Table 3.

Univariable exact logistic regression^* and final multivariable logistic regression (controlling for household clustering) model^† results, examining the effects of predictor variables on the odds of acute gastrointestinal illness in Rigolet, Nunatsiavut, in September 2012

Rigolet September model	n	Univariable results			Multivariable model results
		OR	P	95% CI	OR	P	95% CI
Homeless person staying in the house
No	182	Ref.			Ref.
Yes	38	6·17	0·022	1·30–29·23	4·76	0·021	1·27–17·90
Visited cabin in past month
No	74	1·94	0·100	0·89–4·23	3·33	0·039	1·06–10·44
Yes	152	Ref.			Ref.
Weekly amount spent on obtaining country food
Low (<$150)	89	Ref.			Ref.
Medium ($150–$300)	87	1·68	0·320	0·60–4·70	2·65	0·130	0·75–9·36
High (>$300)	35	3·08	0·062	0·94–10·06	7·18	0·010	1·62–31·90
Storage of drinking water					–	–	–	–
Container in the fridge	122	Ref.			–	–	–	–
Container outside of the fridge	51	2·71	0·077	0·90–8·20	–	–	–	–
No storage	47	3·76	0·028	1·15–12·24	–	–	–	–
Exposure to cats in past month
No	159	Ref.			–	–	–	–
Yes	62	1·83	0·200	0·71–4·73	–	–	–	–
Perceived quality of drinking water
Very poor or poor	32	2·85	0·157	0·67–12·17	–	–	–	–
Fair, good, or very good	187	Ref.			–	–	–	–

OR, Odds ratio; CI, confidence interval.

^*The results from the univariable analysis are presented for those variables with P < 0·20.

^†Likelihood ratio test comparing the model with and without the household-level variable: variance (0·51, 95% CI 0·007–35·138, P = 0·300); intra-class correlation coefficient (0·14, 95% CI 0·002–0·914). Note that due to the structure of the Rigolet data, a random intercept was forced in the model to control for household-level clustering.