Mining And Analysis Of The Data In Severe Burn Patients Based On The Big Data Research Pattern

BackgroundSevere burn is a kind of serious and critical trauma. It is difficult to treat due to its dangerous conditions and complex complications. The most frequency complications of severe burn are sepsis and multiple organ dysfunction syndrome(MODS), which are the primary cause leading to death. Recent studies found that severe sepsis or septic shock are the essence of MODS. Early diagnosis and advanced warning of septic shock or MODS are the key issues in the clinical treatment of severe burn patients. With the development of information technology, big data is gradually becoming one of the important methods in medicine field, especially for clinical medical research. Big data technology could analyze and manage the complex and multifarious clinical data. This may help us to explore the potential rules or changing pattern of severe burn patients and benefit the clinical decisions.ObjectiveMining and analysis of the data in severe burn patients based on the basic idea and technology of big data research. To explore the model characteristics of clinical data in severe burn patients and looking for the prediction model of sepsis shock.MethodsClinical data of 107 severe burn patients has been collected through the electronic medical record system and papery medical records. After the data preprocessing, unsupervised algorithm such as hierarchical clustering or principal component analysis and hierarchical clustering(HCPC) have been used to analyze the data. Then the relationship between the classification and clinical outcome were explored. The centroid algorithm was used for data dimension reduction. On this basis, we extracted the changing trends of centroid before sepsis shock and then established the forecast model of sepsis shock. R i386 3.1.2, Matlab 7. 0, SPSS 18.0 were used for statistical analysis and mathematical calculation. The main variables was filtered by comparing the F value and the prediction model was established by linear random effects model. Finally, the prediction model was made into softwares which were developed by the C programming language.Results:1. The total clinical data acquisition phase points of severe burn patients were 2257, with 58 observation indexes for each phase. And the total amount of data points were 130906.2. The clinical data of severe burn patients could divided into 10 types by HCPC method.3. Urine volume, temperature, diastolic pressure, systolic pressure, respiratory rate, BUN, CR, TBIL, lactate, PLT, CK, p O2, p CO2, HCO3-, oxygenation et al. were the most obvious and important indicators for sepsis shock. The classification criteria could be simplified using the above 15 indicators.4. There is a close relationship between cluster grouping and clinical outcome. The patients in cluster1, cluster2, cluster4, cluster8, cluster9 and cluster10 tended to get better with the respective probability of 91.00%, 89.70%, 60.50%, 66.40%, 71.70% and 80.70%. The patients in cluster3 and cluster5 tended to death with the respective probability of 70.50% and 82.10%. The clinical tendency of patients in cluster6 and cluster7 was sepsis shock with the respective probability of 77.90% and 70.40%.5. The overall variation tendency of the center of mass is opposite for patients with different endings. It gradually rised when the patients tended to get better and declined when the patients was dying.6. By centroiding algorithm and linear random effects model, the prediction model for sepsis shock was established:yij=(0.2527280+ N(0,6.450e-05))+(-0.0251963+ N(0,1.273e-05))x +ijeije is a random variable,ijy is center of mass(CM) for patient i in time j,x is the hospital stays.(0.2527280+ N(0,6.450e-05)) is intercept and(-0.0251963+ N(0,1.273e-05)) is slope of the model. These two sets of data obey the normal distribution.7. The sensitivity, specificity and overall diagnostic accuracy of the model were 75.8%, 67.3% and 78.5% respectively.8. For ease of use in clinical practice, we have developed a kind of relevant software by using the model as the core. There are two types of the software: stand-alone version and online version. The stand-alone version could extract clinical data through electronic medical system. The using of online version is simple and convenient for clinicians.Conclusion:1. The clinical data of severe burn patients could be classified by cluster analysis method. There is a close relationship between different classification and clinical outcome. Through the real-time classification of patients using clinical data, we could predict the prognosis.2. We could simplify the clinical data by the big data technology and 58 kinds of clinical data were replaced by 15 kinds of clinical data.3. The dimensions of clinical data could be reduced by centroid algorithm. There is a kind of characteristic changes in center of mass before sepsis shock. The risk of sepsis shock could be predicted through this trend combining with computer software. In addition, the trend of center of mass associated with clinical outcome closely.