A new and simple model for predicting risk of severe neutropenia in advanced lung cancer could ease the evaluation process and help inform patient management decisions, according to investigators.
The model, based on 10 pretreatment variables, appears to overcome limitations of a comprehensive risk prediction model that is not specific to lung cancer, according to Xiaowen Cao of Duke University, Durham, N.C., and coinvestigators.
“We believe that this model, once validated, will help oncologists accurately identify those patients with lung cancer who are at a high risk of developing severe neutropenia, based on simple, readily available information,” the researchers said in a report on their work, which appears in the journal Lung Cancer.
Oncologists could then make “proactive” decisions about monitoring of high-risk patients, modifying the dose of chemotherapy, and using prophylactic growth factors, the authors added.
Accurate, lung cancer–specific prediction models would be useful to estimate risk of neutropenia, which the investigators acknowledged as a serious chemotherapy-induced toxicity linked to life-threatening infections, dose delays, and reductions that can compromise treatment efficacy, and reduced health-related quality of life.
There are other, previously developed models to predict chemotherapy-induced neutropenia, but those have significant limitations, including development based on small patient sample sizes, according to the researchers.
A comprehensive risk model for neutropenic complications has been developed by Gary H. Lyman, MD, and colleagues, based on a large, prospective cohort including nearly 3,800 patients. That model performs well and had a 90% sensitivity and 96% predictive value; however, it’s not lung cancer specific, and hasn’t been externally validated, according to Ms. Cao and coauthors.
Accordingly, they set out to develop a new risk prediction model based on a lung cancer data set encompassing 11,352 patients from 67 phase 2 or 3 cooperative group studies conducted between 1991 and 2010.
The Lyman model in this data set had an area under the curve (AUC) of 0.8772 in patients with small cell lung cancer (SCLC), but an AUC of just 0.6787 in non–small cell lung cancer (NSCLC), suggesting “much better predictive performance in the SCLC,” the researchers noted.
They used stepwise logistic regression and lasso regression to develop a new model, which was derived based on about two-thirds of the patients, randomly selected, while the validation was conducted using the remaining third.
Variables included in the final model included age, gender, weight, BMI, insurance status, disease stage, number of metastatic sites, chemotherapy agents used, number of chemotherapy agents, planned growth factor use, duration of planned therapy, pleural effusion, presence of symptoms, and performance status.
That model had a good AUC, according to investigators, in both the training set and the testing set (0.8348 and 0.8234, respectively).
“It is worth noticing that our final model compensated for the deficiency of Lyman’s risk model in NSCLC patients,” the researchers noted in a discussion of their results.
The study was supported in part by grants from the National Institutes of Health, the National Center for Advancing Translational Sciences, and the Health and Medical Research Fund of Hong Kong. One study coauthor reported a conflict of interest outside the submitted work related to Genentech.
SOURCE: Cao X et al. Lung Cancer 2020 Jan 5. doi: 10.1016/j.lungcan.2020.01.004.