Using average response time as an outcome variable

Hi all - I am working on regression analyses with timing data as the outcome variable. Specifically, I trying to estimate questionnaire item response time based on item attributes (e.g., item content, number of response options, etc.). I am sure there are respondent factors that impact response time but I won't know the respondents in advance so we only have information on the items themselves. I currently have individual-level response time data for the whole questionnaire and I am unsure of the types of distributions to expect for each item, though at least some of them are loglinear. If these were all simple normal distributions, I would probably just take the mean (i.e., the average completion time) and the 90th percentile (i.e., completion time of some of the slowest respondents) and use those as outcome variables.

My questions:
1. Should I just treat this data is multilevel from the start, or just identify some central tendency measure (e.g., mean) as my outcome variable? If the latter, how do I choose?
2. How would you handle missing cases? I have some instances where people skipped most of the questionnaire and those are straightforward. What about situations where people run out of time and miss the end of the SQ?
3. Any pitfalls or considerations I need to keep in mind for this type of work?
Your data structure is somewhat unclear to me... What is the measurement of your outcome -- is it seconds (or minutes) it takes an individual to answer the whole questionnaire or one single question?