Table 5. Multivariate logistic regression of the risk of persistent anemia, if using iron tablets.

Variable	Odds ratio (95% CI)	Standard error	z test	LR test p value	Wald test p value
Had to stop iron tablets	3.70 (1.68, 8.18)	1.50	3.25	0.0005	0.0012
Iron tablets cause diarrhoea	3.09 (1.45, 6.62)	1.20	2.91	0.0020	0.0036
Had to switch iron tablets	2.24 (1.25, 4.02)	0.67	2.70	0.0059	0.0070
Nasal septal dermoplasty performed	2.35 (1.30, 4.27)	0.71	2.82	0.0038	0.0048
Treatment for gastrointestinal hemorrhage	3.06 (1.61, 5.81)	1.00	3.42	0.0003	0.0006
Treatment for skin telangiectasia	1.70 (1.06, 2.72)	0.41	2.21	0.0266	0.0273

The final model presenting all variables making a significant contribution to the risk of persistent anemia, once adjusted for the presence of other variables, in HHT online survey respondents using iron tablets. The model details 424 observations providing a pseudo r² of 0.14, and overall model p value of <0.0001. P values were calculated post estimation, both by likelihood ratio (LR) tests which assume independence of observations within a cluster (an assumption that was not met with these data), and the non parametric Wald test which does not make such assumptions. There was no clear relationship between persistent anemia and iron tablet-induced nausea, constipation, or abdominal pain (likelihood ratio test p values 0.14, 0.11, and 0.09 respectively). There was also no relationship with age, gender, other otorhinolaryngologic treatments [37] either combined or individually, or other reported HHT treatments for pulmonary, cerebral or hepatic arteriovenous malformations (data not shown).