- Daniel Y T Fong, assistant professor1,
- Judy W C Ho, consultant surgeon2,
- Bryant P H Hui, research assistant3,
- Antoinette M Lee, assistant professor4,
- Duncan J Macfarlane, associate professor5,
- Sharron S K Leung, assistant professor1,
- Ester Cerin, associate professor5,
- Wynnie Y Y Chan, public health nutritionist6,
- Ivy P F Leung, dietitian7,
- Sharon H S Lam, part time lecturer (nutritional science)8,
- Aliki J Taylor, honorary senior clinical research fellow9,
- Kar-keung Cheng, professor9
- 1School of Nursing, Li Ka Shing Faculty of Medicine, University of Hong Kong, Hong Kong, China
- 2Division of Colorectal Surgery, Department of Surgery, Li Ka Shing Faculty of Medicine, University of Hong Kong, Hong Kong
- 3Department of Applied Social Sciences, Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong
- 4Department of Psychiatry, Li Ka Shing Faculty of Medicine, University of Hong Kong, Queen Mary Hospital, Pokfulam, Hong Kong
- 5Institute of Human Performance, University of Hong Kong, Pokfulam, Hong Kong
- 6College of Life Sciences and Technology, Dietetics, Food and Nutritional Sciences, School of Professional and Continuing Education, University of Hong Kong, Pokfulam, Hong Kong
- 7Department of Dietetics, Queen Elizabeth Hospital, Kowloon, Hong Kong
- 8College of International Education (CIE), Hong Kong Baptist University, Shek Mun, Shatin, Hong Kong
- 9Department of Public Health, Epidemiology and Biostatistics, University of Birmingham, Public Health Building, Edgbaston, Birmingham B15 2TT, United Kingdom
- Correspondence to: J W C Ho, Division of Colorectal Surgery, Department of Surgery, Queen Mary Hospital, 102 Pokfulam Road, Hong Kong SAR, China
- Accepted 8 November 2011
Objective To systematically evaluate the effects of physical activity in adult patients after completion of main treatment related to cancer.
Design Meta-analysis of randomised controlled trials with data extraction and quality assessment performed independently by two researchers.
Data sources Pubmed, CINAHL, and Google Scholar from the earliest possible year to September 2011. References from meta-analyses and reviews.
Study selection Randomised controlled trials that assessed the effects of physical activity in adults who had completed their main cancer treatment, except hormonal treatment.
Results There were 34 randomised controlled trials, of which 22 (65%) focused on patients with breast cancer, and 48 outcomes in our meta-analysis. Twenty two studies assessed aerobic exercise, and four also included resistance or strength training. The median duration of physical activity was 13 weeks (range 3-60 weeks). Most control groups were considered sedentary or were assigned no exercise. Based on studies on patients with breast cancer, physical activity was associated with improvements in insulin-like growth factor-I, bench press, leg press, fatigue, depression, and quality of life. When we combined studies on different types of cancer, we found significant improvements in body mass index (BMI), body weight, peak oxygen consumption, peak power output, distance walked in six minutes, right handgrip strength, and quality of life. Sources of study heterogeneity included age, study quality, study size, and type and duration of physical activity. Publication bias did not alter our conclusions.
Conclusions Physical activity has positive effects on physiology, body composition, physical functions, psychological outcomes, and quality of life in patients after treatment for breast cancer. When patients with cancer other than breast cancer were also included, physical activity was associated with reduced BMI and body weight, increased peak oxygen consumption and peak power output, and improved quality of life.
Cancer survivors who have successfully completed their primary cancer treatment often expect to resume their work or daily life at a level similar to that before the cancer diagnosis. While cancer treatment has been shown to be effective in prolonging survival, it can be intensive and can lead to increased fatigue, decreased physical activity, and a reduction in quality of life.1 2 3 In addition, these unwanted effects of treatment can be prolonged and hinder the patients’ return to normal life.4 5 6
Physical activity is a potentially appealing intervention that could alleviate sequelae related to cancer and assist patients in returning to the health status they had before treatment. A systematic review published in 2005 summarised the evidence supporting the recommendation of physical activity during and after treatment related to cancer7; and a meta-analysis published in 2006 reported more favourable outcomes when physical activity was carried out after treatment.8 In a recent study published in 2011, starting an exercise programme after the completion of treatment was shown to be acceptable to over three quarters of patients.9 Several randomised controlled trials have assessed the efficacy of physical activity on indicators of physical and mental health in patients after cancer treatment,8 and these trials reported significant improvement after physical activity.
The effects of physical activity on cancer have been examined in nine meta-analyses, with three focusing on breast cancer10 11 12 and six on any type of cancer.8 13 14 15 16 17 The six meta-analyses on any cancer type did not uniformly examine sources of study heterogeneity,14 15 16 assess publication bias,14 15 16 17 or limit inclusion of randomised controlled trials to those with physical activity intervention only after cancer treatment.8 13 17 Moreover, more randomised controlled trials have been published since the publication of the last meta-analysis in 2011.18 We updated the most recent meta-analysis by including studies published more recently and included only randomised controlled trials. Using data from randomised controlled trials, we evaluated the best current evidence for the effects of physical activity on physical functions, physiological parameters, body composition, psychosocial outcomes, and quality of life in adult patients after they had completed their main treatment related to cancer.
Search strategy for identification of relevant studies
We identified relevant studies by systematically searching PubMed, which included Medline, CINAHL, and Google Scholar with the last search being in September 2011. The search terms used were (cancer OR tumour OR tumor OR neoplasm OR carcinoma OR chemotherapy OR radiotherapy OR bone marrow transplant) AND (physical activity OR exertion OR exercise therapy OR physical education OR physical training OR physical fitness OR walking) AND (intervention OR trial OR adherence OR compliance OR patient compliance OR patient education OR health promotion OR health education OR health behavior OR health behaviour OR behavioral therapy OR behavioural therapy OR behavioral change OR behavioural change OR positive reinforcement OR cognitive therapy) AND (controlled clinical trial OR randomised OR randomized OR placebo OR randomly OR group). We also searched the Cochrane Library for systematic reviews and meta-analyses on the effects of physical activity in cancer survivors. The reference lists of all selected studies, systematic reviews, and meta-analyses were then examined for relevant studies. We obtained the full papers of studies identified as potentially eligible based on the titles and abstracts, and two of the authors (JWCH and BPHH), who served as raters, independently confirmed their eligibility.
The studies included were those that adopted a randomised controlled trial design, included adult patients (aged ≥18), included patients diagnosed with cancer, involved patients who had completed their main treatment for cancer but might be still undergoing hormonal treatment, and assessed the effect of physical activity on health indicators.
Data extraction and methodological quality assessment
JWCH and BPHH independently extracted relevant data using a standardised Excel template. From each study, they extracted general study information, details of intervention and control groups, means and standard deviations of outcomes after the intervention, side effects, and compliance/adherence rates. The summary statistics, when not reported, were calculated when sufficient data were available. Any discrepancies between data extracted by the two raters were discussed and consensus was achieved. A biostatistician (DYTF) also checked all extracted statistics.
Two authors (JWCH and DYTF) independently assessed the methodological rigour of the selected studies using a quality assessment checklist developed by the Scottish Intercollegiate Guidelines Network.19 They assessed the internal validity of each selected study using 10 criteria and rated the overall quality on a scale according to the likelihood that each unmet criterion would alter the study conclusions: ++ (very unlikely), + (unlikely), and − (likely). The two reviewers met to compare their quality assessment results for each study; all discrepancies were discussed and resolved.
We planned to assess the effects of physical activity on cancer outcomes (that is, rates of survival and recurrence), psychological outcomes, and physical outcomes. None of the eligible studies provided data on cancer outcomes, and some of the other outcomes were not assessed by multiple studies. We therefore performed a meta-analysis on an outcome if it was assessed in at least two studies. For each outcome, we first assessed study heterogeneity with Cochrane’s χ2 test, with P<0.10 indicating evidence of heterogeneity. The degree of heterogeneity was measured by the I2 statistic, with I2 ≥50% indicating substantial heterogeneity.20 In the presence of heterogeneity, we searched for its sources by random effects meta-regression. We considered the following study characteristics for heterogeneity: study quality (++, +, and −), publication year, percentage of female participants, mean age of participants, number of subjects analysed, duration of intervention, type of physical activity, and percentage of patients with breast cancer. Pooled estimates and their 95% confidence intervals were obtained with the random effects method.21 Publication bias was also examined with Egger’s linear regression method, with P<0.10 taken as an indication of publication bias.22
A random effects meta-regression was performed in the SAS version 9.2 (SAS, NC) and the other meta-analyses were conducted in Stata version 11 for Windows (StatCorp, TX). The nominal level of significance was 5% in all tests unless otherwise specified.
We identified 1505 records from the database search and 387 additional records from other sources (fig 1⇓). After screening, 54 articles met our inclusion criteria and we assessed their methodological quality. Of these, 39 articles from 34 studies had at least one common outcome, and there were sufficient data for inclusion in the meta-analysis. Seven studies (21%) and two articles from another two studies had not been included in previous meta-analyses (table 1⇓).23 24 25 26 27 28 29 30 31 Twenty two studies (65%) included only patients with breast cancer, three (9%) included only patients with colorectal cancer, one (3%) included only patients with endometrial cancer, and the eight (27%) remaining included patients with different types of cancer. Three studies had more than one intervention group and thus contributed more than one comparison with their control groups to the meta-analysis. Another two studies published both the full and subgroup analyses. Among 43 intervention-control comparisons, the median number of patients analysed was 93 (range 14-641) with a mean age of 55 (39-74). We used 27 (63%) intervention-control comparisons to assess the effects of an aerobic exercise programme, and six (14%) studies had patients who additionally underwent resistance or strength training. The other comparisons were made for strength or tailored training based on preference of patients. The median duration of the physical activity intervention was 13 weeks (range 3-60). Only 13 studies stated the intensity level of physical activity; 11 were of moderate intensity and two were of vigorous intensity. Most control groups were considered sedentary or were assigned no exercise. We assessed 48 outcomes at least twice in the 43 intervention-control comparisons (tables 2 and 3⇓ ⇓).
Meta-analysis of physiological markers
Physiological markers were assessed only in breast cancer studies. Four physiological markers were included in the meta-analysis: insulin-like growth factor-I, insulin, glucose, and homeostatic model assessment. There was no study heterogeneity in these physiological markers, with I2 being 28% at most, and there was no evidence of publication bias (P≥0.47). In three studies, of which two assessed aerobic exercise and the third assessed weight/strength training, physical activity was associated with significantly reduced insulin-like growth factor-I (−12.0 ng/mL, 95% confidence interval −23.3 to −0.5; P=0.04; fig 2⇓) more than the controls) (table 4⇓). No effect was shown for the other physiological markers. Forest plots for all physiological markers can be found on bmj.com.
Meta-analysis of body composition
There was no study heterogeneity for any of the six body parameters assessed. Physical activity was associated with slightly reduced body mass index (BMI) (−0.4, −0.6 to −0.2; P<0.01) and body weight (−1.1 kg, −1.6 to −0.6 kg; P<0.001) (fig 3⇓) than the control condition (table 4⇑). Publication bias was evident only for the waist:hip ratio (P=0.06) because of one of the three studies reporting a more substantial reduction than the other two. The effect on waist:hip ratio, however, remained insignificant after we removed this study from the meta-analysis. Forest plots for all markers of body composition can be found on bmj.com.
Meta-analysis of physical functions
Except for bench press and leg press, which were assessed only in patients with breast cancer, all other physical functions were assessed in studies on several types of cancer. Substantial heterogeneity was found for leg press, left handgrip, and right handgrip (table 2⇑). The higher quality study reported a smaller increase in leg press, whereas the studies involving older patients or published in recent years reported larger increases in left handgrip and right handgrip. Otherwise, neither heterogeneity nor publication bias was detected (P≥0.12). Physical activity was associated with significantly increased peak oxygen consumption (2.2 mL/kg/min, 1.0 to 3.4; P<0.01), peak power output (21.0 W, 13.0 to 29.1; P<0.01), the distance walked in six minutes (29 m, 4 to 55; P=0.03; fig 4⇓), bench press weight (6 kg, 4 to 8; P<0.01), leg press weight (19 kg, 9 to 28; P<0.01), and right handgrip strength by 3.5 kg, 0.3 to 6.7; P=0.03) (table 4⇑). Forest plots for all markers of physical function can be found on bmj.com.
Meta-analysis of psychological outcomes
There was no heterogeneity in any of the psychological outcomes except for fatigue measured by the European Organization for Research and Treatment of Cancer (EORTC) core quality of life questionnaire and the two hospital anxiety and depression scales (table 3⇑). Measured by the revised Piper fatigue scale, physical activity was associated with slightly reduced fatigue (−1.0, −1.8 to −0.1; P=0.03; fig 5⇓) in three comparisons from two studies on breast cancer compared with the control (table 5)⇓. Measured by the Beck depression inventory, physical activity was associated with reduced depression (−4.1, −6.5 to −1.8; P<0.01) in survivors of mixed types of cancer, which was of near clinical importance. No publication bias was found in any of the psychological outcomes (P≥0.82). Forest plots for all markers of psychological outcome can be found on bmj.com.
Meta-analysis of quality of life outcomes
There were 21 quality of life domains included in the meta-analysis and 10 of them exhibited heterogeneity (table 3⇑). Identified sources of heterogeneity were study quality, sample size, type and duration of physical activity, and age. Study quality was strongly and positively associated with sample size, with smaller studies being associated with larger effects of physical activity on emotional function and total domains of the European Organization for Research and Treatment of Cancer. One study in patients with breast cancer found that aerobic plus resistance training was significantly more effective than aerobic training alone on physical, emotional, and functional wellbeing, breast concerns, breast total, and general total scores on the functional assessment of cancer therapy. In addition, the effect of physical activity on breast concerns was significantly greater in younger patients. The age effect, however, was confounded by the duration of physical activity.
In mixed types of cancer survivors, physical activity improved the SF-36 physical function scores by 3.0 points (0.6 to 5.3; P=0.01), social function by 3.4 points (0.4 to 6.4; P=0.03) and mental health by 2.4 points (0.7 to 4.1; P=0.01) (fig 6⇓) compared with the control group (table 5⇑). No publication bias was found in any of the quality of life domains (P=0.21). Forest plots for all markers of quality of life can be found on bmj.com.
Effects of physical activity
In patients who have completed treatment for cancer, physical activity is associated with significantly reduced insulin-like growth factor-I, BMI, body weight, fatigue, and depression; it is also associated with increased peak oxygen consumption, peak power output, distance walked in six minutes, bench and leg press weight, right handgrip strength, and quality of life in the physical function, social function, and mental health domains. Our meta-analysis featured both exploration of sources of heterogeneity and assessment of publication bias that assesses the efficacy of physical activity on health indicators in patients with any type of cancer after completion of their primary cancer treatment. We reviewed 48 outcomes reported from 34 randomised controlled trials in patients with cancer. The effects were clinically important to physical functions and quality of life. We found substantial study heterogeneity, which was attributed to study quality and size, patients’ age, and type and duration of physical activity.
Based on data from four randomised controlled trials in breast cancer,30 32 33 34 physical activity was associated with significantly reduced serum concentration of insulin-like growth factor-I, despite insignificant results being reported in the primary studies. This was because of the increased precision resulting from combining different studies. A recent meta-analysis that pooled three of the studies we included also reported a small reduction in insulin-like growth factor-I.16 Another meta-analysis that examined physiological changes after exercise reported that the effect of physical activity was insignificant.14 That meta-analysis, however, combined different physiological parameters and might have included non-randomised studies. Raised concentration of insulin-like growth factor-I is known to be associated with an increased risk of colorectal cancer.35 Thus, a significant reduction in insulin-like growth factor-I by aerobic or weight/strength training could potentially imply a reduced risk of cancer recurrence. However, there was no clear trend of an effect of physical activity on blood insulin, glucose, and homeostatic model assessment levels.
BMI and body weight were found to be slightly but significantly reduced after physical activity intervention in studies that included patients with any cancer or breast cancer. The same result was found in the most recent meta-analysis, which also reported significant but not clinically important effects.16 None of the other body parameters was significantly improved by physical activity. This finding is consistent with the literature, although previous meta-analyses have shown that percentage of body fat and lean body mass were significantly improved.11 12 16 These meta-analyses, however, also included studies assessing physical activity during cancer treatment or did not show a clinically important effect.
Physical activity after completion of cancer treatment had clear benefits for many physical functions, including peak oxygen consumption, peak power output, distance walked in six minutes, and bench and leg press weights. Except for those of bench and leg presses, these findings have also been consistently reported in all other meta-analyses.8 10 11 12 13 14 The only meta-analysis that reported unstandardised effects showed a slightly larger effect than ours on peak oxygen consumption (3.39 mL/kg/min, 1.67 to 5.10 mL/kg/min) and distance walked in six minutes (35 m, 13 to 58 m).11 That meta-analysis, however, included only studies on patients with breast cancer, who might carry out physical activity during and after treatment.
A significant but small reduction in fatigue was observed in breast cancer studies using the revised Piper fatigue scale but not in studies on any cancer using either of the other two fatigue scales. Both the revised Piper fatigue scale and functional assessment of cancer therapy-fatigue have been shown to perform satisfactorily in measuring cancer related fatigue.36 37 The fatigue scale by the European Organization for Research and Treatment of Cancer has also been shown to have satisfactory psychometric performance but it is not specific to cancer related fatigue.38 The revised Piper fatigue scale, however, was primarily used in patients with breast cancer and thus has limited generalisability to patients with other cancers.39 Nevertheless, it measures the current, rather than recalled, fatigue level of patients and so could be more sensitive in assessing the acute short term effects of physical activity. Only three meta-analyses have concluded that physical activity had significant effects on relieving fatigue.15 16 17 They reported at least small to moderate effects of physical activity. Although the two studies that used the fatigue scale from the European Organization for Research and Treatment of Cancer showed significant improvement, their reported effects were largely different. This led to a large variation that resulted in an insignificant pooled random effects estimate. The same phenomenon was also observed for the total mood scale of the profile of mood states. Depression measured by the Beck depression inventory, however, was significantly reduced at a level of near clinical importance. This finding has not been reported in other meta-analyses.
Quality of life in patients with cancer has been assessed in the physical, mental, social, and overall quality of life domains by using four common instruments. For the physical domain, only the SF-36 survey showed a significant improvement of marginal clinical importance. The study of aerobic plus resistance exercise, however, found significantly larger improvements on both the physical and functional wellbeing scales of the functional assessment of cancer therapy than studies that examined aerobic exercise alone. The effect of the type of physical activity was not examined for the European Organization for Research and Treatment of Cancer and SF-36 scales because they were not measured in studies that used aerobic plus resistance training. The significant improvement shown in the physical function scale of the SF-36 survey is probably because of the inclusion of a recent large study.28 For the mental domain, there was a significant improvement in the emotional wellbeing scale of the functional assessment of cancer therapy and the mental health scale of the SF-36 survey in studies of breast cancer survivors. The European Organization for Research and Treatment of Cancer also has a scale for emotional function, but no significant effect was found except in one study with the smallest sample size. For the social domain, clinically important improvement was found only in the social function scale of the SF 36 survey but not in the social/family wellbeing scale of the functional assessment of cancer therapy. This could be because of the inclusion of a large study.40 These findings have not been reported in other meta-analyses.
Limitations of selected studies and previous meta-analyses
The intensity of physical activity could play an important role in its effects.41 42 The intensity levels in interventions in the selected studies, however, were not consistently reported, which made it difficult to assess their influence on study heterogeneity. Nevertheless, the significantly larger effects of aerobic plus resistance training than aerobic training alone might indicate a potential benefit of higher intensity. Standardisation of reporting of exercise intensity, perhaps with a new method of considering both the gas exchange threshold and the conventional maximal oxygen uptake,43 in future randomised controlled trials of physical activity in patients with cancer is desirable.
The standardised effect sizes of the outcomes across studies were measured by different methods or measurement scales in all but one meta-analysis. This resulted in previously unidentified important effects of physical activity on quality of life. Four meta-analyses that assessed the effects of physical activity in patients of any cancer type included both randomised and non-randomised controlled trials.8 13 14 16 This might have increased study heterogeneity or publication bias. In addition, although these meta-analyses reported that physical activity improved physical function or cardiorespiratory fitness, the measures used to define these two health indicators were unclear. Only one meta-analysis clearly described physical function as a set of aerobic capacity or timed walking distance.13 Although these meta-analyses standardised the mean difference between groups before pooling the estimates, the estimates represented a combination of different outcomes, such as peak oxygen consumption and distance walked in six minutes, which measured distinct domains of physical performance. The same concern applies to other health outcomes, such as fatigue, which might have been measured with different scales. In the sequel, although the other two meta-analyses included randomised controlled trials only, they also pooled different measures of fatigue.15 17 Mixing studies with outcomes measured with different instruments has long been known to potentially pose a threat to the validity of the resulting conclusions, although researchers might still statistically combine diverse study outcomes.44 Importantly, it is difficult to interpret the resulting pooled estimate because it represents a mixture of outcomes under distinct domains.45 Indeed, the Cochrane Handbook has generally emphasised the importance of due consideration when combining different variables, and researchers should not combine outcomes that are too diverse.46 These suboptimal meta-analytic protocols were probably used because of the limited number of relevant randomised controlled trials in the literature. Thus, this meta-analysis included only randomised controlled trials and pooled studies for each outcome measured on the same scale.
We included only published randomised controlled trials in our meta-analysis, which could have increased the risk of publication bias by not including non-randomised studies or unpublished studies. The risk of publication bias might have been further increased by searching only three electronic databases and not contacting other experts for possible inclusion of more relevant studies. Non-randomised studies had been included in previous meta-analyses because of the lack of published randomised controlled trials.8 13 14 16 Non-randomised studies, however, can overestimate treatment effects by 30-41%47; thus, their inclusion in our study might have yielded overly optimistic effects of physical activity. In addition, unpublished studies, such as abstracts or reports, often report smaller treatment effects,48 are often of poor quality, and do not provide sufficient data for statistical synthesis. Furthermore, we carefully examined all relevant meta-analyses, including the most recent three published in 2010 and 2011. We were not aware of any relevant unpublished studies when this study was conducted. Moreover, most outcomes did not exhibit publication bias; when it was found, it did not influence the conclusions.
Although we did not impose any language restriction in our systematic search, we eventually included only studies published in English. Although the exclusion of non-English language studies might result in smaller intervention effects, the language bias is generally small.49 In addition, there might also be bias from not blinding the abstractors and reviewers to authors, institutions, and journals. There has been considerable debate on whether these details should be blinded. One empirical study did not advocate blinding as there is only a small risk of bias, but masking the details requires a large amount of administration time.50
Substantial study heterogeneity was found in several outcomes. This can limit the interpretability of the pooled estimates. Subgroup analyses by the identified sources of study heterogeneity, however, largely reduced the degree of heterogeneity, which became statistically insignificant.
The median duration of physical activity interventions in the included studies was only 13 weeks (range 3-60 weeks). Most mainstream studies of such interventions often conclude that physical activity needs to become integrated into everyday life to have long term benefits on health and fitness.51 52 The generally short duration of interventions in our selected studies would limit assessment of the long term benefits of physical activity.
Implications for research and practice
Future research should focus on cancers other than breast cancer. Moreover, a standardised use of outcome measures and assessment of the intensity of physical activity are desirable and important in future randomised controlled trials to facilitate more reliable and valid synthesis of results from different studies.
Based on our review of 48 outcomes reported from 34 randomised controlled trials in patients with cancer, physical activity was shown to be associated with clinically important positive effects on physical functions and quality of life in patients who had completed their treatment for cancer. All of these benefits were applicable to patients with breast cancer. When we included studies of other types such as prostate, gynaecological, colorectal, gastric, and lung cancers, there was evidence of clinically important benefits on peak oxygen consumption, peak power output, and quality of life, which included physical and social functioning domains. Further randomised controlled trials on patients with cancers other than of the breast are needed to further assess the efficacy of physical activity on other health outcomes.
What is already known on this topic
Current meta-analyses of the efficacy of physical activity in patients who had completed cancer related treatment might feature increased study heterogeneity because of the mixing of studies with outcomes measured under distinct domains or using different instruments
New randomised controlled trials for assessing the efficacy of physical activity in cancer survivors have been published
What this study adds
Physical activity was associated with important positive effects on physical functions and quality of life, which included physical and social functioning domains, in patients who had completed their treatment for cancer
Efficacy of physical activity could be higher in studies with lower methodological quality, smaller study size, or older patients or that assess a physical activity programme involving both aerobic and resistance training
Cite this as: BMJ 2012;344:e70
Contributors: JWCH and DYTF contributed equally to the study. JWCH, DYTF, EC, WYYC, AJT, SHSL, IPFL, and KC conceived and designed the study. DYTF, JWCH, BPHH, DJM, and SHSL searched the literature. DYTF, JWCH, and BPHH reviewed the literature and extracted data. DYTF and BPHH performed the meta-analysis. DYTF, JWCH, and AML analysed and interpreted the data. DYTF, JWCH, and BPHH drafted the manuscript. DJM, EC, AML, SSKL, WYYC, AJT, SHSL, IPFL, and KC critically revised the manuscript for important intellectual content. All authors approved the final version of the manuscript. JWCH, DJM, AML, SSKL, AJT, and KC obtained funding. JWCH and KC supervised the study. JWCH is guarantor.
Funding: This study was supported by the World Cancer Research Fund International, World Cancer Research Fund UK, and World Cancer Research Fund Hong Kong (grant No 2009/02). The funding body had no input in study design and the collection, analysis, and interpretation of data and the writing of the article and the decision to submit it for publication. All authors are independent from the funding body.
Competing interests: All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work.
Ethical approval: Not required.
Data sharing: No additional data available.
This is an open-access article distributed under the terms of the Creative Commons Attribution Non-commercial License, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the license. See: http://creativecommons.org/licenses/by-nc/2.0/ and http://creativecommons.org/licenses/by-nc/2.0/legalcode.