Study On The Early Warning Value Of Coagulation Indexes In Patients With Sepsis In Medical Information Mart For Intensive Care Database And Data Mining Method Of Deep Burn Patients

BackgroundElectronic health records(EHR)are electronic historical records directly formed in daily health-related activities.HER can be storage and backup expediently and with value for reference.At present multi-channel dynamic and even real-time information collection can be achieved through public health services such as hospitals self-health reporting etc.which can play an important role in the fields of evidence-based medicine public health and drug safety monitoring.Due to the different data formats and various confidentiality requirements in the electronic information files there is still a lack of effective methods to systematically acquire analyze and integrate the massive amounts of data in medical practices and a large amount of valuable data resources are idle and wasted.At present retrospective research on medical data is increasingly used in epidemiology and predictive modeling.How to obtain useful information from massive multi-dimensional medical clinical data to facilitate a deeper understanding of a certain disease and treatment method has becoming the current research hotspot gradually.The Medical Information Mart for Intensive Care database(MIMIC database)is a multi-parameter structured single-center intensive care database published by the Massachusetts Institute of Technology.It records the long-term and comprehensive clinical medical records of the intensive care unit of the Beth Israel Deaconess Medical Center in Boston USA.This database is freely open to the public after deprivation and is available to researchers around the world for academic research.MIMIC-? was released at the end of 2015 and is currently the most used version.It contains the admission records of 49 785 patients and includes the admission records of53 423 ICU patients from 2001 to 2012.The Massachusetts Institute of Technology released MIMIC-IV v0.4 in August 2020 which was updated and improved on the basis of MIMIC-? adopted a modular approach for data structure organization and highlighted data sources for better calling different data sources locally but the updated version of the data is still under development.Severe burn patients have a large wound area a long course of disease and are prone to infection.Sepsis is one of the important causes of death in patients with severe burns.The pathophysiological mechanisms of sepsis include the activation of the systemic coagulation system which leads to the massive consumption of anticoagulant factors the inhibition of the fibrinolytic system the damage of the endothelial system and the uncontrollable inflammatory response.When the coagulation function is severely disturbed it often leads to fatal multiple organ dysfunction and eventually causes death.At present many studies have proved that various coagulation indicators are closely related to the prognosis of sepsis but most of them are single-center small sample studies.At present another research hotspot of clinical medical data mining is the research on the prediction of the outcome of severely ill patients which based on the database with the use of computer or mathematical modeling statistical methods statistical analysis or construction of neural network models etc.has a relatively good prognosis prediction accuracy.Compared with the previous single-center data the MIMIC database has richer structured physiological parameter information provides more comprehensive information on co-variables and is more conducive to singlefactor and multi-factor regression analysis.Therefore the research on the early warning value of early coagulation indicators in patients with sepsis based on the MIMIC database has positive guiding significance for the diagnosis and treatment of sepsis.Deep burns include deep second degree third degree and fourth degree burns with infections.Smaller deep burns require more than 3 weeks to heal.Larger deep burns usually require surgical treatment.It has been observed clinically that burn wounds usually deepen in the early stage and its mechanisms include hypercoagulable state and microthrombosis after burns massive release of inflammatory factors after activation of vascular endothelial cells and release of oxygen free radicals after neutrophils aggregate and proteolytic enzymes.Adequate fluid rehydration anticoagulation therapy and reasonable anti-infection treatment in the early stage of deep burns can effectively improve circulation prevent infection effectively alleviate the deepening of the wound and improve the prognosis.Therefore the study of early clinical indicators and treatment options for deep burns has certain significance for judging the prognosis and providing treatment.The MIMIC database is undoubtedly a better research data that can be used as exploratory data analysis.However there is still no research on the deep burn patients in this database and no discussion on data extraction methods.It is of great clinical significance to explore a set of data mining method to collect various physiological parameters and clinical test results of burn patients.ObjectivesThe 1st Part: Based on the MIMIC-? database we studied the relationship between early coagulation indexes and prognosis of patients with sepsis and judge the early warning value of different indexes.The 2nd Part: Based on MIMIC-? we established a set of data mining methods for deep burn patients extracted the data and performed a preliminary statistical analysis to explore the feasibility of data mining method and analysis of data quality.MethodsThe 1st Part:(1)Extract the data of sepsis patients in the MIMIC-? database;(2)Conduct univariate and multivariate Cox regression tests on the early coagulation data and covariates to study the relationship between early coagulation indicators in sepsis patient's prognostic relationship and draw a survival curve to judge the early warning value of different coagulation indicators.The 2nd Part:(1)After downloading the MIMIC-IV database image we used the Pandas core data analysis library in the Python environment to import and integrate the data.(2)Based on the MIT-LCP/mimic-iv public code in Git Hub we made individualized modifications on several specific parameters required for burn study.And a set of data mining methods for deep burn patients was established suitable for Collecting merging and analyzing data in MIMIC-IV database.(3)Through the use of Pandas + Matplotlib +Seaborn early data of deep burns in the MIMIC-IV database was collected and preliminary analyzed while data visualization is realized.(4)Evaluate the feasibility of this data mining method and discuss the quality of the database.ResultsThe 1st Part:(1)A total of 2794 patients with sepsis who met the inclusion and exclusion criteria were extracted;(2)The relationship between various dependent variables and 28-day survival was determined by analyzing the baseline data.Statistical results showed there are statistical differences in age patterns of admission the 48 hours maximum temperature minimum temperature the lowest oxygen saturation the lowest systolic pressure diastolic pressure and minimum hourly urine output minimum hemoglobin minimum red blood cell volume maximum white blood cell count maximum blood sugar potassium minimum potassium minimum serum sodium minimum blood magnesium maximum serum creatinine maximum blood urea nitrogen maximum Elixhauser score ventilator use renal replacement therapy SAPS? score q SOFA score SOFA score and PT PTT INR and PLT count at the time of admission between the survival group and nonsurvival group of 28-day after admission;(3)The univariate Cox regression and multivariate Cox regression model proved a non-linear relationship between PT PTT INR and PLT at admission and 28-day survival results.(4)PT PTT and INR at admission are risk factors for death in the intensive care unit within 28 days and PLT is a protective factor;(5)PT INR and PLT are more sensitive to the prediction of death in the intensive care unit within28 days and can be used to assess sepsis The severity of the disease and the early warning indicators for judging the prognosis of sepsis.The 2nd Part:(1)Preliminarily established a set of data mining methods running in Python environment suitable for collecting merging and analyzing clinical data of deep burn patients in MIMIC-IV database.(2)By using the data mining method established in this study a total of 106 patients with deep burns severer than 2 degrees were screened out in the MIMIC-IV database.The basic data of the patients and the vital signs laboratory examinations and information such as medication status were collected to perform data visualization through the use of Pandas + Matplotlib + Seaborn.(3)The analysis of the extracted clinical data shows that the clinical data format and early clinical data characteristics of deep burn patients selected by our methods have no obvious logical errors and the results are basically credible which can be used as a methodological basis and for further research.(4)Although the MIMIC-IV database has richer structured physiological parameter information than conventional single-center data it currently has shortcomings such as a relatively low number cases lots of confusing diagnosis and omissions with unclear treatment processes.The scope of application in the field of burns is still relatively limited.ConclusionsThrough extracting clinical data of sepsis patients in the MIMIC-? database we found there is a correlation between the early coagulation indexes of sepsis and the 28-day death in the intensive care unit which can be used as early predictors to assess the severity and prognosis of sepsis.The mining methods established in our research is suitable for collecting merging and analyzing clinical data of deep burn patients in MIMIC-IV database with stable data quality.The data format and clinical features of deep burn patients extracted from the database can be used as the method basis and data reference for further research.MIMIC-? database can be used with exploratory data analysis research but the application of this database in burn field is still very limited.