Mahboubeh Yousefi 1, Elham Haem 1, Marziyeh Doostfatemeh 2, Laleh Mahmoudi 2, Raziyeh Kheshti 2, Mats O. Karlsson 3, Gustaf Wellhagen 3
1 Department of Biostatistics, School of Medicine, Shiraz University of Medical Sciences (Shiraz, Iran), 2 Department of Clinical Pharmacy, School of Pharmacy, Shiraz University of Medical Sciences (Shiraz, Iran), 3 Department of Pharmacy, Uppsala University (Uppsala, Sweden)
Introduction: Oxaliplatin-based chemotherapy is vital for patients with colorectal cancer [1]. However, the treatment can cause serious side effects such as neuropathy, and affect the quality of life of patients [2]. Melatonin has been proposed to alleviate especially neuropathy [3]. Different questionnaires are available to quantify the side effects. Such data is typically bounded and integer-valued by nature, however the standard practice when modeling is to assume it is continuous. The bounded integer model is a more recent modeling approach which treats such data in accordance with its nature [4].
Objectives: To evaluate the efficacy of melatonin on neuropathy, quality of life (QoL) and adverse events measured through clinical outcome assessments (COA) in oxaliplatin-based chemotherapy, through bounded integer (BI) modeling, as well as comparing against the standard continuous variable (CV) modeling approach.
Methods: Data came from a randomized, placebo-controlled clinical study on oxaliplatin-based chemotherapy, with 40 subjects in the placebo arm, 40 receiving melatonin. Three COA for neuropathy, quality of life and adverse effects were available longitudinally: European Organization for Research and Treatment of Cancer Quality of Life Questionnaire Chemotherapy-Induced Peripheral Neuropathy 20 and European Organization for Research and Treatment of Cancer Quality of Life Questionnaire and an acute adverse effect questionnaire (AAE). Both BI models and CV models were developed for each subscale total score, independently, parameterized with fixed effect parameters baseline and linear slope, both with inter-individual variability and estimated correlation. In a reduced model, both treatment arms were assumed to have the same slope, while a full model included a different fixed effect slope estimate for the treatment group. When necessary, a logit transformation of the CV model was done. The effect of melatonin was assessed through the objective function value (OFV) via the likelihood-ratio test between full and reduced model. Model performance was also evaluated through visual predictive checks (VPC) and residual diagnostics. From these developed models, stepwise covariate modeling was performed to assess potential covariate effects on model parameters. NONMEM 7.4.4 was used for model fitting.
Results: The ∆OFV between full and reduced models for BI and CV models for each subscale was: 0.36 vs. 0.26 (QoL functional), 0.003 vs. 0.25 (QoL symptom), 2.72 vs. 0.15 (QoL QL2), 0.05 vs. 0.57 (Neuropathy sensory), 0.63 vs. 2.16 (Neuropathy motor), 0.10 vs. 0.11 (Neuropathy autonomic) and 0.006 vs. 0.95 (AAE), respectively. Neither BI or the CV approach could identify a significant melatonin effect on neuropathy, QoL or AAE. The BI model approach provided a robust option to model all subscales, while the CV models needed logit transformation of the dependent variable when there was substantial amounts of data on the boundaries, which requires an arbitrary widening of the interval, else the model did not converge. The VPC plots were in favour of BI over CV. The BI approach only identified covariates (age and cancer stage) on the Neuropathy motor subscale, while the CV approach identified prior surgery for cancer on the Neuropathy sensory and motor subscales, and tumor location on the AAE scale. All covariates were on the baseline parameter.
Conclusions: No melatonin effect on neuropathy, QoL or AAE could be identified in this study. The BI approach was robust to data on the boundaries of the scale, did not require logit (or other) transformations and provided more realistic simulations that were naturally within scale boundaries and integer values, while the CV approach can lead to predictions outside the range and requires rounding or other post-processing to produce all integer values. While direct comparison of objective function values between BI and CV models is disputed, the current study shows the benefits of the bounded integer approach. The BI approach was less prone to pick up covariate effects, which could be an effect of a better overall description of data, whereas the CV approach identified covariate effetcs on three subscales, maybe owing to poor description of these scale in the first place due to large amounts of data on the boundaries.
References:
[1] Bleiberg, H., Oxaliplatin (L-OHP): a new reality in colorectal cancer. British journal of cancer, 1998. 77(4): p. 1-3.
[2] Beijers, A.J., et al., Peripheral neuropathy in colorectal cancer survivors: the influence of oxaliplatin administration. Results from the population-based PROFILES registry. Acta Oncologica, 2015. 54(4): p. 463-469.
[3] Alonso-Alconada, D., et al., Neuroprotective effect of melatonin: a novel therapy against perinatal hypoxia-ischemia. International journal of molecular sciences, 2013. 14(5): p. 9379-9395.
[4] Wellhagen, G.J., M.C. Kjellsson, and M.O. Karlsson, A bounded integer model for rating and composite scale data. The AAPS journal, 2019. 21(4): p. 74.
Reference: PAGE 34 (2026) Abstr 12086 [www.page-meeting.org/?abstract=12086]
Poster: Methodology - New Modelling Approaches