Estimation models of human core temperature and metabolic rate for non-invasive real-time monitoring in the occupational field

Climate change is a very actual and relevant issue because it increasingly affects both everyday and working life. Due to climate change, a number of climatic events are increasing in frequency and intensity, and among these, heat waves are a phenomenon that is becoming a particularly serious problem in workplaces that expose workers to heat. The effects of extreme temperatures are dangerous both for outdoor workers and for those working in indoor workplaces where environmental factors cannot be managed. Exposure to extremely hot working environments increases the risk of developing heat strain and is also associated with occupational injuries as well as heat-related health problems, including heat cramps, heat exhaustion, heat stroke and, in the most severe circumstances, even death. Therefore, finding solutions to identify the onset of possible hyperthermia conditions is crucial. In this circumstance, real-time monitoring of the worker's thermal state can be a possible approach. The physiological parameter that is generally considered as an indicator of the thermal state of the human body is core temperature (CT). Another parameter that is useful to take in consideration is the metabolic rate (M) which provides an indication to the intensity of activity. Heavy activities, in fact, imply an increase in internal heat production, which may enhance CT and so contribute to the onset of heat strain. In the case of both CT and M, the measurement methods considered to be gold standards are not feasible for monitoring these parameters directly in the workplace and it is therefore necessary to obtain them indirectly. A more feasible option is therefore to estimate CT and M by using other correlated physiological parameters that may be monitored in a simple and non-invasive way. The general aim of this PhD thesis is to identify new non-invasive techniques that allow accurate and real-time monitoring of CT and M in the occupational field. Two specific objectives were identified to meet the general aim, which involved the development of two different estimation models, one for CT and one for M, based on other physiological parameters that are easier to measure on workers while they perform work activities. In order to achieve the first specific aim, the main real-time CT estimation methods potentially applicable for CT monitoring in the occupational field were identified in the literature. Among these, the model developed by Buller et al. in 2013 that estimates CT from heart rate (HR) measurements with the Kalman filter (KF) appeared to have more appropriate characteristics for occupational application, although it could be improved in some aspects in order to be effectively adopted with a population of real workers and for real work activities. The model was tested in a preliminary study on a sample of middle-aged female workers while simulating real work activities. This study highlighted some features of the model that could potentially be improved for use in the workplace. In fact, the model of Buller et al. is based on an experimental database collected on young and fit male soldiers, whereas the vast majority of workers do not belong to this category. It also uses a single CT (HR) relation to estimate CT in both the rise and fall phases, whereas the latter due to its hysteretic nature behaves differently in the two phases. This PhD thesis presents a new CT estimation algorithm, called Biphasic Kalman-filter-based (BKFB) model, based on HR measurement using the KF, with characteristics more geared to application in the occupational field. Thirteen healthy subjects (six females and seven males) were included in the study to perform three consecutive tasks simulating work activities. These subjects were recruited to provide a more realistic sample of the existing population of workers (heavily weighted towards middle-aged, not very fit and with a considerable fraction of female workers). During each test, an ingestible core body temperature sensor was used to measure CT values, and a heart rate sensor placed on the chest with a band to measure HR. The new BKFB model was developed using the KF method with these data. The BKFB model has a biphasic structure, using two different models (based on KF) during the increasing and decreasing phases of CT, with the capability of automatically switching between one and the other. The results obtained from the CT estimates were compared with the measured CT value and the BKFB model achieved very promising results in terms of overall Root mean square error (RMSE) (0.28 ± 0.12 °C) and squared sum of error (SSE) (0.10 ± 0.07 °C), both lower in comparison with the same ones obtained using the Buller et al. model (SSE =0.14 ± 0.14 °C and RMSE=0.34 ± 0.16 °C). The BKFB model provides CT estimates that are reasonably comparable to the data obtained from measuring intra-abdominal temperature but is more practical and easier to use for a real-time monitoring system of the workers' thermal state. To achieve the second objective of this PhD thesis, an innovative strategy for estimating M based on the measurement of one or more human body accelerations was identified in the scientific literature. Studies presenting models for estimating M using acceleration data has already been published in the scientific literature but focused more on daily living activities or socially relevant activities such as walking, running, and cycling, while to estimate M in the occupational field, a specific calibrated model on occupationally relevant activities is required. Ten healthy subjects (5 males, 5 females) were enrolled in six tasks simulating specific work activities. During the tests, oxygen consumption rate (VO2), carbon dioxide production rate (VCO2) and accelerations at three body positions were measured. VO2 and VCO2 were used to obtain M and triaxial accelerations to calculate vector magnitude (VM). Three mono-linear zero-intercept regression models based on center-of-mass VM were developed. A Global model was obtained by applying a zero-intercept functional relation between M and VM to the whole data set. Two other models (Vertical and Horizontal) were also developed by separating tasks with different kinematics. The results revealed a low R2 for the Global model (R2=0.30) due to collecting in the same sample many tasks that have little in common. A large R2 improvement is achieved when the dominant direction of movement is considered (R2 = 0.56 for Vertical model and R2 = 0.61 for Horizontal models). The proposed method is based on a non-invasive easy- to-measure physiological parameter and can be used to track the activity level during work activity. In conclusion, both these new techniques for estimating CT and M can be employed for a non-invasive and reasonably accurate real-time monitoring of these two parameters in the occupational field.