2.8. Statistical Methods

All analyses were on incident first-time outcomes, e.g., in the analysis of ophthalmologist consultation, only patients without previous ophthalmologist consultations were included. This meant that in the absolute risk analyses, patients with outcomes before ICI were excluded, and in the analyses of relative rates, patients with outcomes before cancer diagnosis were excluded.

The risk time (time from initial ICI administration to event) was summarized in medians with 25 and 75 percentiles (p25–p75).

The Aalen–Johansen estimator with competing risk of all cause death was used for estimating absolute risks at 30 days, 6 months, and 1 year after initial ICI administration. Kaplan–Meier estimates were used for the risk of all cause death at 30 days, 6 months, and 1 year.

Hazard ratios were modeled in multivariable Cox regression models. For each of the sub cohorts (all patients with cancer, patients with lung cancer, and patients with malignant cutaneous melanoma, respectively) a cox model was used for analyzing each outcome (ocular inflammation, uveitis, and first-time ophthalmologist consultation, respectively). ICI exposure was included as a time-updated variable. The dataset was split on the date of first ICI administration and a proxy variable was created, enabling an analysis where patients contributed to risk time in the non-exposed group from cancer diagnosis to the date before first ICI administration and to risk time in the exposed group from the date of first ICI-exposure to end of follow up. Likewise, age and calendar time were included as time-updated variables. Age was included as categorical variable with the levels ≥25, 26–50, 50–75, and >75 years of age. In modeling ocular inflammation, linearity could not be assumed for the age variable. For this reason, age was not included as a variable in these models. Calendar time was included as years since study entry, which was equal to years since cancer diagnosis. Additionally, the models were adjusted for sex. In the Cox model analyzing the association between ocular inflammation and ICI in patients with malignant cutaneous melanoma, interaction with sex could not be ruled out, and an interaction term with ICI treatment and sex was included.

The statistical analyses were performed using SAS Software version 9.4 (SAS Institute Inc.) and R: A language and environment for statistical computing (version R-4.0.3) [24].