Predicting phenotypes of asthma and eczema with machine learning

BMC Medical Genomics

Table 3 Comparison of random forest performance using selected input domains.

outcome	feature set	AUROC	p-value*	sensitivity (at 90% specificity)	sensitivity (at 80% specificity)	accuracy
Doctor's Diagnosed Eczema	allergens	0.62 (0.03)	0.34	0.22 (0.06)	0.37 (0.06)	0.79 (0.02)
	lung functions	0.56 (0.04)	0.08	0.13 (0.05)	0.24 (0.06)	0.78 (0.02)
	genetic	0.56 (0.04)	0.11	0.14 (0.05)	0.25 (0.06)	0.78 (0.02)
	demographic/environ.	0.56 (0.04)	0.05	0.12 (0.05)	0.24 (0.07)	0.78 (0.02)
	all	0.65 (0.04)	reference	0.2 (0.07)	0.35 (0.08)	0.79 (0.02)
Current Asthma	allergens	0.79 (0.04)	0.11	0.43 (0.08)	0.64 (0.07)	0.86 (0.02)
	lung functions	0.76 (0.04)	0.04	0.44 (0.08)	0.6 (0.09)	0.86 (0.02)
	genetic	0.54 (0.04)	<0.0001	0.12 (0.05)	0.23 (0.07)	0.83 (0.02)
	demographic/environ.	0.62 (0.04)	<0.0001	0.2 (0.08)	0.38 (0.07)	0.83 (0.02)
	all	0.84 (0.03)	reference	0.56 (0.09)	0.73 (0.08)	0.87 (0.02)
Current Wheeze	allergens	0.75 (0.04)	0.35	0.34 (0.09)	0.54 (0.1)	0.88 (0.02)
	lung functions	0.72 (0.05)	0.19	0.42 (0.09)	0.55 (0.08)	0.89 (0.02)
	genetic	0.5 (0.05)	0.0002	0.11 (0.06)	0.21 (0.08)	0.88 (0.02)
	demographic/environ.	0.6 (0.05)	0.006	0.17 (0.07)	0.32 (0.09)	0.88 (0.02)
	all	0.77 (0.04)	reference	0.5 (0.09)	0.62 (0.07)	0.89 (0.02)

Performance of random forest on different outcomes using specific variable subsets and the full set of demographic, environmental, genetic (single nucleotide polymorphisms), allergen sensitisation, and lung functions variables. Results are mean (standard deviation) estimated from out-of-bag distributions across 100 bootstrap runs.
^*testing the hypothesis of difference in AUROC significantly shifted from zero as compared to that of a random forest model using all variables with a corrected paired t-test.
AUROC: area under the receiver operating characteristic curve.

ISSN: 1755-8794