Results
A total of 371 participants were enrolled in the study from five sites, with a preponderance of men (59%) and an average age of 51 years (SD 18) (Table 1). From these, 153 participants had sinus CT images available.
Step 1
The endoscopic assessment presented an average ND of 9.8 cm (SD 0.9) and CT distance of 8.4 cm (SD 0.9) for the total sample. In turn, the external facial distances assessed in this study showed an average curved distance of 11.8 cm (SD 0.9) and a perpendicular distance of 9.6 cm (SD 1). All four variables were significantly higher for men (Table 2).
Considering the endoscopic ND assessment as a reference, Δ Perpendicular averaged -0.1 cm (SD 0.65, 95% CI -0.2 to -0.06), representing an underestimation of true ND. Conversely, Δ Curved averaged 2.0 cm (SD 0.83, 95% CI 1.9 to 2.1), representing an overestimation. In turn, Δ CT scan averaged -1.2 cm (SD 0.8, 95% CI -1.3 to -1.07). The overarching estimations held true on subgroup analysis for men and women and therefore, inferences of ND error estimation for each of these three variables persist regardless of participant sex (Table 2).
The perpendicular distance was the most strongly correlated with ND (P<0.001, R=0.775). Nonetheless, curved and CT distances also showed a significant correlation to the endoscopic ND (P<0.001, R=0.598, and P=0.001, R=0.541, respectively). Self-reported ethnicity or age did not exert statistical significance on any variable assessed on sub-group ANOVA (men and women).
Step 2
The stepwise multiple regression analysis to predict ND based on the external facial distances controlled for age and sex and resulted in three regression models and showed that 64% of the variance in ND can be accounted for by the perpendicular distance alone (F(1,269) = 493.31, p<0,0001) – model 1. Similarly, perpendicular and curved distances were listed on model 2, being responsible for 66% of the variance in ND (F(2,268) = 263.65, p<0,0001). On model 3, sex and both facial distances were included, accounting for 66% of ND variance (F(3,267)= 179.98, p<0,0001). Age was removed as a variable in all regression models.
The three models were then tested against the data of the remaining 100 participants, as ND = Perpendicular distance*0.773 + 2.344 (model 1), ND = Perpendicular distance*0.672 + Curved distance*0.171 + 1.28 (model 2), and ND = Perpendicular distance*0.662 + Curved distance*0.147 + Sex*1.7 + 1.56 (model 3). The ND prediction error using model 1 was 0.03 cm (SD 0.61; 95% CI -0.08 to 0.16), while model 2 presented an average overestimation of 0.03 cm (SD 0.58; 95% CI - 0.78 to 0.15). In turn, model 3 resulted in a mean of 0.9 cm (SD 1; 95%CI 0.7 to 1.1).