Key Issues And Technology Of Latent Variable Modeling In Clinical Medical Data Research

With the rapid development of digital medical data analysis techniques and strong demands of clinical diagnosis and teaching, clinical data analysis techniques have become a hot topic in the field of clinical medicine. Clinical medicine is a kind of science which mainly researches the reason, diagnosis, treatment and prognosis of diseases, and improves clinical treatment and promotes human health. Clinical medical data can well reflect the symptoms and signs of human complex systems, which are widely present in many disciplines such as surgery, obstetrics and gynecology, oncology and so on. Currently, how to effectively extract the relevant important information from clinical medical data to analyze has become a significant topic for majority of health workers and researchers.In the field of clinical medical research, some often related variables that are latent variables can’t be directly observed. These latent variables have abstract concepts and cannot be accurately measured for various reasons. At present, latent variables are used in a variety of disciplines and more and more applied in the clinical field. In this paper, patients with non-small cell lung cancer and pregnant women during pregnancy are used for the study. Without limiting interventions, firstly we studied cross-sectional clinical data by complex system entropy partition method, and then built latent variable model based on cross-sectional clinical data via analyzing patients’ clinical signs and digging out their existing latent variables through clinicians’ medical conclusions.Based on these work, we explored the latent variable modeling method based on longitudinal clinical data. Finally, we summarized and prospected the research work. Explorations and researches are as follows:(1) Based on the investigation and analysis of clinical data characteristics, we proposed the overall research program of latent variable modeling, described related clinical data characteristics and preliminarily analyzed data.(2) According to the characteristics of complex human systems and clinical data, we applied the complex system entropy partition method on the cross-sectional clinical data, and proposed complex system entropy partition method with prior knowledge to reduce the complexity of partition method. Then we effectively partitioned combinations of different symptoms for clinical data, and excavated reasonable latent variables of clinical data by combining with the knowledge of clinical medicine that provided some reasonable medical explanation.(3) On the basis of a complex system entropy partition and characteristics of cross-sectional clinical data, we firstly built latent variable model of clinical data via one-order confirmatory factor analysis. Next we analyzed the correlation between latent variables and symptom signs, and verified the validity of entropy partition method. And then we proposed two-order confirmatory factor analysis latent variable modeling for TCM(Traditional Chinese Medicine) clinical data, and analyzed inherent relationship between two levels of latent variables in non-small cell lung cancer data. The results are not basically contradictory with the professional clinicians’ conclusions so that they can provide some reasonable medical explanations and treatment strategies support for doctors.(4) Based on latent variable modeling of cross-sectional clinical data, we analyzed the limitations of latent variable growth curve model on longitudinal clinical data. Then we proposed the latent variable modeling method for longitudinal clinical data by analyzing their characteristics. Firstly, we analyzed and compared changing trends of latent variables and individual differences via building models of survival and dead patients for non-small cell lung longitudinal clinical data. In addition, we carried out a latent variable modeling analysis for three groups’ longitudinal clinical data of pregnancy-induced hypertension. The results showed the proposed method’s reasonable applicability in longitudinal clinical data and provided some support for doctors’ diagnosis and treatment strategies.