Longitudinal Modeling in Surgical Oncology Research: A Primer Examining Patient-Reported Outcomes

Jacob Levy; Uche Amakiri; Ronnie L Shammas; Francis D Graziano; Jonas A Nelson; Lillian A Boe

doi:10.1002/jso.70006

Longitudinal Modeling in Surgical Oncology Research: A Primer Examining Patient-Reported Outcomes

J Surg Oncol. 2025 Jun 18. doi: 10.1002/jso.70006. Online ahead of print.

Authors

Jacob Levy¹, Uche Amakiri¹, Ronnie L Shammas¹, Francis D Graziano¹, Jonas A Nelson¹, Lillian A Boe²

Affiliations

¹ Plastic and Reconstructive Surgery Service, Department of Surgery, Memorial Sloan Kettering Cancer Center, New York, New York, USA.
² Department of Epidemiology and Biostatistics, Memorial Sloan Kettering Cancer Center, New York, New York, USA.

PMID: 40530785
DOI: 10.1002/jso.70006

Abstract

Background: Oncologic research increasingly prioritizes patient-reported outcomes (PROs) to support patient-centered care. Long-term evaluation of PROs requires longitudinal data analysis, which traditional cross-sectional methods, such as linear regression, cannot adequately address. Advanced statistical models, including linear mixed-effects (LME) and generalized estimating equations (GEEs), are essential to capture the complexity of longitudinal data. This study aims to provide a framework for applying LME and GEE models to analyze longitudinal PROs in surgical oncology research using a postmastectomy breast reconstruction cohort example.

Methods: A retrospective review was conducted on patients who underwent autologous or implant-based postmastectomy reconstruction from 2018 to 2021. Using longitudinally collected BREAST-Q data for up to 5 years, the study analyzed demographic and surgical factors associated with satisfaction with breasts (SAT) and sexual well-being (SEX) scores. Through the application of LME and GEE models, key features of longitudinal data were explored, the methodology and assumptions for each model were detailed, and their practical application was demonstrated.

Results: The analysis included 3269 patients. In both models, Asian race [LME, p = 0.002; GEE, p < 0.001], neoadjuvant [LME, p = 0.007; GEE, p = 0.007], and adjuvant radiation [p < 0.001; p < 0.001] were linked to lower scores for SEX. Complications were negatively associated with SEX as well [p = 0.044; p = 0.007]. Similar results were observed for SAT. Autologous reconstruction was linked to higher SAT and SEX scores at all postoperative time points compared to implant-based reconstruction.

Conclusions: This study demonstrates the utility of LME and GEE models in analyzing longitudinal PROs, providing insights into factors influencing breast reconstruction outcomes. Such models allow for practical analysis in surgical oncology, supporting the development of personalized patient care.

Keywords: linear mixed‐effects and generalized estimating equations; longitudinal data; surgical oncology.

Grants and funding

This study was partially funded by the NIH/NCI Cancer Center Support Grant P30 CA008748, which funds the research infrastructure at Memorial Sloan Kettering Cancer Center.