Skip to main content
NCBI home page
JAMA Network logoLink to JAMA Network
. 2026 Oct 16;9(10):e2639240. doi: 10.1001/jamanetworkopen.2026.39240

Clinical Decision Support for Fall Prevention in Primary Care

The eSTEPS Cluster Randomized Clinical Trial

Nancy K Latham 1,2,✉, Erin Hommel 3, Mackenzie Kiesman 1, Stuart R Lipsitz 1,2, David W Bates 1,2, Diane L Carroll 4, Alexandra C Frost 4, Pamela M Garabedian 4, Julia V Loewenthal 1,2, Michael Sainlaire 1, Lipika Samal 1,2, Wenyu Song 1,2, Matthew Wien 1, Patricia C Dykes 1,2
PMCID: PMC13675817  PMID: 42886484

Key Points

Question

Does an electronic health record–integrated clinical decision support intervention reduce falls among older adults in primary care?

Findings

In this cluster randomized clinical trial of 11 945 patients, there was a modest but statistically significant reduction in the rate per 100 patient-years of 1 or more International Statistical Classification of Diseases and Related Health Problems, Tenth Revision–coded falls (including falls with injuries) over the study period in patients from intervention vs control practices.

Meaning

The findings of this study suggest that an electronic health record–integrated clinical decision support intervention may enable scalable implementation of fall-prevention strategies in routine primary care.

This cluster randomized clinical trial assesses whether an electronic health record–integrated clinical decision support intervention reduces falls among older adults at increased risk in primary care.

Abstract

Importance

Falls are a leading cause of injury among older adults. Although exercise reduces fall risk, implementation in primary care remains limited.

Objective

To assess whether an electronic health record–integrated clinical decision support intervention, Electronic Strategies for Tailored Exercise to Prevent Falls (eSTEPS), reduces falls among older primary care patients at increased risk.

Design, Setting, and Participants

This matched-pair cluster randomized clinical trial was conducted from June 9, 2023, to May 8, 2025, with 12 to 23 months of follow-up at primary care practices within an integrated health system in the US. Adults aged 65 years or older with a positive 3-item fall-risk screen at a primary care visit were included.

Intervention

eSTEPS decision support embedded in the electronic health record with tailored exercise recommendations, optional physical therapy referral, and patient-facing materials (handouts, website, and app).

Main Outcomes and Measures

The primary outcome was the rate of patients with 1 or more falls identified using electronic health record–derived International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) diagnostic codes. Secondary outcomes were the rate of patients with 1 or more fall-related injuries identified by validated ICD-10 algorithms and mortality.

Results

Among 11 945 participants (mean [SD] age, 77.05 [7.43] years; 8044 females [67.3%]), of whom 5681 (47.6%) were in the intervention group and 6264 (52.4%) were in the control group, at 22 primary care practices, the rate of 1 or more ICD-10–coded falls (including falls with injuries) per 100 patient-years over the study period was 19.3 (95% CI, 18.6-20.1) in patients from intervention practices vs 21.2 (95% CI, 20.2-22.1) in patients from control practices (rate ratio [RR], 0.91 [95% CI, 0.88-0.95]; P < .001). The rate of ICD-10–coded fall-related injuries per 100 patient-years was 15.7 (95% CI, 15.0-16.5) from intervention practices vs 16.7 (95% CI, 16.1-17.4) from control practices over the study period (RR for fall injuries per 100 patient-years, 0.94 [95% CI, 0.90-0.97]; P = .001). No differences were observed in mortality. For intervention patients, eSTEPS exercise prescriptions were issued for 1255 patients (22.1%). Overall, 215 clinicians (77.9%) in intervention practices used the clinical decision support to deliver the exercise program at least once over the study period.

Conclusions and Relevance

In this cluster randomized clinical trial of an electronic health record–integrated decision support, falls and fall-related injuries were modestly reduced in older adults in primary care. These results suggest support for scalable, workflow-integrated fall prevention.

Trial Registration

ClinicalTrials.gov Identifier: NCT04993781

Introduction

Falls are a leading cause of injury and death among older adults in the US, accounting for substantial morbidity and mortality, health care use, and costs.1,2,3 Exercise reduces fall rates by approximately 25% and is recommended in fall-prevention guidelines, yet implementation in primary care remains limited.4,5,6,7,8

Primary care clinicians are well positioned to identify and address fall risk but face competing demands that constrain preventive care delivery.9,10,11,12 Scalable, workflow-integrated approaches are needed to support fall prevention without increasing clinician burden. Although hospital-based fall-prevention clinical decision support (CDS) systems have demonstrated effectiveness in reducing falls,13,14,15 there is limited evidence of comparative tools for use in primary care, especially those targeting community-dwelling older adults. Using a participatory design approach,16,17 we developed Electronic Strategies for Tailored Exercise to Prevent Falls (eSTEPS), an electronic health record (EHR)–integrated CDS tool with patient-facing resources,12,18,19,20 and evaluated its effectiveness in reducing falls in a pragmatic cluster randomized clinical trial.

Methods

Study Design and Setting

We conducted a matched-pair cluster randomized clinical trial, in which primary care practices were initially randomized in matched pairs to intervention or usual care (Figure). All Mass General Brigham primary care practices were eligible to take part in the trial except for 2 practices that had participated in a pilot phase of the study. Matching and randomization were conducted by the study biostatistician (S.R.L.), using SAS software, version 9.4 (SAS Institute Inc). Practices were first stratified by practice type (academic vs community) and then ordered within each stratum according to practice size (the average number of eligible patients during the prior 2 years). Practices were matched in pairs by sequentially pairing adjacent practices within each stratum after sorting by size, thereby creating pairs with similar practice type and patient volume. Within each matched pair, 1 practice was randomly assigned to the intervention group and the other to the control group using computer-generated random numbers in SAS. Prior to patient enrollment, 1 practice and its matched pair were excluded because of lack of EHR data access; no patients were enrolled at excluded practices (Figure). Data analysts (M.S.) who extracted the data were blinded to treatment group. Clinicians and patients were not able to be blinded due to the pragmatic study design. Initial registration took place on June 22, 2021. The trial was conducted from June 9, 2023, to May 8, 2025, with 12 to 23 months of follow-up. The trial protocol, statistical analysis plan, and protocol changes are in Supplement 1. Deidentified patient data from the trial will be made available through the Vivli clinical research data sharing platform. The institutional review board of the Mass General Brigham Human Research Protection program approved the study and granted a waiver of informed consent because the study was considered minimal risk. A substudy was conducted that involved interviews with patients and clinicians (to be described in a separate study). Oral informed consent was obtained from substudy participants. This present study followed the Consolidated Standards of Reporting Trials (CONSORT) reporting guideline for cluster randomized clinical trials.21

Figure. CONSORT Diagrams of Primary Care Practices and Patients Enrolled in the Trial.

Two-panel CONSORT flowchart of practice and patient enrollment and follow-up. Two stacked panels labeled A and B, each a gray and white flow diagram with rectangular boxes, one oval, and connecting arrows. Panel A title at upper left: Primary care practice diagram. Top center box: 48 Practices approached for study. A rightward arrow leads to a large right-side box: 24 Excluded, with three indented lines reading 11 Declined to participate, 11 Received no response, and 2 Chosen for pilot, and a parenthetical note not participating in R C T. A vertical line from the top box continues to a centered oval: 24 Randomized. Two diagonal lines branch to left and right. Left branch box: 12 Randomized to intervention. A left-side small box connected by an arrow: 1 Withdrew participation before R C T end. A downward arrow continues to a box: 11 Enrolled for duration of R C T, then a downward arrow to: 11 Included in the final analysis. Right branch box: 12 Randomized to control. A right-side small box connected by an arrow: 1 Withdrew participation before R C T end. A downward arrow continues to: 11 Enrolled for duration of R C T, then to: 11 Included in the final analysis. Panel B title at left: Patient diagram. Top center box: 55 132 Patients assessed for eligibility by E H R, with parenthetical text indicating patients aged greater than or equal to 65 years and enrolled in M G B practices. A rightward arrow leads to a large right-side box: 43 187 Excluded based on E H R data, with two indented lines reading 27 128 Not at risk for falling, based on fall screener, and 16 059 Not assessed for fall risk, with parenthetical fall screen not done. A vertical line continues to a centered oval: 11 945 Randomized. Two diagonal branches lead to lower boxes. Left: 5 681 Enrolled in practices randomized to intervention. A small left box: 27 Lost to follow-up. A downward arrow to a larger box: 5 654 Included in the primary analysis, and a second line: 433 Died. Right: 6 264 Enrolled in practices randomized to control. A small right box: 21 Lost to follow-up. A downward arrow to a larger box: 6 243 Included in the primary analysis, and a second line: 402 Died.

Open in a new tab

EHR indicates electronic health record; MGB, Mass General Brigham; RCT, randomized clinical trial.

Participants

Eligible participants were community-dwelling adults aged 65 years or older who screened positive for fall risk during routine primary care visits using a 3-item screener with the following questions: (1) Have you fallen and hurt yourself in the past year? (2) Have you fallen 2 or more times in the past year? (3) Are you afraid that you might fall because of balance or walking problems?22 Patients were included if they answered yes to 1 or more of the fall screener questions at their index visit or in any primary care visit in the previous 12 months.

Intervention

The eSTEPS CDS was embedded in the Epic EHR as a noninterruptive care-gap alert.12,19 For eligible patients, clinicians could access an associated order set (eg, Epic SmartSet) that provided tailored strength and balance exercise recommendations based on patient characteristics (eFigure in Supplement 2). The exercises were simplified versions of exercises that were included on the Otago Exercise Programme (fall-prevention program).23 All of the exercises were designed to improve balance and lower-limb strength. All exercises could be performed at home with no additional equipment. Each session was designed to be completed in approximately 10 minutes. There were 4 levels of difficulty to allow progression over time. The default option was for clinicians to recommend beginner-level exercises or sitting exercises if the patient could not safely stand. Clinicians could prescribe exercises and optionally refer patients to physical therapy. Primary care staff provided the exercise materials via handouts, and information was integrated into the after-visit summaries and the patient portal with links to a website and mobile app providing videos and optional self-monitoring. Use was at clinician discretion. Patients in the control practice sites received usual care. Usual care included the patients being asked the 3 fall-risk screening questions at each primary care visit and the opportunity for referral to physical therapy. The control practices did not have the eSTEPS CDS support, which helped clinicians to deliver the eSTEPS exercise tools.

Outcomes

The primary outcome was the rate of patients with 1 or more falls, identified using EHR-derived International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) diagnostic codes. Secondary outcomes were the rate of patients with 1 or more fall-related injuries and mortality. Hospitalization (≥1 hospitalization recorded in the EHR) was analyzed as a post hoc outcome. Fall and injury data were obtained from EHR-derived diagnostic codes (ICD-10) using a validated fall and fall-related injury algorithm.24 The algorithm was developed to detect falls and fall-related injuries from administrative data using ICD codes (initially the International Classification of Diseases, Ninth Revision and updated to the ICD-10) and was found to have an area under the curve of 0.79 (95% CI, 0.78-0.81) in a recent study that used self-reported data from a large randomized clinical trial as the external reference standard.24 Patient-reported falls were ascertained during routine primary care and other health system encounters when clinic staff administered the 3 standardized fall-risk screening questions as part of routine clinical care and documented responses in the EHR. The screening questions asked patients whether they had experienced a fall or a fall-related injury during the preceding 12 months. To ensure that reported events occurred after the index visit, analyses included only fall-screening assessments completed 12 or more months after the index visit through the end of the study period.

Serious adverse events (deaths and hospitalizations) were reviewed monthly by an independent safety monitor, and the trial was overseen by an independent data and safety monitoring board. Although 1 interim analysis was specified in the protocol, the data and safety monitoring board recommended that it not be conducted because there were no safety concerns, and trial progress was satisfactory. This preserved the integrity of the final analysis.

All demographic data, including race and ethnicity, were obtained from the EHR. Race categories included Black or African American, White, and other race (American Indian or Alaska Native, Asian, or Native Hawaiian or Other Pacific Islander); ethnicity categories included Hispanic or not Hispanic. Race and ethnicity were included to provide descriptive information about the study sample.

Statistical Analysis

Analyses used SAS, version 9.4. Follow-up was from index visit to death, withdrawal, or study end. Patients who did not remain under observation for the full follow-up period contributed all available person-time prior to censoring. Balance in patient characteristics across arms was assessed using standardized differences.25 We adjusted the models for a prespecified set of baseline covariates: age, race and ethnicity, Charlson Comorbidity Index score,26 and baseline fall injury risk score (scores range from 0 to 1, with higher scores indicating greater fall risk), based on our previously developed and validated data-driven prediction algorithm.27 We adjusted for 2 practice-level covariates: the prior-year practice hospitalization rate and prior-year practice mortality rate. These variables were selected before the primary outcome analyses after baseline practice characteristics were reviewed and were included to account for baseline practice-level heterogeneity and to improve the precision of the estimated intervention effect. Because ascertainment of falls depended on documentation during routine primary care encounters, all models also included the number of completed fall screening assessments as a covariate and the log of follow-up time as an offset. The number of completed fall screening assessments was included to account for differences in surveillance intensity and opportunity for fall ascertainment across participants rather than as a confounder. Follow-up time reflects the duration of observation, whereas the number of completed screening assessments reflects the intensity of outcome ascertainment during that observation period. Because the association between screening frequency and fall ascertainment was not known a priori, screening frequency was modeled as a covariate rather than incorporated as an offset.

The primary outcome was analyzed at the patient level using Poisson regression with a log link to estimate rate ratios (RRs). As aforementioned, models included the number of completed fall screening assessments as a covariate and the log of follow-up time as an offset. Generalized estimating equations were used to account for the matched-pair cluster-randomized design. The marginal correlation structure explicitly reflected the study hierarchy, with patients nested within practices and practices nested within matched pairs. This structure permitted 1 correlation among patients within the same practice and a second correlation among patients from different practices within the same matched pair. Each matched pair (consisting of 2 practices) served as the highest-level clustering unit, and observations arising from different matched pairs were assumed to be independent. Analyses were conducted both without and with adjustment for the prespecified baseline patient-level and practice-level covariates. Because the number of independent matched pairs was small, inference was based on the model-based covariance estimator with small-sample degrees of freedom adjustments for hypothesis tests and 95% CIs.

To estimate the adjusted probability of experiencing at least 1 fall within 1 and 2 years in each treatment group, we used the relationship between Poisson event rates and the exponential distribution for time to first event. Adjusted event rates from the Poisson model were converted to 1-year and 2-year fall probabilities using the exponential survival function. The number needed to treat at 2 years was calculated as the reciprocal of the difference in adjusted 2-year fall probabilities between the intervention and control groups.

A sensitivity analysis was conducted to explore whether cluster size or eSTEPS intervention uptake had an impact on treatment effect. The primary results are reported as RRs with 95% CIs. A 2-sided P < .05 indicated statistical significance.

The planned sample size of 8352 patients (4176 per group) provided 90% power to detect a 20% relative reduction in falls (from 24.8% to 19.7%), assuming a 10% loss to follow-up and accounting for clustering within practices (intraclass correlation coefficient, 0.01) and matched pairs (intraclass correlation coefficient, 0.005). The final sample size exceeded the planned enrollment.

Results

There were 24 primary practices that were initially randomized, and 22 practices in the final analysis. A total of 11 945 patients were included (mean [SD] age, 77.05 [7.43] years; 8044 females [67.3%]; 3901 males [32.7%]), of whom 5681 (47.6%) were in the intervention group, and 6264 (52.4%) were in the control group. Among the total participants, 945 were Black or African American (7.9%), 9428 were White (78.9%), and 430 were of other race (3.6%); 1047 were Hispanic (8.8%), and 10 423 were not Hispanic (87.3%). While most baseline characteristics were similar between groups, differences were observed; the intervention group included more people who were Black or African American, with lower median income, and with a lower fall rate in the previous year and fewer people who were college graduates. At the practice level, intervention practices had higher mortality in the 1 year prior to the study initiation (Table 1). The median (range) cluster size was 393 (24-1950) overall patients; the median (IQR [range]) cluster size in the control group was 395 (103-655 [24-1950]) and in the intervention group was 390 (297-592 [297-1281]).

Table 1. Baseline Characteristics of Patients and Primary Care Practices.

Characteristic No. (%)a
Total Control Intervention
No. of patients 11 945 6264 5681
Age, mean (SD), y 77.05 (7.43) 77.33 (7.37) 76.74 (7.48)
Sex
Female 8044 (67.3) 4191 (66.9) 3853 (67.8)
Male 3901 (32.7) 2073 (33.1) 1828 (32.2)
Race
Black or African American 945 (7.9) 360 (5.7) 585 (10.0)
White 9428 (78.9) 4992 (79.6) 4436 (78.1)
Otherb 430 (3.6) 299 (4.8) 131 (2.3)
Missing 1142 (9.6) 613 (9.8) 529 (9.3)
Ethnicity
Hispanic 1047 (8.8) 567 (9.1) 480 (8.4)
Not Hispanic 10 423 (87.3) 5436 (86.8) 4987 (87.8)
Missing 475 (4.0) 261 (4.2) 214 (3.8)
Insurance
Managed 3573 (29.9) 1790 (28.6) 1783 (31.4)
Medicare 8207 (68.7) 4393 (70.1) 3814 (67.1)
Other 165 (1.4) 81 (1.3) 84 (1.5)
College graduate
No 4710 (39.4) 2220 (35.4) 2490 (43.8)
Yes 5991 (50.2) 3313 (52.9) 2678 (47.1)
Declined or missing 1236 (10.0) 724 (12.0) 512 (9.0)
Income (based on zip code), median (SD), $ 127 431.02 (45 556.92) 132 669.09 (47 977.23) 121 666.09 (41 990.48)
Fall injury score, mean (SD) 0.61 (0.18) 0.61 (0.19) 0.60 (0.18)
Charlson Comorbidity Index, mean (SD) 5.39 (3.51) 5.31 (3.48) 5.47 (3.55)
Charlson Comorbidity Index score ≥2 9744 (81.6) 5083 (81.1) 4661 (82.0)
Hospitalization in previous year 3731 (31.2) 1899 (30.3) 1832 (32.2)
Fall screening question at index visitc
Have you fallen and hurt yourself in the past year? 1874 (29.6) 998 (33.2) 876 (26.2)
Have you fallen ≥2 in the past year? 675 (10.6) 354 (11.8) 321 (9.6)
Are you afraid that you might fall because of balance or walking problems? 2674 (42.2) 1298 (43.2) 1376 (41.2)
No. of practices 22 11 11
Academic medical practices 14 (63.6) 7 (63.6) 7 (63.6)
Mortality rate in previous year for eligible patients, mean (SD) 13.7 (3.2) 12.3 (3.2) 15.2 (2.6)
Hospitalization rate in previous year, mean (SD) 31.1 (6.6) 31.3 (7.1) 31 (6.4)
Cluster size, median (range) 393 (24-1950) 395 (24-1950) 390 (297-1281)
Dual-eligible 1067 (8.9) 509 (8.1) 558 (9.8)
Open in a new tab
a

Data are reported as No. (%) unless otherwise indicated.

b

Included American Indian or Alaska Native, Asian, or Native Hawaiian or Other Pacific Islander.

c

The denominators (total: n = 6344, control: n = 3005, and intervention: n = 3339) represent the number of people who had a positive fall screening question at the index visit; they do not include those who met the screening eligibility criteria because they had a positive fall screening question in the past 12 months.

The mean (SD) length of follow-up was 622.3 (140.6) patient-days in the intervention group and 622.0 (148.5) patient-days in the control group. A total of 835 patients died, and 48 were lost to follow-up (Figure).

The rate per 100 patient-years of patients with 1 or more ICD-10–coded falls (including falls with and without injuries) over the study period was 19.3 (95% CI, 18.6-20.1) in patients from intervention practices vs 21.2 (95% CI, 20.2-22.1) in patients from control practices (RR, 0.91 [95% CI, 0.88-0.95]; P < .001). The proportion of patients with 1 or more falls over 2 years was 32.0% (95% CI, 31.1-33.0) in the intervention group vs 34.5% (95% CI, 33.3-35.7) for the control group. The number needed to treat to prevent 1 fall over 2 years was 28.6 (95% CI, 19.0-42.9) patients.

Self-reported falls or fall injuries that were collected as part of routine screening of older adults during primary care patient visits also had a significant reduction in the intervention compared with the control group (intervention: 17.5 [95% CI, 16.9-18.0]; control: 19.2 [95% CI, 18.7-19.6]; RR, 0.91 [5% CI, 0.87-0.95]; P < .001). The rate of ICD-10–coded fall-related injuries per 100 patient-years was 15.7 (95%, CI 15.0-16.5) from intervention practices vs 16.7 (95% CI, 16.1-17.4) from control practices over the study period (RR for fall injuries per 100 patient-years, 0.94 [95% CI, 0.90-0.97]; P = .001) (Table 2).

Table 2. Outcomes by Treatment Group.

Outcome First events of the available patient pool, No. (%) Rate per 100 patient-years (95% CI)a RR per 100 patient-years (95% CI) Rate difference per 100 patient-years (95% CI) P value
Intervention (n = 5681) Control (n = 6264) Intervention (n = 5681) Control (n = 6264)
Fall rate using ICD-10 codes (any ICD-10 fall code or fall injury code) 1937 (34.1) 2170 (34.6) 19.3 (18.6 to 20.1) 21.2 (20.2 to 22.1) 0.91 (0.88 to 0.95) −1.90 (−3.03 to −0.77) <.001
Fall injury rate using ICD-10 codes (ICD-10 fall injury code)b 1619 (28.5) 1769 (28.2) 15.7 (15.0 to 16.5) 16.7 (16.1 to 17.4) 0.94 (0.90 to 0.97) −1.00 (−1.60 to −0.40) .001
Self-reported fall or fall-related injury (reported during routine clinical screening)c 1923 (33.8) 2120 (33.8) 17.5 (16.9 to 18.0) 19.2 (18.7 to 19.6) 0.91 (0.87 to 0.95) −1.70 (−2.71 to −0.69) <.001
Overall death rate (mortality)b 433 (7.6) 402 (6.4) 5.9 (5.0 to 7.0) 5.6 (5.1 to 6.1) 1.06 (0.89 to 1.27) 0.30 (−0.55 to 1.15) .49
Hospitalization rateb 1767 (31.1) 1787 (28.5) 16.4 (16.1 to 16.6) 16.8 (16.6 to 17.0) 0.97 (0.95 to 0.99) −0.40 (−0.66 to −0.14) .003
Open in a new tab

Abbreviations: ICD-10, International Statistical Classification of Diseases and Related Health Problems, Tenth Revision; RR, rate ratio.

a

Data (mean [SD]) for the intervention group include follow-up patient-days (622.3 [140.6]) and fall screens (11.0 [8.8]); data for the control group include follow-up patient-days (622.0 [148.5]) and fall screens (9.5 [7.6]).

b

Time from immediately after index visit to end of study follow-up.

c

Time from 12 months after index visit to end of study follow-up.

In the subgroup analysis by age, the intervention was effective in reducing falls in both groups (younger patients: RR, 0.94 [95% CI, 0.88-0.99], P = .02; older patients: RR, 0.92 [95% CI, 0.86-0.97]; P = .006). For fall-related injuries, the intervention only had a significant effect in older patients (younger patients: RR, 0.98 [95% CI, 0.93-1.04]; older patients: RR, 0.92 [95% CI, 0.88-0.96]; P < .001). Formal interaction testing demonstrated significant treatment-by-age interactions for ICD-defined fall injury outcomes but not for the remaining outcomes.

In the sensitivity analyses to explore the effect of cluster size, the treatment effect was not significantly different across practice size. There was also no significant interaction effect between the rate of eSTEPS referral and the effect of the treatment on falls or fall injuries.

Unadjusted analyses are reported in the eTable in Supplement 2. Findings for self-reported falls were statistically significant, but other outcomes found no treatment effect for the intervention. After adjustment for the aforementioned prespecified patient-level and practice-level covariates, significant reductions were observed for ICD-10–coded falls, fall injuries, self-reported falls, and hospitalizations. These differences likely reflect baseline imbalances between study groups, as evidenced by the standardized differences shown in Table 1.

No between-group differences were observed for mortality (Table 2). In an exploratory post hoc analysis, hospitalization rate (the rate of ≥1 hospitalizations) was slightly reduced in the intervention compared with the control group (RR per 100 patient-years, 0.97 [95% CI, 0.95-0.99]; P = .003). No serious adverse events that were associated with the intervention and no unanticipated events in the trial were found from safety monitoring.

Intervention Uptake

eSTEPS exercise prescriptions were issued for 1255 intervention patients (22.1%) and 8 control participants (0.01%). Physical therapy referrals were placed for 3166 intervention patients (55.7%) during the study period and for 3597 control participants (57.4%). Among intervention clinicians, 215 (77.9%) used the CDS to deliver the exercise program at least once, and 253 (91.4%) referred a patient to physical therapy. Uptake varied substantially across practices, with 50.9% to 99.5% of clinicians in the intervention practices using the CDS at least once.

Discussion

This cluster randomized clinical trial found that an EHR-integrated CDS intervention for fall prevention resulted in a modest but statistically significant reduction in the rates of falls and fall-related injuries among older adults in primary care. There was also a small but statistically significant reduction in hospitalization.

These findings extend prior evidence on exercise-based fall prevention by demonstrating that integrating exercise promotion into routine primary care workflows may produce modest but clinically meaningful reductions in falls and fall-related injuries. Unlike structured exercise programs that require dedicated personnel or stand-alone delivery, the eSTEPS intervention leveraged existing clinical workflows, enhancing its potential for scalability and broad implementation. A recent systematic review found limited evidence that implementation of fall-prevention guidelines in community settings improves clinician behavior or reduces falls.28 Prior CDS interventions for fall prevention have largely been evaluated in inpatient settings.13,14,15 Although primary care–based CDS approaches have been piloted, large-scale randomized trials remain limited.28,29 The observed reduction in fall-injury rates observed in this study (approximately 9%) is consistent with clinically meaningful effects reported in prior exercise trials.30,31 Our findings are also consistent with the reduction in fall-related injuries observed in the multicomponent intervention evaluated in the STRIDE trial.22 The highly pragmatic design of the eSTEPS trial enabled enrollment of a large population, providing sufficient power to detect modest treatment effects. Although the absolute reduction was approximately 2 fewer falls per 100 patients annually (from 21.2 to 19.3), this effect could translate into significant reductions in fall-related morbidity and health care utilization when implemented at scale across health care systems serving older adults.

Despite incomplete uptake, reductions in falls and fall-related injuries were observed, which may reflect increased clinician awareness of fall risk or selective use of the intervention in higher-risk patients. Additionally, the presence of the CDS may have influenced broader care processes, including more consistent fall-risk screening or counseling, even when exercise prescriptions were not formally ordered. These findings suggest that CDS interventions may have effects beyond direct use, although due to its pragmatic nature, the relative contributions of these mechanisms cannot be determined from this study. The greater effect observed among adults aged 76 years or older compared with those aged 65 to 75 years is consistent with prior evidence that exercise-based fall-prevention interventions may have larger effects in populations at higher baseline risk.23

Because the intervention was embedded within routine clinical workflows and required minimal additional time from clinicians, it may be well suited for integration into existing primary care quality improvement and preventive care initiatives. The approach aligns with current efforts to leverage EHR-based tools to support guideline-concordant care and may be adaptable to other preventive interventions. Wider dissemination will require attention to implementation strategies that improve uptake and sustainability across diverse practice settings.

Strengths and Limitations

Strengths of this study include a pragmatic design embedded into primary care, a large and diverse population, and practice-level randomization. The intervention used EHR integration, which supports scalability across health systems using similar infrastructure. The use of routinely collected EHR data enabled evaluation of outcomes at scale with minimal additional burden on participants or clinicians.

This study also has several limitations. First, reliance on EHR-derived diagnostic codes may have underestimated fall frequency and may have been subject to differential detection if documentation or reporting differed between groups. The intervention did not directly impact ICD-10 coding of falls or injuries, but the CDS might have prompted the clinician to ask about these events and potentially could have increased the detection of these events in the intervention group. If this occurred, a reduction in the observed treatment effect would have resulted. Similar treatment effects were observed for outcomes that were less vulnerable to ascertainment bias, including self-reported falls obtained through routine screening of all patients and hospitalizations. These findings suggest that differential detection was unlikely to have materially influenced the study results. Second, uptake was variable across practices and clinicians, which may have influenced the generalizability of the observed effect. Third, the study was conducted within a single integrated health system and relied on a single EHR platform, which may have limited generalizability to other settings, although the intervention was implemented using standard configuration tools implemented using a widely used EHR system. Fourth, data on patient uptake and adherence to prescribed exercise interventions were not available, limiting our ability to understand the mechanisms underlying the observed treatment effect. Although rates of physical therapy orders were similar between the intervention and control groups, we could not determine whether patients initiated or adhered to the recommended interventions.

These findings suggest that EHR-integrated CDS tools may support scalable fall prevention in primary care. Future work should focus on improving uptake, optimizing workflow integration, and evaluating effectiveness across diverse settings with multiple EHR systems.

Conclusions

In this cluster randomized clinical trial, an EHR-integrated CDS intervention embedded within routine primary care was associated with reduced falls and fall-related injuries among older adults. These findings demonstrate that evidence-based fall prevention can be successfully integrated into routine primary care workflows, providing a scalable strategy for reducing falls in large populations of older adults.

Supplement 1.

Trial Protocol, Statistical Analysis Plan, and Trial Protocol Changes

jamanetwopen-e2639240-s001.pdf (887.1KB, pdf)
Supplement 2.

eFigure. Intervention Components and Workflow

eTable. Unadjusted Outcomes

jamanetwopen-e2639240-s002.pdf (413.1KB, pdf)
Supplement 3.

Data Sharing Statement

jamanetwopen-e2639240-s003.pdf (16KB, pdf)

References

  • 1.Hartholt KA, Lee R, Burns ER, van Beeck EF. Mortality from falls among US adults aged 75 years or older, 2000-2016. JAMA. 2019;321(21):2131-2133. doi: 10.1001/jama.2019.4185 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Haddad YK, Miller GF, Kakara R, et al. Healthcare spending for non-fatal falls among older adults, USA. Inj Prev. 2024;30(4):272-276. doi: 10.1136/ip-2023-045023 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Hoffman G, Franco N, Perloff J, Lynn J, Okoye S, Min L. Incidence of and county variation in fall injuries in US residents aged 65 years or older, 2016-2019. JAMA Netw Open. 2022;5(2):e2148007. doi: 10.1001/jamanetworkopen.2021.48007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Sherrington C, Fairhall N, Kwok W, et al. Evidence on physical activity and falls prevention for people aged 65+ years: systematic review to inform the WHO guidelines on physical activity and sedentary behaviour. Int J Behav Nutr Phys Act. 2020;17(1):144. doi: 10.1186/s12966-020-01041-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Sherrington C, Fairhall NJ, Wallbank GK, et al. Exercise for preventing falls in older people living in the community. Cochrane Database Syst Rev. 2019;1(1):CD012424. doi: 10.1002/14651858.CD012424.pub2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Sherrington C, Michaleff ZA, Fairhall N, et al. Exercise to prevent falls in older adults: an updated systematic review and meta-analysis. Br J Sports Med. 2017;51(24):1750-1758. doi: 10.1136/bjsports-2016-096547 [DOI] [PubMed] [Google Scholar]
  • 7.Montero-Odasso M, van der Velde N, Martin FC, et al. ; Task Force on Global Guidelines for Falls in Older Adults . World guidelines for falls prevention and management for older adults: a global initiative. Age Ageing. 2022;51(9):afac205. doi: 10.1093/ageing/afac205 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Nicholson WK, Silverstein M, Wong JB, et al. ; US Preventive Services Task Force . Interventions to prevent falls in community-dwelling older adults: US Preventive Services Task Force Recommendation Statement. JAMA. 2024;332(1):51-57. doi: 10.1001/jama.2024.8481 [DOI] [PubMed] [Google Scholar]
  • 9.Ashman JJ, Santo L, Okeyode T. Characteristics of office-based physician visits by age, 2019. NCHS National Health Statistics Reports. US Centers for Disease Control and Prevention; National Center for Health Statistics . April 18, 2023. Accessed March 1, 2026. https://stacks.cdc.gov/view/cdc/125462 [Google Scholar]
  • 10.Bodenheimer T. Revitalizing primary care, part 1: root causes of primary care’s problems. Ann Fam Med. 2022;20(5):464-468. doi: 10.1370/afm.2858 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Massachusetts Health Policy Commission . A dire diagnosis: the declining health of primary care in Massachusetts and the urgent need for action. January 2025. Accessed March 1, 2026. https://masshpc.gov/publications/policyresearch-brief/dire-diagnosis-declining-health-primary-care-massachusetts-and
  • 12.Rice H, Garabedian PM, Shear K, et al. Clinical decision support for fall prevention: defining end-user needs. Appl Clin Inform. 2022;13(3):647-655. doi: 10.1055/s-0042-1750360 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Dykes PC, Carroll DL, Hurley A, et al. Fall prevention in acute care hospitals: a randomized trial. JAMA. 2010;304(17):1912-1918. doi: 10.1001/jama.2010.1567 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Cho I, Kim M, Song MR, Dykes PC. Evaluation of an approach to clinical decision support for preventing inpatient falls: a pragmatic trial. JAMIA Open. 2023;6(2):ooad019. doi: 10.1093/jamiaopen/ooad019 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Jung H, Park HA, Lee HY. Impact of a decision support system on fall-prevention nursing practices. J Patient Saf. 2023;19(8):525-531. doi: 10.1097/PTS.0000000000001168 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Bødker S, Dindler C, Iversen OS, Smith RC. What is participatory design? In: Participatory Design. Synthesis Lectures on Human-Centered Informatics. Springer Cham; 2022:5-13. [Google Scholar]
  • 17.Massoud MRNK, Schall MW, Sevin C. A Framework for Spread: From Local Improvements to System-Wide Change. Institute for Healthcare Improvement; 2006. [Google Scholar]
  • 18.Czuber NK, Garabedian PM, Rice H, Tejeda CJ, Dykes PC, Latham NK. Human-centered design and development of a fall prevention exercise app for older adults in primary care settings. Appl Clin Inform. 2024;15(3):544-555. doi: 10.1055/a-2267-1727 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Tejeda CJ, Garabedian PM, Rice H, Samal L, Latham NK, Dykes PC. Development and usability testing of an exercise-based primary care fall prevention clinical decision support tool. AMIA Annu Symp Proc. 2024;2023:699-708. [PMC free article] [PubMed] [Google Scholar]
  • 20.Xealth. From “what if” to “what’s next.” Accessed July 8, 2025. https://www.xealth.com/
  • 21.Hopewell S, Chan A, Collins GS, et al. CONSORT 2025 Statement: Updated Guideline for Reporting Randomized Trials. JAMA. 2025;333(22):1998-2005. doi: 10.1001/jama.2025.4347 [DOI] [PubMed] [Google Scholar]
  • 22.Bhasin S, Gill TM, Reuben DB, et al. ; STRIDE Trial Investigators . A randomized trial of a multifactorial strategy to prevent serious fall injuries. N Engl J Med. 2020;383(2):129-140. doi: 10.1056/NEJMoa2002183 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Wang C, Kim SM. The Otago Exercise Program’s effect on fall prevention: a systematic review and meta-analysis. Front Public Health. 2025;13:1522952. doi: 10.3389/fpubh.2025.1522952 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Ganz DA, Esserman D, Latham NK, et al. Validation of a rule-based ICD-10-CM algorithm to detect fall injuries in Medicare data. J Gerontol A Biol Sci Med Sci. 2024;79(7):glae096. doi: 10.1093/gerona/glae096 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Normand ST, Landrum MB, Guadagnoli E, et al. Validating recommendations for coronary angiography following acute myocardial infarction in the elderly: a matched analysis using propensity scores. J Clin Epidemiol. 2001;54(4):387-398. doi:10.1016/s0895-4356(00)00321-8 [DOI] [PubMed] [Google Scholar]
  • 26.Charlson ME, Carrozzino D, Guidi J, Patierno C. Charlson Comorbidity Index: a critical review of clinimetric properties. Psychother Psychosom. 2022;91(1):8-35. doi: 10.1159/000521288 [DOI] [PubMed] [Google Scholar]
  • 27.Song W, Latham NK, Liu L, et al. Improved accuracy and efficiency of primary care fall risk screening of older adults using a machine learning approach. J Am Geriatr Soc. 2024;72(4):1145-1154. doi: 10.1111/jgs.18776 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Haley MN, Sherrington C, Lawler K, et al. Falling short on implementation of fall prevention guidelines in health services: a systematic review with meta-analysis. Age Ageing. 2025;54(10):afaf307. doi: 10.1093/ageing/afaf307 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Casey CM, Parker EM, Winkler G, Liu X, Lambert GH, Eckstrom E. Lessons learned from implementing CDC’s STEADI falls prevention algorithm in primary care. Gerontologist. 2017;57(4):787-796. doi: 10.1093/geront/gnw074 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Choudhary PK, Choudhary S, Saha S, Katanić B, İlbak İ, Tornóczky GJ. Effectiveness of balance- and strength-based exercise interventions for fall prevention in community-dwelling older adults: a systematic review of randomized controlled trials. Life (Basel). 2025;16(1):41. doi: 10.3390/life16010041 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Zhu Y, Zhang Y, Li X, Du Z. Effects of exercise interventions on physical function, cognitive function and quality of life of frail older adults in nursing homes: a systematic review and meta-analysis. Front Psychol. 2025;16:1679734. doi: 10.3389/fpsyg.2025.1679734 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement 1.

Trial Protocol, Statistical Analysis Plan, and Trial Protocol Changes

jamanetwopen-e2639240-s001.pdf (887.1KB, pdf)
Supplement 2.

eFigure. Intervention Components and Workflow

eTable. Unadjusted Outcomes

jamanetwopen-e2639240-s002.pdf (413.1KB, pdf)
Supplement 3.

Data Sharing Statement

jamanetwopen-e2639240-s003.pdf (16KB, pdf)

Articles from JAMA Network Open are provided here courtesy of American Medical Association

RESOURCES