Encephalitis is an acute clinical syndrome of the central nervous system (CNS), often associated with fatal outcome or permanent damage, including cognitive and behavioral impairment, affective disorders and epileptic seizures. It is a rare disease with annual incidence ranging between 3.5 and 7.4 cases per 100,000 persons, worldwide. It is more common in children, the elderly and people with a weakened immune system (e.g. those with HIV/AIDS or cancer).
Although encephalitis has been studied extensively, there have not been any comparisons made among these studies mainly because there have not been any standard statistical methods used to describe and analyze the encephalitis data sets.
The main objective of this paper, therefore, is to perform exploratory cluster analysis using the England human encephalitis data set with the aim of achieving a better understanding of human encephalitis and generate hypotheses about etiology. In particular, we aim to: 1) identify noise variables that have little or no contribution and filter out these variables from the data; 2) identify or determine subgroups of encephalitis; 3) identify major clinical and laboratory features/characteristics associated with encephalitis, and determine which of these variables distinguish a particular cluster from others; 4) identify major risk factors associated with encephalitis.
Research
The research data is related to a prospective multicenter study where 268 patients are recruited from 24 hospitals/neurological centres in three geographical locations (South West, London, North West) across England.
Variable Selection
A total of 209 encephalitis patients and 35 variables (shown in Figure 1) are included in our initial analysis although some are filtered during the variable selection step. Complete linkage hierarchical clustering with the simple matching distance matrix is used in constructing the heatmap. However, similar results were also obtained the Euclidean and Jaccard distance matrices. The 209 by 35 dimensional matrix of encephalitis data set is displayed on the heatmap where each column represents the standardized binary measurements for a given patient. The variables that are similar for more than 80% of patients are excluded (12 variables). Variables for which a significant number of values are missing have also been removed from the analysis. The list includes Abnormal EEG, Fever, Abnormal WCC, PBChange, Abnormal Protein, Headache, Lethargy, Animal Contact, Mosquito Bite, Rash, untreated water etc.
Cluster Analysis
Clustering of our encephalitis data set using agglomerative hierarchical clustering approach revealed six major clusters. An additional small cluster, consisting of 5 encephalitis patients, was also observed. All of these 5 patients had seizures. Abnormal EEG and fever are observed in 4 (80%) of the patients. No other symptom was observed for more than one patient; moreover, most of the exposure variables are zero for these patients. The only exception is water exposure where 3 out of the 5 patients were exposed to untreated water.
The dendogram revealed marked within cluster heterogeneity and subgroups within Cluster 1. Further investigation of this cluster to identify more homogenous sub-clusters might, therefore, be of interest. With a slightly lower cut-off point, Cluster 1 can be subdivided into two sub-clusters. The first sub-group is characterized by abnormal WCC, lethargy, headache and fever; whereas, the major characteristics of the second sub-group are abnormal protein, abnormal WCC, lethargy, headache and gastro-intestinal symptoms. It is important to note that focal neurological abnormalities were observed in all of the patients in the second sub-cluster whereas only a small proportion of patients in the first sub-cluster had this symptom. Animal contact was identified as a major characteristic of sub-group 2, although only a small proportion of patients in the first sub-group were exposed to this risk factor. Another important difference between the two sub-clusters is fever.
Discussions & Conclusion
Hierarchical cluster analysis revealed six major clusters in the England encephalitis data set. An additional small cluster consisting of 5 patients was also observed in the data. It is in general assumed and is common practice to group encephalitis cases according to disease etiology. However, our results indicate that patients are clustered with respect to mainly symptom and laboratory variables rather than causal agents. In fact, disease etiology is distributed across all clusters where the majority of the patients in all the clusters are of unknown etiology. There is some intuitive sense to the clustering which deserves further exploration in relation to clusters of symptoms and laboratory variables which seem to exclude certain causes. Tuberculosis (MTB) only appears in 3 of the 6 clusters; varicella zoster virus (VZV) does not appear in clusters 4 or 5; "other viral" not in 4; acute disseminated encephalomyelitis (ADEM) is not in 5; other bacterial not in 5 or 6; and antibody-mediated encephalitis (ANT) is not found in cluster 6. Furthermore, exposure variables appear to be non-informative towards discriminating among the clusters. These similarities and/or differences with respect to symptom and laboratory measurements might, therefore, be attributed to other factors that are not included in our data set.
Among the exposure variables, however, animal contact and exposure to recent infection have only been identified as possible risk factors. Exposure to untreated water and sick person contact are also linked to encephalitis suggesting that these variables might be potential risk factors. However, further statistical analysis focused more on variable selection is required to identify potential risk factors. This might help researchers to devise new preventive measures.
The main objective of this article is mention the importance of cluster analysis is to identify the variables, sub-groups and observing the impact of the variables in human encephalitis. Through this article I have tried to portray the exact procedure followed in that process and the outcomes of the same post cluster analysis.
Source:
Dipayan Kabiraj
12133
No comments:
Post a Comment