Topic modelling of preterm birth

ELIAS, DARIO E; GIMENEZ LG; KRUPITZKI HB; PIOLA M; POLETTA FA; LOPEZ CAMELO JS

Mendoza

Congreso; 10th Argentinian Congress of Bioinformatics and Computational Biology; 2019

La Nave Cultural, Universidad Nacional de Cuyo

Background Preterm birth is defined as birthbefore 37 gestation weeks. It is the leading cause of perinatal morbimortalityworldwide. It is estimated that 10% of all births in the world are preterm.Nowaday preterm birth etiology is unknown, it is considered to be amultifactorial phenomenon. The aim of this work is to describe the mainconcerns of preterm birth and its temporal variation with a topic modellingapproach.Results In this study 57 thousand abstractslinked to preterm birth were analyzed, published between 1970 and 2018. LatentDirichlet allocation was used for modeling topics from 1292 unique bigrams ofnoun phrases. The topic number that presented the least perplexity was 110.Each topic was labeled with the terms that showed the greatest posteriorprobability. Moreover, the individual variationof each topic was analyzed, for example, the topics linked to molecularanalysis (ej. polymerase chain reaction, growth factor, genetic factor) had amedian prevalence of 0.02, they showed a growth from 1995 reaching the maximum(0.03) between 2005 and 2010. On the other hand, the median prevalence of sociodemographictopics was 0.04, in 1979 they presented their maximum prevalence (0.15) anddecreases reaching the minimum (0.004) in 2017. This decrease could beassociated to sociotechnological changes and to topic number increase. Conclusions The results suggest that the mainconcerns about preterm birth would be associated with issues of epidemiologicalsurveillance, prematurity consequences, risk factors and analytical andmolecular methods. Furthermore, concerns would have varied over time, whichwould correspond to technological advance.