TL;DR

Introduction

The literature now suggests that the total volume of physical activity is more important than how this physical activity is accumulated. Due to this finding, the physical activity guidelines in the United Kingdom (UK) now allow individuals to meet the physical activity guidelines by performing physical activity of at least moderate intensity in bouts of any duration. As discussed in the previous section of this report, there are a large number of both observational and RCT based studies that have collected accelerometer measured physical activity data alongside a range of health outcomes data. However, there are no published meta-analyses that focus on short bouts of physical activity and their impact upon health outcomes. This could be due to the relatively small number of studies that have reported the effect of short bouts of physical activity on comparable health outcomes. Furthermore, the majority of studies examining short bouts have either been interventional or secondary analysis of previously collected data with a large majority of these using data from National Health and Nutrition Examination Survey (NHANES) due to the ease of access to raw accelerometer files allowing reprocessing to derive the unbouted volume of physical activity performed by each participant.

To bring the results of such studies together, a meta-analysis can be performed. A meta-analysis involves the combining of aggregate data from two or more studies that address the same research question. They are generally considered to be the highest standard of evidence as they provide a quantitative assessment of a body of research. In recent years, Individual Participant Data (IPD) meta-analysis has emerged as a new gold standard in reviewing. Rather than extracting summary/aggregate data from a study, an IPD meta-analysis seeks to harmonise and analyse original research data from two or more studies. Such a review can both improve the quality of the data contained within the review and allows a broader range of statistical methodologies to be utilised.

This style of review can be particularly useful when accelerometry measured physical activity data is contained within the studies to be reviewed. To obtain useful insights into an individual’s physical activity from accelerometry data a range of subjective analytic decisions need to be made. These decisions can, and often do, differ between studies in terms of the number of valid days that must be achieved to produce a valid set of participant data, the number of hours that constitute a valid day and the cut points used to determine the intensity of physical activity. However, if the individual participant data can be obtained, along with the raw accelerometer files, then all the accelerometer data can be analysed using a consistent analysis procedure. Furthermore, it allows a range of previously uncalculated physical activity variables to be explored. This is particularly relevant for Snacktivity as short bouts of physical activity are still under explored in previous research. Therefore, many studies that have collected accelerometer data only report physical activity that was completed in bouts lasting 10 minutes or over. However, by accessing the raw accelerometer files, we can derive a series of physical activity bouts of any chosen duration.

Harmonising of a large number of original research datasets can become a challenge. Data harmonisation is the process of improving the comparability of data of similar measures collected by different studies. This is a vital step in an IPD meta-analysis that is not required in a traditional meta-analysis. There have been a number of notable data harmonisation efforts in the field of physical activity. Firstly, the International Children Accelerometer Database (ICAD) has harmonised studies which have measured physical activity using a waist worn Actigraph (Actigraph, LLC, Pensacola, Florida) in children. ICAD included 20 observational studies in the original ICAD dataset with data collected in North American, Brazil, Europe and Australia. The ICAD dataset has been used to answer a range of research questions with 27 publications resulting from the dataset to date. These ICAD publications have highlighted the potential benefits of harmonising datasets from many geographic locations, as well as the extensive challenges encountered when attempting to harmonise physical activity and health outcomes data. Another notable study published in 2019 has used a harmonised meta-analysis to examine the impact of physical activity on cardiovascular disease and cancer mortality. A total 850,060 participants were included within the analysis for cardiovascular disease mortality involving data from nine studies collected in Europe, North America and Japan. Both the harmonised datasets referred to above illustrate how combining datasets provides the opportunity to substantially increase the sample size in subsequent research studies and increase the diversity of the study sample.

Although conducting harmonisation presents a range of additional challenges, it will result in the production of a higher quality final paper. Therefore, we will conduct a harmonised compositional isotemporal substitution analysis of datasets that have collected accelerometer measured physical activity and a range of health outcomes to assess whether substituting sedentary time with short bouts of MVPA produces cardiometabolic health benefits.

Proposed methodology

The exact methodology for this study is still being devised. The primary reason for this is that this study will be heavily informed by the results of the scoping review. This scoping review will inform both the wear location and device(s) that the meta-analysis will focus on. The scoping review will also help to inform the inclusion and exclusion criteria for the meta-analysis (e.g. sample size, data access status).

Data collection

The datasets will be identified from the scoping review described in Study 1. Based on the findings of the scoping review, a precise research question will be determined for the meta-analysis. The main features of this will be which device(s) and wear locations the IPD meta-analysis will focus on as well as the minimum sample size that will be required for a study to be included. In terms of the device(s) to focus on, this will be determined by which device(s) have been used to collect the largest proportion of the available data. In terms of the minimum sample size that will be included, we must ensure that the IPD meta-analysis is feasible to conduct. Therefore, the minimum sample size will be calculated to ensure as much data as possible is included that contains information about as diverse a group of individuals as possible whilst keeping the study feasible to conduct.

Once the desired datasets have been identified, the data will be sought from these studies. If the datasets are open access, then the data will be retrieved. If the data are not open access, the primary researcher on the study will be contacted via email to ask whether they are willing to contribute their data to the study. If they agree, a data sharing agreement will be drawn up and signed by both parties before the data are transferred securely. The exact methodologies for this step are still being developed however all relevant data protection policies will be followed.

Data harmonisation

Once all the available data have been collected, it will be harmonised. Accelerometer files will be reanalysed using Kinesoft (Loughborough, UK) or GGIR software. A quality control procedure will be performed on each accelerometer file. This will include identifying and flagging any spurious data files that may represent a technical error in the device. The exact methodology for this process is still being developed however a spurious file could be identified by a prolonged period of zero counts or counts value higher than would be deemed plausible to represent human movement. Studies have previously outlined potential methodologies to identify spurious data.

Daily wear time will then be calculated alongside the total number of valid days of data collected for each participant. The number of valid days that will constitute a valid file will be determined from both previous literature and by analysing how much data would be removed as the valid day criteria becomes stricter. Non-wear time has previously been defined as 120 minute of consecutive zero values. Previous studies have suggested that approximately 10 hours of data constitutes a valid data and 4 days of valid data constitute a valid week. However, other studies have suggested as few as 2 days could be considered valid.

The primary outcome of interest will be bouts of physical activity of at least moderate intensity lasting between 2 and 5 minutes as this meets the definition of an activity snack. However, multiple bout durations including unbouted, under 5 minutes, under 10 minutes and over 10 minutes will also be extracted at the full range of intensity thresholds (low, moderate and vigorous). In order to harmonise the demographic and health outcomes information, a similar procedure will be used to that employed by the International Children’s Accelerometer Database (ICAD). An excel template will be developed to act as a “data dictionary”. Within this template, each variable measured in each study will be given a name, short label, long description, unit and format and will then be assigned to a variable group. For example, it is likely ethnicity will be recorded using different categories depending on the country in which the data were collected. All the ethnicity variables will have these five key pieces of information recorded and all would be assigned to the variable group ethnicity. Once this has been completed for all studies, each variable group will be examined, and a harmonisation approach will be developed. It may be necessary that multiple variables are created for each variable group to allow full harmonisation. These harmonisation decisions will be discussed with the supervisory team and wider members of the Snacktivity group as necessary. This harmonisation approach is in line with that utilised by ICAD and described by Atkin et al.

Data analysis

Once the data have been harmonised, separate analysis will be conducted on both each individual study as well as all the studies combined to determine the impact that each individual study has on the final model. Firstly, summary statistics will be produced on key demographic, physical activity and health outcome variables. Relationships between variables will be explored using a combination of statistical methods and data visualisation. Linear regression analysis will be constructed to examine the relationship between total physical activity accumulated in bouts of varying lengths. All models will control for a range of moderators and mediators (where these have been collected) that have previously been identified in the literature including age, sex, BMI, socio-economic status, highest educational level, smoking status and clinical diagnosis. A moderator has been defined as a variable that explains under what conditions an explanatory variable is related to an outcome variable whilst a mediator explains how or why an outcome variable is related to an explanatory variable. The results from these models will be compared to determine whether the length of activity bouts has an impact on the health outcome of interest. Where possible, linear regression will be used. However, if certain outcomes are only provided in a categorical format, then logistic regression may also be used. Linear regression if preferred as a choice due to the increased data resolution available when handling continuous rather than categorical data.