Appendix 1 (diagnosis accuracy):
For each patient, the EP could choose 1 to 3 intended diagnoses among 16possible categorical diagnoses. The EP had to note his/her confidence level for each diagnosis in the 3 questionnaires (pre-CT, post-unenhanced-CT and post-final-CT), and the sum of these confidence levels had to equal 100 for each questionnaire.To clarify matters, in this appendix, we only consider 2 questionnaires called pre- and post-CT, but in the study, there were 3 questionnaires and the same reasoning applies.
Example for a patient X:
EP diagnosis before CT:
-diverticulitis 60%
-fecal impaction 30%
-other 10%
The sum of the confidence level in the questionnaire before CT equalled 100%, idem for the post-CT diagnosis.
Example for a patient Y:
EP diagnosis before CT:
-diverticulitis 40%
-fecal impaction 40%
-other 20%
The sum of the confidence level in the questionnaire before CT equalled100%, idem for the post-CT diagnosis.
We considered the pre- and post-CT diagnosis as accurate if the intended diagnosis with the highest confidence level was the same as the gold standard diagnosis. In case of multiple ex-aequointended diagnoses with the highest and same confidence level, we tookinto consideration all of these ex-aequo intended diagnoses to determine the accuracies (see example below).
Let us go back to our examples:
-For patient X, both pre-CT and post-CT intended diagnoses were considered accurate as the diagnosis with the higher confidence level was the same as that of the gold standard (diverticulitis 60% and diverticulitis 100%, pre- and post-CT,respectively).
-For patient Y, the pre-CT intended diagnosis was considered accurate as one of the two highest confidence level diagnoses was the same as that of the gold standard. By contrast, the post-CT diagnosis was considered non-accurate as the diagnosis with the highest confidence level (diverticulitis 80%) was different from that of the gold standard.
The number of accurate diagnoses was then compared between each conditional strategy:
-for the clinical strategy, the number of accurate intended diagnosesin pre-CT evaluations for the whole population study;
-for the current practice strategy, the number of accurate intended diagnoses in pre-CT evaluations for patients for whom no CT wasrequested, and the number of accurate diagnoses in post final-CTevaluations for patients for whom CT was requested;
-for the systematic unenhanced CT strategy, the number of accurate intended diagnosesin post-unenhanced-CT evaluations for the whole population study.
Appendix 2 (changes in diagnosis):
We estimated the change in diagnosis in each conditional strategy using a mathematical distance (L1) for each patient, in order to incorporate the diagnosis uncertainty and the possibility of multiple diagnoses, as follows:
L1= Σi=1to16 | GS diagi – EP diagi |
Where:
-Σi=1to16 corresponds to the sum for the 16 categorical diagnosis.
-“GS diagi” corresponds to the gold standard diagnosis for the pathology code “i” which was equal to 100 for the gold standard diagnosis and 0 for the others;
-“EP diagi” corresponds to the confidence percentage for the diagnostic code “i” by the EP.
This distance was thus calculated for each patient in the three questionnairesbut with a view to clarifying matters, we used two questionnaires called pre- and post-CTin this appendix.
Examples are given below:
Example 1:
Patient XEmergency Diagnosis / Gold standard / L1 distance
Pre-CT / Post-CT / pre-CT / post-CT
Diverticulitis
40% / Diverticulitis 100% / Diverticulitis 100% / 120 / 0
Fecal impaction 40% / Fecal impaction
0% / Fecal impaction
0%
Other
20% / Other
0% / Other
0%
… diagnosis ifrom 4 to 16 / 0% / 0%
Sum / 100 / 100 / 100
Fig.A1 and Fig.A2: Equations details allowing calculating L1 pre- and post-CT for the patient X.
Example 2:
Patient YEmergency Diagnosis / Gold standard / L1 distance
Pre-CT / Post-CT / pre-CT / post-CT
Fecal impaction
0% / Fecal impaction60% / Fecal impaction
100% / 200 / 80
Diverticulitis
60% / Diverticulitis
0% / Diverticulitis
0%
Bowel obstruction 30% / Bowel obstruction0% / Bowel obstruction
0%
Other
10% / Other
40% / Other
0%
… diagnosis i from 6 to 16 / 0% / 0%
Sum / 100 / 100 / 100
Fig. A3 and Fig. A4: Equations details allowing calculating L1 pre- and post-CT for the patient Y.
The results of this mathematical model could be interpreted as follows:
-if L1 = 0 then the EP diagnosis was single and accurate with 100% confidence;
-If L1 ϵ ]0,100] then the gold standard diagnosis was proposed by the EP, with a confidence level ≥ 50%, regardless of the number of intended diagnoses proposed;
-If L1 ϵ ]100,200] then the gold standard diagnosis was proposed by the EP, with a confidence level 50%, regardless of the number of intended diagnoses proposed and so the EP’s diagnostic confidence was very uncertain;
-if L1 = 200, then the gold standard diagnosis was notproposed for the EP intended diagnosis;
These distances were compared between each conditional strategy with the Wilcoxon signed rank test. A radar chart was created that shows the L1averagedistance for each conditional strategy according to each categorical diagnosis.