational	80.6	NA	Community-dwelling	Yes
Soysal 50	2021	−12	Turkey	86	Observational	81	Questionnaire	NA	No
Suttanon 17	2012	−12	Australia	25	NRS	81	Self-report	Community-dwelling	No
Suttanon 21	2011	−12	Australia	14	Observational	79.6	Caregiver reports	Community-dwelling	No
Suttanon 36	2013	12	Australia	15	NRS	80.9	Self-report/caregiver reports	Community-dwelling	Yes
Suttanon 35	2013	−6	Australia	40	RCT	81.9	Self-report/caregiver reports	Community-dwelling	No
Tchalla 20	2013	−12	France	96	RCT	86.6	Caregiver reports	Community-dwelling	Yes
Tsujimoto 22	2022	−12	Japan	2,490	NRS	79	Self-report/caregiver reports	NA	No
Weller 51	2004	−12	Canada	528	Observational	83.2	Survey	Health institutions	No

^*: A positive/negative number indicates that the fall data was collected prospectively/retrospectively.

AD = Alzheimer’s disease. NA = Not Available.

RCT = Randomized Control Trial. NRS = Non-randomized Controlled Study. U.K. = the United Kingdom. U.S. = the United States

Among the included studies, 14 studies defined a fall as “an unexpected event in which the person comes to rest on the ground, floor, or lower level” ^{14, 18–20, 33, 35, 37, 38, 40, 43–45, 48, 49}. This definition was consistent with the one developed by the World Health Organization ⁵⁴. The rest 17 studies did not provide a specific definition of falls (Table 1).

The fall tracking duration stretched over 4 ^{41, 45}, 6 ^{35, 38, 39, 43, 49}, 12 ^{14–16, 19–22, 32–34, 36, 37, 40, 42, 44, 46}, ^{48, 50–53}, 24 ⁴⁷, 36 ³⁰, and 48 ³¹ months (Table 1). Various approaches (such as daily fall diaries, monthly telephone calls or fall calendars, bimonthly fall diaries, or biannual calendars) were used by participants ^{22, 35, 36, 52}, or caregivers ^{18, 20, 21, 44, 48} to collect fall data. For observational studies, 12 reported retrospective falls ^{14–16, 21, 37, 39, 41, 46, 48–51}, and 10 reported falls prospectively ^{19, 30, 31, 38, 40, 42–45, 47}. For RCT and NRS, the pre-training fall risk data was used ^{17, 18, 20, 22, 32–35} except for one study ³⁶, in which the fall risk data was prospectively gathered. Out of the 26 studies that reported fall prevalence-related data, fall prevalence was directly provided in some studies 16, 19, ^{20, 22, 34, 38, 41, 43, 45, 46, 48, 50, 53} or calculated based on the number of fallers and sample sizes for others ^{14, 15, 21, 30, 31, 35–37, 39, 40, 42, 44, 47, 49, 51, 52}. Nine ^{14–16, 18, 19, 21, 32, 33, 40} and 12 ^{14, 16, 20, 21,37–39, 42, 44, 49, 52, 53} studies respectively provided information regarding the number of falls and rate of recurrent fallers (Table 1). The information for the rate of injured fallers was only provided in two studies ^{42, 45}. One was based on a 12-month tracking duration ⁴² and the other used the data collected over 4 months ⁴⁵.

3.3. Primary meta-analysis: Fall prevalence

Fourteen studies recorded the fall history over the past 12 months ^{14–16, 20–22, 34, 37, 46, 48, 50–53}, while five studies tracked fall incidences over one year prospectively ^{19, 36, 40, 42, 44}. Of the 4,022 participants in these 19 studies, 1,571 were fallers. The annual fall prevalence exhibited considerable variation ranging from 25% ¹⁹ to 73.96% ²⁰ (Fig. 3A–1). The meta-analysis yielded a pooled annual fall prevalence of 44.27% (95% CI=[38.84%, 49.70%]). The removal of the study with the largest fall prevalence ²⁰ decreased the overall fall prevalence by 2.61% (41.66%), and the removal of the study with the smallest fall prevalence ¹⁹ increased the overall fall prevalence by just 1.13% (45.40%). The composite fall prevalence over 6 (based on five studies ^{35, 38, 39, 43, 49}) and 4 months (based on two studies ^{41, 45}) was 46.11% (95% CI=[33.45%, 58.76%], Fig. 3A–2) and 15.90% (95% CI=[0.45%, 31.35%], Fig. 3A–3), respectively. Since the fall prevalence over 24 ⁴⁷, 36 ³⁰, and 48 ³¹ months each was reported in a single study, no meta-analyses were performed for these three durations. Their reported fall prevalence was respectively 28.87% (95% CI=[10.85%, 37.89%]), 17.09% (95% CI=[10.27%, 23.91%]), and 36.36% (95% CI=[22.15%, 50.57%]). The funnel plot indicates a low publication bias (Fig. 4).

Figure 3.

Forest plots of A) the fall prevalence over 1) 12 months, 2) 6 months, and 3) 4 months, B) the annual average number of falls, and C) the rate of recurrent fallers in 1) 12 months and 2) 6 months among older adults with Alzheimer’s disease.

Figure 4.

The funnel plot based on 19 studies reporting the annual fall prevalence. The Y-axis shows the standard error of fall prevalence (%), and the X-axis indicates the fall prevalence (%). Each of the 19 open circles represents the observed point estimate for an individual study. The vertical dashed line represents the annual pooled fall prevalence for the 19 studies. All observed point estimates are almost symmetrical around the pooled fall prevalence, and most studies are inside the inverted funnel region, indicating a low publication bias.

3.4. Secondary meta-analyses: Other fall-related metrics

The number of falls information was available for nine studies over a 12-month interval ^{14–16, 18, 19, 21, 32, 33, 40}. The average annual number of falls exhibited a wide range of 0.3 ³³ to 3.76 ¹⁵ falls/person (Fig. 3B). The collective yearly average number of falls was 1.30/person (95% CI=[0.74, 1.86]). Another study reported this metric for the past 6 months (0.95, 95% CI=[0.21, 1.69]) ³².

Nine studies reported recurrent faller data over 12 months ^{14, 16, 20, 21, 37, 42, 44, 52, 53}. The meta-analysis derived an overall rate of recurrent fallers of 42.08% (95% CI=[27.80%, 56.36%]) (Fig. 3C–1). The meta-analysis of three studies that described the recurrent faller data over a 6-month period indicated that 49.86% (95% CI=[29.22%, 70.50%]) of fallers experienced multiple falls (Fig. 3C–2) ^{38, 39, 49}.

Two studies reported data about injured fallers: one was over 12 months ⁴² and the other over 4 months ⁴⁵. The reported 12-month and 4-month injured faller rates were 45% (95% CI=[29.58%, 60.42%], Fig. 3D) and 60% (95% CI=[46.55%, 73.45%]), respectively.

DISCUSSION

Although it has been conceptually recognized that people with AD fall more than their cognitively healthy counterparts, the reported fall risk in people with AD varied drastically. The present meta-analysis aimed at summarizing the fall risk data, including fall prevalence, the number of falls, the rate of recurrent fallers, and the rate of injured fallers, in people with AD based on the currently available data in the literature.

Our meta-analysis indicated an annual fall prevalence of 44.27% among 4,022 people with AD (Fig. 3A–1). This value could more closely and credibly reflect the fall prevalence in people with AD than previous studies, which reported vastly diversifying fall prevalence between 25% ¹⁹ and 73.96% ²⁰ over limited sample sizes ranging from 14 ²¹ to 2,490 ²². Our pooled annual fall prevalence among people with AD is about 1.7 folds of that in older adults without cognitive impairment (44.27% vs. 26.5%) ^{55, 56}. This finding agrees with most previous studies. Specifically, among the 19 studies that reported the annual fall rate in people with AD, 18 documented a fall prevalence higher than the one in older adults without cognitive impairment (28.57%−73.96%, Fig. 3A–1). Despite the wide range of fall prevalence amongst studies, the values reinforce the notion that people with AD are at a higher fall risk than older adults without cognitive impairment.

Falls are a complex and multifactorial phenomenon involving the interactions between intrinsic and extrinsic risk factors ^{57, 58}. Although the factors increasing the fall risk among people with AD are not fully understood, people with AD have some underlying fall mechanisms in common with older adults without cognitive impairment ⁵⁹. Research demonstrated that the brain regions necessary for cognitive functioning and memory play a role in coordinating and maintaining mobility, balance, and gait ⁶⁰. Cognitive function influences gait through judgment, reaction time, and psychomotor control ⁶¹. Patients with AD exhibit more compromised balance dysfunctions, postural instability, mobility deficits, and sensory system impairments compared to healthy controls ⁶². As mobility impairments, balance dysfunctions, and muscle weakness are well-established fall risk factors ⁶³, these would explain why people with AD show a higher risk of falls than older adults without cognitive impairment.

Also, people with AD could have unique fall risk factors that are not present in cognitively healthy older adults. Dementia has been identified as an independent fall risk factor ^{55, 64}. Previous studies suggested that the severity of dementia is positively correlated with fall risk ^{65, 66}. In addition, dementia-related psychological factors, such as verbally disruptive and attention-seeking behaviors, depressive symptomatology, and a higher level of emotional stress in caregivers, were also associated with an increased fall risk among persons with dementia ⁶⁵. People with AD also exhibit a significantly higher fear of falling than cognitively healthy older adults ¹⁶. The discrepancy in the annual fall prevalence between people with AD and healthy older adults not only reflects that the mechanisms of falls differ between individuals with and without dementia but also implies that any fall preventive strategies targeting people with AD should consider such between-population differences in fall mechanisms.

This meta-analysis also examined other important fall-related metrics, including the average number of falls, rate of recurrent fallers, and rate of injured fallers in people with AD. Our results showed that the average number of falls among 358 people with AD over a 12-month timeframe was 1.30 falls/person (Fig. 3B). Concurring with previous studies ^{16, 42}, this value is higher than the one among older adults without cognitive impairment (0.3 falls/person). Among AD fallers, about 42.08% fell more than once (Fig. 3C), and 45% experienced at least one fall-related injury ⁴². The ratio of recurrent fallers in this population is also higher than the reported proportion of multiple fallers in healthy older adults (12.5%) ¹⁸. These supplementary measurements could offer extra meaningful information to characterize the fall risk in people with AD. For instance, the average number of falls and the rate of injured fallers could help estimate the total fall incidences and injurious falls for people with AD, which could be valuable for allocating medical resources for mitigating fall-related injuries. Additionally, the information on recurrent fallers may be helpful to investigate further how fall history is related to falls in people with AD. Moreover, the higher fall incidences and proportion of recurrent fallers in persons with AD than in older adults without cognitive impairment reinforce the notion that the fall mechanisms and preventive strategy could differ between these populations ⁶⁷.

Considering the increasing number of studies on people with AD, it is essential to standardize results reporting methods associated with falls. Our results demonstrate that the current literature has adopted various paradigms for collecting and reporting fall data. For example, the fall incidence tracking duration varied between 4 ^{41, 45} and 48 ³¹ months; the average number of falls was based on either the number of fallers ¹⁵ or the sample size ^{14, 16, 21, 32, 33}; the injured body sites resulting from falls were not specified ^{42, 45}; the methods of collecting fall data were different; and the terminologies used were not consistent. The unstandardized methods of collecting and reporting fall incidents could negatively affect the communication between researchers and the dissemination/deployment of the findings. Therefore, uniform terminologies and reporting systems are highly desired to facilitate fall prevention efforts for this population.

This meta-analysis has limitations. First, although several databases were searched, some articles could be missed because of the restricted language. Second, included studies in this meta-analysis were from diverse settings, including community, special care units, and residential aged care facilities. The diversity in living situations could bring biases to our findings. Third, this meta-analysis did not check if other factors (e.g., gender, race, ethnicity, age) would impact the fall risk in people with AD due to the small sample size of the studies. Additionally, our cutoff age excluded those with AD who were younger than 65 years. However, given that about 5–10% of the reported AD cases are early-onset AD ⁶⁸, our findings could still reflect the fall risk among the majority of older people with AD. Fourth, fall tracking durations and tools could also bias our findings. For example, the retrospective fall recall or prospective fall tracking with a long recall duration by either participant themselves or their caregivers could introduce inaccuracy or subjectivity to fall data. Therefore, more reliable and objective procedures built upon new technologies should be used to track falls in people with AD.

In summary, this meta-analysis concluded that the annual fall prevalence is about 44.27% among people with AD, which is higher than cognitively healthy older adults. Other fall risk-related measurements were also greater in people with AD compared to their cognitively normal counterparts. Consistent approaches are needed to collect and report fall-related data for individuals with AD across studies.