*2.5. Safety*

Safety was evaluated at each visit by means of a physical examination (monitoring of blood pressure, clinical ear, nose, and throat examination to check the nasal septum and other side e ffects on nasal mucosa), and assessment for adverse events (AE). All AE were coded using the Medical Dictionary for Regulatory Activities (MedDRA). Adverse events were classified by the investigators as unrelated, dubitable, or possibly, probably or certainly related to the treatment. Monitoring the safety of administration of the product, motivated by the iatrogenic risks, justified the setting up of a specific independent monitoring and safety committee. The committee met in particular in the case of the occurrence of serious adverse events and gave its recommendations on the continuation of the study after collection of adverse events and observance of the treatment after 30 days for the first 8 patients included. It was composed of a specialist of the disease not involved in the study, an ENT specialist, and a statistician specialized in the methodology of clinical trials.

Systemic absorption of tacrolimus was evaluated by means of FK506 dosages in blood samples 8, 22, and 43 days after the beginning of the treatment. No systemic absorption and e ffects were expected, but, in case of a positive dosage of 5 ng/mL or more, it was decided that the laboratory would immediately inform the investigator in order to ask the patient to stop the treatment.

#### *2.6. Sample Size Calculation*

We hypothesized that 60% of patients would be improved in the treatment group against 15% in the placebo group. It was therefore necessary to include 22 patients in each group to reach an 80% power with a 5% alpha (bilateral), leading to 44 patients overall (Fisher exact test).

Taking into account early withdrawal and patients who may be lost to follow-up, we planned to include 24 patients in each group, that is to say, a total of 48 patients.

## *2.7. Randomization*

The randomization process was centralized. Patients were randomized by blocks of 4 and unstratified. Allocation of a randomization arm to an included patient was made by IWRS (Interactive Web Response System), on the basis of a unique randomization list for all investigation centers. The randomization list was pre-established, by the "Pole IMER" at the Hospices Civils de Lyon–Clinical Research Unit. Clinsight software version 7.1 (Ennov Clinical®, Paris, France) was used to manage this study. After verifying the inclusion criteria, the investigator connected to the platform to create the list of patients. Once the inclusion criteria had been validated, the patient was randomized and a treatment code was allocated by the system. The treatment was then dispensed by the pharmacy at the Hospital Center. This was a double-blind study in which neither the patient nor the investigator was aware of the nature of the treatment administered.

#### *2.8. Statistical Methods*

Populations: 2 populations were defined. The per protocol population, which was set at 70% adherence, consisted of all patients receiving at least 60 ointment treatments of the 84 planned. The intention-to-treat (ITT) population consisted of all randomized patients starting the treatment, and patients were considered in their randomization group. All analyses were performed on the ITT populations; the main outcome was also analyzed on the per protocol population.

Initial characteristics of the patients were summarized by means of descriptive statistics (number, average, standard deviation, median, minimum, and maximum for the quantitative variables, and numbers and percentages for the qualitative variables).

Analysis of the main outcome: the percentage of patients experiencing improvement in their nosebleeds was computed in each group. The percentage was compared between groups using a Chi<sup>2</sup> test (or Fisher exact test if the conditions for Chi<sup>2</sup> were not fulfilled), and the analysis was performed on the intention-to-treat population and on the per protocol population. Patients who stopped the treatment but who had filled in epistaxis grids were analyzed using their data. Patients who withdraw from the study before completing the follow-up were considered as failures.

Analysis of the secondary outcomes: the percentage of patients with at least one adverse event was computed and compared between the 2 groups. Quantitative parameters were presented as mean ± standard deviation and median (minimum and maximum) for all groups and were compared using the Student *t*-test (or Mann–Whitney test in case of non-normality). Qualitative parameters at inclusion were presented in terms of number (percentage) and compared using the Chi<sup>2</sup> test (or Fisher exact test where conditions for the Chi<sup>2</sup> test were not fulfilled). Mixed models were produced to compare evolution between the groups.

All analyses were performed using SAS software version 9.4 (SAS Institute Inc., Cary, NC, USA). Effect sizes were computed as risk difference (Chan–Zhan 95% CI) and relative risks for binary outcomes, and as Cohen's d for quantitative outcomes.

#### *2.9. Missing Data*

The main outcome was based on grids that were filled in daily. If one day was missing, the value was replaced by an average of the 4 values before and the 4 values after the missing value. This strategy was applied up to 7 missing values over 6 weeks (i.e., 10%). If more than 7 days and less than 21 days (included) were missing, a daily average was computed from the data available (from the 6-week period evaluated) and multiplied by 42 to estimate epistaxis duration. If a patient was lost to follow-up or refused to communicate his nosebleed grids or had more than 21 days missing on his grids, the result for the patient concerned was considered as a failure.
