. 2018 Sep 19;8(9):e022041. doi: 10.1136/bmjopen-2018-022041

Table 3.

Logistic regression models predicting risky use of social drugs

Dependent variable	Model χ² (7df), sig value	Variance explained^*	Significant predictors^†	OR (95% CI)
High caffeine	41.81, p<0.001	0.017–0.026	Sleep-promoting medication	1.49 (1.12 to 1.98)
			Night-shift work	1.42 (1.16 to 1.74)
			Gender	1.29 (1.02 to 1.63)
			Age	1.02 (1.01 to 1.03)
Alcohol misuse^‡	Overall model not significant
Smokers who smoke to stay awake^§	23.759, p=0.001	0.008–0.028	Sleep-promoting medication	1.97 (1.06 to 3.64)
Smokers who smoke to stay awake^§	23.759, p=0.001	0.008–0.028	Night-shift work	2.12 (1.34 to 3.36)

*Estimates here represent Cox & Snell R-Square and Nagelkerke R-square values.

†Predictors and levels entered into the model: wake medications: used in the past month versus not used, sleep medications: used in the past month versus not used, medication with sleepiness as a side effect: used in the past month versus not used, night shifts: worked versus not. Only variables significantly contributing to the model are included in the table. Model controlled for age and gender.

‡Users who exceeded NIH/NIAAA limits for past week use.

§Smokers who reported engaging in this behaviour in order to remain alert.