The preparatory work for the revision of the Canadian guidelines has led to the identification of numerous questions for which adequate answers are not available in the literature and that warrant further research. Many of the questions raised here are of potentially great interest to extending our understanding of the relationship between PA and health. This section has restricted proposals for future research to questions that were felt to provide results that will improve the basis for future revisions of the PA guidelines.
General methodological issues
Because guidelines for PA aim at improving PA in segments of the population where the it is inadequate, there is clearly a need for more studies that provide solid evidence for how to make changes in the best way, and evidence for the health effects of these changes. Many of the studies that have formed the evidence for these guidelines did not directly address these fundamental questions and, moreover, were of rather poor or suboptimal quality for a variety of reasons. Thus, there is a need for more high-quality, large-scale, randomized intervention trials testing methods of increasing PA and testing the health effects of increasing PA where applicable.
For research questions that realistically cannot be addressed in randomized trials, such as the longterm effects of behavioural changes at the population level on morbidity and mortality, there is a similarly strong need for high-quality, large-scale, long-lasting, observational prospective cohort studies with repeated assessment of PA, allowing the identification of groups who have changed PA, confounders, and health outcomes. In both randomized trials and observational cohort studies, there is a need to apply far more precise and accurate measurement tools, especially for PA, but also for several health outcomes and potential confounders. Also, reporting of randomized clinical trials should be in accordance with the CONSORT guidelines or of non-randomized trials using the TREND guidelines .
Considering the heterogeneity of the population in Canada, investigations of corresponding differences between groups within the population are needed with regard to the distribution of PA, conditions for improving PA, and the health effects of given amounts and intensities PA, all of which may justify different guidelines for various population groups. Such groups could be defined, for example, by sex, age range, ethnicity, socioeconomic status, family setting, disability, puberty, or pregnancy. With advances in genomic research, there is a prospect of defining groups by genetic profile. Groups may also be defined by characteristics that themselves can be manipulated or modified, such as other lifestyle factors. The best possible epidemiological-statistical tools (e.g. interaction analysis) should be applied to investigate whether it is really necessary to make distinctions between groups.
It must be taken into account that the more groups into which the population is divided, the more difficult it is to validly demonstrate differential effects. It is a major challenge to identify the optimal cost-benefit balance between interventions customized to particular groups--or even tailored to particular individuals--versus fewer and more broadly applicable interventions. A need for different interventions may apply to individuals who already suffer from various diseases, but PA in such groups should be separated into PA aimed at general preventive effects among these individuals and PA aimed at specific therapeutic effects for these individuals as patients (clinical guidelines).
PA is related to numerous health outcomes, many of which constitute the core of the evidence for the PA guidelines. However, there are also many other outcomes of potential relevance for which the evidence is too weak to provide a compelling argument for recommendations. These include, for example, mental health and cognitive ability, especially in children, adolescents, and the elderly; common musculoskeletal disorders that also create great demands on the health care sector, such as osteoarthritis and osteoporosis with related fractures; and chronic medical disorders such as chronic heart failure and chronic obstructive pulmonary disease. Obviously, the guidelines that are applicable to a single individual should not differ for different health outcomes. The task is to integrate the evidence for different health outcomes and thereby allow recommendations that are generally applicable irrespective of differences in relationships to different health outcomes.
Research that translates PA-related health outcomes to quality-of-life measures is needed with regard to immediate, short-term and long-term effects. This applies to all age groups and draws attention to definitions of good and poor health, which may need to be specific for different age groups, especially for young children and the oldest adults. The consistent finding of an inverse association between allcause mortality and PA should be extended to quality-adjusted life years or disability-free life years.
Characterization of PA requires measurements of frequency, duration, intensity, volume (intensity × duration), and mode, all that vary over time intervals from seconds to years. Most studies so far have been based on self-reported (for children, parent-reported) PA, often very crudely categorized data, and this information about PA obviously suffers from several types of severe measurement error. This leads to misclassifications of individuals' PA and hence to biases in the assessment of the relationship between PA and health outcomes, between putative determinants of PA and PA, and between interventions aimed at changing PA and achieved changes in PA. It is a great challenge to obtain evidence based on better measurement technology. The current options are various types of devices that continuously monitor body movements (e.g., accelerometers); however, although these do provide a more objective measure of PA, they still do not cover all aspects of PA. Cardiorespiratory fitness has been used as a proxy for the cumulative effects of PA in the preceding period, but is also dependent on other factors (e.g., genetic constitution). Furthermore, there is a need to study the impact of a sedentary lifestyle as a separate entity from typically assessed PA, based on the suspicion that there are both determinants of and effects of a sedentary lifestyle that are not simply equivalent to "low PA" [21, 22]. NEAT (non-exercise activity thermogenesis) may play an important role in mediating, moderating or determining the relationship between PA or sedentary behaviour and health outcomes and deserves further investigation .
Effect measurement and dose-response
The magnitude of effects and the dose-response relationship, both for given levels of PA and for changes in PA, needs more investigation based on better measurement technology. When assessing effects in general and their dependence on the starting level of risk, it is of particular importance to pay attention to the measures of effects (relative versus absolute effect, where the former is heavily dependent on the reference level of risk that defines the denominator of the measure). Assessment of dose-response functions requires specification of the definition of "dose" (better characterization of PA) and of "response," including the time relationship (e.g., immediate, delayed, or cumulative investigations of the dose-response relationships should also address various compositions of the "dose," for example whether very-low-intensity PA for longer periods can a substitute high-intensity PA for short periods. Statistically convincing distinctions between different shapes of dose-response curves, including determination of threshold effects, will require very large sample sizes.
In the context of analytical epidemiological research (in both observational and interventional studies) addressing cause-effect relationships, a mediator is defined as a modifiable factor or variable that is influenced by the cause in question and thereby induces the effect that through the mediator is attributable to the cause. Obviously, causes may induce effects both through and beyond mediation of a particular mediator. From a biological or psychological point of view, the mediation process can be considered as the "mechanism" or "pathway" by which the PA is linked upstream to determinants of PA and downstream to health effects. Understanding the mediation process may enhance the opportunities for interventions to improve PA, and for increased PA to improve health, not least because different types of PA may be related to different mediators [24, 25]. The current evidence on mediation is rather limited, and there is a great need for more research, both regarding the mediation of interventions aimed at changing PA (particularly psycho-behavioural factors, e.g., self-regulation and self-efficacy) and the mediation of changes in PA on health outcomes (e.g., via biological coronary heart disease risk factors, cardiorespiratory fitness, energy expenditure). Demonstration of the associations between a putative mediator and the cause, and between this mediator and the effect, may not suffice as evidence for mediation.
Hazards and adverse effects
Although the evidence so far indicates that there are few hazards or adverse effects of increasing PA or keeping a high level of PA, the possibility of adverse effects should be kept in mind in conducting future studies; for the sake of transparency, developing clinical practice guidelines and when building evidence for increasing PA.
In addition to the usual type of biases that affect randomized trials and cohort studies, some special biases need consideration--namely, those based on continuously operating behaviours, such as PA, as determinants of concurrent or later chronic disease health outcomes, such as obesity and coronary heart disease, respectively. Reverse causation implies that the disease process is already ongoing and, although not yet clinically diagnosed, may lead to changes in PA or associated confounders. For a condition such as obesity, there is the additional problem that this condition tends to fluctuate (statistically with the effect of regression toward the mean) and behavioural lifestyle factors (e.g., PA) may change in response to this fluctuation [26, 27]. Therefore, an apparent association may be created between these lifestyle factors and subsequent changes in the opposite direction.
Systematic review process
The systematic review process is of fundamental importance to the synthesis of evidence forming the basis for guidelines [11, 28]. Where the outcome of a review can be supported by a meta-analysis providing pertinent quantitative effect estimates, this should be done. Adherence to internationally well-established standards for systematic reviews and meta-analyses, for example as developed and implemented by the Cochrane Collaboration, will increase the value and trustworthiness of the output of the process. Particular attention needs to be paid to the correspondence between the aim of the review and the criteria for literature selection in order to avoid missing potentially informative, high-quality studies. Where available, previously performed systematic and high quality reviews should be taken as the starting point so that the review process can take advantage of what has already been done by others, while of course maintaining a critical analytical approach .
The application and interpretation of scoring systems for the quality and message of the evidence to the output should be based on scales that are adapted to the type of evidence obtainable. The current scoring systems are typically based on an assessment of evidence that can be generated in large-scale therapeutic randomized trials (e.g., drug trials). However, it is not possible to use such trial designs when addressing questions about the long-term effects on mortality of behavioural preventive measures such as PA. Future research should explore how existing medical-oriented evidence assessment procedures require modification to accommodate population and lifestyle behavioural modifications. One example of such modifications is that assessment of the intervention by comparative trial design may have to be based on randomisation of a few clusters in which the intervention is set up by facilitating or encouraging physical activity in various ways in institutions or regions. Another possibly necessary modification is that the evaluation of the outcome rather than being based on changes in risk of clinically relevant endpoints, namely morbidity and mortality, may have to be based on what else would be considered as surrogate endpoints such as well being, improved fitness and strength, improved risk factor profile, e.g. improved blood pressure, blood lipids, glucose tolerance.
Harmonization of physical activity guidelines
Guidelines for PA are being prepared by governmental and non-governmental organizations in many countries (e.g., US, UK) and by several international institutions (e.g., the World Health Organization). Future revisions of guidelines in all countries should take advantage--alongside the systematic reviews being conducted conditional on previous reviews--of comparisons between different guidelines in order to identify the reasons for differences (e.g., evidence, cultural, autonomy) and use this information to contribute to the improvement of future guidelines. In the process of such harmonization it would also be of great value to make the rationale and criteria applied for choosing particular thresholds for duration and intensity of recommended PA of particular types explicit, given that the available evidence seldom leads to unambiguous thresholds and that feasibility in the target populations necessarily must be considered as well.
Healthy lifestyle approach
Lifestyle risk factors, such as PA, diet, smoking, alcohol intake and sleep, are clustered and possibly interact, both in terms of their determinants (lifestyle habits may substitute for each other) and effects on health. There is an obvious need for a healthy lifestyle approach in researching these factors and thereby forming the basis for more comprehensive guidelines that may be adequate for large segments of the population [30, 31].