Machine learning in asthma research: moving toward a more integrated approach

Figures & data

Figure 1. The relationship between statistical and computer science disciplines

MS – Multivariate statisticsML – Machine learningAI – Artificial intelligence

Figure 2. Annual publication production

Number of publications on asthma per year for the three scientific disciplines are shown separately and jointly.MS – Multivariate statistics; ML&AI – Machine learning and artificial intelligence; BS – Bayesian statistics

Table 1. Summary information on retrieved asthma studies, 1965–2020

Description
Years	1965–2020
Documents	2395
Authors	12,607
Authors of single-authored documents	95
Authors of multi-authored documents	12,512
Documents per Author	0.19
Authors per Document	5.26
Coauthors per Documents	7.83
Collaboration Index	5.46
Average citations per documents	30.38
Sources	775
Keywords Plus (ID)	11,450
Author’s Keywords (DE)	4546

Table 2. The 10 most productive authors in the application of Multivariate statistics (MS), Machine learning and artificial intelligence (ML&AI) and Bayesian statistics (BS) to asthma research

Author	Number of articles	Times Cited	Year
Custovic A	34	1992	2008
Sterk P	30	1765	2004
Weiss S	30	882	2000
Chung K	27	2345	2007
Bousquet J	26	1173	1996
Lee S	26	174	2003
Simpson A	26	1473	2008
Bisgaard H	22	536	2009
Kim H	21	234	2012
Kim S	19	323	2009

Table 3. The 10 most productive countries in the application of Multivariate statistics (MS), Machine learning and artificial intelligence (ML&AI) and Bayesian statistics (BS) to asthma research SCP: single country publications; MCP: multiple country publications

Rank by # of articles	Country	Articles	Frequency	SCP	MCP	Rank by # of citation	Country	Total Citations	Average Article Citations
1	USA	599	34.9%	596	3	1	USA	23,828	39.8
2	United Kingdom	163	9.5%	163	0	2	United Kingdom	9101	55.8
3	the Netherlands	82	4.8%	82	0	3	Canada	3161	50.2
4	France	77	4.5%	77	0	4	Australia	3034	44.0
5	China	70	4.1%	70	0	5	Switzerland	2785	174.1
6	Australia	69	4.0%	68	1	6	the Netherlands	2751	33.6
7	Canada	63	3.7%	63	0	7	France	2087	27.1
8	Korea	58	3.4%	58	0	8	Italy	1951	36.1
9	Italy	54	3.1%	54	0	9	Sweden	1599	50.0
10	Germany	53	3.1%	53	0	10	Georgia	1537	512.3

Figure 3. Themes in asthma research fields

Panel a) The network was generating using co-occurrence patterns among the 50 top keywords. The size of the nodes is proportional to the strength of the keyword and colors show the nodes’ cluster membership. Panel b) The map was generated considering 150 keywords with minimum frequency of 3. The size of the cluster is proportional to the number of occurrences of the keywords and, consequently, to the number of linked articles. Only the 4 most predominant keywords are displayed.

Figure 4. Word-clouds for the eight topics in asthma research

Figure 5. Distribution of the big data analytics approaches in the eight topics

MS – Multivariate statistics; ML – Machine learning; BS – Bayesian statistics

Table 4. Distribution of the major methodologies across the selected articles

Discipline	Methodology	Relative frequency (%)
Multivariate statistics (MS)	Cluster analysis	849/2935 (28.9%)
	Factor analysis	372/2935 (12.7%)
	Principal component analysis	244/2935 (8.3%)
	Discriminant analysis	149/2935 (5.1%)
	Latent Class/Transition models	126/2935 (4.3%)
	SEM	78/2935 (2.7%)
	Markov models	67/2935 (2.3%)
	Mixture models	24/2935 (0.8%)
	Canonical Correlation	19/2935 (0.6%)
	Correspondence analysis	17/2935 (0.6%)
	Latent Variable models	11/2935 (0.4%)
	Multidimensional Scaling	8/2935 (0.3%)
	Profile Regression	1/2935 (0.0%)
Machine learning and artificial intelligence (ML&AI)	Decision Trees	159/2935 (5.4%)
	Machine Learning	145/2935 (4.9%)
	Artificial Neural Networks	76/2935 (2.6%)
	Artificial intelligence	74/2935 (2.5%)
	Unsupervised Learning	66/2935 (2.2%)
	Random Forest	50/2935 (1.7%)
	Support Vector Machine	42/2935 (1.4%)
	Deep Learning	25/2935 (0.9%)
	Supervised Learning	18/2935 (0.6%)
Bayesian statistics (BS)	Bayesian Models	156/2935 (5.3%)
	Spatial and Temporal models	39/2935 (1.3%)
	Bayesian Networks	35/2935 (1.2%)
	Naïve Bayes	4/2935 (0.1%)
	Bayesian Hierarchical models	28/2935 (1.0%)
	Bayesian meta-analysis	27/2935 (0.9%)
	Bayesian Analysis	25/2935 (0.9%)