Abstract
Cancer patients may undergo lengthy and painful chemotherapy treatments, comprising several successive regimens or plans. Treatment inefficacy and other adverse events can lead to discontinuation (or failure) of these plans, or prematurely changing them, which results in a significant amount of physical, financial, and emotional toxicity to the patients and their families. In this work, we build treatment failure models based on the Real World Evidence (RWE) gathered from patients' profiles available in our oncology EMR/EHR system. We also describe our feature engineering pipeline, experimental methods, and valuable insights obtained about treatment failures from trained models. We report our findings on five primary cancer types with the most frequent treatment failures (or discontinuations) to build unique and novel feature vectors from the clinical notes, diagnoses, and medications that are available in our oncology EMR. After following a novel three axes - performance, complexity, and explainability - design exploration framework, boosted random forests are selected because they provide a baseline accuracy of 80% and an F1 score of 75%, with reduced model complexity, thus making them more interpretable to and usable by oncologists.
Key findings
- 1Treatment failure, defined: a chemotherapy plan that is discontinued, or changed prematurely, because of treatment inefficacy or other adverse events.
- 2Built on real-world evidence: the models draw on patient profiles in CureMD's oncology EMR/EHR system.
- 3Five primary cancer types: the ones with the most frequent treatment failures or discontinuations.
- 4Novel feature vectors: engineered from the clinical notes, diagnoses and medications held in the oncology EMR.
- 5Boosted random forests selected: after a novel three-axes design exploration of performance, complexity and explainability, they delivered 80% baseline accuracy and a 75% F1 score with reduced model complexity, making them more interpretable and usable for oncologists.
Why this matters
A chemotherapy plan that is discontinued or changed early costs the patient physically, financially and emotionally. This paper shows that the likelihood of that outcome can be estimated from data an oncology EHR already holds when the regimen is chosen, and that the model chosen to do it can be one an oncologist can read.
It is the foundational method behind CureMD's AI Chemotherapy model, and the first of a series of CureMD Research papers on treatment failure, explainability and disease surveillance listed below.
Keywords
- Boosting
- Random Forests
- Chemotherapy
- Treatment Failure
- Feature Engineering
Editors and affiliations
Jose M. Juarez
University of Murcia, Murcia, Spain
Mar Marcos
Universitat Jaume I, Castellón de la Plana, Spain
Gregor Stiglic
University of Maribor, Maribor, Slovenia
Allan Tucker
Brunel University London, Uxbridge, UK
References
- 1.Arezzo, F., et al.: A machine learning tool to predict the response to neoadjuvant chemotherapy in patients with locally advanced cervical cancer. Appl. Sci. 11(2), 823 (2021). doi:10.3390/app11020823
- 2.Jabbari, F., Villaruz, L.C., Davis, M., Cooper, G.F.: Lung cancer survival prediction using instance-specific Bayesian networks. In: Michalowski, M., Moskovitch, R. (eds.) AIME 2020. LNCS (LNAI), vol. 12299, pp. 149 to 159. Springer, Cham (2020). doi:10.1007/978-3-030-59137-3_14
- 3.Elixhauser, A., Steiner, C., Harris, D.R., Coffey, R.M.: Comorbidity measures for use with administrative data. Medical Care, pp. 8 to 27 (1998). doi:10.1097/00005650-199801000-00004
- 4.French, J., et al.: Identification of patient prescribing predicting cancer diagnosis using boosted decision trees. In: Riaño, D., Wilk, S., ten Teije, A. (eds.) AIME 2019. LNCS (LNAI), vol. 11526, pp. 328 to 333. Springer, Cham (2019). doi:10.1007/978-3-030-21642-9_42
- 5.Kashef, A., Khatibi, T., Mehrvar, A.: Treatment outcome classification of pediatric acute lymphoblastic leukemia patients with clinical and medical data using machine learning: a case study at Mahak hospital. Inform. Med. Unlocked 20, 100399 (2020). doi:10.1016/j.imu.2020.100399
- 6.Kluwer, W.: Medi-span Generic Product Identifier (GPI) (2019)
- 7.Malakouti, S., Hauskrecht, M.: Predicting patient's diagnoses and diagnostic categories from clinical-events in EHR data. In: Riaño, D., Wilk, S., ten Teije, A. (eds.) AIME 2019. LNCS (LNAI), vol. 11526, pp. 125 to 130. Springer, Cham (2019). doi:10.1007/978-3-030-21642-9_17
- 8.Mishina, Y., Murata, R., Yamauchi, Y., Yamashita, T., Fujiyoshi, H.: Boosted random forest. IEICE Trans. Inf. Syst. 98(9), 1630 to 1636 (2015). doi:10.1587/transinf.2014OPP0004
- 9.Moghadas-Dastjerdi, H., Sha-E-Tallat, H.R., Sannachi, L., Sadeghi-Naini, A., Czarnota, G.J.: A priori prediction of tumour response to neoadjuvant chemotherapy in breast cancer patients using quantitative CT and machine learning. Sci. Rep. 10(1), 10936 (2020). doi:10.1038/s41598-020-67823-8
- 10.Ribba, B., et al.: A review of mixed-effects models of tumor growth and effects of anticancer drug treatment used in population analysis. CPT: Pharmacometr. Syst. Pharmacol. 3(5), 1 to 10 (2014). doi:10.1038/psp.2014.12
