Review inclusion/exclusion criteria (esp unplanned ICU admission)

[docsteveharris]

https://github.com/tharusan/snap2_final/blob/c23bbf0d922fbb363c7d1097f9cb3a430a0bf451/snap2_syntax1.Rmd#L242 This is a key question

We are defining the patient population on the following grounds

Exclude low acuity who would never be offered a routine bed

• obstetric patients
• ...

Exclude high acuity

• protocolised (i.e. those who always come)
• patients already in critical care
• qn:?? here we exclude patients who deteriorate in theatre

Need to be clear how we make that final definition

• [ ] completely document here the inclusion/exclusion
• [ ] annotate in the code how we define those inclusioin/exclusion categories
• [ ] specific attention to the unplanned ICU admission (i.e. deteriorating in theatre)

[dannyjnwong]

See my previous comment in the issue #7 (https://github.com/tharusan/snap2_final/issues/7). I think we should focuss on planned ICU admission as the intervention.

[docsteveharris]

OK. So if we use an RCT as a mental model for this then our 'trial' would

• randomise at the end of theatre
• patients who had deteriorated in theatre will be excluded
• patients on a protocolised admission pathway will be excluded (e.g. cardiothoracics)
• very low risk will be excluded (i.e. day case, obstetrics)

our study population are middling/high risk patients the intervention is an admission to ICU directly (or via recovery) the instrument is 'occupancy' we don't care if the patient was booked or not

??

[dannyjnwong]

Alternatively, you can randomise earlier. In our dataset, we also have this question: "3.27. Has this patient been referred for postoperative critical care? (Y/N)" this was asked intraoperatively, before the end of surgery.

The variable name is S03ReferredForPostoperativeCriticalCare

[rmoonesinghe]

Hi all - sorry to be coming to this late. I guess we want to stick to the stuff which will be most helpful in 'real life'. In summary, I would:

• exclude obstetrics (too low risk)
• consider excluding SORT/clinical-judgement model prediction <1% mortality risk (as too low risk)
• exclude patients with absolute indications for critical care (from our dataset I think this would be patients who were receiving level 2 and 3 support on arrival in anaesthetic room - i.e. Q2.4a)

In terms of Steve's mental model, MY RESPONSES IN CAPS

randomise at the end of theatre - YES patients who had deteriorated in theatre will be excluded - NO - WE WON'T HAVE ENOUGH INFO TO KNOW IF IT WAS A MAJOR OR MINOR DETERIORATION patients on a protocolised admission pathway will be excluded (e.g. cardiothoracics) - I GUESS. I WOULD TEST THIS BY LOOKING AT COMPLIANCE WITH CRITICAL CARE ADMISSIN FOR THIS GROUP OF PATIENTS. CARDIAC ONLY RATHER THAN THORACIC very low risk will be excluded (i.e. day case, obstetrics) - SHOULDN'T BE ANY DAY CASE. DEFO EXCLUDE OBS. POSSIBLY EXCLUDE SORT/CLINICAL JUDGEMENT PREDICTION <1%

our study population are middling/high risk patients - YES the intervention is an admission to ICU directly (or via recovery) - YES the instrument is 'occupancy' - YES we don't care if the patient was booked or not - AGREE

I think we need to randomise at the end of surgery; Danny's suggestion is good if we want an 'intention to treat'; analysis with all the challenges of getting patients into critical care, but we will run into precisely the practical problems we were trying to avoid by doing an IV analysis rather than actual RCT.

CRF HERE FOR EASE Appendix 1 - CRF main EPICCS v1.2 20170201.docx

[docsteveharris]

Thanks Rams So a treatment is defined as 

admission to critical care following major surgery without first going to the ward we don't care if they were booked we understand this group will include patients who deteriorated in theatre  we are doing our best to avoid super low risk / protocolised routine to create a group where there was 'choice'