Looking desperately for advice regarding the statistically correct analysis of dependent data in medical research


my name is Andy and I am racking my brain with the following research question:
Let's assume you are a surgeon who wants to remove a tumor which is exactly in the middle of the abdomen. The tumor is surgically accessible from a left flank approach or right flank approach in general.
Unfortunately, some patients have an intraabdominal obstacle (e.g., a scar) that hinders the access to the tumor and this obstacle cannot be foreseen in advance. If the obstacle is present you have to temporarily stop the procedure, prepare the other access side and make an incision there and hope that the obstacle is not present on the other side as well. Although the presence of this obstacle is rare, patients usuallly do not appreciate having scars on both flanks and therefore I want to proof that one side (e.g. right) is superior in terms of this obstacle being present less frequently.
In summary, the obstacle can theoretically be encountered solely during the left side approach, the right side approach, at both sides or not being existent at all.
The outcome is whether your access is blocked by an obstacle or not (-> binary outcome). Apart from the access side, there are others risk factors for the occurrence of this obstacle.
I want to built a logistic regression model, after explanatory univariate analysis, to evaluate the relationship between the access side (left vs. right) and the outcome, after adjusting for confounders. Explanatory univariate analysis is used to decide which other predictor variables to include in the logistic regression model.
I am planning to retrospectively analyze 2000 consecutive procedures performed on 1700 individual patients, where this particular tumor removal was performed without prior randomization to a left- or right-sided approach respectively. However, some patients, but not all patients, had multiple operations due to incomplete tumor removal necessitating a second, third or even fourth operation at a subsequent point in time. In some of the patients, who had repeated surgeries, the access side was always the same while in others it alternated between left and right. In some patients, the access was blocked on one side and therefore the other side could be evaluated as well with respect to the obstacle.
I was planning to include only outcomes which were documented during the first approach (left or right) within a single procedure to avoid dependent data. However, I still rack my brain over how to include or exclude the repeated measurements at different time points (e.g. first operation and second operation 6 weeks later). I have had the following ideas:

A) Include all 2000 procedures for univariate analysis and multivariate analysis respectively, completely ignoring dependence due to repeated measurements.
B) Include all 2000 procedures for univariate analysis and apply a generalized estimating equation (GEE model) with the patient's ID as "exchangeable" to account for intraclass correlation at least in the multivariate analysis.
C) Include only the first procedure for each access side per individual patient respectively for both, univariate and multivariate analysis.
D) Include only the observation from the first procedure per individual patient for univariate analysis and the first procedure for each access side per individual patient for multivariate analysis, resulting in a maximum of two observations per cluster in the GEE model.
E) Only include exactly one observation for each individual. This might be the first operation of a certain individual or an operation randomly drawn from all operations an individual patient had during the study period resulting in a total of 1700 operations to be included for further analysis (univariate and multivariate analysis, with the later being a "usual" logistic regression model since there dependent data has been "removed".

Which approach is statistically sound in your opinion? I personally oppose option E) because statistical power is possibly reduced, if I exclude a considerable amount of observations.

Many thanks in advance for your help!