Walking Improves Sleep in Individuals With Cancer
The authors' meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Moher, Liberati, Tetzlaff, & Altman, 2009). Relevant studies were identified through searches of the following databases: China Knowledge Resource Integrated Database, CINAHL®, Cochrane Central Register of Controlled Trials, EMBASE, PsycINFO®, PubMed, Wanfang Data, and Web of Science. The search terms used were "sleep OR sleep disturbance OR sleep quality OR insomnia," "cancer OR tumour OR tumor OR neoplasm OR chemotherapy OR radiotherapy," and "home-based walking exercise OR walking exercise." The date range was from the earliest publication date available in each database to May 2014. To confirm whether any relevant studies were published since the author's initial search, the search was updated on July 15, 2014.
Studies involving individuals who had been diagnosed with any type of cancer and were aged 18 years or older were eligible for inclusion in the current study. In addition, studies in which walking had been used as the intervention were included, as were studies that included an alternative treatment group or an inactive control group (e.g., wait list, no treatment, usual care or exercise style).
Studies that assessed a self-reported sleep outcome using validated scales (e.g., Pittsburgh Sleep Quality Index [PSQI], symptom Numeric Rating Scale, Symptom Assessment Scale, European Organisation for the Research and Treatment of Cancer Quality-of-Life Questionnaire–Core 30 [EORTC QLQ-C30]) were included. The PSQI is a 19-item scale that evaluates sleep quality during a one-month period. It has seven components that can be summed to obtain a global sleep quality score ranging from 0–21. A global PSQI score of greater than 5 is indicative of poor sleep quality. The PSQI exhibits good reliability and validity; the Cronbach alpha is 0.83, and concurrent validity is r = 0.33 (Buysse, Reynolds, Monk, Berman, & Kupfer, 1989). The 11-point symptom Numeric Rating Scale is a valid sleep measure, with a concurrent validity of r = 0.85 (Paice & Cohen, 1997). Higher scores indicate better sleep quality. The Symptom Assessment Scale measures sleep using a series of straight 100 mm lines, with higher scores reflecting worse sleep quality (Wewers & Lowe, 1990). The scale is considered to be reliable and valid; the test-retest reliability has been found to be 0.95–0.99, and the criterion-related validity is 0.42–0.91. The EORTC QLQ-C30 comprises five function and nine symptom subscales (one item assesses sleep) measuring sleep quality, with higher scores representing poorer sleep quality. The reliability and validity of the questionnaire have been established; the Cronbach alpha is 0.43–0.68, and the concurrent validity is 0.78–0.81 (Fredheim, Borchgrevink, Saltnes, & Kaasa, 2007; Groenvold, Klee, Sprangers, & Aaronson, 1997). Studies using a prospective RCT design that were published or accepted for publication in English or Chinese by a peer-reviewed journal were included.
Two investigators independently screened the titles and abstracts of articles identified using the search strategy previously described. After removing duplicate publications using Thomson Reuters EndNote X7, the remaining articles were reviewed in full. Only studies fulfilling the selection criteria were included in the current meta-analysis.
Two investigators developed a data extraction sheet and independently extracted the data from each study, including (a) characteristics of the selected studies (e.g., authors' names and year of publication), (b) characteristics of the patient populations (e.g., type of cancer, patient age, number of patients in each group, percentage of women in the sample), (c) characteristics of the intervention (e.g., type, frequency, length, and intensity of exercise), and (d) outcome measures. Quantitative data were extracted to calculate the effect size. When assessment time points were greater than one, the immediate postintervention measure was selected. Discrepancies were rechecked by the corresponding author of the current article and consensus was achieved by discussion.
The following domains were assessed in relation to their risk of bias (Higgins & Green, 2011): (a) random sequence generation, (b) allocation concealment, (c) blinding of participants and staff, (d) blinding of outcome assessment, (e) incomplete outcome data, and (f) selective reporting. Each domain was rated as having "low," "unclear," or "high" risk of bias. Two reviewers independently performed the assessment of potential bias for each study, with a third reviewer serving as the arbitrator.
Quantitative data were entered into Biostat Comprehensive Meta-Analysis, version 2.0. Two-sided p values were calculated, with p < 0.05 set as the level of statistical significance. First, pre- to post-test change scores were derived for the intervention group and control group from each included study. Then, the effect size for the difference between the intervention and control groups was calculated for each study. Hedges' g was used as the measure of the effect size. It was calculated by finding the difference between the intervention and control group means (d), divided by their pooled standard deviation and multiplied by a factor (J) that corrects for underestimation of the population standard deviation. A forest plot was used to present the effect size of all of the included studies. An inverse variance random-effects model was applied to analyze the data because it is more conservative than a fixed-effects model (DerSimonian & Laird, 1986).
To establish whether the selected studies differed significantly, the authors of the current study first examined whether the interstudy heterogeneity was statistically significant by evaluating the Cochran Q statistic (Higgins, Thompson, Deeks, & Altman, 2003), with p < 0.05 indicating significant heterogeneity. The magnitude of heterogeneity was measured using the I2 statistic, with I2 of 50% or greater indicating substantial heterogeneity across studies. A sensitivity analysis was also performed by removing the study with the largest effect size to determine its contribution to the overall effect size in the current meta-analysis.
Subgroup analyses were conducted by dividing the studies into groups according to (a) type of intervention, (b) type of cancer, (c) whether sleep was the primary outcome, (d) stage of cancer treatment at enrollment, (e) whether random sequence generation was appropriately executed (risk of selection bias), and (f) whether allocation concealment was appropriately executed (risk of selection bias). Moderator analyses were performed to explore possible reasons for the observed heterogeneity. To ensure sufficient data for analyses, each moderator analysis was limited to instances in which groups were represented by at least three studies. For categorical moderators, a mixed-effect model was used to compare differences among the effect sizes in each comparison (Lipsey & Wilson, 2001). Metaregression was used for the analyses of continuous moderators (Lipsey & Wilson, 2001).
Begg's rank correlation (Begg & Mazumdar, 1994) and Egger's intercept (Egger, Davey Smith, Schneider, & Minder, 1997) assess potential publication bias, with p > 0.05 indicating significant publication bias. The trim-and-fill method (Duval & Tweedie, 2000) was applied using a funnel plot to further assess potential publication bias. The overall effect size was adjusted by taking into account the estimated effect sizes of missing studies.
Methods
Search Strategies
The authors' meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Moher, Liberati, Tetzlaff, & Altman, 2009). Relevant studies were identified through searches of the following databases: China Knowledge Resource Integrated Database, CINAHL®, Cochrane Central Register of Controlled Trials, EMBASE, PsycINFO®, PubMed, Wanfang Data, and Web of Science. The search terms used were "sleep OR sleep disturbance OR sleep quality OR insomnia," "cancer OR tumour OR tumor OR neoplasm OR chemotherapy OR radiotherapy," and "home-based walking exercise OR walking exercise." The date range was from the earliest publication date available in each database to May 2014. To confirm whether any relevant studies were published since the author's initial search, the search was updated on July 15, 2014.
Selection Criteria
Studies involving individuals who had been diagnosed with any type of cancer and were aged 18 years or older were eligible for inclusion in the current study. In addition, studies in which walking had been used as the intervention were included, as were studies that included an alternative treatment group or an inactive control group (e.g., wait list, no treatment, usual care or exercise style).
Studies that assessed a self-reported sleep outcome using validated scales (e.g., Pittsburgh Sleep Quality Index [PSQI], symptom Numeric Rating Scale, Symptom Assessment Scale, European Organisation for the Research and Treatment of Cancer Quality-of-Life Questionnaire–Core 30 [EORTC QLQ-C30]) were included. The PSQI is a 19-item scale that evaluates sleep quality during a one-month period. It has seven components that can be summed to obtain a global sleep quality score ranging from 0–21. A global PSQI score of greater than 5 is indicative of poor sleep quality. The PSQI exhibits good reliability and validity; the Cronbach alpha is 0.83, and concurrent validity is r = 0.33 (Buysse, Reynolds, Monk, Berman, & Kupfer, 1989). The 11-point symptom Numeric Rating Scale is a valid sleep measure, with a concurrent validity of r = 0.85 (Paice & Cohen, 1997). Higher scores indicate better sleep quality. The Symptom Assessment Scale measures sleep using a series of straight 100 mm lines, with higher scores reflecting worse sleep quality (Wewers & Lowe, 1990). The scale is considered to be reliable and valid; the test-retest reliability has been found to be 0.95–0.99, and the criterion-related validity is 0.42–0.91. The EORTC QLQ-C30 comprises five function and nine symptom subscales (one item assesses sleep) measuring sleep quality, with higher scores representing poorer sleep quality. The reliability and validity of the questionnaire have been established; the Cronbach alpha is 0.43–0.68, and the concurrent validity is 0.78–0.81 (Fredheim, Borchgrevink, Saltnes, & Kaasa, 2007; Groenvold, Klee, Sprangers, & Aaronson, 1997). Studies using a prospective RCT design that were published or accepted for publication in English or Chinese by a peer-reviewed journal were included.
Study Selection
Two investigators independently screened the titles and abstracts of articles identified using the search strategy previously described. After removing duplicate publications using Thomson Reuters EndNote X7, the remaining articles were reviewed in full. Only studies fulfilling the selection criteria were included in the current meta-analysis.
Data Extraction and Methodology Quality Assessment
Two investigators developed a data extraction sheet and independently extracted the data from each study, including (a) characteristics of the selected studies (e.g., authors' names and year of publication), (b) characteristics of the patient populations (e.g., type of cancer, patient age, number of patients in each group, percentage of women in the sample), (c) characteristics of the intervention (e.g., type, frequency, length, and intensity of exercise), and (d) outcome measures. Quantitative data were extracted to calculate the effect size. When assessment time points were greater than one, the immediate postintervention measure was selected. Discrepancies were rechecked by the corresponding author of the current article and consensus was achieved by discussion.
The following domains were assessed in relation to their risk of bias (Higgins & Green, 2011): (a) random sequence generation, (b) allocation concealment, (c) blinding of participants and staff, (d) blinding of outcome assessment, (e) incomplete outcome data, and (f) selective reporting. Each domain was rated as having "low," "unclear," or "high" risk of bias. Two reviewers independently performed the assessment of potential bias for each study, with a third reviewer serving as the arbitrator.
Data Analysis
Quantitative data were entered into Biostat Comprehensive Meta-Analysis, version 2.0. Two-sided p values were calculated, with p < 0.05 set as the level of statistical significance. First, pre- to post-test change scores were derived for the intervention group and control group from each included study. Then, the effect size for the difference between the intervention and control groups was calculated for each study. Hedges' g was used as the measure of the effect size. It was calculated by finding the difference between the intervention and control group means (d), divided by their pooled standard deviation and multiplied by a factor (J) that corrects for underestimation of the population standard deviation. A forest plot was used to present the effect size of all of the included studies. An inverse variance random-effects model was applied to analyze the data because it is more conservative than a fixed-effects model (DerSimonian & Laird, 1986).
To establish whether the selected studies differed significantly, the authors of the current study first examined whether the interstudy heterogeneity was statistically significant by evaluating the Cochran Q statistic (Higgins, Thompson, Deeks, & Altman, 2003), with p < 0.05 indicating significant heterogeneity. The magnitude of heterogeneity was measured using the I2 statistic, with I2 of 50% or greater indicating substantial heterogeneity across studies. A sensitivity analysis was also performed by removing the study with the largest effect size to determine its contribution to the overall effect size in the current meta-analysis.
Subgroup analyses were conducted by dividing the studies into groups according to (a) type of intervention, (b) type of cancer, (c) whether sleep was the primary outcome, (d) stage of cancer treatment at enrollment, (e) whether random sequence generation was appropriately executed (risk of selection bias), and (f) whether allocation concealment was appropriately executed (risk of selection bias). Moderator analyses were performed to explore possible reasons for the observed heterogeneity. To ensure sufficient data for analyses, each moderator analysis was limited to instances in which groups were represented by at least three studies. For categorical moderators, a mixed-effect model was used to compare differences among the effect sizes in each comparison (Lipsey & Wilson, 2001). Metaregression was used for the analyses of continuous moderators (Lipsey & Wilson, 2001).
Begg's rank correlation (Begg & Mazumdar, 1994) and Egger's intercept (Egger, Davey Smith, Schneider, & Minder, 1997) assess potential publication bias, with p > 0.05 indicating significant publication bias. The trim-and-fill method (Duval & Tweedie, 2000) was applied using a funnel plot to further assess potential publication bias. The overall effect size was adjusted by taking into account the estimated effect sizes of missing studies.
SHARE
![Pediatric Supportive Care (PDQ®): Supportive care - Patient Information [NCI]-Overview](https://cdnus.oss-us-west-1.aliyuncs.com/libs/images/NewsJpg/cate_6/072_215x160.jpg)
