2.3. Statistical Analysis

The main outcome of the present study was the comparative analysis of the amount of sCD14 in samples of both groups (H0: sCD14 in healthy subjects = sCD14 in decayed patients). Sample size calculation was performed using α = 0.05 (5%) and an 80% sample power. Standard deviation of sCD14 content in saliva of decayed children was 37.67 ng/mL, as obtained in previous papers [19]. The minimum significant value considered was 5.6 ng/mL. On the basis of these data, the needed number of subjects to be enrolled resulted 86 for each group.

Descriptive analysis. In both groups, the mean and standard deviation of clinical parameters and of sCD14 content were computed.

Inferential statistics. To assess if differences in the salivary sCD14 level were present between heathy and decayed subjects, data of two groups (H and D) were compared by t-test.

We performed chi-square tests and tests for trends in proportions with an alpha of 0.05 on ordered categorical variables when effects were successively bigger or smaller on sex, age in three developmental age-groups (6–9, 10–13, 14–16 years), oral hygiene, and sCD14 levels in quintiles.

In particular we selected the age groups considering the dentition and pubertal development: 6–9 fist mixed dentition and pre-pubertal stage, 10–13 completion of permanent dentition and developing sexual characteristics, 14–16 permanent dentition and completion of pubertal development [20,21].

We calculated Odds Ratios (OR) and 95% Confidence Intervals (CI) for the binary outcome Healthy vs. Diseased using both univariate and multivariate (adjusting for sex and continuous age) unconditional logistic regression.

After examining the sCD14 distribution between number of carries (0–9) and the results from the logistic models, a hurdle model appeared to be the most appropriate model to examine and describe the data [22]. Briefly, hurdle models are used to evaluate count data with an overabundance of zeros (healthy patients with zero caries in this study). A hurdle model is a 2 steps model that first evaluates a right censored logistic regression to evaluate the probability that an outcome is a zero or not, if this hurdle is passed a left censored count data model is applied to the data to evaluate counts larger or equal to 1. We applied a poisson hurdle model with only sCD14 as a continuous variable since from Vuong tests [23] age, sex and oral hygiene variables, or using a negative binomial distribution in the count part of the model did not give statistically better fits.

The R-project statistical software was used to carry out the analyses [24].