Multi-index Neural-Network Prediction Of Susceptibility To Acute Mountain Sickness And Application Research

Acute mountain sickness (AMS) is characterized by the presence of headache and atleast one of the following symptoms: loss of appetite, nausea, vomiting, fatigue/weakness,dizziness/light-headedness, and insomnia. It occurs when people arrive at an altitude above2500m, and can be classified into acute mild high-altitude sickness, high-altitudepulmonary edema, and high-altitude cerebral edema. With a high incidence and greatseverity, AMS is the main threat to the health and life of the serviceman who quicklyascend to high-altitude areas and it is also the main cause for their weakened battle force.Studies have showed that susceptibility to AMS exists evidently，but no feasible method isavailable to predict it at present. Therefore, we try to build a prediction model ofsusceptibility to AMS, by combining multiple indices and the basic principles of neuralnetwork method. In this study, a new approach was adopted, which breaks the boundariesbetween disciplines, in order to provide a scientific basis for medical support for servicemenwho quickly ascend to high altitude.Main research contentFirst, the needs analysis of prediction of susceptibility to AMS.Second, the preliminary screening of the risk factors of susceptibility to AMS andMeta-analysis.Third, the research on the relationship between the risk factors of susceptibility toAMS and AMS before exposure to high altitude.Fourth, the research on the building of a neural-network prediction model ofsusceptibility to AMS and its application.Main research methodsIn our study, the following methods were adopted.First, we collected data by searching and reading through relevant literature,investigating the subjects’ basic demographics, experiences of ascending to high altitude,and drinking and smoking histories, and assessing their mental status or symptoms. Second, the method of system analysis was used to build a preliminary index system ofsusceptibility to AMS, according to the principles of evidence-based medicine.Third, field experiments were performed to acquire the data for screening andoptimizing indices, building the model, and verifying the model.Fourth, statistical methods were adopted to optimize the index system.Fifth, the neural network method was used to build the prediction model ofsusceptibility to AMS.Main research resultsFirst，the needs analysis of prediction of susceptibility to AMS.Using statistical methods including descriptive statistics, curve estimation and so on,the relationship between the rate of personnel losses due to AMS and altitude was identified.The rate of personnel losses due to AMS remained high at high altitudes even with labor ofno intensity, and it was positively correlated with altitude. From3,500m up to4,000mabove see level, the rate of personnel losses was in the range of2.64to8.83, while from4,000m upwards, it was in the range of9.72to25.34. Thus, it is necessary to carry outresearch on susceptibility to AMS of the population who suddenly enter high altitude. Withthis undertaking, it is hopeful to find ways to reduce the incidence of AMS and provide ascientific basis for medical support for the mass population who suddenly enter highaltitude.Second, the preliminary screening of the risk factors of susceptibility to AMS andMeta-analysis.The main risk factors of susceptibility to AMS come from four aspects, i.e., physiology,biochemistry, mentality, and gene. Referring to previous studies and applying Meta-analysis,we preliminarily determined18indices to be used for research on susceptibility to AMS,including stature, weight, body mass index (BMI), thoracic cage volume, arterial oxygensaturation (SaO2), blood pressure change after cold stimulation，cortisol of plasma, vitalcapacity (VC), forced vital capacity (FVC), heart rate variability(HRV), mental factors,breathing-holding time, levels of plasma reactive oxygen species, level of plasmanorepinephrine, fraction of exhaled nitric oxide (FENO), fraction of exhaled carbonmonoxide (FECO), smoking and drinking. This selection was also based on facility,maneuverability, and security of experiment on mass population. Third, the research on the relationship between the risk factors of susceptibility toAMS and AMS before exposure to high altitude.A total of314healthy young male recruits were voluntarily enrolled. Before thesubjects ascending to the altitude of4,300m, the data concerning their stature, weight, BMI,thoracic cage volume, SaO2, blood pressure change after cold stimulation，blood samples,VC, FVC, HRV, breath-holding time, FENO and FECO values, demographic factors, anddrinking and smoking history were obtained. Mental health scores were also obtained usinga self-rating anxiety scale, a self-rating depression scale and the symptom checklist90(SCL-90). We stayed with the subjects for the first week after reaching the high altitude, toobtain their Lake Louise Score (LLS) on each day of the first week. In the subjects withLLS>4, headache and at least one other symptom were diagnosed as the symptoms of AMS.The highest LLS of each individual during the first7days were considered as the final LLS.The data showed that the recruits had psychological stress before exposure to high altitude.The total score of anxiety symptoms, depression symptoms, and SCL-90was associatedwith AMS score. The AMS incidence of the subjects with anxiety, depression and SCL-90symptoms was significantly higher than that of those without these symptoms（2=10.944，p <0.05；2=20.355， p <0.05；2=3.987，<0.05）. Both FENO and FECO were foundto be significantly associated with AMS. The AMS group had lower FENO(p=0.003) andFECO(p<0.001) values, and a lower smoking rate (p<0.001) than non-AMS group. MeanFENOand FECOvalues was11.03ppb [95%confidence interval (CI)9.07-12.98] and4.39ppm [95%CI3.76-5.02], respectively, in AMS group, and14.74ppb [95%CI13.25-16.23]and6.10ppm [95%CI5.49-6.72], respectively, in non-AMS group (<0.0001). FECOwas strongly correlated with the variables associated with smoking behavior. The AMSgroup had a significantly lower smoking rate than the non-AMS group (p<0.001). However,in the study, the FENO value of smokers was significantly lower than that of non-smokers(p<0.001). Therefore, FECOvalue was negatively correlated with max-LLS and most-dayLLSs, but it did not indicate that the higher FECOvalue is, the lower the risk of AMS wouldbe. We think the effect of smoking is bidirectional. In some range, FECOhas a protective rolefor AMS. VC was negatively correlated with AMS. As for the other factors, such as age,BMI, HVR, thoracic cage volume, blood pressure change after cold stimulation,breathing-holding time, level of cortisol in plasma, and drinking behavior, etc. were not significantly associated with AMS, but the level of cortisol in plasma significantly increased.So, we think that FENO and FECO values before exposure to high altitude were also riskfactors of susceptibility to AMS.Fourth, the research on the building of a neural-network prediction model ofsusceptibility to AMS and its applicationAccording to the characteristics of AMS susceptibility indices and neural networktheory, we adopted the learning vector quantization (LVQ) and the back propagation (BP)neural network method to build the prediction model of susceptibility to AMS. Using thecomparative method and trial and error method, and with the framework and parameters ofthe network being continually amended, the average correct-prediction precision of theLVQ model ultimately reached72.22%, which can lead to an effective preliminaryscreening of susceptibility to AMS.Based on the above-mentioned results, we applied the LVQ model to the screening ofAMS susceptibility, according to which the medical support project can be designed for themass population who quickly ascend to high altitude areas.