Fitness measurement framework

Abstract

A computer-implemented method and corresponding device for measuring an individual's fitness are presented. The method comprises obtaining individual performance data and exercise intensity data for a physical activity performed by the individual; processing the exercise intensity data and population performance data to determine an individual performance model; comparing the individual performance data to the individual performance model to determine an estimated fitness parameter of the individual; processing the estimated fitness parameter using a demographic characteristic of the individual to generate a physiological plausibility parameter; and iteratively re-evaluating the estimated fitness parameter using the physiological plausibility parameter as a soft constraint to obtain a final fitness parameter measuring the fitness level of the individual.

Description

[0001]Fitness measurement framework FIELD OF THE INVENTIONThe invention generally relates to measuring fitness levels, especially by computationalmeans. BACKGROUND TO THE INVENTIONCardiorespiratory fitness, often simply referred to as fitness, is a powerful predictor of all-cause mortality and is associated with a variety of health conditions, includingcardiovascular disease and type 2 diabetes. As a result, there is interest in measuringthe fitness of individuals based on their response to exercise.The gold-standard measurement for fitness is maximal oxygen consumption, referred to as VO2max, which is the fastest rate of oxygen consumption an individual is capable of achieving during physical exertion. VO2max is difficult to measure directly, and so there have been attempts to estimate VO2max indirectly using an individual’s heart rate response to exercise. SUMMARY OF THE INVENTIONMethods of estimating VO2max from exercise heart rate response may rely onmeasurements taken during steady-state exercise. “Steady-state” may be defined as aperiod of at least four minutes where heart rate has stabilised at a given exercise intensity. VO2max is then estimated by linear regression of the exercise intensity and heart rate relationship. When exercise intensity increases linearly, for example during ramped exercise, this method cannot be used to estimate VO2max unless heart rate data are first adjusted using methods specific to the mode of exercise. During non-steady- state exercise, such as a combination of steady-state and self-selected intensities, heartrate data may be segmented into sections that conform to a steady-state relationship forlinear regression analysis. Across all these approaches, distinct analytic techniques would be required to calibrate exercise intensity for different exercise modalities, such as over-ground walking, treadmill exercise, and stationary cycling. The inventors have recognised several issues with the existing paradigm of fitness measurement and estimation. Steady-state exercise can be long and inefficient depending on the number of steady-state intensities, making it impractical for populationswith low exercise tolerance. This causes fitness estimation to be limited to those withhigher exercise tolerance. Ramped exercise, while potentially more time-efficient,presents its own analytical challenges. Heart rate response at a given intensity duringramped exercise is typically lower than heart rate response during steady-state exercise.This leads to potential bias in VO2max estimation. Segmenting steady-state sections ofheart rate data from non-steady-state exercise requires longer exercise durations to ensure enough steady-state periods for analysis. This would limit fitness estimation tothose who can exercise for sustained periods. The need for exercise modality-specificanalytic techniques to calibrate exercise intensity makes comparison of fitness estimatesacross different modalities difficult.Additionally, some approaches may rely on demographic factors such as age, height,weight, and gender to estimate fitness without incorporating information from exercise. While these methods might be useful for broad population-level estimates, they are likely inaccurate for individual-level assessments due to significant variability in fitness levelswithin a given population phenotype.The inventors have therefore developed a method to estimate fitness that departs fromconventional linear regression approaches. The method can accommodate informationcollected during non-steady-state exercise and uses all exercise information without theneed for segmentation. Further, the method allows different exercise modalities to beinterrelated within individual, enabling fitness to be estimated across those modalities.The method also uses modelled relationships between demographic factors and fitnessto refine individual-level estimates.According to an aspect of the disclosure there is presented a computer-implementedmethod of measuring a fitness level of an individual, comprising: obtaining time-seriesexercise data relating to a physical activity performed by the individual, the time-series exercise data comprising individual performance data and exercise intensity data, the exercise intensity data characterising the physical activity and representing an expected demand imposed by the physical activity on a physiology of a subject individual, the individual performance data comprising data for a biomarker that represents a responseof the individual to physical activity; processing the exercise intensity data and populationperformance data to determine an individual performance model characterising an expected biomarker behaviour of an individual undertaking the physical activity, the population performance data comprising expected behaviour of the biomarker duringphysical activity; comparing the individual performance data to the individualperformance model to determine an estimated fitness parameter of the individual; processing the estimated fitness parameter using a demographic characteristic of the individual to generate a physiological plausibility parameter that represents theplausibility of the individual having the estimated fitness parameter; and iteratively re-evaluating the estimated fitness parameter using the physiological plausibility parameter as a soft constraint to obtain a final fitness parameter, wherein the final fitness parameter measures the fitness level of the individual. This method provides the benefit that exercise intensity data may be processed regardless of the exercise modality it arises from, providing a more general indication of an individual’s fitness and allowing more reliable comparison between values obtained from different modalities. In particular, in embodiments, the method may be well suitedto handling both steady state and ramped exercise tasks. Additionally, it can manage acombination of these exercise types, which is common in semi-structured exercise tasks. This flexibility makes fitness estimation more efficient by reducing the exercise time needed compared to steady-state exercise and increasing the range of exercise tasks from which fitness estimates can be derived. The physical activity may comprise using an exercise machine, and the exercise intensitydata may comprise time-series data representing a measure of performance of theexercise machine. This has the advantage that exercise machines provide more detailed and more easilyquantifiable data on exercise tasks compared to an individual’s self-report.The measure of performance of the exercise machine may comprise wattage, speed of operation, or inclination. Determining the individual performance model may comprise: processing the exerciseintensity data to calculate a net energy expenditure of the individual; calibrating the netenergy expenditure to a target performance of the biomarker; and processing the targetperformance of the biomarker to obtain expected time-series data for the biomarker in light of the net energy expenditure. This has the advantage that the time-series biomarker data provides a highly detailedmodel of the individual’s expected performance across different exercise modalities,allowing for a detailed comparison with the individual performance data and therefore a more precise optimisation and fitness measurement. The final fitness parameter may comprise a maximum rate of oxygen consumption, a maximum heart rate, or a calibration coefficient for energy expenditure across different exercise modalities. The demographic characteristic may describe one of age, gender, height, weight, or resting heart rate.This has the advantage that these particular demographic characteristics are readilyavailable, easy to measure, and may provide a strong indication of an individual’sexpected exercise performance.Generating the physiological plausibility parameter may comprise comparing theestimated fitness parameter to a distribution of a corresponding fitness parameter for individuals having the demographic characteristic. The method may further comprise injecting noise into the distribution of the corresponding fitness parameter. This has the advantage of mitigating the effect of sample bias in the distribution, increasing the flexibility and applicability of the distribution and thereby providing a more reliable physiological plausibility parameter. The biomarker may be heart rate. This has the advantage that heart rate is simple to measure and provides a strong indication of an individual’s physiological response to exercise.Iteratively re-evaluating the estimated fitness parameter may comprise using particleswarm optimisation.This has the advantage that particle swarm optimisation is both computationally efficientand parallelisable, allowing optimal fitness estimates to be obtained more rapidly.The method may further comprise storing the final fitness parameter in a memory or transmitting the final fitness parameter to a computing device. Comparing the individual performance data to the individual performance model to determine an estimated fitness parameter of the individual may comprise iteratively revising at least one model parameter until a predetermined level of agreement is reached between the individual performance model and the individual performance data. Iteratively re-evaluating the estimated fitness parameter may comprise: updating thevalue of the estimated fitness parameter; updating the value of a model parameter basedon the updated fitness parameter; recalculating an updated individual performancemodel based on the updated model parameter; comparing the individual performance data to the updated individual performance model to determine an updated fitnessparameter of the individual; and generating an updated physiological plausibilityparameter based on the demographic characteristic and the updated fitness parameter. Using the physiological plausibility parameter as a soft constraint may comprise weighting model parameters such that parameter values closer to an expectation for the demographic characteristic are more favoured. This has the advantage of taking an individual’s demographic or other information into account to provide a more reliable estimate of fitness. According to a further aspect of the disclosure there is presented a computer program comprising instructions that, when executed on a processor of a computing device, cause the processor to execute the above-described method. The instructions may be provided on one or more carriers. For example there may be one or more non-transient memories, e.g. a EEPROM (e.g. a flash memory) a disk, CD- or DVD-ROM, programmed memory such as read-only memory (e.g. for Firmware), one or more transient memories (e.g. RAM), and / or a data carrier(s) such as an optical or electrical signal carrier. The memory / memories may be integrated into a corresponding processing chip and / or separate to the chip. Code (and / or data) to implement embodiments of the present disclosure may comprise source, object or executable code in a conventional programming language (interpreted or compiled) such as C, or assembly code, code for setting up or controlling an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array), or code for a hardware description language. According to a further aspect of the disclosure there is presented a computing devicecomprising a processor configured to: obtain time-series exercise data relating to aphysical activity performed by the individual, the time-series exercise data comprising individual performance data and exercise intensity data, the exercise intensity data characterising the physical activity and representing an expected demand imposed by the physical activity on a physiology of a subject individual, the individual performance data comprising data for a biomarker that represents a response of the individual tophysical activity; process the exercise intensity data and population performance data todetermine an individual performance model characterising an expected biomarker behaviour of an individual undertaking the physical activity, the population performancedata comprising expected behaviour of the biomarker during physical activity; comparethe individual performance data to the individual performance model to determine anestimated fitness parameter of the individual; process the estimated fitness parameterusing a demographic characteristic of the individual to generate a physiological plausibility parameter that represents the plausibility of the individual having theestimated fitness parameter; and iteratively re-evaluate the estimated fitness parameterusing the physiological plausibility parameter as a soft constraint to obtain a final fitness parameter, wherein the final fitness parameter measures the fitness level of the individual. This device shares the advantages of the method described above. The computing device may be a wearable device. This has the advantage that the device may be worn during exercise, such that resultsmay be made conveniently accessible to the individual wherever and whenever theyhave performed an exercise bout (e.g. outside or at a gym).The computing device may be a mobile device.This has the advantage that fitness data may be conveniently gathered, processed, andaccessed by the user while they are out and about, on a mobile device that they may routinely have with them regardless. According to a further aspect of the disclosure there is presented an exercise machine comprising the computing device. These and other aspects will be apparent from the embodiments described in the following. The scope of the present disclosure is not intended to be limited by this summary nor to implementations that necessarily solve any or all of the disadvantages noted. BRIEF DESCRIPTION OF THE DRAWINGS For a better understanding of the present disclosure and to show how embodiments may be put into effect, reference is made to the accompanying drawings in which: Fig.1 shows a system for measuring an individual’s fitness; Fig.2 is a flow diagram illustrating a computational system of measuring an individual’s fitness; Fig.3 is a flow diagram illustrating a computational model of heart rate dynamics; Figs. 4A and 4B show graphs illustrating calibration of exercise intensity to energy expenditure;Fig. 5 shows graphs illustrating heart rate kinetics;Fig.6 is a flow chart illustrating a method of measuring an individual’s fitness; and Fig.7 shows a comparison of the performance of methods described herein to existing methods. DETAILED DESCRIPTIONFig. 1 shows a system 100 for measuring the fitness of an individual 102. The systemcomprises an exercise machine 104 and a computing device 106. The computing device 106 comprises a processor 108, which may be connected to a memory 110 and a communication interface 112. The computing device 106 may be integrated into a wearable device that is worn by the user 102. Alternatively, the computing device 106 may be integrated into the exercise machine 104, or may be an external device in communication with the exercise machine104 and / or a device of the user 102, e.g. via the communication interface 112. Inparticular, methods described herein may be executed on a phone, smartwatch, or other mobile device of the user 102. The exercise machine 104 may be any known type of exercise machine, such as a cycling machine, a treadmill, or a rowing machine. Generally speaking, it is envisioned that the individual 102 wishes to have their fitnesslevel (e.g. their VO2max or maximum heart rate HRmax) measured by the computingdevice 106. This is achieved by the individual performing an exercise task, which may for example involve the exercise machine 104. Data from any or all of the exercise machine 104, the user 102, and / or another computing device associated with the user may then be provided to the computing device 106, for example via the communication interface 112, for measuring fitness. Alternatively the individual 102 may exercise without using an exercise machine 104. For example, the individual 102 may exercise by overground walking. In this case, the individual 102 may provide the computing device 106 with a measure of the speed of walking, and / or a measure of the distance walked and the time taken to do so. Additionally or alternatively, the computing device 106 may comprise a GPS module configured to determine the walking speed of the individual 102. The computing device 106 measures the fitness of the individual 102 by methodsdescribed herein (this may also be referred to as estimating or assessing the fitness ofthe individual 102). Once the computing device 106 determines a fitness parameter of the individual 102, it may report this parameter to the individual 102, and / or may broadcast the fitness parameter via the communication interface 112 to another computing device, such as a computing device belonging to the user 102, a medicalprofessional associated with the user, and / or some other third party. Additionally oralternatively, the computing device 106 may store the determined fitness parameter in the memory 110.Fig.2 shows a schematic flow diagram of a computational framework 200 for measuringthe fitness of an individual such as the individual 102. The framework may, for example, be implemented on the processor 108 of the computing device 106.The framework 200 may be considered as having three major parts as shown in thefigure: a first part 202 that deals with measured exercise data, a second part 204 wherein models are applied to the measured data, and a third part 206 wherein the models are optimised to obtain the best results. At 208, time-series data are obtained for an individual such as the individual 102 performing an exercise task, which may be performed using the exercise machine 104. The time-series data comprises exercise intensity data 210 and a heart rate response of the individual 212. It will be appreciated that, while the framework 200 uses heart rate response, alternative individual performance data may be used. In particular, a response of any biomarker thatrepresents the physiological response of the individual 102 to physical activity may beused instead of heart rate. The exercise intensity data 210 characterises the physical activity performed. For example, the exercise intensity data may include an indication that the individual 102 exercised by using a cycling machine.The exercise intensity data 210 may further comprise modality-specific data, i.e. data ofa form corresponding to the type of exercise performed. For example, if the exercise uses an exercise machine 104, the exercise intensity data 210 may include time-seriesdata of a wattage of the exercise machine 104 recorded as a time series during exercise.Additionally or alternatively, the exercise intensity data 210 may include time-series dataof a speed or inclination of a treadmill and / or a stair climbing cadence of the individual102. As a further example, if the exercise is overground walking, the exercise intensitydata 210 may include time-series data of a speed of walking, and / or a distance walkedtogether with the time taken to walk that distance. The preceding examples may beconsidered “indirect” measures of exercise intensity, as the physiological response of the individual 102 is not directly measured.Additionally or alternatively, the exercise intensity data 210 may include a “direct”measure of exercise intensity obtained by measuring a physiological response of the individual 102 to the exercise. For example, the exercise intensity data 210 may includetime-series data of respiratory gas exchange of the individual 102 during exercise, or anyother physiological measurement that is characteristic of demand imposed by the exercise activity on the individual 102.In some cases, there may be gaps or omissions in time series data of the exerciseintensity data 210 where data is not recorded for particular times or time intervals. For example, such gaps or omissions may arise due to measurement errors or data corruption. In some embodiments these gaps or omissions may interpolated prior to further processing of the exercise intensity data 210. For example, gaps in the exerciseintensity data 210 may be interpolated by linear interpolation, Gaussian processregression, or any other appropriate technique. The heart rate response data 212 may be a time-series measurement of the individual’sheart rate over the time when the exercise task was performed and during exerciserecovery. Preferably, the heart rate response data 212 can be correlated with the exercise intensity data 210 so that heart rate and exercise intensity can be determined at each moment of exercise. The exercise intensity data 210 is provided to a first model 214 that calibrates the exercise intensity data 210 to a net energy expenditure of the individual 102. That is, the first model 214 determines, based on the exercise intensity data 210, how much energy the individual 102 expended during the exercise task. The net energy expenditure is provided to a second model 216 that determines a targetheart rate, which is the heart rate that would be expected to be attained by an individual performing exercise at the net energy expenditure determined by the first model 214. Determining the target heart rate requires making some assumption about the fitness of the individual 102, for example by assuming a value for at least one model parameter. This model parameter may be a fitness parameter of the individual 102 such as VO2maxand HRmax. The model parameter used may not initially be specific to the individual 102,but may, for example, be based on general demographic data, or on a presumed average individual. The target heart rate is provided to a third model 218 that models heart rate dynamics toobtain a time series 220 of an individual’s expected heart rate during the exercise task.This time series 220 is otherwise referred to herein as an individual performance model220. As with the second model 216, obtaining the individual performance model 220requires model parameters that implicitly assume aspects of the individual 102, and which initially may be based on an average individual and / or population data. The third model 218 may also take account of historic individual-level data, such as historic heart- rate response data of the individual 102, in obtaining the individual performance model 220. At step 222, the individual performance model 220 is compared to the heart rateresponse data 212. If the model parameters used by the three models 214, 216, 218were a perfect description of the individual 102, the individual performance model 220 would be identical to the heart rate response data 212. However, it is generallyenvisioned that this will not initially be the case. Consequently, at step 224 the modelparameters are adjusted (estimated, optimised) to obtain a new set of estimated model parameters 226 intended to give a closer fit to the heart rate response data 212. The revised model parameters 226 are provided to the three models 214, 216, 218, which repeat their respective determinations to produce a new individual performance model 220 for comparison 222 and further optimisation 224. This process may be repeated until a certain required level of agreement is reached between the individual performance model 220 and the heart rate response data 212. The optimisation at step 224 may be performed using particle swarm optimisation. The step of parameter estimation and optimisation 224 is further informed by individual- level characteristics 228 (otherwise referred to herein as individual characteristics or demographic characteristics) which describe at least one characteristic of the individual102. For example, the individual-level characteristics 228 may include any or all of theindividual’s age, gender, height, weight, or resting heart rate.The individual characteristics 228 are used at step 230 to obtain a measure of theplausibility of the current values of the model parameters determined at step 224, otherwise referred to herein as a physiological plausibility parameter. For example, aparticular value V0 of VO2max may be average for a 25-year-old, but exceptionally highfor a 70-year-old. If V0 is the current estimate of VO2max being used by the models 214,216, 218, this value may be accepted if the individual 102 is 25 years old, but revised to a lower value if the individual 102 is 70 years old. As part of the determination 230 of the physiological plausibility parameter, a distribution of expected values of particular model parameters in a population of individuals having particular demographic characteristics may be used. For example, a distribution of values of VO2max among individuals of a particular age may be used to reach the conclusions described in the preceding paragraph. Such a distribution may be generalised using noise injection techniques as described above, thereby increasing theflexibility and applicability of the distribution. For example, noise may be injected into adistribution of parameters by adding random noise to a mean of the distribution and / orapplying covariant shrinkage to a covariance of the distribution. This may mitigate theeffects of any bias in the initial distribution, thereby allowing subsequent optimisation to arrive at a more generally applicable result. In examples where a multivariate distribution is used, this technique may be generalised by adding random noise to a mean vector of the distribution, and applying covariant shrinkage to a covariance matrix of the distribution to bring the covariance matrix closer to the identity matrix.In embodiments, the physiological plausibility parameter may be used as a soft constrainton the optimisation step 224, resulting in the estimated model parameters 226 beingcloser to what would be expected for a population phenotype represented by theindividual characteristics 228. In other words, the physiological plausibility parametermay be used to encourage fitness parameter estimates to be closer to values expected for a given population phenotype, iteratively re-evaluating the estimated fitnessparameters 226 until the optimisation step 224 converges to an optimal solution.In particular, in embodiments the soft constraint may be a penalisation factor that issubtracted from a main objective function at step 222. This may preferentially weightmodel parameters that are closer to an expectation for a given set of demographic characteristics. Once a set of model parameters is found which provides a high enough level of agreement between the heart rate response data 212 and the individual performance model 220 to meet a predetermined threshold, it is determined that the model parameters provide a sufficiently accurate description of the individual 102. A fitness parameter suchas VO2max that has been optimised as a model parameter may then be taken as theindividual’s estimated VO2max, providing a measure of the fitness of the individual 102. Once a final value has been reached for the fitness parameter in this manner, the value may be communicated to the individual 102, or may be used in any other way. For example, the value may be stored in the memory 110 of the computing device 106, and / or may be transmitted to another computing device, such as a computing device belonging to a healthcare professional. Additionally or alternatively, the value may bestored in cloud storage. In embodiments, fitness values for a particular individual 102measured at different times and / or by different devices 106 may be aggregated by a webportal or computer application for ease of viewing.Heart rate dynamics may be described by the following set of coupled differentialequations: Δ^^^^^ = ^^^^^^^ − ^^^^^^^^^^^^^^ = ^^^Δ^^^^^^^^^^ − ^^^^^^^In these equations, HR represents a current heart rate of the individual 102, while ΔHRis a change in this heart rate. HRtarget is a target heart rate given by ^^^^^^^^^^= + ^^^^^^^, where HRmin and HRmax indicate the individual’s minimum and maximum heart rates respectively, VO2max is the VO2max of the individual 102 as defined above scaledby total body mass, and Enet is net energy expenditure. Din, Dout, c1, c2, and c3 are time-dependent model parameters. The time dependences of Din and Dout are shown in the equations, while the other parameters are given by: ^^^ = ^^^^^^^^^^^^ = (1 − ^^^)^^^^^^ + ^^^^^^ = ^^^^^^^^^where γ1, γ2, and γ3 are constant coefficients and HRR is fractional heart rate reservegiven These equations may be used to propagate a change in net energy expenditure (Enet) to a change in heart rate (ΔHR).Fig. 3 is a conceptual flow chart illustrating the behaviour of the above model of heartrate dynamics.At 302, an incoming signal (the change in net energy expenditure Enet) is converted to achange in target heart rate (HRtarget) at the signal translation stage.The translated signal is combined at an adder 304 with a delayed copy of the signal produced by a signal delay stage 306. The delayed signal is fed into a further adder 308to be combined with the output of a signal decay stage 310, which decays the signal bya fixed proportion for each time interval. The output of the adder 308 is fed into amultiplication stage 314. The other input of the multiplication stage 314 is from asubtraction stage 312, which accepts the target heart rate obtained from the second model 216 as one input, and subtracts from this the current heart rate (obtained as indicated below).The output of the multiplication stage 314 is the change in heart rate arising from thechange in exercise intensity, which is added to the current heart rate at a further adder 316. The output of this adder 316 is the new current heart rate, which is fed back into the signal translation stage 302, signal delay stage 306, signal decay stage 310, and theadders 312 and 316. The responsiveness of each of the signal translation stage 302, thesignal delay stage 306, and the signal decay stage 310 is thus controlled in proportionto the current heart rate output by the adder 316.This model allows heart rate kinetics to emerge in response to a change in exerciseintensity (and therefore net energy expenditure Enet).Figs.4A and 4B show a series of plots 402, 404, 406, 408, 410, 412 demonstrating the calibration of exercise intensity to net energy expenditure performed by the first model 214 in a simulated example. Fig. 4A shows three plots 402, 404, 406 that are examples of time-series exerciseintensity data. In 402 and 404 exercise intensity is expressed on the vertical axis aswattage of a cycle ergometer (an example of an exercise machine 104), with time inminutes on the horizontal axis. Plot 402 shows a steady-state cycle wherein exerciseintensity is stepped between fixed levels, spending four minutes at each step. Plot 404shows a ramped cycle wherein exercise intensity increases linearly with two different gradients. Finally, plot 406 shows treadmill exercise where treadmill speed (solid line)and treadmill fractional grade (dotted line) change at different times during the exercisebout. Fig.4B shows three plots 408, 410, 412 which show simulated net energy expenditures of an individual completing the exercise tasks represented by plots 402, 404, 406 respectively. Net energy expenditure is represented as rate of oxygen consumption. It will be seen that, generally speaking, net energy expenditure is proportional to exerciseintensity in accordance with individualised model parameters estimated for the firstmodel 214. In particular, plot 412 illustrates how two different expressions of treadmillexercise intensity (speed and grade) may be harmonised into a single expression of net energy expenditure. In embodiments, such harmonisation can allow differing exercise modalities to be interrelated.Fig.5 shows three plots 502, 504, 506 of heart rate response data 212 (represented bypoints) taken from an individual completing the three respective exercise tasks represented by plots 402, 404, and 406, plotted against time in minutes. Each graph alsoshows an individual performance model 220 (represented by a curve) generated by thethree models 214, 216, 218 using the methods described herein.Fig. 6 shows an exemplary method 600 of measuring an individual’s fitness level basedon the disclosure herein. For example, the method 600 may be implemented on the processor 108. At step S602, the method 600 comprises obtaining time-series exercise data relating toa physical activity performed by an individual, the time-series exercise data comprisingindividual performance data and exercise intensity data, the exercise intensity data characterising the physical activity and representing an expected demand imposed by the physical activity on a physiology of a subject individual, the individual performance data comprising data for a biomarker that represents a response of the individual to physical activity. For example, the physical activity may be performed by the individual 102 using theexercise machine 104. The individual performance data may be the measured exerciseheart rate response 212, and the exercise intensity data may be the exercise intensitydata 210. At step S604, the method 600 comprises processing the exercise intensity data and population performance data to determine an individual performance model characterising an expected biomarker behaviour of an individual undertaking the physical activity, the population performance data comprising expected behaviour of the biomarker during physical activity.For example, the individual performance model may be the individual performance model220 obtained using the three models 214, 216, 218 as described herein.At step S606, the method 600 comprises comparing the individual performance data tothe individual performance model to determine an estimated fitness parameter of the individual. For example, this may correspond to the comparison step 222 described above with reference to Fig.2. At step S608, the method 600 comprises processing the estimated fitness parameter using a demographic characteristic of the individual to generate a physiological plausibility parameter that represents the plausibility of the individual having the estimated fitness parameter. For example, the demographic characteristic may correspond to the individual-levelcharacteristics 228 described above with reference to Fig.2.At step S610, the method 600 comprises iteratively re-evaluating the estimated fitness parameter using the physiological plausibility parameter as a soft constraint to obtain a final fitness parameter, wherein the final fitness parameter measures the fitness level of the individual.For example, this may correspond to the optimisation step 224 described above withreference to Fig.2 Fig. 7 shows three graphs 702, 704, 706 providing examples of the performance of methods described herein (referred to as “framework”) compared to two existingmethods, Alternative Method 1 (AM1) and Alternative Method 2 (AM2). AlternativeMethods 1 and 2 calibrate exercise intensity differently from each other, however bothestimate fitness by applying linear regression to an assumed steady-state relationshipbetween heart rate and exercise intensity. In models adjusted for age, sex, and resting heart rate, one-standard deviation higher VO2max estimated using methods disclosed herein was associated with 10% lower 2-hour glucose 702, 37% lower fasting insulin 704, and 9.4% higher high-densitylipoprotein (HDL) cholesterol 706. By contrast, one-standard deviation higher VO2maxestimated using Alternative Method 2 was associated with 4.4% lower 2-hour glucose702, 12% lower fasting insulin 704, and 2.6% higher HDL 706. Methods disclosed hereintherefore show the strongest association with each of these key cardiometabolic riskfactors. Although the subject matter has been described in language specific to structural features and / or methodological acts, it is to be understood that the subject matter definedin the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims.

Claims

CLAIMS 1. A computer-implemented method of measuring a fitness level of an individual, comprising: obtaining time-series exercise data relating to a physical activity performed by the individual, the time-series exercise data comprising individual performance data and exercise intensity data, the exercise intensity data characterising the physical activity and representing an expected demand imposed by the physical activity on a physiology of a subject individual, the individual performance data comprising data for a biomarker that represents a response of the individual to physical activity; processing the exercise intensity data and population performance data to determine an individual performance model characterising an expected biomarker behaviour of an individual undertaking the physical activity, the population performance data comprising expected behaviour of the biomarker during physical activity; comparing the individual performance data to the individual performance model to determine an estimated fitness parameter of the individual; processing the estimated fitness parameter using a demographic characteristic of the individual to generate a physiological plausibility parameter that represents theplausibility of the individual having the estimated fitness parameter; anditeratively re-evaluating the estimated fitness parameter using the physiologicalplausibility parameter as a soft constraint to obtain a final fitness parameter, wherein thefinal fitness parameter measures the fitness level of the individual.

2. The method of claim 1, wherein the physical activity comprises using an exercisemachine and the exercise intensity data comprises time-series data representing ameasure of performance of the exercise machine.

3. The method of claim 1 or 2, wherein the measure of performance of the exercisemachine comprises wattage, speed of operation, or inclination.

4. The method of any preceding claim, wherein determining the individual performancemodel comprises: processing the exercise intensity data to calculate a net energy expenditure ofthe individual;calibrating the net energy expenditure to a target performance of the biomarker; and processing the target performance of the biomarker to obtain expected time-series data for the biomarker in light of the net energy expenditure.

5. The method of any preceding claim, wherein the final fitness parameter comprises amaximum rate of oxygen consumption, a maximum heart rate, or a calibration coefficientfor energy expenditure across different exercise modalities.

6. The method of any preceding claim, wherein the demographic characteristic describes one of age, gender, height, weight, or resting heart rate.

7. The method of any preceding claim, wherein generating the physiological plausibility parameter comprises comparing the estimated fitness parameter to a distribution of a corresponding fitness parameter for individuals having the demographic characteristic.

8. The method of claim 7, further comprising injecting noise into the distribution of the corresponding fitness parameter.

9. The method of any preceding claim, wherein the biomarker is heart rate.

10. The method of any preceding claim, wherein iteratively re-evaluating the estimated fitness parameter comprises using particle swarm optimisation.

11. The method of any preceding claim, further comprising storing the final fitnessparameter in a memory or transmitting the final fitness parameter to a computing device.

12. The method of any preceding claim, wherein comparing the individual performance data to the individual performance model to determine an estimated fitness parameter of the individual comprises iteratively revising at least one model parameter until a predetermined level of agreement is reached between the individual performance model and the individual performance data.

13. The method of any preceding claim, wherein iteratively re-evaluating the estimated fitness parameter comprises:updating the value of the estimated fitness parameter; updating the value of a model parameter based on the updated fitness parameter; recalculating an updated individual performance model based on the updated model parameter; comparing the individual performance data to the updated individual performance model to determine an updated fitness parameter of the individual; and generating an updated physiological plausibility parameter based on the demographic characteristic and the updated fitness parameter.

14. The method of any preceding claim, wherein using the physiological plausibilityparameter as a soft constraint comprises weighting model parameters such thatparameter values closer to an expectation for the demographic characteristic are more favoured.

15. A computer program comprising instructions that, when executed on a processor of a computing device, cause the processor to execute the method of any preceding claim.

16. A computing device comprising a processor configured to: obtain time-series exercise data relating to a physical activity performed by the individual, the time-series exercise data comprising individual performance data and exercise intensity data, the exercise intensity data characterising the physical activity and representing an expected demand imposed by the physical activity on a physiology of a subject individual, the individual performance data comprising data for a biomarker that represents a response of the individual to physical activity; process the exercise intensity data and population performance data to determine an individual performance model characterising an expected biomarker behaviour of an individual undertaking the physical activity, the population performance data comprising expected behaviour of the biomarker during physical activity; compare the individual performance data to the individual performance model to determine an estimated fitness parameter of the individual; process the estimated fitness parameter using a demographic characteristic of the individual to generate a physiological plausibility parameter that represents theplausibility of the individual having the estimated fitness parameter; anditeratively re-evaluate the estimated fitness parameter using the physiologicalplausibility parameter as a soft constraint to obtain a final fitness parameter, wherein thefinal fitness parameter measures the fitness level of the individual.

17. The computing device of claim 16, wherein the computing device is a wearable device.

18. The computing device of claim 16, wherein the computing device is a mobile device.

19. An exercise machine comprising the computing device of claim 16.

Images