Category | Score | No-PNI (N = 52) | PNI (N = 25) | OR | P |
---|---|---|---|---|---|
Tumor enhancement > 2/3 | 0 | 8 (72.7%) | 1 (20%) | 2.42 | 0.213 |
1 | 1 (9.1%) | 3 (60%) | |||
2 | 2 (18.2%) | 1 (20%) | |||
Soft tissue stranding | 0 | 31 (62%) | 7 (30.4%) | 1.93 | 0.069 |
1 | 13 (26%) | 13 (56.5%) | |||
2 | 6 (12%) | 3 (13%) | |||
Perivascular stranding | 0 | 25 (50%) | 8 (34.8%) | 1.53 | 0.216 |
1 | 18 (36%) | 10 (43.5%) | |||
2 | 7 (14%) | 5 (21.7%) | |||
Peritumoral stranding | 0 | 27 (54%) | 10 (43.5%) | 1.78 | 0.143 |
1 | 21 (42%) | 9 (39.1%) | |||
2 | 2 (4%) | 4 (17.4%) |
Correlation of image findings with cholangiocarcinoma perineural invasion
Imaging findings vs perineural invasion (PNI)
Imaging categories:
Tumor enhancement greater than 2/3
Soft tissue stranding
Perivascular stranding
Peritumoral stranding
Because each category is rated by two readers, we assign a score 0, 1, or 2 by the number of readers rating positive. We then assess the association between the scores with the status of perineural invasion (PNI). For each category, an odds ratio (OR) is calculated from a (univariate) logistic regression of PNI vs score. The following table summarizes the results.
The following figure shows the prevalence of PNI within each level of the score (0, 1, 2).
The receiver operating characteristic (ROC) curve for PNI against each score is plotted below, with the area under the curve (AUC) and its 95% confidence interval labelled on the graph. Again, the results show that soft tissue stranding and peritumoral stranding are most predictive of PNI.
Between-reader agreement
For soft tissue stranding and peritumoral stranding, the two readers’ rating are cross-tabulated below.
Meg 0 | Meg 1 | |
---|---|---|
Steve 0 | 38 | 15 |
Steve 1 | 11 | 9 |
Meg 0 | Meg 1 | |
---|---|---|
Steve 0 | 37 | 29 |
Steve 1 | 1 | 6 |
The corresponding kappa coefficients (a measure of concordance) are calculated below.
Kappa | P | |
---|---|---|
Soft tissue stranding | 0.157 | 0.176 |
Peritumoral stranding | 0.150 | 0.035 |
Pathology vs PNI
To look at the association between tumor grade and PNI, we plot the prevalence of PNI for each grade level below, with OR (95% CI) calculated from a logistic regression model for PNI against tumor grade. There is a fairly strong positive association, but the association is not statistically significant.
Survival analysis
Proportion of patients who died by PNI status is plotted below.
The Kaplan–Meier curves for survival time from surgery to death are plotted below by PNI status.