Analysis of an electronic radiology-pathology correlation process for breast imaging
Lu Mao (lmao@biostat.wisc.edu)
2024-07-26
Data processing
The original dataset contains 5015 records. Among them 48 have “Breast Aspiration Cyst” as the first procedure, all of which are followed immediately by ultrasound and are thus assigned as such. Besides, there are 47 mammograms not followed by any of MR, stereo, or ultrasound. These records are deleted, leaving a dataset of N=4968 for analysis.
Comparison between pre- and post-RadPath
Variables are compared between pre- and post-RadPath (starting 2/1/2022) in Table 1 below. Categorical variables are summarized by \(N(\%)\) and tested by the chi-square test; Quantitative variables are summarized by median (inter-quartile range; IQR) and tested by the Wilcoxon rank sum test.
| Overall (N=4968) | Pre (N=1820) | Post (N=3148) | p-value | |
|---|---|---|---|---|
| Procedure - MR | 350(7%) | 119(6.5%) | 231(7.3%) | 0.328 |
| Procedure - Stereo | 1466(29.5%) | 556(30.5%) | 910(28.9%) | |
| Procedure - US | 3152(63.4%) | 1145(62.9%) | 2007(63.8%) | |
| Lesion - Benign | 3077(61.9%) | 1090(59.9%) | 1987(63.1%) | 0.07 |
| Lesion - High Risk | 215(4.3%) | 80(4.4%) | 135(4.3%) | |
| Lesion - Malignant | 1676(33.7%) | 650(35.7%) | 1026(32.6%) | |
| Path to call (days) | 1 (1, 1) | 1 (1, 1) | 1 (1, 1) | 0.066 |
| Path to addendum (days) | 1 (1, 5) | 2 (1, 6) | 1 (1, 4) | <0.001 |
| Path to BIP (days) | 7 (4, 8) | 7 (4, 8) | 7 (4, 8) | 0.129 |
Class of lesion is borderline significant, with more benign and less malignant cases post RadPath. Time from path result to addendum is significantly lower post RadPath. (The difference in path to call is harder to see because data points are clustered at 1. But post actually tends to be lower than pre).
Figure 1. Boxplots of times from path result to call, addendum, and BIP
by pre- and post-RadPath.
Variations over time
Temporal trends of times from path result to call, addendum, and BIP are plotted below. Data points are grouped by month. Solid line: median; dashed line: mean; shaded area: inter-quartile range (IQR).
Without BIP
Comparison between pre- and post-RadPath
| Overall (N=4072) | Pre (N=1407) | Post (N=2665) | p-value | |
|---|---|---|---|---|
| Procedure - MR | 256(6.3%) | 73(5.2%) | 183(6.9%) | 0.004 |
| Procedure - Stereo | 1169(28.7%) | 443(31.5%) | 726(27.2%) | |
| Procedure - US | 2647(65%) | 891(63.3%) | 1756(65.9%) | |
| Lesion - Benign | 2758(67.7%) | 1011(71.9%) | 1747(65.6%) | <0.001 |
| Lesion - High Risk | 100(2.5%) | 48(3.4%) | 52(2%) | |
| Lesion - Malignant | 1214(29.8%) | 348(24.7%) | 866(32.5%) | |
| Path to call (days) | 1 (1, 1) | 1 (1, 1) | 1 (1, 1) | 0.116 |
| Path to addendum (days) | 1 (1, 4) | 1 (1, 5) | 1 (0, 3) | <0.001 |
Variations over time
Temporal trends of times from path result to call and addendum are plotted below. Data points are grouped by month. Solid line: median; dashed line: mean; shaded area: inter-quartile range (IQR).
