Table 4.

Logistic Regressions: Outcome is the Risk of Any Complication Associated with Dantrolene

Full Dataset (N=368)
Variables	Coefficients	S.E.	P Value	Odds Ratio	(95% CI)
Dose Factor *	0.363	0.101	0.000	1.29	(1.12-1.48)
Fluid Administration	0.893	0.289	0.002	2.44	(1.37-4.35)
Oral Surgery	−1.341	0.581	0.021	0.26	(0.08-0.84)
Neurosurgery	−1.760	0.647	0.006	0.17	(0.05-0.63)
Constant	−3.142	0.599	0.000
Reduced Dataset (N=349)
Dose Factor *	0.317	0.111	0.004	1.25	(1.07- 1.45)
Obstetrics/ Gynecology	1.906	0.608	0.002	6.72	(1.99- 22.7)
Furosemide	−0.822	0.317	0.009	0.44	(0.23- 0.82)
Constant	−2.766	0.614	0.000

^*Dose Factor is the natural log of the total dose of dantrolene or of the initial dose if a total dose was not reported.

The logistic regression for the full dataset describes a 29% increase in risk of any complication when the total dantrolene dose was doubled, a 144% increase in risk when fluid administration was part of treatment, an 83% decrease in risk in the presence of neurosurgery and a 74% decrease in risk in the presence of oral surgery.

The logistic regression for the reduced dataset describes a 25% increase in risk of any complication when the total dantrolene dose was doubled, a 572% increase in risk in the presence of obstetric or gynecologic surgery, a 56% decrease in risk if furosemide was given and no relationship with fluid administration or other types of surgery.

As is common in statistical analysis, some but not all of the assumptions of this technique are met by the data. The outcome variable, presence or absence of complications reported after dantrolene administration, is binary, but it is a composite of all reported complications. The independent variables are not linear combinations of each other. Observations are independent because each case is independent. There may be error in the measurement of independent variables. As described in Methods we used stepwise entry of variables into the regression to judge that no important variables were omitted and no extraneous variables were included. We assume that the logit function is appropriate to describe the outcome variable. For all the above reasons we do not propose that these regressions are sufficient to predict the occurrence of complications. Our goal is to use the available data to alert the clinician to conditions which have been associated with complications after administration of dantrolene. For example, the logistic regression for the full model describes an 18.7% chance of any complication after administration of 100 mg of dantrolene in the absence of fluid administration, neurosurgery or oral surgery. The dose of 100 mg dantrolene is the 1st quartile of the distribution of doses and 600 mg is the 3rd quartile. Increasing the dose to 600 mg is associated with 30.6% chance of any complication and in the presence of fluid administration the chance of any complication increases to 51.8%. Similarly the logistic regression for the reduced model describes a 21.0% chance of any complication after administration of 94.5 mg of dantrolene in the absence of furosemide or obstetric-gynecologic surgery. The dose of 94.5 mg dantrolene is the 1st quartile of the distribution of doses and 600 mg is the 3rd quartile in the reduced dataset. Increasing the dose to 600 mg is associated with 32.3% chance of any complication and in the presence of furosemide the chance of any complication decreases to 17.4%.