In Assessing Factors Associated with Axillary Lymph Node Involvement in Breast Cancer Patients: An Application of Zero Inflated Models
by
Alok Kumar Dwivedi
Department of Biostatistics, All India Institute of Medical Sciences, Ansari Nagar, New Delhi-110029, India
Coauthors: S.V.S. Deo, S.N. Dwivedi, Arvind Pandey

Objectives: Axillary lymph node involvement is one of the most significant prognostic factors for breast cancer patients. The observed form of nodal involvement follows count data having positively skewed distribution with high proportion of zeros. Hence, even transformation approach may not help. Further, its categorization may loose important information. Therefore, this presentation aims to find out factors associated with node involvement using zero inflated models.

Data and Methodology: The patients’ records collected during 1993 to 2005 in the Department of Surgical Oncology, Institute Rotary Cancer Hospital (IRCH), AIIMS, New Delhi were extracted from their computerized database. The patients who underwent surgery with axillary lymph node dissection were included in the study. However, cases with unknown primary site, recurrence, metastasis and bilateral involvement and male breast cancer patients were excluded from the study. Zero inflated Poisson and negative binomial regression models were used.

Results: A total of 1152 cancer patients could be included in the final analysis. The mean and standard deviation of age was reported 47.73 + 11.09 (Range 20, 86 years). The mean number of total node dissected was found 14.33 (standard deviation 5.54). The prevalence of nodal positivity was 705 (61.2%).The mean number of positive nodes was 3.86 (standard deviation 5.56) Therefore, the variance and mean ratio of number of node positivity was 8.00, which revealed wide over-dispersion. In the presence of excess zeros, zero inflated Poisson model was used which could not be able to explain over-dispersion completely and finally zero inflated negative binomial model was found to be appropriate for this dataset. The factors mainly parity, skin changes, primary site and pathological tumor size were found to be associated with axillary lymph node involvement.

Conclusion: The developed model may be useful in identifying nodal involvement leading to appropriate management of the patients

Date received: January 12, 2008