Our institutional review board approved this study.
The study design comprised 3 phases,
as summarized in the flow chart (Figure 1).
CXR selection phase:
A chest radiologist (expert 1 with 10 years of experience in chest imaging),
selected a set of 40 CXR from our local Picture Archiving and Communication System (PACS).
Each CXR was chosen to fall into one of these 3 categories: CXR expected to mobilize detection skills (n=16),
CXR expected to mobilize interpretation skills (n=17) and normal CXR (n=7) (Figure 2).
After deidentification,
the chest radiographic images were copied from PACS,
converted into JPEG format and loaded onto a PowerPoint (Microsoft Corp,
Redmond,
Washington) slide show in a random order without mention of their category.
CXR validation phase
The 40 CXR were read by 2 independent radiologists (expert 2 and expert 3 with 19 and 8 years of experience in chest imaging,
respectively) during a single one-hour session.
For each radiograph,
the experts were asked to answer 3 questions on a printed survey sheet: 1.
Do you consider this radiograph normal or abnormal? 2.
Does it require a complementary CT study? 3.
What is your final diagnosis? The answers to the first and second questions were binary while the answer to the third question was open ended.
The readers were informed that some CXR could be normal.
CXR experiment phase:
A total of 81 radiology residents from 6 university hospitals of the West of France (Angers,
Brest,
Nantes,
Poitiers,
Rennes,
Tours) participated to the study on the occasion of two inter-regional training courses on chest CT imaging hold at the university hospital of Nantes (France) the 6th of March 2015 and the 5th of April 2017.
During these two days,
an identical experiment session was organized gathering 36 and 45 residents,
respectively.
The residents were first asked to fill in a 5-minutes questionnaire including the following items: university hospital of origin,
year of residency,
attendance at at least one training course on CXR during their residency,
the average number of CXR read per week,
and personal need for further CXR training course.
Then,
a one-hour reading session was organized in the same way as for experts,
therefore including the same 40 CXR displayed in the same order.
Statistical analysis:
Only CXR having received a full consensus of the 3 experts for the 3 questions were included in the analysis.
For each question of each CXR,
experts’ consensus was used as the reference against which residents’ answers were evaluated.
Residents’ mean success rates were expressed in percentage ± standard error for all CXR,
for normal CXR (normal CXR category) and for abnormal CXR including those expected to mobilize detection skills (detection CXR category) and those expected to mobilize interpretation skills (interpretation CXR category).
Distribution of percentages across residents was assessed by calculating the median and interquartile range (IQR).
Residents’ answers were assessed according to their year of residency (postgraduate year (PGY) 1-2 versus PGY 3-5),
the attendance at at least one CXR training course during their residency,
and the average number of CXR read per week (≤10 versus >10).
Comparisons of success rates were performed using Wilcoxon signed rank test.
Statistical analyses were performed using SPSS software (SPSS Inc.
Released 2006.
SPSS for Windows,
Version 15.0.
Chicago: SPSS Inc.).
P values less than 0.05 were considered significant.