Development and validation of prediction models for health-related quality of life outcomes after breast cancer surgery and reconstruction

Verheul, E.M. and Karsten, M.M. and Gebert, P. and Doppelbauer, L. and Borstnar, S. and Siesling, S. and Lingsma, H.F. and Vrancken-Peeters, J.M.C. and Stiggelbout, A.M. and Mureau, M.A.M. and Koppert, L.B. and van Klaveren, D. and Bak, M. and de Wreede, L.C. and Frank, K. and Hartman, L. and Hedayati, E. and Hackmann, T. and Hallsson, L.R. and Engelberts, Y. and Liu, Y. and Lal, D.M. and Korfage, I.J. and Nwosu, A.C. and Oliveira, C.C. and Perić, B. and Pazo-Cid, R. and Romero-Piqueras, C. and Rietjens, J. and Snelders, D. and Siebert, U. and Sroczynski, G. and Steyerberg, E. and van Mulligen, E. and Wit, T. and Wouters, M.W. (2026) Development and validation of prediction models for health-related quality of life outcomes after breast cancer surgery and reconstruction. European Journal of Surgical Oncology, 52 (4): 111466.

Abstract

Background Predictions of Health-Related Quality of Life (HRQoL) outcomes could support realistic recovery expectations after breast cancer (BC) surgery. We aimed to develop and validate prediction models for HRQoL outcomes after BC surgery. Methods We used three datasets of BC patients from Berlin, Germany; Ljubljana, Slovenia; and Rotterdam; Netherlands. We included non-metastasised patients who were surgically treated for an initial diagnosis of BC and completed pre- and postoperative validated questionnaires. We used linear mixed models to analyse 15 domains of the EORTC QLQ-C30 and EORTC QLQ-BR23 over a two-year horizon. Baseline domain score (measured pre-operatively), age, BMI, smoking, TN stage, receptor status, neoadjuvant chemotherapy, axillary surgery and surgery type (breast-conserving, mastectomy, and immediate implant-based reconstruction) were included as predictors. Predictive performance at validation was assessed by the proportion of variance explained (marginal R2; mR2). Results We included N = 795 patients from Germany for development and N = 623 from Slovenia and N = 417 from Netherlands for validation. The largest proportion of variance was explained by the prediction models for sexual functioning (SF, mR2 35%), physical functioning (PF, mR2 29%), body image (BI, mR2 26%), and cognitive functioning (CF, mR2 25%). The models captured meaningfully different trends over time for different outcomes and surgery types. The predictive performance of the models was largely driven by the baseline domain score. Performance was reasonable at external validation, with r2 values of 19–33% for PF, 10–17% for CF, 15–18% for BI, and 22–28% for SF, although some other outcomes (e.g. breast symptoms and role functioning) showed miscalibration, indicating a need for recalibration. Conclusion HRQoL after breast cancer surgery can be predicted using simple models with baseline domain scores and surgery type, demonstrating a new opportunity for Patient-Reported Outcome Measures (PROMs) in personalized care.