Patient selection consensus

[RobinGuillard]

Dataset description: 17 patients with tinnitus, 6 healthy subjects.
50 recordings in all, 41 nights of tinnitus patients.
We have at least 2 recordings per patient. But some have some issues: long electrode disconnection, absence of masking volume completion, unsleepy patient, partial recordings of a night.

Here are my advices for patient selection.

• One patient slept only 1h30 on the whole habituation night. He should be excluded.
• For one patient (SCHM), I have 8 night of recording, but 1 night is partial, and one night has important impedance issue, and only one night has masking volume before and after (but I have EVAs every time). It would be very sad to exclude this subject, all the more as I have partial masking volumes for some nights (only morning or only evening). I think that we can "interpolate" the masking volume missing of one night and include this subject. It would be then a normal subject with 2 nights and associated measurements.
• for one patient, I have 3 nights, but I know one is missing 1h30 of sleep at the end. We can say we exclude this night, out of caution in first intention, and include the subject with 2 nights.
• For some patients, I have electrode disconnection issues. For at least one patient, the electrodes were gone the major part of one night. For others, we still have the major part of the night. What is important in this case is to have a good representation of all sleep states. It is more bothering to have 3h of bad signal concentrated at the end of the night (REM time) than 5h of lost signal equally distributed. I would advise to say "if we have more than 30-40% of data deleted either at the beginning or at the end of the recording, we suppress the night, if the data loss is equally distributed it's ok to go up to 55-60% of the night lost).

[RobinGuillard]

The preceding message applies for the correlation study between masking volume and episodes distribution. Dealing with EVAs VS episode distribution, the only modification is we should include the 5 other nights of patient SCHM.

[lkorczowski]

Before going to specific, here are my general proposal:

Valid data

• patient has sleep correctly through the night (no insomnia, max 2 episodes awaken)
• data are mostly valid through the night (only random and short hardware problem), i.e. 90% of the total signal should be valid
• all patients should have exactly the same setup to be evaluated (e.g. two nights, valid masking volume, etc.). I don't believe that "interpolating" data is valid. Yeah it is sad.

Note that these general rules applies for a given evaluation, we could have several, e.g.:

• one evaluation intra-subject with only two nights respecting drastically valid data (removing all patients not meeting this criterion)
• one evaluation with inter-subject where the number of night might be less important (therefore having 1, 2 or more nights patients)