Interobserver Variability of Lymph Node Count in Pelvic Lymph Node Dissection for Prostate Cancer
Cheryl A Mather, Maria Westerhoff, Suzanne Dintzis, Rodney Schmidt, Lawrence True. University of Washington, Seattle, WA
Background: The prognostic significance of lymph node (LN) status is well established in prostate cancer. Data suggest that examination of greater than 10 LNs is associated with a 15% lower risk of prostate cancer death, even amongst patients with negative nodes. This study examines the variability of pathologist's role in LN counts for pelvic lymph node dissection (PLND).
Design: LN slides from 12 radical prostatectomy cases were distributed to four pathologists. Each pathologist was instructed to report the number of LNs per case. From this data interobserver agreement was calculated. Concomitantly, 696 prostatectomy cases from our database were analyzed. From these data, multiple regression was used to determine what factors were most significant to the number of LNs reported. The average number of LNs per pathologist was also calculated. From this data pathologists were grouped into quartiles. Average LN count was compared by quartile.
Results: Interobserver agreement of LN count was very poor (Fleiss kappa -0.12). Multiple regression analysis of LN counts using independent variables of pathologist, surgeon, organ weight, patient age, Gleason score, presence of metastases, extracapsular extension, extensive perineural invasion, lymphovascular space invasion, surgical stage and the number of LN packets submitted found that only the number of node packets was a statistically significant factor in the number of LNs counted (p=<0.0001). Quartile analysis of all pathologists showed that the lowest quartile (Q1) reported 5.5 fewer LNs on average than the top quartile (Q4, p=<0.0001, 95% CI 3.4 to 7.6 nodes). In the cases where only two LN packets were submitted, Q1 reported 3.5 fewer LNs when compared to Q4 (p=0.0009, 95% CI 1.5 to 5.5). Q2 and Q3 also reported significantly fewer nodes than Q4 for all cases (p=0.001 and 0.007, respectively) and for those cases where only two LN packets were submitted (p=0.0012 and 0.026, respectively).
Conclusions: Interobserver agreement in LN counts in PLND is very poor. Analysis of LN counts by all pathologists at our institution indicate that counting may be biased, since the top quartile of pathologists reports significantly more nodes than the other three, even when controlling for the most significant factor affecting LN counts (number of packets submitted). These data suggest that pathologists do not use the same criteria for LN counts and calls into question studies use LN counts as a metric of clinical outcome or quality of surgery or pathology. We recommend developing a national consensus for how pelvic LNs are counted.
Category: Quality Assurance
Monday, March 19, 2012 1:00 PM
Poster Session II # 267, Monday Afternoon