Long-term Outcomes of Exercise: Follow-up of a Randomized Trial in Older Women With Osteopenia

Background  Long-term evidence from randomized trials of the effectiveness of exercise in preventing disability and fall-related fractures in elderly people has been lacking.

Methods  We performed extended follow-up of 160 women (aged 70-73 years at baseline) with osteopenia in a population-based, randomized, controlled exercise trial. The trial was conducted from April 1 through April 30, 2001. Follow-up was conducted from May 1, 2001, through December 31, 2005. Mean total time in observation was 7.1 years. Primary outcome measures were femoral neck bone mineral density, postural sway, and leg strength. Secondary outcome measures were hospital-treated fractures and functional ability measures. Outcomes were measured annually using masked assessors.

Results  There was a significant difference between groups in favor of exercise in terms of postural sway (group × time interaction, P = .005), walking speed (group × time interaction, P < .001), and Frenchay Activities Index score (group x time interaction, P = .001). The bone mineral density decreased similarly across time in both groups. The incidence rate of fractures during the total follow-up among women in the exercise group vs women in the control group was 0.05 vs 0.08 per 1000 person-years (Poisson incidence rate ratio, 0.68; 95% confidence interval, 0.34-1.32). There were no hip fractures in the exercise group, whereas 5 hip fractures occurred in the control group. One woman in the exercise group and 8 women in the control group died (Poisson incidence rate ratio, 0.11; 95% confidence interval, 0.01-0.85).

Conclusion  Mainly home-based exercises followed by voluntary home training seem to have a long-term effect on balance and gait and may even protect high-risk elderly women from hip fractures.

Trial Registration  clinicaltrials.gov Identifier: NCT00655577

Fall-related fractures are associated with long-term pain, functional impairment, and increased risks of institutionalization and death in elderly women. Hip fractures place the greatest demands on resources and have the greatest effect on patients because they are associated with high mortality rates and increased morbidity.^1-4

Falls are responsible for at least 90% of all hip fractures.⁵ Recent meta-analyses concluded that exercise is successful in reducing fall risk and preventing falls in elderly people.^6-8 Evidence from prospective cohort studies indicates that a more active lifestyle is associated with a reduced risk of fractures in elderly women and that the risk goes down as physical activity level goes up.^9-12 Several randomized controlled trials also have shown the effect of exercise on surrogate end points of efficacy, such as bone mineral density (BMD), balance, and lower extremity strength.^13-18

Individuals with osteoporosis are at increased risk for fractures, not only because of low BMD but also owing to their decreased balance and muscle strength.¹⁹ Although there is evidence of the effectiveness of exercise in the prevention of falls in healthy populations, the effectiveness of exercise in elderly women with osteoporosis has been less studied. To our knowledge, a randomized controlled exercise trial with fractures as an outcome measure has never been performed. This is largely a result of methodological challenges such as study sample sizes, participants who drop out of a study, compliance, and masking. However, there is an urgent need to develop effective public health strategies for preventing fractures. Of all the methods of fracture prevention, regular physical activity is the only one that provides other considerable health-related benefits that may have a positive, albeit indirect, effect on fall and fracture risk in older adults.

Previously, we conducted a 30-month, population-based, randomized, controlled exercise trial aimed at reducing the risk factors for fractures in elderly women with osteopenia.^13,14 The aim of the present extended, 7-year follow-up of that trial was to assess the long-term effect of the chosen exercise regimen on balance, strength, and functional performance and its effect on the risk of fractures in elderly women. On the basis of the results of the original trial, we hypothesized that the women in the exercise group would maintain their higher performance level compared with the control group and that they would have fewer fractures.

The study flowchart is presented in the Figure. The original exercise trial was conducted between October 1, 1998, and March 31, 2001. The follow-up visits were performed annually at 4 years, 5 years, and 6 years from the beginning of the trial. The last follow-up measurements were performed in spring 2004. Data on fractures and causes of death were obtained from the baseline measurement in 1998 until December 31, 2005. The study design, recruitment process, and trial procedures have been described in detail previously.^13,14,20 The follow-up protocol was approved by the local ethics committee. All participants gave written informed consent at the beginning of the study and again in 2001 for the extended follow-up period.

Originally, we used the Finnish Population Register Centre to identify all women living in Oulu in 1997 and born between 1924 and 1927 (n = 1689) (Figure). The women were sent a questionnaire about their lifetime medical and lifestyle factors.²⁰ The women were screened for weight, height, and distal radius and hip BMD. Women who had a femoral neck (n = 270) and distal radius (n = 696) BMD value of at least 2 SDs lower than the reference value were excluded. Other exclusion criteria were health- or medication-related factors (acute illness or unstable chronic illness, malignant neoplasm, continuous use of oral corticosteroids, use of hormone therapy, and use of osteoporosis medication; n = 84), use of walking aid devices other than a stick (n = 4), severe cognitive impairment (n = 4), bilateral hip joint replacement (n = 1), and involvement in other interventions (n = 3). Women who had a femoral neck and distal radius BMD value less than 2 SDs lower than the reference value and who fulfilled the inclusion criteria (n = 160) were included in the trial in August 1998. The women were randomly assigned to an exercise group (n = 84) and a control group (n = 76) using computer-generated random numbers. Each participant received sequentially, according to the original identification numbers, the next random assignment in the computer list. Randomization was performed after recruitment, and it was conducted by a technical assistant not involved in the trial.

A detailed description of the intervention has been previously published.¹⁴ In brief, the women in the exercise group were asked to attend supervised balance, leg strength, and impact training sessions once a week for a 6-month period from October through March each year. In addition, the participants were asked to train 20 minutes daily at home following a program that consisted of similar patterns of exercise to those in the supervised sessions. From April through September, the exercises took place only at home. The participants in the control group were given general health information at baseline and were asked to continue their daily routine activities.

All women who had participated in the intervention phase of the study were enrolled in the postintervention follow-up phase, as well as those who had withdrawn earlier (15 in the intervention group and 9 in the control group). During the follow-up, all study participants had a yearly visit with the same physiotherapist and nurse who had performed the measurements during the trial. The same questionnaire that had been sent to the women before the first screening visit was administered by a nurse at each visit. Fracture data were obtained for all 160 women who were enrolled in the original trial. Data on 7-year incidence of hospital-treated fragility fractures were collected from the national hospital discharge register. The Finnish register is the oldest nationwide discharge register in the world, and its coverage has been shown to be accurate, particularly for severe injuries such as bone fractures.²¹ To avoid the bias of recording multiple hospitalizations for the same fracture, a comprehensive screening of patients' medical records was performed manually in 2006. For patients with multiple hospitalizations due to the same fracture, only the first hospitalization was analyzed to avoid overcounting of events. If more than 1 fracture occurred at the same time, these were recorded as separate fractures. Fractures occurring in connection with motor vehicle crashes and bicycle accidents were not included. Information on the date and cause of death was obtained from the Cause of Death Register located at Statistics Finland (Helsinki) by using the underlying and immediate causes of death.

Outcome measures were assessed annually. A detailed description of the measurement protocol has been published previously.^13,14 The visits included the same procedures as during the intervention period and were similar for all participants irrespective of their former randomization group. The assessors in direct contact with participants during the study did not know to which group they had been assigned.

A clinical examination was performed by the same experienced nurse at baseline and at 12 months, 24 months, 30 months, 4 years, 5 years, and 6 years from the beginning of the study. The same questionnaire that had been sent to the women before the first screening visit was administered by a nurse at each visit.²⁰ To evaluate the activities of daily living, the self-report version of the Frenchay Activities Index²² (FAI) was used. Symptoms of depression were ascertained with the Geriatric Depression Scale (GDS),²³ and cognitive function was assessed with the Mini-Mental State Examination.²⁴ At each visit the following measurements were obtained: body weight and height, body mass index (BMI; calculated as weight in kilograms divided by height in meters squared), percentages of fat and lean mass (Body stat 1500; Bodystat Ltd, Douglas, Isle of Man), Timed Up and Go (TUG) test score,²⁵ walking speed,²⁶ walking endurance,²⁷ and physical activity level.¹³ Postural sway measurements were performed in standardized conditions using an inclinometry-based method.¹³ Maximum isometric leg extensor strength was assessed bilaterally with a computerized strain gauge dynamometer (Newtest Ltd, Oulu), and grip strength was measured with a hand dynamometer (Newtest Ltd).

Areal BMD and bone mineral content (BMC) measurements were performed annually at the left proximal femur using dual energy x-ray absorptiometry (DEXA; GE Lunar Corp, Madison, Wisconsin). The BMD of the dominant distal radius was measured at baseline, at 30 months, and at 6 years using peripheral dual energy x-ray absorptiometry (Osteometer DTX-200; Osteometer MediTech, Roedovre, Denmark). Assessment of calcaneal broadband ultrasound attenuation (decibels per megahertz) and speed of sound (meters per second) was performed at baseline, at 30 months, and at 6 years using quantitative ultrasound (Sahara Clinical Bone Sonometer; Hologic, Bedford, Massachusetts). The calibration and measurement, as well as quality control procedures, have been described in detail previously.¹³

All the analyses were performed using SPSS for Windows, version 15.0 (SPSS Inc, Chicago, Illinois), and StatsDirect, version 2.6.5 (StatsDirect Ltd, Altrincham, England), statistical software. The trial data were analyzed on an intention-to-treat basis. Originally, we estimated that at a 5% significance level we would require 64 women in each group to give 80% power to detect a 0.02-g/cm² difference in femoral BMD (with an SD of 0.04 g/cm²) between groups. The study also had 80% power at the 5% significance level to show a difference of 2.0 cm (15.0 cm vs 13.0 cm; SD, 3.5) in postural sway at 30 months between groups. The primary outcome measures for the extended follow-up were femoral neck BMD, postural sway, and leg strength. The secondary outcome measures were hospital-treated fractures and functional ability measures (walking speed, walking endurance, and TUG test; FAI, GDS, and Mini-Mental State Examination scores). Thus, the original trial was not powered for detecting a difference in antifracture efficacy between groups. The fracture incidence rate was calculated as the number of fractures during the total amount of person-years in each group. Person-years was calculated as the time that the women were undergoing observation in the study. Fractures distal to the elbow and knee were considered as distal fractures, and fractures proximal to the elbow and knee were considered as proximal fractures. Poisson regression analysis was used to analyze the fracture and mortality rate data. Poisson incidence rate, incidence rate ratio (IRR), and P values for the incidence rate difference between groups are presented. We performed multivariate stepwise logistic regression analysis using all variables associated with any fracture in univariate analyses to evaluate the determinants of having 1 or more fractures in the pooled groups. Odds ratios with 95% confidence intervals (CIs) are presented.

All the BMD and performance variables were normally distributed. The absolute and percentage changes from baseline were calculated for each characteristic. The means with 95% CIs were calculated for change within a group and for difference between groups. The independent-sample t test was used to compare the exercisers with the control group. For the dichotomous variables, the χ² test was used to evaluate the significance of the differences.

The changes in bone and performance variables during the trial and posttrial follow-up period were examined using multiple linear regression analyses with generalized estimating equations (GEEs) to account for correlations among 7 repeated measurements. The GEE method extends standard regression analysis, taking into account the baseline value and correlation between repeated measurements. This method can handle cases with missing data without the need for imputation,²⁸ and data for a person at a certain time during follow-up are included regardless of whether data for that person are missing at other times. Also, GEE analysis takes into account the status or changing value of covariates at each visit. The bone analyses were adjusted for BMI change from baseline, calcium intake, and osteoporosis medication use. The performance models were adjusted for BMI change.

Baseline data on the exercise and control groups are given in Table 1. Sixty-nine women (82.1%) in the exercise group and 67 (88.2%) in the control group completed the trial. Fifty-five women (65.5%) in the exercise group were available for the final follow-up measurements and 45 (59.2%) in the control group. Of the 160 women, 74 (46.2%) had complete data at all 7 time points during the trial and extended follow-up. Twenty-six women (16.2%) had missing data, and 60 (37.5%) were lost to follow-up either during the trial or after the intervention. The reasons for withdrawal during the trial were unwillingness to continue (n = 12), new or worsening health problems (n = 7), and medication included in the initial exclusion criteria (n = 5). The reasons for not completing the extended follow-up were death (n = 9), health problems or family reasons (n = 13), and unknown reasons (n = 13). Participants who dropped out of the study were similar to those who continued to participate with respect to baseline characteristics. The changes in BMD and BMC, postural balance, strength, and functional capacity, as well as the frequency of falls and fall-related fractures during the trial, were reported previously.^13,14

The mean (SD) total time in observation was 7.1 (0.2) years in the exercise group and 6.9 (0.7) years in the control group. There were 17 hospital-treated fractures during follow-up in the exercise group, whereas 23 fractures occurred in the control group. The incidence rate of fractures during the total follow-up among the exercisers vs women in the control group was 0.05 vs 0.08 per 1000 person-years (IRR, 0.68; 95% CI, 0.34-1.32; P = .22). When the fracture model was adjusted for changes in FAI score, gait speed, and postural sway from baseline to the final visit, gait speed was the only independent predictor of fracture incidence rate (IRR, 0.13; 95% CI, 0.02-0.75; P = .02). There were no hip fractures in the exercise group, whereas 5 hip fractures occurred in the control group (incidence rate difference, P = .02). Distribution of fractures according to anatomical location is indicated in Table 2. Fractures were proximal in 52.2% of the control group and 17.6% of the exercise group (P = .02). Moderate lifelong physical activity decreased the overall risk of having any fracture during the total follow-up period (odds ratio, 0.22; 95% CI, 0.1-1.0).

By the end of 2005, 1 woman (1.2%) in the exercise group had died, whereas 8 women (10.5%) in the control group had died, giving a crude death rate of 0.003 per 1000 person-years in the exercise group and 0.03 per 1000 person-years in the control group (Poisson IRR, 0.11; 95% CI, 0.01-0.85; P = .01). The woman in the exercise group died of lung cancer, whereas the main acute or underlying causes of death in the control group were cancer (n = 2), cardiovascular disease (n= 5), and an external cause (bicycle accident; n = 1).

The absolute values on the physical performance tests and the significance of the difference between groups in change across time appear in Table 3.

Postural sway increased in both groups from baseline to 6 years, with the increment being more pronounced in the control group (group × time interaction, P = .005). When only the 3 last postintervention measurements were included in the analyses, the difference in change between groups was of borderline significance (P = .06).

The exercise group demonstrated a significant gain compared with the control group in mean leg strength during the trial. The difference between groups diminished until the 4-year follow-up visit (P = .41 for the change from baseline). During the postintervention follow-up, the exercise group had a more pronounced decrease in mean maximal leg strength than the control group (group × time interaction, P = .04).

The exercisers maintained their gait speed at the baseline level, whereas the women in the control group had a decrease in gait speed during the total 6-year follow-up (group × time interaction, P = .001). The difference between groups was also seen during postintervention follow-up (group × time interaction, P = .03).

There was a significant difference in change from baseline in the 2-minute walking test score between exercisers and the control group (3.3 m vs 16.4 m; P = .06; group × time interaction, P = .005). The exercisers improved their performance until the 4-year visit. In both groups, most of the decrease in walking performance occurred between years 5 and 6. The difference between groups diminished after the trial and was of borderline significance during the extended follow-up (P = .06). The difference between groups in TUG test score was practically the same at baseline and at the final visit. Group × time interaction was significant during total follow-up (P < .001) and postintervention follow-up (P = .04), mainly owing to improvement of the exercisers during the intervention.

In the trial, activities of daily living assessed with the FAI showed a significant and similar decrease within the exercise and control groups, but during the total 6-year follow-up there was a group × time interaction in favor of the exercise group (group × time interaction, P < .001). During the postintervention follow-up, the women in the exercise group slightly increased their score compared with the control group (group × time interaction, P = .006). There was no clinically significant difference in change in GDS or Mini-Mental State Examination scores between groups across time.

Table 4 gives the absolute BMD and BMC values measured with dual energy x-ray absorptiometry in the beginning of the trial and during the postintervention follow-up period, as well as the significance of the difference between groups in multiple linear regression analyses with GEEs. The BMD and BMC decreased across time, and the time trends were similar in both groups.

This population-based postintervention follow-up study showed for the first time to our knowledge that supervised, mainly home-based weight-bearing exercise was successful in reducing important risk factors for falls and fractures in elderly women with osteopenia. Despite a small sample size, the data also showed a promising effect of training on hip fractures.

The strengths of the study include its design and long-term follow-up. The target population was a homogeneous, stable, and representative sample of older Finnish women obtained from the Finnish Population Register Centre, which has 100% coverage. The chosen exercise regimen is easily applicable at the population level.

Randomized controlled trials have suggested, and systematic reviews have confirmed, that strength and balance training for healthy elderly people can reduce the risk of falls by 15% to 50%.²⁹ Previous systematic reviews of randomized controlled exercise trials reporting the number of fractures in elderly populations have failed to find evidence of the efficacy of exercise for preventing osteoporotic fractures.^30,31 A meta-analysis by Lock et al³¹ evaluated the clinical effectiveness of 4 exercise trials^32-35 in which the outcomes measured differed between trials: osteoporotic fractures at any site were reported in some trials,³³ and spine^32-35 or wrist³³ fractures were found in other trials. Exercise was not associated with risk of fractures in that analysis. None of the included trials adequately reported the concealment of allocation of participants to treatment arms, and only 2 trials^32,34,35 reported masking of the outcome assessor. Only 1 trial^34,35 described participants who withdrew from treatment.

To our knowledge, our study is the first to report the long-term effect of exercise on fractures in elderly women with osteopenia. Our results are in concordance with an up-to-date review including the most recent high-quality trials.⁷ The authors concluded that exercise can reduce falls, fall-related fractures, and several risk factors for falls in individuals with low BMD. Our data also confirm the promising results of prospective observational cohort studies showing that moderate lifelong physical activity is associated with a reduced risk of fragility fractures in elderly women.^9,10 However, the best evidence for the effect of exercise in preventing fragility fractures would be to conduct several population-based randomized controlled trials in individuals with osteopenia with fractures as an end point. So far there have been no such trials, and more high-quality randomized controlled trials recording fractures at all sites are needed to evaluate the effect of exercise on risk of fractures. The fractures were located more proximally in the control group than in the exercise group, indicating that the type of fall may have been different in the exercisers. Forces applied to the hip during a fall depend on the fall characteristics, self-protective responses, and thickness of soft tissues overlying the hip.³⁶ During the trial, the women in the exercise group had a significant improvement in many physiological characteristics that may have modified the fall dynamics and improved additional absorptive mechanisms.

Hip fracture is easy to diagnose, and practically all hip fractures require hospital treatment. Therefore, the hospital discharge registers are a good source for the identification of patients with hip fracture in Finland. However, a potential bias in using hospital discharge records is related to multiple hospitalizations of a single patient and possible readmissions due to the same fracture. To avoid this bias, fractures in all original 1689 women were confirmed from patient records manually in 2006. Data on fractures were collected only from hospital admissions, which for some fracture types (such as vertebral and wrist fractures) represent only a proportion of the fractures in that population. However, it is unlikely that the treatment of fractures was dependent on group assignment in this study, and our results are not unbiased in regard to the effect of exercise on risk of fracture.

Each activity of daily living, such as rising from bed, requires a certain amount of postural control and strength. Many elderly people live just beyond the threshold of the capacity needed for such tasks. The reserve in performance capacity may be so slight that even a small additional decline in balance or strength can cause serious difficulties. On the contrary, a minor increase in capacity may help maintain independence. To be able to move about their residences and communities, elderly people should be able to walk 300 m. The speed required to cross the street safely has been suggested as 1.1 m/s,³⁷ and a walking speed of 1 m/s or less was shown to be indicative of poor health outcomes.³⁸ Those who perform the TUG test in less than 20 seconds appear to be able to go outside alone safely.²⁵ In our study, the chosen exercise regimen had no effect on BMD or BMC. However, the women in the exercise group preserved their gait speed, postural control, and functional ability and had fewer fractures compared with the women in the control group. Change in gait speed was the only significant independent predictor of fracture incidence rate. These results suggest that these women may have had an increase in performance capacity reserve large enough to prevent loss of independence and future fractures.

We also found a significant decrease in all-cause mortality within the exercise group. In observational studies, physical activity has been shown to reduce all-cause mortality in elderly women,^39,40 and there seems to be an inverse linear dose-response relationship between the volume of physical activity and all-cause mortality. However, the small sample size limits the conclusions that can be drawn in this study.

Several limitations in our study need to be considered. Of the 1689 originally eligible women, 467 (27.6%) did not participate in the study. These women seemed to be the frailest group in our cohort. Of the nonparticipants, 194 women (41.5%) died during follow-up, which is a significantly higher mortality rate than in the participants. Hip fractures were also more common in the nonparticipants than in the participants. Therefore, the results of this study cannot be generalized to very ill or institutionalized elderly people. To make these observations more generalizable, the findings must be replicated in institutionalized women and other age groups of women. In the exercise trial, it was not possible for obvious reasons to mask patients to the intervention. Finally, because of incomplete information on falls and physical activity during the postintervention period, these data were not reported.

In conclusion, 30 months of supervised, mainly home-based exercises followed by voluntary home training had a positive long-term effect on balance and gait in high-risk elderly women. The exercise program also seemed to decrease the risk of hip fracture. Lifelong physical activity was associated with reduced risk of fractures. Furthermore, mortality was significantly lower in the exercise group than in the control group during the extended follow-up period. Regular daily physical activity should be recommended to elderly women with osteopenia.

Correspondence: Raija Korpelainen, PhD, Department of Sports and Exercise Medicine, Oulu Deaconess Institute, Kajaaninkatu 17, 90100 Oulu, Finland (raija.korpelainen@odl.fi).

Accepted for Publication: February 12, 2010.

Author Contributions: Drs R. Korpelainen, Nieminen, and J. Korpelainen had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: R. Korpelainen, Keinänen-Kiukaanniemi, Väänänen, and J. Korpelainen. Acquisition of data: R. Korpelainen, Heikkinen, and J. Korpelainen. Analysis and interpretation of data: R. Korpelainen, Keinänen-Kiukaanniemi, Nieminen, and J. Korpelainen. Drafting of the manuscript: R. Korpelainen, Keinänen-Kiukaanniemi, and Nieminen. Critical revision of the manuscript for important intellectual content: Keinänen-Kiukaanniemi, Nieminen, Heikkinen, Väänänen, and J. Korpelainen. Statistical analysis: R. Korpelainen and Nieminen. Obtained funding: R. Korpelainen and Väänänen. Administrative, technical, and material support: Keinänen-Kiukaanniemi and J. Korpelainen. Study supervision: Keinänen-Kiukaanniemi, Heikkinen, and Väänänen.

Financial Disclosure: None reported.

Funding/Support: This study was funded by the Finnish Ministry of Education, the Finnish Cultural Foundation, the Juho Vainio Foundation, the Miina Sillanpää Foundation, the Research Foundation of Orion Corporation, and the Northern Ostrobothnia District Hospital.

Role of the Sponsors: The sponsors of the study had no role in the design or conduct of the study; the collection, analysis, or interpretation of the data; or the drafting of the manuscript.