Effectiveness of Physical Activity Interventions among Rural Cancer Survivors: A Systematic Review and Meta-Analysis

Over 19 million new cancer cases were reported worldwide in 2020 (1, 2), and this number is expected to surpass 30 million by 2040 (3, 4). The rapid increase in cancer incidence can be partially attributed to population growth and aging (5, 6). Advances in early detection and treatment have contributed to the growing number of cancer survivors and an improved long-term survival rate, particularly in high-middle income countries (6, 7). Thus, cancer control efforts are needed to reduce the risk of cancer recurrence and the development of comorbidities and to improve long-term health outcomes and quality of life in cancer survivors (8).

In addition to reducing cancer risk, physical activity (PA) reduces the risk of cancer recurrence and comorbidities and improves physical health and psychological wellbeing after a cancer diagnosis (9–12). The numerous health benefits of PA include reduced risk of all-cause and cardiovascular disease mortality, hypertension, type 2 diabetes, and certain cancers (13). Despite the well-documented benefits of engaging in regular PA, less than 20% of adults meet PA recommendations worldwide (13, 14), and fewer adults with a history of cancer meet exercise guidelines for cancer survivors (11, 15). Persistent adverse effects of cancer treatment, such as fatigue, psychosocial distress, insomnia, chemotherapy-induced peripheral neuropathy, and pain, are commonly cited exercise barriers reported by cancer survivors (12, 16). However, there is strong to moderate evidence that PA can help manage these treatment-related adverse effects and improve quality of life among cancer survivors (11, 12), warranting innovative intervention and implementation strategies to help cancer survivors move more and sit less.

Rural cancer survivors, or cancer survivors residing in nonmetropolitan or remote areas, are more likely to be physically inactive than cancer survivors residing in urban or metropolitan areas and face social and environmental barriers to exercise in addition to those related to the adverse effects of cancer treatment (17–19). Geographic isolation, inadequate transportation, and low access to health care and supportive oncology services and resources contribute to rural–urban differences in PA and physical and mental health outcomes among cancer survivors (20–23). Previous reviews have highlighted the lack of availability and accessibility of exercise programs for rural adults (24–28) and rural cancer survivors (29). Among rural adults with no history of cancer, mixed findings among intervention studies suggest that evidence-based exercise programs have not yet been effectively translated and implemented within rural communities (24). However, no study, to our knowledge, has explored the effectiveness of PA interventions among rural cancer survivors.

This systematic review and meta-analysis goes beyond previous reviews by examining the effectiveness of PA interventions among rural cancer survivors. The purposes of this study were to summarize the characteristics and results of PA interventions among adult rural cancer survivors and to estimate the effectiveness of interventions for cancer survivors living in rural or remote settings. In addition, we summarized measures of rurality across studies, as there is no single global classification system.

This systematic review was registered with PROSPERO prospective register of systematic reviews (registration number CRD42021229290) at the Center for Reviews and Dissemination, University of York, UK (https://www.crd.york.ac.uk/prospero/), and adheres to the Preferred Reporting of Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guidelines (http://www.prisma-statement.org/; refs. 30, 31). The PRISMA checklist is available as Supplementary Table S1.

We included controlled or uncontrolled trials [e.g., randomized controlled trial (RCT), non-randomized trial, quasi-experimental, or pre–post] evaluating a PA, exercise, or fitness intervention (any type or setting) in adults (ages ≥18 years) diagnosed with any type of cancer and residing in a rural area. Residence in a rural area was considered as defined or described by the study authors (e.g., self-identified or proclaimed or designated as rural using a defined classification system, such as Rural–Urban Continuum Codes or Rural–Urban Commuting Area Codes). We considered any comparator (e.g., standard or usual care, active control, inactive control) or participants as their own control (e.g., pre–post), and trials of any sample size were included. Studies were excluded if the full-text article was not available in English or the intervention did not assess and report PA at pre- and post-intervention.

Three databases, PubMed Medline (January 1996–July 25, 2020), CINAHL (1961–July 25, 2020), and PsychINFO (1887–July 25, 2020), were systematically searched. The search was restricted to original articles published in English from each database's inception through July 25, 2020. The reference lists of a recent scoping review (29) and all included articles were further hand searched to identify additional studies and companion articles, or articles relevant to the primary study that may include additional intervention details (e.g., articles describing study protocols published separately from study outcomes).

The search strategy was developed in consultation with a health sciences librarian, and detailed search strings for each database are presented in detail in Supplementary Table S2. Briefly, we used a combination of keywords for cancer survivorship, rural health, and PA or exercise to identify relevant publications. Citation details (e.g., authors, title, journal name, year of publication, volume, issue number, and page numbers) were downloaded and compiled into a single database. Duplicate articles were identified, reviewed, and removed from the database.

Two coders (S.K. Mama and H.J. Leach) independently screened titles and abstracts. Agreement between coders for title and abstract reviews were 92.9% and 94.9%, respectively, and inter-rater reliabilities, calculated using Cohen's κ, were 0.67 for title and 0.87 for abstract review. The full texts of remaining articles were independently reviewed against inclusion and exclusion criteria by two coders (S.K. Mama and H.J. Leach), and agreement between coders was 93.3% (Cohen's κ = 0.87). Disagreements between reviewers were resolved by consensus, and reasons for exclusion were documented and are shown in Fig. 1.

A coding tool adapted from previous systematic reviews conducted by the study team was used by two coders (S.K. Mama and H.J. Leach) to independently extract and code data from included studies (available from the author upon request; refs. 24, 32). Disagreements between coders were discussed until consensus was reached.

Extracted data included citation details and companion article citation details, measures of rurality (e.g., rural setting and classification) and study characteristics (e.g., primary outcome, target population, theory, study design, delivery mode), participant characteristics (e.g., sample size, demographics), intervention characteristics (e.g., study setting, delivery personnel, number of contacts, contact duration, and overall time), measurement characteristics (e.g., assessment time points, PA measure and type), and outcomes (e.g., means, standard deviations, adherence percent, attrition, summary of findings). In studies that had multiple follow-ups or post-intervention assessments, data were extracted from the follow-up or assessment time point closest to the cessation of the intervention.

Risk of bias was assessed using the Effective Public Health Practice Project's Quality Assessment Tool for Quantitative Studies (33), which assesses study features to generate risk of bias ratings in six domains: Selection bias, study design, confounders, blinding, data collection methods, and withdrawals and dropouts. Judgments for each risk of bias domain and the overall study quality are expressed as “weak,” “moderate,” or “strong.”

We analyzed data as reported in the studies. Measures of rurality across studies and intervention characteristics data were narratively synthesized. For quantitative data, we determined proportion of participants achieving intervention-specific PA goals using the number of participants who achieved the goal at post-intervention as numerators. For the denominator, we considered all participants included in the study. When studies compared intervention versus control groups at different time points, the proportion of participants achieving PA goals was calculated only for the intervention arm at the end of the intervention (as opposed to the last follow-up assessment). When data were unclear or not provided for a given outcome, the study was not included in the analysis. Also, when outcomes were reported using both self-report and device-based measures within an individual, we used self-reported data to increase consistency in measures across studies. We calculated the relative risk (RR) to compare dichotomous outcomes, the mean difference (MD) for continuous outcomes, and the 95% confidence intervals (CI).

Given the methodological variation among included studies, we used a random-effects model to calculate a pooled proportion of participants achieving PA goals and its 95% CI. We used the Freeman–Tukey arcsine transformation to stabilize variances and conducted a meta-analysis using inverse variance weights. Resulting estimates and CI boundaries were back-transformed into proportions. We performed a Mantel–Haenszel meta-analysis with a random-effects model of studies providing data for both groups. Separate analyses were performed for before and after studies using the Cochrane methodology to pool paired MDs. When studies did not report mean, we used the median value. When a study did not report standard deviation (SD) of within-participant differences between before and after measurements, SD_diff was estimated using SDs at baseline and post-intervention in addition to the within-groups correlation coefficient. An imputed conservative correlation of 0.8 was used when the within-groups correlation coefficient was not reported (34). Statistical heterogeneity between studies was assessed using Cochran's Q and I², with an I² value greater than 50% representing substantial heterogeneity between studies (35). We interpreted the results in terms of magnitudes of associations and precision of the risk estimates conveyed by 95% CIs, rather than using P values as measures of significance. All analyses were performed using STATA 15 (StataCorp LP).

A funnel plot and an Egger regression asymmetry test was planned if more than 6 studies reported data on the same outcome to assess publication bias and small-study effects in the meta-analysis. However, none of the reported outcomes met the criteria.

The literature search identified 366 studies, and 5 additional studies were identified after handsearching the reference list of a recent scoping review (29). Of the 281 titles and abstracts screened, 15 full-text articles were reviewed for eligibility. Seven studies met inclusion criteria and were included in this review (Fig. 1).

Study and participant characteristics are summarized in Table 1. The median sample size was 91 (range, 23 to 160). Four studies were based in rural areas in the United States (36–39), and three studies were based in rural areas in Australia (40–42). Five studies conducted RCTs (37–41), and the remaining two studies used a pre–post design (36, 42). Four studies included female breast cancer survivors exclusively (36, 37, 39, 40), one study included breast, prostate, and colorectal cancer survivors (38), and two studies were open to all cancer types (41, 42). In three studies, some or all of the samples were currently undergoing chemotherapy or radiotherapy (39, 40, 42), and four studies included participants who had completed treatment (36–38, 41).

Measures of rurality or classification schemes used to identify rural areas varied across studies and by country. In the U.S., three out of four studies used Rural–Urban Commuting Area (RUCA) codes (43) to classify U.S. census tracts nested within counties into rural and urban categories using measures of population density, urbanization, and daily commuting (36–38), and one study did not specify the definition or classification system used (39). In Australia, one study used a standard classification system (Australian Standard Geographical Classification), similar to RUCA codes, to classify residents within a postal code into rural and urban categories (40), one used population density (42), and one did not specify the definition or classification system used (41).

Social cognitive theory was the most commonly cited theoretical framework used to guide intervention development. Additional theories cited included self-regulation theory, goal setting theory, and the Chronic Disease Self-Management Model. Most (57.1%) interventions were delivered individually, 28.6% were group-based, and 28.6% used a combination of individual and group-based delivery. Three studies reported a single intervention delivery method (36, 39, 42). Intervention delivery by telephone (71.4%) was the most commonly used method followed by print or mail (42.9%). Most (83.3%) studies used a trained research assistant or health professional (e.g., counselor) to deliver the intervention, and one study did not use a delivery agent (e.g., completely phone/device or mail based).

The median intervention duration was 24 weeks (range, 6 to 52 weeks), and median total contact time was 467.5 minutes (range, 246 to 1,586 minutes). Three studies focused on aerobic PA exclusively (37, 41, 42), three incorporated aerobic and muscle-strengthening activities (36, 38, 40), and one did not specify type of PA (39). Intervention adherence, or the percentage of intervention sessions attended, ranged from 60% to 85%, and attrition ranged from 4% to 24%. Four studies assessed PA maintenance (37, 39, 41, 42), and the median follow-up time post-intervention was 12 weeks (range, 6 to 26 weeks).

Most (71.4%) studies assessed PA outcomes using questionnaires (36, 38–40, 42), one study used pedometers (41), and one study used both questionnaires and accelerometers (37). Five studies reported overall improvement in PA (36, 37, 40–42). Of the five RCTs included in this review, one study reported improvements in PA compared with the control group (41), and one study reported improvements in strength-training activity compared with the control (40).

Of the seven studies included in the review, four studies were considered to have moderate risk of bias (36, 37, 40, 41), and three studies had a high overall risk of bias (38, 39, 42). A summary of judgments for each domain is shown in Fig. 2, and judgments for each domain for each included study are available in Supplementary Table S3. The four studies with moderate risk of bias were subsequently used for the meta-analysis.

A meta-analysis of the four RCTs, consisting of 401 participants, showed that 39% of individuals receiving an intervention achieved the study's PA goal (95% CI, 18%–62%; I² = 95%; Fig. 3). We observed that the proportion of patients achieving the goal increased to >50% when the goal was more stringent (≥180 to 225 minutes per week compared with ≥150 minutes of PA). Only one study compared the proportion of patients achieving a PA goal at post-intervention (26 weeks) and follow-up (52 weeks; ref. 40). No statistically significant differences were observed in the number of patients achieving the study goal between groups (at 26 weeks: RR, 1.4; 95% CI, 0.87–2.2; at 52 weeks: RR, 1.3; 95% CI, 0.88–1.9).

The mean time spent per week exercising (aerobic PA) increased from baseline to post-intervention, which ranged in duration from 6 to 52 weeks (MD = 97.7 minutes; 95% CI, 75.0–120.4; I² > 99%). When subgrouped by study design and intervention, results remained similar. A statistically significant increase in the time spent (in minutes) per week on aerobic PA was observed at post-intervention compared with baseline (MD = 215.5 minutes; 95% CI, 185.9–245.1; I² < 58%; Table 2). The MD was lower for the study with long-term data at 52 weeks (MD = 27.1; 95% CI, 26.4–27.9; ref. 38). Only one study compared PA minutes per week against controls with no statistically significant difference found between the time reported per group (MD = 0.65; 95% CI, −0.27–1.6). Another study compared the number of steps per day between groups at post-intervention (12 weeks) and found a statistically significant increase in the number of steps per day in participants assigned to the intervention group compared with those in the control group (MD = 1775.0; 95% CI, 357.0–3,193.0; ref. 41).

Only two studies provided data on this outcome (38, 39). Participants in one study reported an increase in time spent on strength training at post-intervention, which was 52 weeks (MD = 22.7 minutes; 95% CI, 22.3–23.1). In a controlled study, participants in the intervention group reported longer time spent in self-directed physical therapy than the control group at 6 weeks (MD = 1.7; 95% CI, 0.48–3.0). However, the overall pooled estimate did not reach statistical significance (I² = 99%; Table 2).

Five studies provided data on this outcome, as shown in Supplementary Fig. S4. The pooled proportion of participants in the intervention group who remained in the program until the last follow-up assessment (range 6 to 78 weeks) was 82% (95% CI, 69%–92%; I² = 85%).

This systematic review identified only seven intervention studies to promote PA among rural cancer survivors worldwide and is the first, to our knowledge, to estimate the effectiveness of interventions to increase PA among cancer survivors living in rural areas. Findings from this review suggest modest increases in PA, and a lack of available controlled interventions among rural cancer survivors. Furthermore, we found that discrepancies remain in the operational definition or classification scheme used to categorize areas as rural or urban in both the U.S. and Australia, expanding on findings from a previous review (24).

This review conducted a quantitative synthesis of the effects of interventions for increasing PA among cancer survivors living in rural areas, which builds on a previously published scoping review by Smith-Turchyn and colleagues (29). Smith-Turchyn and colleagues identified 13 studies representing eight unique exercise interventions (29), four of which met our inclusion criteria and were included in the current review. The current review identified seven unique PA interventions and found that interventions moderately increased PA, equivalent to approximately 97.7 minutes per week of aerobic exercise and 12.2 minutes per week of strength training. However, the clinical meaningfulness of these findings remains uncertain due to the low number of eligible RCTs, small sample sizes and small to moderate intervention effects on PA within studies, and moderate to high risk of bias and low overall quality of evidence across studies.

Most interventions included in this review were delivered by telephone, print or mail, did not include a face-to-face component, and were delivered individually versus group-based, and all the RCTs included in the meta-analysis used distance-based delivery. Previous reviews have found negligible to small effects for distance-based PA behavior change interventions that have relied on print and telephone modes of intervention delivery among cancer survivors (44). Furthermore, group-based strategies are effective for increasing PA behavior for most populations and in most settings, and may be more efficacious than individually delivered programs when the appropriate group dynamic principles are used (32, 45, 46). Despite being the most effective approach for PA behavior change compared with mediated delivery (e.g., email or telephone; ref, 47), only one study used face-to-face and group-based delivery, but was not an RCT and thus excluded from the meta-analysis (42). Although this study (Frensham and colleagues; ref. 41) used primarily web-based intervention delivery, they included two face-to-face workshops where participants were instructed on using the study website, logged their steps, received feedback, shared their experiences and received peer support, and had access to other health information (e.g., healthy eating) and resources (e.g., community centers, events, etc.). This relatively small face-to-face component may help explain the success of the intervention in increasing PA (MD = 1,775.0 steps per day), which was the largest reported among the studies included in this review (41). Given the rapid rise in the use of digital health platforms and telehealth due to COVID-19 (48, 49), additional research is needed to assess the potential for current technologies (e.g., Zoom, Skype, Microsoft Teams, FaceTime, etc.) to overcome the limitations of previous distance-based approaches while to providing the benefits and support associated with face-to-face delivery to increase PA in rural cancer survivors (50, 51).

These findings highlight the important tradeoff between reach and effectiveness when it comes to PA promotion efforts in rural settings. Distance-based, unsupervised approaches to promote PA may reduce cost and barriers to engaging in PA programs, thereby enhancing the reach of programs to underserved groups, including cancer survivors residing in rural and remote areas. Conversely, face-to-face approaches may be more efficacious for increasing PA (52). This suggests a need for implementation strategies to test whether established evidence-based approaches remain effective in rural and remote areas, particularly once adaptations to delivery mode to increase reach are implemented. Community engaged research approaches, which include rural community stakeholders in the design, adaptation, and implementation process, may further increase the saliency and sustainability and improve effectiveness of PA interventions in rural communities and for rural cancer survivors.

The major strengths of this study are the systematic and comprehensive search strategies, and rigorous quantitative synthesis of outcomes. Limitations include a limited number of studies included in the meta-analysis, moderate to high risk of bias among included studies, and lack of studies using objective measures of PA. Objective measures of PA such as accelerometers can capture intensity and amounts of PA, with greater precision for light intensity activity, and activity outside of structured leisure time PA (53). These facets of PA may be of particular relevance to rural populations who may spend more time in occupational or household activities (54, 55). The limited number of studies included in this review and meta-analysis restricted subgroup analyses by cancer type, stage of disease, age, social determinants of health, and other characteristics that are known to impact health behaviors, such as PA (15). Additional research is needed to explore barriers to PA that are relevant to rural cancer survivors and to incorporate those in future reviews. Finally, the current review was limited to articles published in English. Given that we aimed to assess interventions conducted worldwide, this criterion may have excluded relevant articles published in other languages.

This systematic review and meta-analysis identified only 7 interventions worldwide that described or evaluated a PA program for rural cancer survivors. Although the interventions demonstrated moderate increases in aerobic and resistance exercise among rural cancer survivors, the clinical meaningfulness of findings from this review remains uncertain due to the continued lack of availability of rigorous PA interventions designed for or adapted to rural cancer survivors. As the number of cancer survivors continues to rise in rural communities, there is a growing need to adapt and test evidence-based interventions to address the unique PA needs of rural cancer survivors. Future research is needed to explore the use of new technologies (e.g., telehealth and videoconferencing) and approaches (e.g., using lay health educators and local fitness professionals) to improve reach while maintaining effectiveness.

S.K. Mama reports grants from National Cancer Institute during the conduct of the study. M.A. Lopez-Olivo reports grants from National Cancer Institute and grants from Rheumatology Research Foundation outside the submitted work. No disclosures were reported by the other authors.

S.K. Mama (K07 CA222335) and M.A. Lopez-Olivo (K08 CA237619) are partially supported by career development awards from the National Cancer Institute, and H.J. Leach is partially supported by a career development award (131629-MRSG-18-021-01-CPPB) from the American Cancer Society.