The EF program provides multiple types of physical activity in a structured format. Classes meet for 60 minutes, 3 times a week, and are taught by certified instructors. Project Enhance strongly recommends affiliate organizations’ instructor candidates have a primary fitness certification or experience in physical therapy, occupational therapy or nursing prior to attending the 12-hour in-person EF New Instructor Training certification. EF certification includes 3 pre-training modules and a 1.5-day in-person certification. EF Master Trainers support EF instructors through new instructor training, mentoring and fidelity monitoring. Pre-requisites to EF Master Trainer training include 1 year teaching EF to fidelity and a National Commission for Certifying Agencies (NCCA) accredited fitness certification or degree. Continuing education for all EF instructors entitled "Essentials for EF Instructors" is available through Project Enhance’s American Council on Exercise portal (http://www.projectenhance.org/EnhanceFitness/Enact.aspx), available online only to certified EF Instructors. The purpose of continuing education is to support the continued development of skills, and maintain certification. Most EF classes are taught on an ongoing basis, but some are offered in 16-week increments (hereafter referred to as a “program cycle”). Classes include cardiovascular, strength, flexibility and balance exercises. Exercises may be done sitting, standing, or standing with support. The program measures 3 areas of physical function, each via its own test: upper-extremity strength (measured by arm curls), lower-extremity strength (measured by chair stands) and a composite test of abilities, including strength, dynamic balance, mobility, and gait (hereafter shortened to ‘balance & mobility’; measured by eight-foot up-and-go).²⁰

The dataset consisted of EF program data collected regularly from January 2005 to June 2016. Certified EF program instructors collect participant data, which are entered into an online database, the Online Data Entry System (ODES). ODES is maintained by the licensor of EF, Sound Generations (https://soundgenerations.org), a non-profit social services organization in Seattle, WA. This study was conducted under the approval of the University of Washington Institutional Review Board; the study used fully de-identified secondary data and was classified as exempt.

The initial dataset had 25,584 participants. Our exclusion criteria addressed several potential problems with the data, and retained 30% of the original sample. First, we excluded participants who did not have any follow-up data after baseline (n=14,050; 54.9%). There was also extreme variation in the time between the first and second physical function tests, ranging from 4 weeks to 11 years. Therefore, we excluded participants whose second physical function test date was more than 6 weeks deviant from the EF-defined program cycle (n=4,052; 15.8%). For example, we dropped participants with more than 22 weeks between tests. Last, because there was a large drop in attendance after the second program cycle, we included only the first 3 physical function tests. Put differently, sample participants had at least 2 test scores (baseline and the first program cycle), but may have had up to 3 test scores (a second program cycle). The final sample includes 7,483 (29.2%) participants.

Such heavy exclusions eliminated the possibility of analyzing the benefit of long-term involvement, and may have introduced selection bias (discussed further in the limitations section). To test for this bias, we compared the final sample to the excluded participants using logistic regression.

Most of the 7,483 sample participants (Table 1) were between the ages of 65 and 75 years (52%), female (81%), White (62%), and not married or partnered (44%). Most did not have a disability (84%), were in at least good health (83%), and had not experienced any falls in the last 4 months (71%). The mean attendance for the first program cycle, and the physical function test scores for baseline and first follow-up are presented by demographic variables in Table 2. Some degree of improvement was observed in every subgroup, but many of the listed changes did not meet the MDC until the second follow-up.

We found statistically significant, but weak (OR<1.1) associations between being in the final sample, and not being disabled, being more active, and being in better health. We found a stronger association (OR: 2.2) between Asian race and being in the final sample.

The mean number of complete weeks of EF attendance (i.e., the number of weeks the participant attended 3 sessions) in the first program cycle was 10.6.

The mean performance on the physical function tests at baseline was 16.5 arm curls (of which 48% were in the normal range for their age-sex category);²³ 12.4 chair stands (57% in normal range), and 8.8 seconds for the eight-foot up-and-go (40% in normal range) (Table 3). Total mean improvement was 2.7 arm curls, 2.1 chair stands, and 1.1 seconds on the eight-foot up-and-go tests. These improvements were statistically significant in all 3 tests (p<0.001) at each follow-up (Figure 1).

Normative data showed shifts to higher-performing categories in all physical function tests (Table 3). For example, for arm curls from baseline to the first program cycle (4 months), the number of people in the below-normal category shrank from 14% (n=1,055) to 7% (n=513), those in the normal category shrank from 48% (n=3,551) to 40% (n=2,757); while those in the above-normal grew from 35% (n=2,587) to 50% (n=3,440).

In the regression models, every additional complete week of EF class attendance per program cycle (each week of reaching the DHHS and CDC recommended physical activity levels) was statistically associated with an improvement in upper extremity strength (0.16 additional arm curls, p<.001), lower extremity strength (0.12 additional chair stands, p<.001) and balance & mobility (0.08 second reduction in time for the eight-foot up-and-go, p<.001) (Table 4).

Statistically significant associations between changes in each physical function test (the dependent variables in 3 different regression models that included attendance) and person-level predictors are presented below. We focused on those that were significantly associated in the desired direction (i.e., increased for arm curls and chair stands, decreased for eight-foot up-and-go) with at least 2 of the 3 physical function tests (Table 4) to emphasize the findings with the strongest evidence.

Greater EF attendance was associated with improved physical function. Results were consistent with previous EF research,^16,27 and the well-documented association between exercise attendance and physical function.^28–30

The mean attendance for the first program cycle was 10.6 complete weeks out of 16 possible. This adherence rate of 66% was similar to mean rates of other interventions, which a systematic review noted range from 58% to 77%.³¹ Although this sample was generally adherent to the EF program, there was still a significant association between improved physical function and attendance. This was important because our inclusion criteria constrained variation in attendance among participants. This suggests that even small increases in attendance of EF and other types of structured exercise programs, can benefit physical function.

Results appeared to be clinically significant, as evidenced by 2 metrics: the MDC and MCID values, and improvements in normative performance categories. First, for the MDC and MCID metric, the total mean improvement on arm curls (2.7 repetitions) met the MDC criteria of 2.3 repetitions on the second follow-up. The total mean improvement on chair stands (2.1 repetitions) met the MDC criteria of 2.0 repetitions on the second follow-up, and the MCID criteria of 2.0 repetitions on the second follow-up. The total mean improvement on the eight-ft up-and-go (1.1 seconds) approached the MDC criteria of 1.4 seconds on the second follow-up. There are currently no MCID available for the arm curl and eight-foot up-and-go tests. Second, for the age and sex normative categories metric, EF participants shifted to higher performance categories across each program cycle. For example, those in the above normal category for arm curls grew by 14.9% (n=853), while those in the normal category shrank by 7.9% (n=794). Complete details of these categorical shifts are presented in Table 3.

The baseline performance category did not consistently predict improvement; that is, those performing below norms at baseline did not necessarily experience less improvement than higher performing participants. This finding adds evidence to the potential of EF and similar types of structured exercise programs to benefit older adults with all levels of function. It also suggests that those in the lower-functioning category could improve at levels comparable to their better-functioning peers. These continued gains are important because some programs for older adults are not associated with a reduction in functional decline.¹²

Our results were not surprising, as the demographics and health conditions we found to be associated with less improvement have an intuitive association with lower physical function. Further, the self-report of worse health, falls, and disability has been associated with lower function¹. Our results were not likely due to differences in attendance or baseline performance, because we controlled for these in the models. We also controlled for a list of comorbidities. However, it is possible that these variables may not have fully accounted for the potential effects of the health condition. Nonetheless, groups with worse health, older age, more falls, and a disability still appeared to show improvement, as evidenced by the weak or non-significant interactions with time, and the improved mean scores. This improvement is important. Two systematic reviews found some interventions were not associated with functional improvements among older adults in worse health, and those that were linked to improvement had highly variable effect sizes.^11,12 Unexplained differences in EF effectiveness may be due to discomfort with the environment or the instructor (known as proxy efficacy³²), a lack of individual engagement during class sessions, and health or knowledge barriers not captured in the model. These results likely apply to other types of structured exercise programs.

We found identifiable subgroups who have the most to gain from EF. Targeted efforts to provide additional support for subgroups of older age, and with specific demographics and health conditions that participate in EF and similar programs could yield high dividends in improved physical function.

Our study had several strengths, including using a large sample of routinely collected program data from real-world settings, and looking at the same individual over time with multiple measurements. We included only participants with completed intervention cycles to avoid categorization bias. However, this limited the results to a select sample, and we could not draw conclusions about participants who did not provide a follow-up fitness check. However, the difference between those in the analytic sample and those who only contributed baseline data is likely to be minimal in the real-world setting, as the only statistically significant differences were very weak (with the exception of Asian race). We also could not control for reverse causation, and some variables lacked context. For example, the falls variable did not include a cause; so participants could have fallen for a multitude of reasons, such as vestibular disease, vision impairment, decreased strength or reflexes, or even from high-intensity athletic activities. One additional consideration is variable program fidelity, however a previous study found that EF maintains core elements of the tested intervention and has mechanisms to track and enhance fidelity (e.g., instructor certifications, participant program manuals, and periodic reviews with Master trainers).⁴⁰

Future qualitative research could attempt to learn how to better support participants throughout their time in the program, and identify site-specific barriers to effectiveness, which are likely highly variable. Evaluation modifications could include collecting data to directly identify underlying barriers to the program’s effectiveness. Additional analysis may use methods such as structural equation models to see the complex web of associations more accurately, or examine specific improvement trajectories of subgroups. Another analysis may determine if test days deter attendance, which may bias results. Finally, not all the physical function tests had MCID values available for the general population of older adults. Research should be conducted to establish MCID values so clinicians and researchers have a threshold though which to identify clinically important changes.