Removing home hazards for older adults living in affordable housing: A stepped-wedge cluster randomized trial

Abstract

Background:

Falls are the leading cause of injury, disability, premature institutionalization, and injury-related mortality among older adults. Home hazard removal can effectively reduce falls in this population but is not implemented as standard practice. This study translated an evidence-based home hazard removal program for delivery in low-income senior apartments to test whether the intervention would work in the “real world.”

Methods:

From May 1, 2019 to December 31, 2020, a stepped-wedge cluster randomized trial was used to implement the evidence-based Home Hazard Removal Program (HARP) among residents with high fall risk in 11 low-income senior apartment buildings. Five clusters of buildings were randomly assigned an intervention allocation sequence. Three-level negative-binomial models (repeated measures nested within individuals, individuals nested within buildings) were used to compare fall rates between treatment and control conditions (excluding a crossover period), controlling for demographic characteristics, fall risk, and time period.

Results:

Among 656 residents, 548 agreed to screening, 435 were eligible (high fall risk), and 291 agreed to participate and received HARP. Participants were, on average, 72 years, 67% female, and 76% Black. Approximately 95.4% of fall prevention strategies and modifications implemented were still used 3 months later. The fall rate (per 1000 participant-days) was 4.87 during the control period and 4.31 during the posttreatment period. After adjusting for covariates and secular trend, there was no significant difference in fall rate (incidence rate ratio [IRR] 0.97, 95% CI 0.66–1.42). After excluding data collected during a hiatus in the intervention due to COVID-19, the reduction in fall rate was not significant (IRR 0.93, 95% CI 0.62–1.40).

Conclusions:

Although HARP did not significantly reduce the rate of falls, this pragmatic study showed that the program was feasible to deliver in low-income senior housing and was acceptable among residents. There was effective collaboration between researchers and community agency staff.

INTRODUCTION

Falls remain the leading cause of injury, long-term disability, premature institutionalization, and injury-related mortality in older adults.^1–4 Over 25% of community-dwelling adults aged 65 years and older fall each year,⁵ and those over age 70 have an especially high fall risk.⁶ Home hazards are associated with an increased risk of falling among older adults in the US.⁷

Home hazard removal is effective in reducing falls among older adults. A recent review of trials comparing home hazard removal with usual care or control found a 26% reduction in the rate of falls among older adults overall and a 38% reduction among older adults at high risk of falling.⁸ Home hazard removal is also cost effective.⁹ Despite these findings, implementation of home hazard removal is not standard practice in the US. Although there has been tremendous investment in the discovery of effective fall prevention interventions for older adults at home,^{8, 9} this work has not translated into scalable programs that reduce falls in the community.

Low-income senior apartments, such as those supported by the Supportive Housing for Elderly Program Section 202, are home to older adults with a high fall risk profile (eg, older, functional limitations).¹⁰ Despite the presence of accessibility features in the residential units, fall hazards (eg, slippery rugs, clutter, unstable beds) are still present. Given that fewer than half of primary care physicians screen older patients for fall risk^{11, 12} and the lower rate of preventive health care among this low-income population, implementing home hazard removal programs in such settings could be an effective way to translate evidence-based fall prevention programs into the community.

The number of older households eligible for rental assistance is projected to increase by 1.3 million between 2020 and 2030.¹³ There is an immediate need for programs to reduce home hazards for vulnerable older adults. The Home Hazard Removal Program (HARP) is an intervention, based on a competence/press theoretical framework,¹⁴ that targets home hazard identification and removal. HARP previously demonstrated a 38% reduction in the number of falls for older adults with high fall risk.¹⁵ In this study, we translated and tested the effectiveness of this evidence-based fall prevention program (HARP) for delivery in low-income senior apartments to determine whether the intervention can be implemented in a “real-world” community setting. A stepped-wedge design was selected in this implementation trial because equipoise was tipped in favor of the HARP intervention. Although there are tradeoffs with regard to bias controlled by the experimental design, the stepped-wedge design allows for improved rigor in the analysis phase. For this study, the stepped-wedge design also provided the opportunity to examine important implementation outcomes and allowed for recruitment of participants with the intention of providing the intervention for all eligible individuals residing in the senior apartment buildings. We hypothesized that the implementation of HARP within the low-income senior apartments would be feasible and would decrease the rate of falls in this population.

METHODS

Study design

We performed a stepped-wedge cluster randomized controlled trial^{16, 17} from May 1, 2019 to December 31, 2020 (clinical trials.gov NCT03780777). The rationale for studying fall prevention in affordable housing^{18, 19} includes a need for pragmatic effectiveness in fall prevention studies,¹⁹ strong face validity,²⁰ strong evidence for home hazard removal to reduce falls among high-risk fallers like those who live in Section 202 housing,²¹ and minimal risk associated with the intervention. The study was granted a waiver of consent by the Washington University Institutional Review Board because this was an evaluation of an evidence-based program primarily implemented by the partnering agency who owned the buildings. While there was no formal consent or assent process, the program was explained to every resident during each component of the intervention, and each resident was allowed to refuse or opt out of participation at any time. Reporting of this study follows Consolidated Standards of Reporting Trials (CONSORT) guidelines, using the extension for stepped-wedge cluster randomized trials.²²

This trial was underway when the COVID-19 pandemic began. Washington University closed in-person study operations from March 20, 2020 to June 7, 2020. Although we attempted to deliver the intervention remotely, participants were not familiar with Zoom, and telephone delivery proved too difficult for home hazard removal; thus, intervention delivery was halted during this time. Fall monitoring was completed via phone during this period.

Participants

The stepped-wedge cluster randomized trial was conducted in 11 low-income senior apartment buildings, 10 of which were federally subsidized under Section 202, located in the urban and suburban St. Louis metropolitan area (in Missouri and Illinois). Residents are eligible for Section 202 housing if the household includes at least one person aged 62 years or older and the household income is less than 50% of the area median income.²³ All buildings were owned and managed by the same faith-based, not-for-profit organization. Buildings were yoked in groups of 2 (one group of 3) based on geographical proximity into 5 clusters (Figure 1). The 5 clusters included 586 total housing units (apartments). All residents were eligible to participate in screening when their site was randomly assigned to cross over into the intervention period.

Figure 1.

Design of stepped-wedge cluster-randomized trial

Randomization

The study statistician randomly ordered the 5 clusters using a computer-generated random allocation sequence. Agency staff (including building managers) were provided with a schedule that reflected the study timeline and randomized sequence of intervention delivery. After a 17.5-week baseline fall monitoring period for all buildings, the first cluster of 2 buildings began the intervention. All other clusters continued fall monitoring. This procedure continued at approximately 10-week intervals until all 5 clusters crossed over into intervention (Figure 1). The 2 units that converted to treatment just prior to March 2020 were delayed midtreatment because of the COVID-19 pandemic. Thus, one step had a longer intervention period (15.6 weeks). Due to the nature of the study design and intervention, there were no blinding procedures.

Intervention

The intervention had 3 components designed to reduce falls among residents: (1) fall risk awareness for staff and residents, (2) resident fall risk screening, and (3) a home hazard removal program (HARP) for participants at high risk of falling. Study team occupational therapy (OT) practitioners delivered the core intervention elements at each site with assistance from building staff. All units received the same fall risk awareness/educational elements, fall risk screening, and intervention.

Fall risk awareness.

A fall risk education program was provided to staff (building and case managers) from each building 3 months before the trial start date. When a building unit crossed over to treatment, education/awareness activities for residents were implemented and maintained over the course of the monitoring period (described in Table 1).

Table 1.

Fall risk education programs designed to raise fall risk awareness

Element	Description	Delivery method	Target
Educationa	Fall prevention program introduction (fall risks, fall prevention evidence)	Brown bag sessions and didactic sessions. Two sessions delivered over 2 months.	Apartment managers and service coordinators
Fall prevention awareness/ bingo46	Fall Prevention Bingo includes 75 fall prevention facts delivered in an approximately 1-hour bingo session	In-person in each facility; offered to all residents as each building crossed over to treatment.	Residents
Wall calendar	Evidence-based fall prevention tips provided on wall calendar	Hand-delivered to each resident	Residents
Public awareness posters	Evidence-based fall prevention tips and facts	Posted in high-traffic areas in facility (near mailboxes, in the elevator)	Residents and staff

^aDelivered prior to intervention implementation.

Fall risk screening and eligibility for HARP.

When the building crossed over into the intervention period, all residents were invited to participate in a fall risk screening to determine their eligibility for HARP using the Stopping Elderly Accidents, Deaths, and Injuries (STEADI) algorithm.²⁴ Residents were asked 3 questions: (1) Have you fallen in the past year? (2) Do you feel unsteady when walking? and (3) Are you worried about falling? Endorsing any of the questions resulted in offering gait screening using the Timed Up & Go,^{25, 26} a gait speed test that assesses the time to stand, walk 10 meters, turn, and return to the chair and sit back down. Residents with a time greater than 12 seconds were referred to HARP. Residents with low risk were offered the Home Safety Self-Assessment tool,²⁷ an evidence-based self-assessment of home hazards, and fall prevention resources and were instructed in how to reach the fall prevention team if they believed their fall risk changed over time.

Home Hazard Removal Program (HARP) is a fully manualized intervention including 1–2 home hazard removal visits and a 3-month booster delivered by trained and licensed OT practitioners.²⁸ HARP is described elsewhere,^{15, 29} but briefly, the intervention elements consist of: (1) a comprehensive assessment of the individual, their behaviors, and the environment; (2) a home hazard removal plan; (3) remediation of hazards; and (4) a booster session 3 months after the intervention. The primary mechanisms for resolving barriers in HARP are minor home repair (eg, grab bars), adaptive equipment, task modification, and education and self-management strategies to improve awareness of fall risks.³⁰ During Session 1, environmental hazards and unsafe behaviors are identified using the Westmead Home Safety Assessment.³¹ A tailored barrier removal plan is developed. In Session 2 and an optional Session 3, the OT practitioner facilitates home modifications (eg, placing nonskid strips in the bathtub or engaging building maintenance personnel to install grab bars) and provides education and training on their use. Three months after the initial visit, participants receive an in-person booster session to identify and remediate new home hazards and address any issues from the initial set of strategies and modifications. All modifications were provided at no cost to study participants. If the participant exhibited cognitive impairment, the OT practitioner used their clinical reasoning skills to adapt the intervention, specifically the active elements, while ensuring that the essential elements of HARP were delivered with consistency. For example, a participant with a cognitive impairment might require instructions in a picture format or reminders/signs placed in the environment, or a caregiver might be included in the education.

Demographic characteristics

Demographic characteristics including gender, age, and race (Black, White, Asian, Latinx, and race not listed) were obtained from participants or building staff (if needed).

Outcomes

Process outcomes were measured to address the acceptance of HARP in the community and included the proportion of residents who agreed to participate in each element of the program (education/awareness session, screening, and home hazard removal). Adherence was calculated as the number of recommendations that were in use at the booster visit divided by the total number of recommendations suggested.

The primary outcome measure was rate of falls, defined as the number of falls per 1000 participant-days during the observation period. Falls were defined as coming to rest on the ground or below knee level unexpectedly.³² An exploratory outcome was rate of serious fall-related injury, defined as the number of falls requiring an emergency department visit per 1000 participant-days. Participants were given a monthly calendar to help keep track of their falls. Participants received an automated phone call monthly to report any falls during the previous month. The automated call verified that the person on the phone was the participant and specifically asked, “A fall is an unexpected event in which a person comes to rest on the ground, floor, or a lower level. In the last month, how many falls, including slips or trips, in which you lost your balance and landed on the floor, ground, or lower level have you had? Please type in the number of falls on your keypad.” Staff blinded to treatment status contacted participants who could not or did not want to receive the automated phone call. When a fall was reported, a blinded rater called the participant again to verify the details of the fall by asking how many times the participant fell in the past month, the date and time of the fall, what they were doing when they fell, whether they visited the emergency department, and other circumstances surrounding the fall (eg, location, direction they fell, what surface they fell on). If the participant had a cognitive impairment and a caregiver was available, staff would verify that the fall occurred and details of the fall with the caregiver. Participants who missed an automated monthly phone call were called by research staff multiple times as needed to complete their monthly fall call. If this was unsuccessful and staff were in the participant’s building completing other participants’ treatment visits, they visited the participant in person to complete any missing months for fall reporting. Participants received a $5 credit on a reloadable gift card for each month they reported their falls.

Several additional measures of fall risk were assessed during the first visit. Falls have multifactorial risk³³; thus, we used a composite fall risk measure that includes several validated fall risk assessments for intrinsic (eg, medications, cognition, depression, substance use, functional status) and extrinsic (ie, environmental hazards) fall risk factors. This follows similar methodology to that of our prior randomized controlled trial of HARP¹⁵ and uses criteria for practical physiological assessments³⁴ and measures with good prediction of future falls.³⁵ Participants were asked whether they were taking medications that could increase fall risk (eg, opioids, antidepressants/anti-anxiety medications, sleep aids, blood pressure/heart medications, muscle relaxants). Cognitive impairment was assessed with the Short Blessed Test (SBT³⁶; score ≥10 is associated with significant impairment and was considered a fall risk. Concern about falling was measured using the 7-item Short Falls Efficacy Scale International (Short FES-I³⁷; score ≥9 indicates moderate or high concern and was considered a fall risk). Depression was measured using the 15-item Geriatric Depression Scale Short Form (GDS-SF³⁸; score ≥5 indicates probable depression and was considered a fall risk). Alcohol use was measured using the 10-item Short Michigan Alcoholism Screening Test–Geriatric version (SMAST-G^{39, 40}; score ≥2 indicates an alcohol problem and was considered a fall risk). Functional status was assessed using the 14-item Older American Resources and Services Activity of Daily Living scale (OARS ADL⁴¹; moderate to total impairment was considered a fall risk). Home hazards were assessed using the Westmead Home Safety Assessment, which identifies 72 environmental home hazards for older adults at risk of falling³¹ (≥4 hazards was considered a fall risk). A fall risk composite score was calculated as the percentage of the above fall risk assessments that indicated fall risk completed by the participant.

Sample size calculation

An a priori sample size calculation was undertaken based on preliminary data and the literature, with a plan to enroll 10 buildings with 39 subjects in each. We projected a 75-day (2.5 months, or 10.7 weeks) baseline observation/fall monitoring period (all in control condition), after which 2 buildings would be randomly selected to transition to the intervention (5 steps). Each step was projected to be 75 days (2.5 months, or 10.7 weeks) for outcome measurements. Our previous data showed the rate of falls in the control condition to be 6.6 per 1000 person-days. The clinically meaningful intervention effect was defined as a 30% reduction in the rate of falls, based on prior trials of fall prevention interventions,^{42, 43} resulting in a fall rate of 4.6 per 1000 person-days in the intervention condition. Under 3 different configurations of between-building and within-building ICCs, we projected to have sufficient power (all >80%) to demonstrate a statistically significant intervention effect with a two-sided alpha of 5%. Power calculations were conducted using simulation with the R package SWSamp.

Statistical analysis

Study data were managed using Research Electronic Data Capture (REDCap), a secure, web-based application designed to support data capture for research studies.⁴⁴ Data were exported to SPSS version 27 and SAS 9.4 for analysis. Comparisons of demographic characteristics by acceptance of the intervention were conducted with Pearson chi-square or Fisher’s exact tests for categorical variables and independent samples t-tests for continuous variables. All primary analyses were conducted using an intention-to-treat paradigm. Additional analyses excluded data collected during the COVID-19 intervention hiatus. We calculated the fall rate (number of falls per 1000 participant-days) with corresponding 95% confidence intervals (CIs) for building by period, as well as the overall fall rate and 95% CI in posttreatment and control conditions. Three-level negative-binomial models were used to compare the fall rate between posttreatment and control conditions (crossover period excluded from analysis) with individual repeated measures as the first level, individual as the second level, and building as the third level. First, unadjusted models included only the intervention indicator (coded 0 before the time of intervention, 1 after the intervention) as a fixed effect. Then, adjusted models added time period and other covariates (age, gender, and fall risk composite score) as fixed effects. The effect of time period in addition to the intervention indicator was included because there can be confounding effects of general time trends during the study period in stepped-wedge designs. Models were performed using GLIMMIX in SAS software. According to the sample size calculation above, 390 participants were needed to detect the targeted effect size with two-sided alpha of 5%. Our sample size for analysis was 291.

RESULTS

Participant flow

Participant flow overall and by cluster is shown in Figure 2. Between May 1, 2019 and December 31, 2020, 656 residents spent at least 1 month in one of the 11 buildings, with 548 residents (83.5%) agreeing to screening. Of those who were screened, 435 (79.4%) were eligible and were offered HARP. Of those offered the intervention, 309 (71.0%) agreed to participate. Reasons for declining participation included no perceived benefit (n=54), not wanting to contribute to research (n=21), poor health (n=5), lack of time (n=3), became ineligible before treatment began (eg, moved; n=10), or miscellaneous other reasons (n=30; 3 were unknown). Those who declined the intervention were more likely to be male than those who accepted (46.0% vs. 33.3%, Χ²_df(1)=6.2, p=0.013).

Figure 2.

Participant flow. CTL=control; CO & TRT=crossover and treatment; Post-TRT=posttreatment

Among the 309 who agreed to the intervention, 5 did not agree to monthly fall monitoring, and 13 did not complete an initial assessment visit (ie, unable to contact/schedule the appointment [n=7], no longer cognitively able to participate [n=2], deceased [n=2], or moved [n=2]). Thus, 291 participants (66.9% of those eligible) completed an initial assessment visit and agreed to monthly fall monitoring and were included in analysis.

Of the 291 participants, 1 opted out of fall calls and did not re-enroll (cluster 3) and 37 dropped out (12 in cluster 1, 11 in cluster 2, 6 in cluster 6, 6 in cluster 4, and 2 in cluster 5) due to death (11), moving to an assisted living or skilled nursing facility (5), moving to another independent living facility (9), moving to the home of their caretaker (8), being evicted (1), or for other unknown reasons (3). In addition, 22 (6 in cluster 1, 3 in cluster 2, 8 in cluster 3, 2 in cluster 4, and 3 in cluster 5) opted out of fall calls for a short period of time but then subsequently re-enrolled in fall calls. Rates of dropout did not significantly differ between participants who fell versus those who did not fall during the study period. The average number of months of fall calls per participant was 16.9 (SD 4.8; median 20, interquartile range 15–20).

Participant characteristics

Demographic and fall risk characteristics among participants who received the intervention (n=291) are presented in Table 2. The mean age was 72.1 years (SD=10.4). About two-thirds (67.4%) were female, and 76.2% were Black. Approximately 77.4% were taking medications that could increase their fall risk, 29.0% had significant cognitive impairment (≥10 on the SBT), 80.8% had at least a moderate concern about falling (≥9 on the Short FES-I), 25.8% had possible depression (≥5 on the GDS-SF), and 11.8% had a possible alcohol problem (≥2 on the SMAST-G). In addition, 70.2% had at least moderate ADL impairment, and 71.1% had at least 4 environmental fall risk hazards. The average fall risk composite score (percentage of fall risk assessments completed that indicated fall risk) was 0.53 (SD 0.20).

Table 2.

Characteristics of high-risk participants who received home hazard removal

Characteristic	Cluster 1 n=67	Cluster 2 n=70	Cluster 3 n=65	Cluster 4 n=42	Cluster 5 n=47	Total n=291
Age, M (SD)	75.4 (8.0)	68.7 (9.4)	75.6 (8.9)	65.2 (14.5)	73.7 (8.0)	72.1 (10.4)
Range	62–93	30–95	57–101	19–98	61–91	19–101
Gender, n (%)
Female	53 (79.1)	33 (47.1)	50 (76.9)	31 (73.8)	29 (61.7)	196 (67.4)
Male	14 (20.9)	37 (52.9)	15 (23.1)	11 (26.2)	18 (38.3)	95 (32.6)
Race, n (%)a
Black	58 (93.6)	65 (92.9)	30 (46.2)	21 (50.0)	44 (93.6)	218 (76.2)
White	1 (1.6)	3 (4.3)	35 (53.9)	21 (50.0)	2 (4.3)	62 (21.7)
Race not listed (not Black, White, Asian, Latinx)	3 (4.8)	2 (2.9)	0 (0.0)	0 (0.0)	1 (2.1)	6 (2.1)
Taking medications that increase fall risk, n (%)b
No	7 (11.1)	15 (23.1)	18 (29.5)	9 (22.5)	12 (29.3)	61 (22.6)
Yes	56 (88.9)	50 (76.9)	43 (70.5)	31 (77.5)	29 (70.7)	209 (77.4)
Cognitive impairment (SBT), M (SD)	6.5 (5.7)	7.0 (5.2)	5.4 (5.5)	7.1 (5.8)	8.1 (7.0)	6.7 (5.8)
Range c	0–24	0–23	0–22	0–23	0–28	0–28
Fall efficacy (Short FES-I), M (SD)	13.6 (5.5)	14.1 (5.6)	13.9 (4.8)	14.4 (5.9)	14.0 (6.7)	14.0 (5.6)
Range d	7–28	7–28	7–26	7–28	7–28	7–28
Depression (GDS-SF), M (SD)	2.6 (2.6)	3.1 (2.8)	3.0 (2.6)	5.0 (3.1)	2.3 (1.9)	3.1 (2.7)
Range e	0–12	0–11	0–11	0–12	0–8	0–12
Alcohol (SMAST-G), M (SD)	0.4 (1.1)	0.9 (1.8)	0.4 (1.1)	0.6 (1.8)	0.4 (1.1)	0.5 (1.4)
Range f	0–6	0–8	0–5	0–9	0–5	0–9
Function (OARS ADL), n (%) g
Excellent/Good	7 (10.6)	4 (6.0)	3 (4.7)	6 (14.3)	4 (8.7)	24 (8.4)
Mild impairment	12 (18.2)	12 (17.9)	12 (18.8)	11 (26.2)	14 (30.4)	61 (21.4)
Moderate impairment	25 (37.9)	24 (35.8)	26 (40.6)	6 (14.3)	17 (37.0)	98 (34.4)
Severe impairment	8 (12.1)	15 (22.4)	10 (15.6)	5 (11.9)	6 (13.0)	44 (15.4)
Total impairment	14 (21.2)	12 (17.9)	13 (20.3)	14 (33.3)	5 (10.9)	58 (20.4)
Environmental barriers (Westmead Home Safety Assessment), M (SD)	5.8 (3.1)	6.6 (3.4)	5.2 (3.0)	7.8 (5.0)	4.4 (3.5)	5.9 (3.7)
Range h	0–11	0–15	0–13	1–22	0–12	0–22
Fall risk composite, M (SD)	0.53 (0.17)	0.56 (0.20)	0.51 (0.17)	0.58 (0.17)	0.44 (0.24)	0.53 (0.20)
Range i,j	0.14–0.86	0.00–1.00	0.14–0.86	0.14–1.00	0.00–1.00	0.00–1.00

SBT=Short Blessed Test; Short FES-I=Short Falls Efficacy Scale International; GDS-SF=Geriatric Depression Scale Short Form; SMAST-G=Short Michigan Alcoholism Screening Test–Geriatric version; OARS ADL=Older Adult Resources Services Activity of Daily Living scale.

^an=286

^bn=270

^cn=269

^dn=287

^en=279

^fn=289

^gn=285

^hn=284

ⁱn=289

^jFall risk composite score is the percentage of fall risk assessments completed that indicated fall risk (taking medications that increase fall risk, SBT ≥10, Short FES-I ≥9, GDS-SF ≥5, SMAST-G ≥2, OARS ADL ≥ moderate impairment, Westmead Home Safety Assessment ≥4 environmental barriers).

Process measures

Of 214 participants with documentation on whether they attended a fall risk education/awareness session in the building, 180 (84.1%) attended at least 1 education/awareness session. Among all 291 participants, the HARP intervention (initial assessment and environmental hazard remediation) occurred in an average of 2.4 visits (SD 0.7) over an average of 2.8 weeks (SD 3.0). An average total time of 98.8 minutes (SD 43.4) was spent with each participant. Two hundred sixty-nine participants (92.4%) completed the booster session, which occurred in an average of 1.7 visits (SD 0.7) over an average of 1.2 weeks (SD 2.2). Total treatment time at the booster was an average of 61.8 minutes (SD 33.9). Participants had an average of 5.7 (SD 3.5) recommended home modifications. The total number of modifications across all participants was 1650, and the large majority of these were adaptive equipment (1283, 77.8%), followed by task modification (166, 10.1%) and architectural modifications (113, 6.8%). Average adherence, measured by the proportion of recommended home modifications in use at the 3-month booster session, was 95.4%.

Falls outcomes

Of the 291 participants who received HARP, 158 fell at least once during the control or posttreatment period. A total of 300 falls occurred among 110 participants during the control period, and 293 falls occurred among 113 participants during the posttreatment period.

Table 3 includes overall fall rates during control and posttreatment periods with unadjusted and adjusted IRRs from the multilevel negative-binomial models. Results are shown first including all data collected (including during the COVID-19 hiatus period) and then excluding data collected during the COVID-19 hiatus period. When including data collected during the COVID-19 hiatus, the rate of falls among participants was 4.87 falls per 1000 participant-days (95% CI 4.34–5.46) during the control period and 4.31 falls per 1000 participant-days (95% CI 3.83–4.83) during the posttreatment period. While the unadjusted model indicated a significant reduction in fall rate during the posttreatment period (IRR 0.81, 95% CI 0.67–0.99), the model adjusting for temporal trend, age, gender, and composite fall risk score indicated no significant difference in fall rate (IRR 0.97, 95% CI 0.66–1.42).

Table 3.

Fall rates during intervention and control periods, with results of negative-binomial models

Participants who completed initial assessment visit (n=291), including COVID-19 period
			Unadjusted model			Adjusted modela
	Post-treatment Period	Control Period	β (SE)	p	IRR (95% CI)	β (SE)	p	IRR (95% CI)
Number of falls per number of participant-days observed	293/68,056	300/61,547	−0.21 (0.10)	0.040	0.81 (0.67, 0.99)	−0.03 (0.20)	0.881	0.97 (0.66, 1.42)
Fall rate per 1000 participant-days (95% CI)	4.31 (3.83, 4.83)	4.87 (4.34, 5.46)
Participants who completed initial assessment visit (n=291), excluding COVID-19 period b
			Unadjusted model			Adjusted model a
	Posttreatment Period	Control Period	β (SE)	p	IRR (95% CI)	β (SE)	p	IRR (95% CI)
Number of falls/Number of participant-days observed	239/55,387	289/57,183	−0.23 (0.10)	0.024	0.79 (0.65, 0.97)	−0.07 (0.21)	0.734	0.93 (0.62, 1.40)
Fall rate per 1000 participant-days (95% CI)	4.32 (3.79, 4.90)	5.05 (4.49, 5.67)

CI=confidence interval; IRR=incidence rate ratio.

^aAdjusted for time point (secular trend), age, gender, and composite fall risk score.

^bExcludes dates March 20–June 6, 2020 due to intervention hiatus during the COVID-19 pandemic.

When excluding COVID-19 hiatus data, the rate of falls was 5.05 falls per 1000 participant-days (95% CI 4.49–5.67) during the control period and 4.32 (95% CI 3.79–4.90) during the posttreatment period. Similarly, while the unadjusted model showed a significant reduction in fall rate during the posttreatment period (IRR 0.79, 95% CI 0.65–0.97), the adjusted model indicated a non-significant reduction in fall rate (IRR 0.93, 95% CI 0.62–1.40). Fall rates by cluster and building for each study time point can be found in Supplementary Table 1 (includes data collected during the COVID-19 hiatus) and Supplementary Table 2 (excludes data collected during the COVID-19 hiatus).

An exploratory outcome was rate of serious fall-related injury, determined by asking participants who fell whether the fall resulted in an emergency department visit. A total of 48 participants visited the emergency department due to a fall during the control or posttreatment period: 30 during the control period (total of 42 visits) and 27 during the posttreatment period (total of 36 visits). When including data collected during the COVID-19 hiatus, the rate of fall-related injury among participants was 0.68 injuries per 1000 participant-days (95% CI 0.49–0.92) during the control period and 0.53 per 1000 participant-days (95% CI 0.37–0.73) during the posttreatment period. Differences in rates were not significant in unadjusted (IRR 0.85, 95% CI 0.52–1.39) or adjusted analyses (IRR 1.15, 95% CI 0.45–2.97). When excluding data collected during the COVID-19 hiatus, the rate of fall-related injury was 0.72 injuries per 1000 participant-days (95% CI 0.51–0.97) during the control period and 0.52 per 1000 participant-days (95% CI 0.35–0.75) during the posttreatment period, but this was not significant in unadjusted (IRR 0.79, 95% CI 0.48–1.33) or adjusted analyses (IRR 0.94, 95% CI 0.33–2.69).

DISCUSSION

This multisite, stepped-wedge cluster randomized controlled trial of education, fall screening, and a home hazard removal program for older adults at high risk for falls did not significantly reduce the rate of falls or serious fall-related injuries (i.e., falls resulting in emergency department visits) among participants in low-income senior apartments. However, HARP was acceptable among building residents and feasible to deliver in low-income senior housing. This pragmatic study reflected an effective collaboration between researchers and community agency staff, providing important information about the potential implementation of HARP for this population.

Prior trials of home hazard removal programs have shown marked reductions in the rate of falls.⁸ However, many of these studies were conducted in higher-income countries. The pragmatic features of this study (low-income housing units managed by a mission-driven community agency, broad inclusion criteria, delivery with agency staff) shed light on the potential of this intervention for low-income communities in the US. It is encouraging that the large majority (84%) of residents agreed to screening, and two-thirds of those eligible completed an initial assessment visit. Adherence to suggested home modifications (eg, adaptive equipment such as rollators or bath mats, task modifications such as keeping needed items close, or architectural modifications such as grab bars) was also high (95%). The successful implementation of HARP in low-income housing units with collaboration between agency and research staff and acceptance among residents suggests that reaching this vulnerable population is feasible and welcome.

This study occurred amid the global COVID-19 pandemic, causing a hiatus in intervention delivery. The pandemic might have also caused a general shift in fall trends or reporting, as a recent study of high-risk, community-dwelling older adults observed a reduction in self-reported falls during the early months of COVID-19, possibly due to reduced ability to leave home or participate in other activities.⁴⁵ In our study, a sensitivity analysis eliminating COVID-19 hiatus time points indicated an 8% reduction in fall rate, but this was not statistically significant. However, our observed difference in fall rate and sample size/person-time was smaller than in our original power calculation, suggesting that our study was underpowered to detect this effect size.

Although deriving clear policy recommendations from this pragmatic trial is complicated by the COVID-19 pandemic, there are some important lessons that may inform future trials. First, the potential for a reduction in falls is possible with a multistep intervention like HARP in low-income housing. Given the growing demand for low-income housing and the increasing fall rates across the US, HARP is one potential intervention for this population. Instead of relying solely on inconsistent screening of patients for fall risk in the doctor’s office,^{11, 12} delivering the intervention in a housing unit can help reduce barriers to accessing these services.

Several limitations should be considered when interpreting study findings. Non-significant findings could be due to the study being underpowered to detect the observed difference but was also likely impacted by the abrupt changes that occurred during the COVID-19 pandemic. The buildings were located in zip codes with the earliest and highest rates of COVID-19 infections and deaths. It is highly likely that fall reporting was unreliable during this time period. There was also a hiatus in HARP intervention delivery due to COVID-19, because it was not feasible or safe to deliver the intervention remotely. However, we resumed in-person intervention implementation as soon as possible. Secular trends in reported falls during the study period, which must be adjusted for in stepped-wedge trials, likely contributed to our non-significant findings. The observed fall rate during the control period (4.9 falls per 1000 participant-days) was lower than projected during a priori power calculations (6.6 falls per 1000 participant-days), which also reduced our power to detect observed effects. Although dropout rates did not significantly differ between participants who fell and who did not fall, the fall rate after participants dropped out is unknown. It is possible that fall rates were higher among dropouts after withdrawal from the study, which could have contributed to the appearance of reduced unadjusted fall rates over time during the study. In addition, some fall risk factors were not collected systematically and accounted for in analysis, including vision and hearing impairments. Furthermore, results may not be generalizable to other low-income senior apartment buildings.

Conclusion

This pragmatic trial found that HARP, delivered to residents of low-income senior housing with a community partner, did not significantly reduce the rate of falls. Intervention delivery was disrupted by the COVID-19 pandemic, and the study was ultimately underpowered. However, the intervention was acceptable to low-income senior housing staff and residents and feasible to deliver in this “real-world” community setting.

Key Points:

• A stepped-wedge cluster randomized trial of the Home Hazard Removal Program conducted in 11 low-income senior apartment buildings did not have a significant reduction in fall rates.

• Disruptions due to COVID-19 and reduced power may have impacted findings.

• The home hazard removal intervention was feasible to deliver in low-income housing with collaboration from research and agency staff and was accepted among residents.

Why does this matter?

Interventions to prevent falls are urgently needed for low-income, community-dwelling older adults. Our findings support the feasibility and acceptance of a home hazard removal program implemented in low-income senior housing and inform future pragmatic trials that aim to reach this vulnerable population.