Effect of exercise intervention on health-related quality of life in middle-aged and older people with osteoporosis: a systematic review and meta-analysis

Search strategy

We systematically searched six databases, including Web of Science, PubMed, Embase, Cochrane Library, China National Knowledge Infrastructure (CNKI), and Wan Fang Database, for RCTs on the impact of exercise interventions on HRQOL among middle-aged and older osteoporosis patients. The time period was set from the establishment of the database to 31 December 2023, and a combination of subject words and free words was used in terms of the retrieval strategy. Using the PubMed database as an example, subjects and free words were used for retrieval: osteoporosis, aged, middle-aged, quality of life, exercise, resistance training, endurance training and randomized controlled trial. See Supplemental Information 2 for the specific retrieval strategies.

Literature selection

The literature screening criteria was developed according to evidence-based PICOS (participants, intervention, comparison, outcomes, and study design) in medical science (Liberati et al., 2009). The studies included must meet the following criteria: (a) participants were clinically diagnosed with osteoporosis (T-score ≤ −2.5) and aged 45 years or above; (b) exercise interventions were conducted in the experimental group, without limitation on exercise types; (c) the control group did not receive exercise intervention, but maintained their daily activities and/or received routine treatment and health education; (d) the outcome indicators included HRQOL, which was measured by questionnaire; (e) the research adopted a randomized controlled trial as its research design. The exclusion standards included: (a) duplicate literature; (b) articles and abstracts from conferences; (c) books, dissertations and theses; (d) considerable baseline variations observed between the experimental and control groups; (e) studies lack of data for the determination of the effect size; (f) studies not reported in Chinese or English. The literature screening process consisted of three steps: step one, the literature was imported into EndNote X9 literature management software for preliminary processing, and duplicates were removed through automatic and manual methods; step two, the literature that failed to satisfy the inclusion requirements was excluded, which involved title and abstract screening based on the citation information retrieved; step three, the full text of the literature included in the preliminary screening was read one by one to determine whether it should be finally included. Independently, two trained researchers (DG and XL) evaluated the literature in accordance with the inclusion and exclusion criteria. If there was disagreement between the two researchers, which could not be resolved through negotiation, a third researcher (YS) intervened to assist in making a decision. These three reviewers acquire extensive expertise and research experience in the field of sport and exercise science. The study with the most thorough analysis or the biggest sample size was included when data from two or more studies came from the same sample. Interrater reliability for screening was calculated using kappa coefficient (McHugh, 2012).

Data extraction

Two researchers (DG and XL) used an independent double-blind method to collect data from studies that matched the inclusion criteria, entering the data into a pre-made Excel spreadsheet to minimise errors and potential bias. If the data entered by the two researchers was inconsistent, a third researcher (YS) will ensure it to be re-checked and unified. The content of the data extracted was as follows: (a) fundamental literary information (name of the author, date of publication); (b) basic information of the research object (sample size, age, gender, fracture history); (c) intervention plan (category, time, frequency, period); (d) measurement tools and results of outcome indicator HRQOL. In case of two measurement tools adopted in the study for outcome indicators, the outcome data were extracted respectively.

Intervention categories and outcome measurement

The categories of exercise interventions are broadly classified as aerobic exercise, resistance training, balance training, multicomponent exercise, etc. Aerobic exercise is a continuous form of exercise designed to enhance the body’s metabolic function, and includes walking, running, gymnastics, swimming, Tai Chi, and more (Alves et al., 2015). Resistance exercise is a form of exercise that increases muscle strength by overcoming external resistance, which can be formed by using elastic bands, dumbbells, barbells, other instruments, or body weights (Loveless & Ihm, 2015). Balance training refers to functional training aimed at improving the human body’s balance function. It is mostly practiced on a balance board, balance beam, or narrow path by means of walking, body movements, balance exercises, etc. Multicomponent exercise is a combination of two or more training methods, typically consisting of aerobic exercise, resistance training, balance training, or other forms of exercise. If the included studies did not clearly report the category of exercise interventions, we classified them based on the criteria described above.

The measurement of HRQOL is mainly achieved through self-evaluation scales. According to different scoring methods, they can be divided into two forms: one is forward scoring scales, such as Short Form 36 Health Survey (SF-36), Osteoporosis Quality of Life Questionnaire (OQLQ), and EuroQol Five Dimensions Questionnaire (EQ5D), a higher score indicates a better HRQOL of the patients; another is reverse scoring scales, such as Quality of Life Questionnaire of the European Foundation for Osteoporosis (Qualeffo-41), where a higher score indicates a poorer HRQOL of the patients. In order to maintain data consistency and facilitate consolidation for effect size calculation, we converted the mean outcome indicators of the experimental group and those of the control group to negative values when reverse scoring scales were used in the studies included.

Bias risk assessment

Two reviewers independently assessed the methodological quality of the studies included using the Cochrane risk assessment tool in accordance with the principles of evidence-based medical studies (Sterne et al., 2019). The Cochrane risk of bias tool was adopted to evaluate the studies’ quality in six areas: selection bias (random sequence generation and allocation concealment), performance bias (blinding of participants and staff), detection bias (blinding of outcome assessors), attrition bias (incomplete outcome data), reporting bias (selective reporting of results), and other bias. To express their opinions, reviewers used the terms “low risk of bias”, “high risk of bias” or “unclear risk of bias”. In case of inconsistent assessment results, a third researcher was involved in the discussion to reach a consensus, and a bias risk graph was ultimately generated.

Statistical analysis

Statistical analysis was processed with Review Manager 5.4 software to create forest and funnel plots. Due to the fact that the experimental data was a continuous variable and the measurement tools were different, we utilized standardized mean difference (SMD) and 95% confidence interval (95% CI) as the effect scales to merge the effects. The threshold for statistically significant merge effects is P < 0.05. The Q-statistic and I² statistic were used to assess the degree of heterogeneity in the study. The random effects model was adopted for significant heterogeneity (P < 0.10, I² ≥ 50%), whereas the fixed effects model was employed for non-significant heterogeneity (P ≥ 0.10, I² < 50%). In case of significant heterogeneity between studies, a one-to-one exclusion method was used in the sensitivity analysis to explore possible sources of heterogeneity. It should be noted that final measurements of outcome indicators were used when combining the effects, rather than baseline variation values, as not all literature included reported values prior to and after the interventions. In studies where only baseline and change values were available, the research by Shu et al. (2020) was referenced, and the final measurements were able to be estimated using the following formula: $mea{n}_{final}\approx mea{n}_{base}+mea{n}_{change},S{D}_{final}\approx \frac{2\times R\times S{D}_{base}+\sqrt{4\times R^2\times S{D}_{{}_{base}}^2-4\times S{D}_{{}_{base}}^2+4\times S{D}_{change}^2}}{2}$, R = 0.5.

Literature search results

Figure 1 depicts the selection procedure for the literature included in this study. 1,328 articles were obtained by retrieving information from six databases. By importing these articles into literature management software, 469 duplicate articles were removed. Next, to screen the 859 articles remained, their titles and abstracts were read, after which 795 more articles that did not meet the requirements of this study were excluded. Then, by thoroughly reading the full text of the 64 articles that were initially included, 50 articles were finally removed, including those with data concerning duplicate sample (n = 3), unavailable data (n = 8), control group not eligible (n = 6), non-Chinese or English articles (n = 3), no outcomes of interest (n = 6), ineligible interventions (n = 7), ineligible participants (n = 15), and unreported baseline data (n = 2). In the end, 14 studies in total were incorporated into the meta-analysis (Alp et al., 2009; Arnold et al., 2008; Çergel et al., 2019; Evstigneeva et al., 2016; Ferrara et al., 2019; Fu & Fan, 2021; Gibbs et al., 2020; Hongo et al., 2007; Kanemaru et al., 2010; Khalili et al., 2016; Kronhed et al., 2009; Sen, Esmaeilzadeh & Eskiyurt, 2020; Stanghelle et al., 2020; Zhang et al., 2022a). There was strong agreement between the reviewers for the screening records and full texts (Kappa: 0.89).

Basic characteristics of included literature

Table 1 lists the detailed information of the 14 studies included, with all of them published between 2007 and 2022. Next, we describe the features that may be sources of heterogeneity in the results in a brief manner. The sample sizes included in these studies ranged from 34 to 183 people, with a total of 1,214 middle-aged and older osteoporosis patients recruited, the vast majority of whom were women (94.81%, n = 1,151). The majority of participants enrolled in six studies had a history of osteoporotic fractures, while participants enrolled in five studies had no such history. The other three studies did not report relevant information explicitly.

The categories of exercise interventions included in this study exhibited significant differences. According to the previously established classification criteria for exercise interventions, they can be divided into three categories: aerobic exercise (n = 3), resistance training (n = 6), and multicomponent exercise (n = 5). In the category of aerobic exercise, two out of three studies used Tai Chi as an intervention, which is known for its low-intensity, balance, and coordination training; the remaining one study implemented personalized exercise programmes based on individual preferences and characteristics, such as running, dancing, hiking, and swimming. In the category of resistance training, six studies were further subdivided: three studies focused on back extensor strength training programmes aimed at enhancing the strength and stability of core muscle groups; and the other three studies adopted strength training programmes involving multiple muscle groups to comprehensively improve muscle strength and function. As for the multicomponent exercise category, three out of five studies combined resistance training and balance training to promote comprehensive physical fitness and prevent falls; the other two studies adopted a comprehensive approach that included aerobic exercise, resistance training, and balance training, aiming to improve the physical health level of participants in a comprehensive manner.

Different tools were also adopted in these studies to measure outcome indicators, with SF-36 alone was used in six studies, Qualeffo-41 alone was used in three studies, OQLQ was used in two studies, SF-36 and Qualeffo-41 were used in two studies, and EQ5D and OQLQ were used in two studies. Overall, SF-36 was the most commonly utilized assessment tool, as it was employed in eight out of the 14 studies. SF-36 evaluates eight aspects of health-related quality of life (HRQOL), which belong to two major categories: physical component and mental component. Qualefo-41 and OQLQ are survey questionnaires specifically designed to evaluate the quality of life of patients with osteoporosis. The evaluation includes a total score and scores in five domains. EQ5D is a tool used to measure health-related quality of life, published by the European Quality of Life Research Group in 1990. The scale includes five health dimensions.

Risk of bias

The results of the 14 included studies’ bias risk assessment are shown in Fig. 2. All studies were classified as high risk or uncertain in terms of performance bias, eight studies were classified as low risk in terms of detection bias, 12 studies were classified as low risk in terms of follow-up bias and reporting bias, and 10 studies were classified as low risk in terms of other bias. Nine studies were considered as low risk in terms of random sequence generation or allocation concealment. The statistical proportion of each item in the bias risk assessment is shown in Fig. 3.

Main outcome: general HRQOL

Eight studies in total evaluated the effects of exercise interventions on general HRQOL. The results of the meta-analysis of the experimental and control groups’ combined effect size data on the general HRQOL are displayed in Fig. 4. After exercise interventions, the general HRQOL in the experimental group showed remarkable improvements compared to the control group (SMD = 0.79, 95% CI [0.34–1.24], p = 0.0006). Due to significant heterogeneity among the studies (I² = 86%, p < 0.00001), sensitivity analysis was performed using the leave-one-out method, which indicated that the exclusion of any one study did not have a significant impact on the general HRQOL effect size (SMD = 0.68 to 0.92).

Subgroup analysis based on the exercise intervention programmes utilised in the eight trials was carried out in order to further examine the effects of varied exercise interventions on the improvements in overall HRQOL. Figure 5 displays the findings of the subgroup analysis based on exercise type, frequency, and duration. Resistance exercise (SMD = 1.01, 95% CI [0.50–1.52], p = 0.0001) significantly improved general HRQOL compared to multicomponent exercise (SMD = 0.58, 95% CI [−0.08–1.24], p = 0.09), according to the subgroup analysis based on exercise type. Exercise interventions that occurred three or more times per week (SMD = 0.80, 95% CI [0.22–1.38], p = 0.007) significantly improved general HRQOL compared to those that occurred less frequently (SMD = 0.79, 95% CI [−0.13–1.71], p = 0.09), according to the subgroup analysis based on exercise frequency. Also, the subgroup analysis based on exercise duration showed that exercise interventions lasting 13–24 weeks (SMD = 0.85, 95% CI [0.37–1.33], p = 0.0005) had a greater and statistically significant impact on general HRQOL than interventions lasting 12 weeks and below (SMD = 0.82, 95% CI [−0.12–1.76], p = 0.09) or more than 24 weeks (SMD = 0.69, 95% CI [−0.69–2.06], p = 0.33).

Secondary outcome: physical HRQOL

Physical HRQOL includes bodily pain, physical function, role physical, and general health. Regarding the outcomes, bodily pain, physical function, role physical and general health were measured in 11, nine, eight and 11 studies, respectively. Figure 6 shows the results of the meta-analysis of the combined effect sizes data for each component of the physical HRQOL in the experimental and control groups. Combined results showed that the experimental group was superior to the control group in the reduction of bodily pain (SMD = 0.51, 95% CI [0.24–0.78], p = 0.0002), the improvement of physical function (SMD = 0.56, 95% CI [0.21–0.91], p = 0.002), physical (SMD = 0.39, 95% CI [0.14–0.64], p = 0.003), and general health (SMD = 0.68, 95% CI [0.25–1.11], p = 0.002). The results of the sensitivity analysis were similar to those presented above, with no outstanding differences.

Secondary outcome: mental HRQOL

Mental HRQOL includes vitality, social function, role emotion, and mental health. Regarding the results, vitality, social function, role emotion and mental health were measured in 8, 10, 9 and 10 studies, respectively. Figure 7 shows the results of a meta-analysis of the combined effect sizes data for each component of the mental HRQOL in the experimental and control groups. The combined results demonstrated that the group that received exercise interventions improved vitality (SMD = 0.58, 95% CI [0.15–1.01], p = 0.008), social function (SMD = 0.37, 95% CI [0.17–0.58], p = 0.0004), and mental health (SMD = 0.50, 95% CI [0.25–0.74], p < 0.0001) in comparison to the control group. Despite being modestly favorable, the combined effect of role emotion was not statistically significant (SMD = 0.25, 95% CI [−0.03–0.54], p = 0.08). The results of the sensitivity analysis were similar to those presented above, but the differences were not significant.

Publication bias

Because ten or more studies were included in the meta-analysis of the effects of exercise interventions on bodily pain, general health, social function, and mental health, we used a funnel plot to test their publication bias. From Fig. 8, we can see that the scatter is upward distributed, with a fundamental balance between left and right, and no significant publication bias among the studies.