A novel model to label delirium in an intensive care unit from clinician actions

Table 4 Significant differences for three models between four reclassification groups on five clinical measures

	Reclassification Groups
	Double + vs Double─	Double + vs Reclassified +	Reclassified + vs Reclassified─	Double- vs Reclassified─
Expected Relationship	Different p < 0.05^a	Same p > 0.05^a	Different p < 0.05^a	Same p > 0.05^a
Novel Model
Age	< 0.00005	< 0.00005	< 0.00005	< 0.00005
Discharge location	< 0.00005	< 0.00005	< 0.00005	< 0.00005
Death in hospital	< 0.00005	0.00011	0.0452	0.0859
30-day mortality	0.00072	0.115	0.00017	0.0011
one-year mortality	< 0.00005	0.178	0.0017	< 0.00005
Puelle’s Classifier
Age	< 0.00005	0.00097	< 0.00005	0.964
Discharge location	< 0.00005	< 0.00005	0.00062	0.00062
Death in hospital	0.0017	< 0.00005	0.820	0.103
30-day mortality^b	0.0949	–	–	–
one-year mortality	< 0.00005	< 0.00005	0.660	0.0015
Kim’s Classifier^c
Age	< 0.00005	0.0130	–	–
Discharge location	< 0.00005	< 0.00005	–	–
Death in hospital	0.0035	< 0.00005	–	–
30-day mortality^b	0.472	–	–	–
one-year mortality	< 0.00005	0.0010	–	–

Binary classification generates four groups of subjects, including two groups of reclassified hospitalizations, which were compared by biological validation to assess model goodness-of-fit. Double-Positives are expected to differ from Double-Negatives. In the case of successful reclassification, we expect Double-Positives to be similar to Reclassified-Positives, Double-Negatives to be similar to Reclassified-Negatives, and Reclassified-Positives to differ from Reclassified-Negatives. Bolded fields represent p-values concordant with expectations of good model fit
^ap-values were generated from pairwise chi-squared testing with the Holm correction for all measures except age, which was tested with pairwise Tukey multiple comparisons of means
^bPairwise testing was not performed in the event of no significant difference between Double-Positives and Double-Negatives
^cThe Kim model does not generate reclassified negatives, making associated tests unavailable

ISSN: 1472-6947