Delirium misdiagnosis risk in psychiatry: a machine learning-logistic regression predictive algorithm

Table 1 Machine Learning-Logistic regression classifier for accurate delirium diagnosis^a

Variable	B (SE)	Odds ratio (95% CI)	P
Age	- 0.02(0.02)	0.97 (0.93 to 1.02)	0.29
Sex(Female)	- 1.21(0.71)	3.35 (0.83 to 13.42)	0.08
Referral unit(medical vs. surgical)	- 0.21(0.84)	1.23 (0.23 to 6.41)	0.79
Psychiatry history	- 1.69(0.78)	0.18 (0.03 to 0.85)	0.03
Pain^b	- 0.13(0.83)	1.14 (0.22 to 5.88)	0.86
hypoactive delirium	- 0.53(0.84)	0.58 (0.11 to 3.03)	0.52
Death^c	- 0.64(1.10)	0.52 (0.05 to 4.55)	0.55
Hospital stay(days)	- 0.0061(0.007)	1.00 (0.99 to 1.02)	0.40
4-AT^d score prior to referral (filed vs. non filed)	- 1.42 (0.84)	4.16 (0.80 to 21.45)	0.09

For the model: Hosmer-Lemeshow statistic = 10.28 (p = 0.24), B unexponentiated coefficient, SE Standard error
^anumber of patients with accurate diagnosis of delirium on referral was 39 (54%).
^bNumber of patients with referral diagnosis labelled as “pain” was 25
^cNumber of cases where 4-AT score was filed prior to referral was 41. Number of patients with hypoactive delirium subtype was 15
^d7 patients with delirium diagnosis died during hospital admission
Bold some evidence against the null hypothesis of no association between psychiatric history and delirium diagnosis status, with odds ratio confidence interval not crossing the null

ISSN: 1472-6963