R version 3.6

The time interval for overall survival was calculated starting from the date of hospital admission until death from SARS-CoV-2 pneumonia or the date of the last follow-up. An event was considered death due to SARS-CoV-2 pneumonia. Mortality was determined via the vital status of each patient at the end of the study observation period. For data analysis continuous variables were reported as median (IQR) and categorical variables were shown as frequency or percentage. The least absolute shrinkage and selection operator (LASSO) was used for multivariable selection (18 (link)). The area under curve (AUC) value was used to evaluate the accuracy of the vital status prediction. The Cox proportional hazards regression analysis was used to evaluate the prediction of a prognostic model for overall survival. Proportional hazards assumption for the Cox proportional hazards regression model was assessed via the Schoenfeld residuals test. The 95% confidence intervals (CIs) were estimated via 5,000 bootstraps replicates. Propensity score matching (PSM) was performed to adjust demographic factors (including age, sex, comorbidities), survival status and treatments. All statistical analyses were performed using R (version 3.6) and SAS (version 9.4). A p < 0.05 was considered as statistically significant.

Patient characteristics were described as mean and standard deviation or median and IQR for continuous variables and frequency (%) for categorical variables. The time to surgical recurrence was compared between the three cohorts by the Log-Rank test and was displayed on a Kaplan–Meier curve. The factors associated with surgical recurrence were analyzed by the univariate and multivariate Cox regression model. The results were considered significant at the critical rate of 5% (p < 0.05). Calculations were performed using SAS Version 9.4 (SAS Institute, Cary, NC, USA) for statistics and R Version 3.6 for graphics.

The absolute value of the distance from the benchmark score was calculated for dentists and for non-dental caregivers for each photograph. Due to skewness of the distribution, logarithm of the values was used. The direction of distance from the benchmark indicated whether the score assigned to a photograph was lower, equal or higher than the benchmark score.
To quantify differences between caregivers and dentists with regard to the distance from the benchmark as well as the direction of this difference, a linear mixed effects model and a generalized linear mixed effects model were fitted, respectively. Type of assessor (caregiver, dentist) and oral health aspect (denture hygiene, oral hygiene, teeth, gums, tongue, palate/lips/cheeks) were added to the models as random effects.
Mean squared errors were computed to compare the scores provided by non-dental caregivers and dentists with regard to the variance around the benchmark for each oral health aspect. Statistical programs R (version 3.6) and SAS (version 9.4) were used.