Table 1 Description of the wearable-specific indicators of physical activity behaviour (WIPAB)
WIPAB | Description | Interpretation | References |
|---|---|---|---|
Activity intensity | |||
Intensity gradient | Description of the activity intensity distributions across 24 h | A less negative (higher) intensity gradient reflects more time accumulated across the entire intensity spectrum | Rowlands, Edwardson [20] |
MX metric | Quantification of the average acceleration above which the most active x minutes or hours are accumulated | A higher MX metric indicates a more intense PA behaviour for a defined time period | Rowlands, Dawkins [19] |
Accumulation pattern | |||
Power law exponent alpha | Description of the bout distributions according to their duration for a given activity intensity | A higher power law exponent alpha indicates the accumulation of a certain activity intensity with a greater proportion of shorter bouts | Fortune, Mundell [41] |
Proportion of total time accumulated in bouts longer than x | Proportion of time accumulated in bouts longer than a certain length x | A higher proportion of the total time accumulated in bouts longer than a certain length x, reflects a greater imbalance between the number of bouts and their contribution to the accumulated time at that intensity | Chastin and Granat [8] |
Gini index | Description of the bout length distributions for a given activity intensity | Higher values indicate a greater inequality in bout lengths (e.g. a relatively high proportion of long bout lengths that contribute to the activity pattern), whereas lower values reflect an activity pattern with a high number of mainly short bouts of similar length | Chastin and Granat [8] Ortlieb, Dias [42] |
Temporal correlation and regularity in the time series | |||
Scaling exponent alpha | Detection of temporal correlations in the activity fluctuations by means of the detrended fluctuation analysis (DFA) | Scaling exponent alpha values below 0.5 indicate that the time series is anti-correlated, a value of 0.5 indicates no correlation (“white noise”), and values above 0.5 indicate a positive correlation in fluctuations. Alpha values around 1 indicate the highest temporal correlation in the activity fluctuations | Hu, Van Someren [43] Hu, Riemersma-van der Lek [21] |
Autocorrelation at lag k | Quantification of the degree of relationship between observations that are k lags apart | Autocorrelations coefficients that are closer to 1 or -1 indicate a stronger positive or negative correlation, respectively. Thus, in case of the 24 h autocorrelation, such values would indicate that the timings of the daily activities match perfectly between days or are the exact opposite | Chen, Wu [44] Merilahti and Korhonen [45] Taibi, Price [46] |
Lempel–Ziv complexity (LZC) | Quantification of the diversity of subpatterns as well as the dynamics of change between different subpatterns | Higher LZC values indicate a greater chance of the occurrence of new subpatterns in the numeric sequence and, thus, a more complex temporal behaviour | Aboy, Hornero [22] Paraschiv-Ionescu, Perruchoud [23] |
Sample entropy | Quantification of the degree of regularity in a time series by analysing the presence of different subsequences (patterns). Regularity in a time series indicates that similar patterns are repeated across time | Higher sample entropy indicates increased disorder, thus greater complexity, irregularity and unpredictability in a time series. Lower values imply a more regular time series | Hauge, Berle [24] Krane-Gartiser, Henriksen [26] Krane-Gartiser, Asheim [25] Scott, Vaaler [27] Delgado-Bonal and Marshak [47] |
Symbolic dynamics | Quantification of the complexity of a time series by grouping defined subsequences into different pattern families according to the number and types of variations from one symbol to the next | The rates of occurrences of the four families, expressed as percentage of the total number of patterns analysed, indicates the complexity of the time series | Porta, Guzzetti [29] Guzzetti, Borroni [28] |