Understanding the importance of key risk factors in predicting chronic bronchitic symptoms using a machine learning approach

Abstract

Background Chronic respiratory symptoms involving bronchitis, cough and phlegm in children are underappreciated but pose a significant public health burden. Efforts for prevention and management could be supported by an understanding of the relative importance of determinants, including environmental exposures. Thus, we aim to develop a prediction model for bronchitic symptoms. Methods Schoolchildren from the population-based southern California Children’s Health Study were visited annually from 2003 to 2012. Bronchitic symptoms over the prior 12 months were assessed by questionnaire. A gradient boosting model was fit using groups of risk factors (including traffic/air pollution exposures) for all children and by asthma status. Training data consisted of one observation per participant in a random study year (for 50% of participants). Validation data consisted of: (1) a random (later) year in the same participants ( within -participant); (2) a random year in participants excluded from the training data ( across -participant). Results At baseline, 13.2% of children had asthma and 18.1% reported bronchitic symptoms. Models performed similarly within- and across-participant. Previous year symptoms/medication use provided much of the predictive ability (across-participant area under the receiver operating characteristic curve (AUC): 0.76 vs 0.78 for all risk factors, in all participants). Traffic/air pollution exposures added modestly to prediction as did body mass index percentile, age and parent stress. Conclusions Regardless of asthma status, previous symptoms were the most important predictors of current symptoms. Traffic/air pollution variables contribute modest predictive information, but impact large populations. Methods proposed here could be generalized to personalized exacerbation predictions in future longitudinal studies to support targeted prevention efforts. Electronic supplementary material The online version of this article (10.1186/s12874-019-0708-x) contains supplementary material, which is available to authorized users.

Prediction of Tobacco Leave Grades with Ensemble Machine Learning Methods

“…The lower MSE value is, the better. The result of this study was supported by previous research comparing the performance between gradient boosting machine and random forest [15,16].…”

“…The approach evaluated here, using random forest and gradient boosting machine has rarely been used in tobacco plantation except in the health sectors such as predicting chronic bronchitis symptom analysis [15], smoking behaviour [16] and tobacco spending in each household in Georgia [17]. All previous studies using gradient boosting machine showed that the method could provide an excellent performance in predicting and classifying.…”

Machine Learning Prediction of Length of Stay in Adult Spinal Deformity Patients Undergoing Posterior Spine Fusion Surgery

“…XGB and GB algorithms are the most complex algorithms, and therefore, they are thought to have the best predictive power. LR algorithms are less computationally intensive than DT, RF, XGB, and GB algorithms; however, this often comes at the expense of accuracy and predictability [10][11][12][13]. The five algorithms were used to predict patient LOS for the posterior patients with the Sci-Kit Learn package in Python (National Institute for Research in Computer Science and Automation, Rocquencourt, France) [14].…”

Reflection on modern methods: generalized linear models for prognosis and intervention—theory, practice and implications for machine learning

“…Unfortunately, there is evidence to suggest that ML methods are already beginning to be (mis)used in this manner. [39][40][41] Integration of modern causal inference methods into ML applications should be sought and encouraged for answering causal questions. Indeed, there is already promising research being done in this area (e.g.…”