Back

Symptom-based ordinal scale fuzzy clustering of functional gastrointestinal disorders

Mansourian, M.; Marateb, H. R.; Hassanzadeh Keshteli, A.; Daghagh Zadeh, H.; Mananas, M. A.; Adibi, P.

2020-05-18 gastroenterology
10.1101/2020.05.11.20098376 medRxiv
Show abstract

BackgroundThe validity of Rome III criteria for diagnosing functional gastrointestinal disorders (FGIDs) have been frequently questioned in the literature. In epidemiology, when a disease is diagnosed, the existence of a true cluster must be proven. Thus, clustering the common GI symptoms of individuals and comparing the clusters with FGIDs defined by the Rome III criteria could provide insights about the validity of FGIDs defined by those criteria. Well-separated compact clusters were detected in responses to questionnaires of the epidemiological features of different FGIDs in Iranian adults using fuzzy ordinal clustering. The representative sample from each cluster i.e. Cluster Representative (CR) was formed whose corresponding FGID was diagnosed with Rome III criteria. Then, FGID diagnosis was performed for all participants in each cluster and the percentage of cases whose FGID was the same as the clusters identified FGID (agreement) was reported. ResultsFourteen valid clusters were detected in 4763 people. The average membership of the objects in each cluster was 77.3%, indicating similarity of the objects in clusters to their corresponding CRs. Eight clusters were assigned to single FGIDs (irritable bowel syndromes: constipation IBS-C, diarrhea IBS-D and un-subtyped IBS-U; functional bloating FB; functional constipation FC; belching disorder BD. The agreement was higher than 50% in single FGID clusters except those whose diagnosis was IBS-U. ConclusionsIBS-C, IBS-D, FC, BD, and FB defined with Rome III criteria exist in the population, which is not the case for IBS-U.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.