|Home | About | Journals | Submit | Contact Us | Français|
Interventions to increase physical activity among chronically ill adults are intended to improve quality of life as well as reduce disease complications or slow disease progression.
This meta-analytic review integrates quality-of-life outcomes from primary research studies testing interventions to increase physical activity among adults with chronic illness.
Extensive literature searching strategies were employed to locate published and unpublished primary research testing physical activity interventions. Results were coded for studies that had at least 5 participants with chronic illness. Fixed- and random-effects meta-analytic procedures included moderator analyses.
Eighty-five samples from 66 reports with 7,291 subjects were synthesized. The mean quality-of-life effect size for two-group comparisons (treatment vs. control) was 0.11 (higher mean quality-of-life scores for treatment subjects than for control subjects). The treatment group pre-post comparison effect size was 0.27 for quality of life. Heterogeneity was modest in two-group comparisons. Most design and sample attributes were unrelated to intervention effects on quality of life. Studies that exclusively used supervised center-based exercise reported larger quality-of-life improvements than studies that included any educational-motivational content. Effect sizes were larger among unpublished and unfunded studies. The effect size for physical activity did not predict the quality-of-life effect size.
Subjects experience improved quality of life from exposure to interventions designed to increase physical activity, despite considerable heterogeneity in the magnitude of the effect. Future primary research should include quality-of-life outcomes so that patterns of relationships among variables can be further explored.
Most adults with chronic illnesses remain sedentary despite evidence of potential health benefits of increased physical activity (PA). Interventions to increase PA among chronically ill adults are intended to reduce disease complications or slow disease progression as well as improve quality of life (QOL). These potential benefits have contributed to the large body of primary research testing interventions to increase PA. Few previous reviews have addressed QOL outcomes from interventions to increase PA. This quantitative synthesis meets the need to synthesize and integrate the QOL outcome findings to guide practice and inform future research.
Researchers have summarized PA intervention research in numerous narrative reviews and a growing number of meta-analyses. Although reviewers often mention PA’s consequences for QOL, few summarize QOL findings. Instead, reviews address symptoms or health outcomes that are assumed to be related to QOL (Ciccolo, Jowers, & Bartholomew, 2004; Dishman, 2003; Rietberg, Brooks, Uitdehaag, & Kwakkel, 2005). However, neither symptom changes nor health outcomes are adequate proxies for QOL outcomes (Netz, Wu, Becker, & Tenenbaum, 2005). PA may improve QOL beyond symptom and physical function changes (Drewnoski & Evans, 2001; Netz et al. 2005).
Few reviewers have examined QOL outcomes directly. Rejeski, Brawley, and Shumaker’s (1996) narrative review of the link between increased PA and QOL concluded that findings are inconsistent. Some meta-analyses have addressed QOL-related outcomes (e.g. fatigue, depression, anxiety) but have not synthesized outcomes measured by QOL instruments (Devos-Comby, Cronan, & Roesch, 2006; Fox, 2000; Puetz, O’Connor, & Dishman, 2006). The scarce meta-analytic reviews addressing QOL outcomes from PA interventions have been limited to specific diseases or to older adults. They reported mixed outcomes (Netz et al., 2005; Spronk, Bosch, Veen, den Hoed, & Hunink, 2005; Taylor et al., 2005). This meta-analysis was designed to address the need to quantitatively summarize the effects of PA interventions on QOL outcomes in broader participant populations.
This synthesis addressed the following research questions (1) What is the overall mean difference effect size in QOL scores between treatment and control subjects after interventions to increase PA? (2) What is the overall mean difference effect size in QOL scores between treatment subjects prior to versus after PA interventions? (3) Do PA interventions’ effects on QOL outcomes vary depending on characteristics of participants, methodology, or interventions? (4) Do PA behavior outcomes following interventions predict QOL outcomes? (5) For two-group comparisons, do control groups’ post-test outcome measures differ significantly from pre-test values?
Research synthesis methods widely reported in the literature to identify and secure potential primary research reports, evaluate their eligibility, extract data from research reports, meta-analyze primary study characteristics and findings, and interpret meta-analysis results were used. This project is part of a larger study synthesizing PA interventions among chronically ill adults. Further details about methods and results of the findings regarding PA outcomes and disease-specific health outcomes are available in other articles (Conn, Hafdahl, Brown, & Brown, 2008; Conn, Hafdahl, LeMaster et al., 2008; Conn, Hafdahl, Mehr et al. 2007; Conn, Hafdahl, Minor, & Nielsen, 2008; Conn, Hafdahl, Moore, Nielsen, & Brown, 2008; Conn, Hafdahl, Porock, McDaniel, & Nielsen, 2006; Nielsen, Hafdahl, Conn, LeMaster, & Brown, 2006). Several excellent texts are available to readers unfamiliar with meta-analysis methods (Cooper, 1998; Cooper & Hedges, 1994; Lipsey & Wilson, 2001; Sutton, Abrams, Jones, Sheldon, & Song, 2000). The project was approved by the Institutional Review Board for the Protection of Human Subjects as not requiring informed consent.
Diverse search strategies were used to limit the bias introduced by narrow searches (Conn, Isamaralai et al., 2003; Nony, Cucherat, Haugh, & Boissell, 1995). A reference librarian performed computerized searches in 11 databases (MEDLINE, Cochrane Central Register of Controlled Trials, Dissertation Abstracts International, PsychInfo, SportDiscus, HealthStar, Clinical Evidence, Scopus, DARE, ABI/Inform, Cumulative Index to Nursing & Allied Health Literature). Broad search terms were used for intervention (adherence, behavior therapy, clinical trial, compliance, counseling, evaluation, evaluation study, evidence-based medicine, health care evaluation, health behavior, health education, health promotion, intervention, outcome & process assessment, patient education, program, program development, program evaluation, self care, treatment outcome, validation study) and PA (exercise, physical activity, physical fitness, exertion, exercise therapy, physical education & training, walking) (Conn, Isamaralai et al., 2003). The National Institutes of Health database of funded studies was searched. Ancestry searching was conducted on all eligible studies and review articles. Computerized searches were completed on all authors of eligible studies. Conference abstracts were searched. Hand searches of 42 journals with a preponderance of potential primary studies were completed for chronic illness-specific journals (e.g. Diabetologia) and general journals that publish PA research (e.g. Medicine & Science in Sports & Exercise).
Several strategies were used to search for unpublished studies. Dissertation Abstracts International was thoroughly searched. Conference abstracts were evaluated for eligible studies. Author searches were conducted on both Dissertation Abstracts International and conference abstracts to locate published studies. If published reports were not available, abstract authors were contacted. All first authors of studies included in the meta-analysis were contacted to solicit additional published or unpublished studies. These comprehensive search strategies yielded ten unpublished research reports included in this meta-analysis.
Studies were included that attempted to increase PA behavior by using supervised center-based exercise interventions that measured post-intervention PA, educational-motivational interventions designed to increase PA, or both. Studies with diverse QOL measures were included if adequate data were available to calculate effect sizes (e.g. means and variability measures, exact p value from t test, t statistic). Primary study inclusion criteria are reported in Table 1.
A coding frame to assess outcomes of primary studies and characteristics of sources, subjects, methods, and interventions was developed, pilot tested, and revised. Coding elements were derived from attributes coded in previous behavior-change meta-analyses, intervention attributes reported in research literature, suggestions from experts in meta-analysis and PA, and findings from the research team’s previous studies. Dissemination vehicle, year of distribution, and presence of funding were coded as source attributes. Participant characteristics included age, gender and minority distribution, and chronic illness inclusion criteria (e.g. diabetes). Attrition, random assignment, and the length of the interval between the intervention and outcome measurement were coded as methodological characteristics. Intervention information including presence or absence of supervised center-based exercise and of educational-motivational sessions, details about any center-based supervised exercise, behavioral target (PA exclusively vs. PA plus other behaviors), intervention intensiveness, social setting, and recommended PA were coded. Other details of interventions were coded but inadequately reported for moderator analyses. PA behavior outcomes were extracted from studies to enable us to examine the association between PA behavior effects and QOL outcomes. For QOL, measures that primary authors described as addressing life satisfaction, well-being, or QOL were coded. To select among multiple measures in some studies, a priori lists of QOL and PA behavior measures with preference given to widely-used and validated instruments determine which outcomes were coded.
Two extensively trained coders extracted the data. Discrepancies in coding were resolved by the senior author or other member of the research team as appropriate. Data coding was not masked because evidence indicates it does not decrease bias. Staff cross checked all author lists to locate research reports that might contain overlapping samples to ensure that only independent samples were analyzed. Authors were contacted to clarify the uniqueness of samples when reports were unclear.
Table 2 lists important features of the data analyses. A standardized mean difference effect size (ES) for QOL and PA outcomes was calculated. For comparisons between treatment and control groups after the intervention, the ES is the mean of the treatment group minus the mean of the control group, divided by their pooled standard deviation. In two-group comparisons, positive ESs reflect better scores among treatment subjects than control subjects. In single-group studies, positive ESs indicate that subjects scored better after than before the intervention. ESs were weighted such that studies with larger samples had more influence. Homogeneity between studies was assessed with Q. Outliers were examined graphically and statistically. Random-effects analyses were used to estimate the mean and variability of true ESs across studies. (Fixed-effects analyses were also conducted; the report focuses on random-effects results.) The random-effects model is appropriate because heterogeneous studies with varied methods that tested diverse interventions were expected. A Common Language Effect Size (CLES) was calculated to aid interpretation of findings. The ESs could not be converted to an original metric because of the wide variation in QOL measures used and the variation in scoring among studies using the same measure. Moderator analyses were conducted using meta-analytic analogues of ANOVA and regression to determine whether QOL ESs were related to source attributes, methods, intervention characteristics, or PA behavior ESs. Moderator analyses should be considered exploratory given the lack of previous research to suggest hypothesis testing and given the number of studies retrieved. Further information about the analyses is available from the senior author (VC).
Ultimately 85 samples described in 66 studies in which (approximately) 7,291 subjects participated (a list of included studies is available from the senior author) were included. Ten unpublished research reports were included. The independent group analysis included 5,159 subjects. The pre-post comparison analyses included 4,486 treatment subjects and 3,780 control subjects. The most common chronic illness target populations were cardiac (k = 24), cancer (k = 21), diabetes (k = 19), and arthritis (k = 7). (k represents the number of comparisons.) Table 3 provides descriptive information about the included studies. Sample size ranged from 8 to 927 subjects with small and moderate samples being common (median = 58). Attrition was modest from both treatment and control groups among the studies that reported this information (median = .10). Women were well represented in the samples (median = .56). The median of mean age was 61 years with the youngest sample having a mean age of 40 years and the oldest sample’s mean age was 82 years.
Eight studies used supervised exercise. Among the studies that used center-based supervised exercise, typical exercise included 36 sessions of nearly 1-hour duration. Fourteen studies included educational or motivational content designed to increase subjects’ PA. Among studies reporting the information, most comparisons (k = 25) use interventions designed to both increase PA and improve other health behaviors while 17 targeted only PA behavior. Most of the studies reporting the social context of the intervention delivered the interventions to groups (k = 28) while others were delivered to individuals (k = 14). Few interventions (k = 8) recommended a specific form of exercise, more did not make such recommendations (k = 34). Interventions duration varied from 1 week to 52 weeks.
Table 4 contains results from analyses that address the first, second, and fifth research questions. The paper focuses on random-effects results. The overall mean effect in two-group studies was .11 (δ in Table 4). The treatment group’s mean pre- versus post-test ES was .27 for both assumptions regarding the pre-post association (see Table 2 for explanation regarding correlation assumptions). Each type of comparison demonstrated significant ES heterogeneity according to the Q homogeneity test, but for two-group studies this was only barely significant and relatively small as quantified by the between-studies variance component’s square root (i.e., the true ESs’ SD), δ = .071. These findings document that, although interventions’ effects varied somewhat among studies, on average interventions to increase PA improved QOL outcomes, p < 0.001 in every case. In contrast, control subjects experienced little improvement (δ = .05 – .06), with only the comparison under the high-association assumption being statistically significant at .05 < p < .10.
The CLES for the two-group comparisons was .53, indicating that 53% of the time a random treatment subject would have a better QOL value than a random control subject after the intervention (see Table 2 for explanation regarding CLES). The CLES for treatment group pre- and post-intervention comparisons was .58 (assuming no pre-post correlation), indicating that 58% of the time a treatment subject’s QOL score would be better at outcome assessment than at baseline measurement. The CLES for control groups was .51, indicating that control subjects are slightly more likely to have a better outcome QOL score than baseline QOL value. Figure 1 presents a stem-and-leaf display of the 2-group ESs as a visual depiction of findings. The highly diverse measurement of QOL prevented us from converting ES estimates to an original metric. The funnel plots were symmetrical, indicating no obvious evidence of publication bias.
Moderator analyses that address the third and fourth research questions are presented in Tables 5 and and6.6. Dichotomous moderator analyses for two-group comparisons are presented in Table 5. Unpublished studies reported considerably larger mean ESs (δ = .39) than published studies (δ = .09). Studies without external funding reported larger mean ESs (δ = .31) than studies with external funding (δ = .08). No statistically significant differences were observed between studies that randomly assigned subjects versus those that did not; studies that focused only on PA behavior versus those targeting multiple health behaviors; interventions delivered to individuals versus those given to groups; or studies with specific exercise recommendations (including intensity) versus those without such recommendations. The mean ES difference between studies with center-based supervised exercise (δ = .16) and studies without supervised exercise (δ = .06) did not reach statistical significance. Studies that did not use educational-motivational sessions (they used only supervised center-based exercise) reported significantly larger mean ESs (δ = .24) than studies using only educational-motivational sessions or combined educational-motivational sessions with center-based supervised exercise (δ = .02).
Two-group continuous moderator analyses findings are shown in Table 6. Mean QOL ESs were not predicted by participants’ mean age (1 = 0.004); proportion female (1 = −0.021), minority (1 = −1.165), or attrition (1 = −0.360); extent of supervised exercise (1 = 0.559) or educational-motivational contact between interventionists and subjects (1 = 0.056); or PA recommendations (1 = −0.001). In addition, the PA outcome ES was unrelated to QOL outcome ES (PA two-group ES: 1 =0.039; PA pre-post treatment group ES: 1 = −.070; PA pre-post control group ES: 1 = 0.109).
A moderator analysis was conducted to determine if QOL ES differed among the three most common types of chronic illnesses with adequate data (diabetes, cardiac disease, and cancer). The mean QOL ES was .21 for cardiac disease, .15 for cancer, and .04 for diabetes. The results revealed that these differences were not statistically significant (Qbetween = 3.0).
This meta-analysis documented that subjects who receive interventions designed to increase their PA experience better QOL outcomes over their baseline scores and in comparison to control subjects. The magnitude of the ES is difficult to assess because too few studies used any single QOL measure in exactly the same way to allow us to convert the ES to an original metric. The ES magnitude, as calculated and as depicted by CLES scores, seems modest. It is unclear what ES would represent a clinically meaningful improvement in QOL among chronically ill adults. People with major chronic illnesses experience many reasons for declining QOL, including the disease itself and physical or psychosocial sequelae of the disease, and onerous or distressing treatments. Studies may recruit chronically ill study subjects from specialty medical practice settings where the subjects may already be receiving optimal medical care. Even a small change in QOL may be important since QOL is a complex phenomenon likely affected by diverse factors.
It is important to note that these findings were heterogeneous, as expected, though less so for two-group than pre-post comparisons and less so for QOL outcomes than for other health and PA outcomes reported in these primary studies (Conn, Hafdahl, Brown et al., 2008; Conn, Hafdahl, Mehr et al., 2007; Conn, Hafdhal, Minor et al., 2008; Conn, Hafdahl, Moore et al. 2008). Interventions varied dramatically from brief motivational sessions to extended supervised exercise programs. Diverse measures were used to assess QOL and PA. No gold standards exist for interventions or measures of QOL and PA. Other important factors that may affect validity of findings which are infrequently reported in primary studies could not be assessed (e.g. treatment fidelity). As more primary research accumulates, future meta-analyses may be able to determine if ES are related to research methods.
The explanation for QOL changes is unclear. These interventions were designed to change PA behavior, not to directly affect QOL. Mean differences in PA behavior were not associated with QOL mean differences in the moderator analyses. It is possible that people achieved small increases in PA that were not detected by the PA measures but that contributed to increased QOL; however, this is inconsistent with the magnitude of ESs on PA in a related study (Conn, Hafdahl, Brown et al., 2008). Even small improvements in functional status from slight increases in PA may contribute to improved QOL. Previous meta-analyses of disease-specific outcomes (e.g. HbA1c, arthritis functional status) among common chronic illnesses documented improved health outcomes (Conn, Hafdahl, Mehr et al., 2007; Conn, Hafdahl, Minor et al., 2008; Conn, Hafdahl, Moore et al., 2008; Nielsen et al., 2006). These improvements may explain the improvements in QOL. It is also possible that subjects experienced enhanced perceived mastery over their chronic illnesses. Future primary PA research should include QOL measures and report the association between improvements in QOL and PA behavior changes to address this issue and avoid possible ecological fallacy in interpreting meta-analytic findings (Berlin, Santanna, Schmid, Szczech & Feldman, 2002). Research syntheses focused on correlates of and explanatory models for QOL could address the association between PA and QOL more directly.
The exploratory moderator analyses documented some intriguing suggestive findings that future research should examine. The larger ES among unpublished and unfunded studies was somewhat surprising. These findings do not support the pattern of publication bias against studies with small ESs often reported in the literature (Cook et al., 1993; Conn, Valentine, Cooper, & Rantz, 2003). These unpublished and unfunded studies may include projects with extraordinary researcher effort to ensure successful projects, such as graduate student research. It is also possible investigators were more likely to provide study information about unpublished studies if the study reported large ESs. The finding of no association between ES and random assignment of subjects does not support the common assumption of bias toward positive effects in studies without random assignment.
Our finding that behavior target (PA behavior only vs. multiple health behavior) was not associated with ES differences contrasts with previous meta-analyses of PA behavior and health outcomes that have documented better outcomes among studies that targeted only PA behavior (Conn, Valentine, & Cooper, 2002; Conn, Hafdahl, Mehr et al., 2007; Conn, Hafdahl, Brown et al., 2008). The meta-analyses that reported larger effects for interventions that focused exclusively on PA behavior have examined outcomes directly affected by PA behavior. The explanation for these differences may become clearer as more intervention trials include QOL outcomes as well as PA behavior and health outcomes.
The absence of moderator effects for age, gender, and minority distribution suggests that diverse samples may experience modest improvement from interventions to increase PA. People with chronic illnesses may avoid changing PA behavior because they fear further decline in their QOL. Many are dealing with demanding chronic illnesses that require continual self-management. Health care providers and health educators may use these exploratory findings to counter fears that increased PA will necessarily decrease QOL.
These findings suggest that interventions using only supervised center-based exercise may have more impact on QOL than interventions including educational-motivational content, regardless of whether it is accompanied by supervised exercise. These findings contrast with previous work documenting a lack of superiority of supervised exercise for PA behavior and health outcomes (Conn, Hafdahl, Mehr et al., 2007). Although the exploratory moderator analyses are intriguing, they should be interpreted with caution. Relationships documented in the moderator analyses may be confounded by other sample- or study-level characteristics. Further primary studies testing differences within randomized controlled trials are needed.
QOL outcomes as measured by well-being, life satisfaction, and QOL measures were included. Studies that used mood, energy, or fatigue measures as QOL outcomes were excluded. Although findings have been mixed, some research has suggested a link between PA and mood (Conn, Hafdahl, Porock et al., 2006; Rietberg et al., 2005) and between PA and energy/fatigue (Puetz, Beasman, & O’Connor, 2006). Few studies in this meta-analysis addressed mood outcomes; after more primary studies reporting mood outcomes have been conducted, a synthesis of mood outcomes would be valuable.
In conclusion, this meta-analysis documented modest improvements in QOL outcomes among adults with chronic illnesses following interventions to increase PA. These findings should encourage researchers and providers evaluating interventions designed to increase PA to include QOL outcome measures in their projects.
Financial support provided by a grant from the National Institutes of Health (R01NR07870) to Vicki Conn, principal investigator.
Vicki S. Conn, School of Nursing, University of Missouri, Columbia MO.
Adam R. Hafdahl, Washington University, St. Louis MO.
Lori M. Brown, School of Nursing, University of Missouri, Columbia, MO.