Subgroup Discovery Analysis of Treatment Patterns in Lung Cancer Patients

Abstract

Lung cancer is the leading cause of cancer death. More than 235,000 new cases of lung cancer patients are expected in 2023, with an estimation of more than 125,000 deaths. Choosing the correct treatment is an important element to enhance the probability of survival and to improve patient’s quality of life. Cancer treatments might provoke secondary effects. These toxicities cause different health problems that impact the patient’s quality of life. Hence, finding patterns in treatment prescribed to specific patients capable of reduce toxicities while maintaining or improving their effectiveness is an important goal that aims to be pursued from the clinical perspective.In this work, we have analyzed a clinical dataset containing treatments for Lung Cancer patients. The analysis consisted of searching for patterns containing patient characteristics and prescribed treatments, looking at its associated outcomes in terms of disease progression and the presence of toxicities. A well-known method for pattern search is Subgroup Discovery (SD). However state-of-the-art algorithms present some limitations: the need for fine-tuning of key parameters on a dataset basis, usage of a single pattern search criteria with thresholds set by hand, usage of non-overlapping data structures for subgroup space exploration, and impossibility to search for patterns by fixing some relevant dataset variables. All these limitations result in discovered patterns with poor quality metrics and patterns with limited interest or relevance.In order to tackle mentioned limitations, we propose IGSD (InfoGained Subgroup Discovery), a new SD algorithm for pattern discovery that combines Information Gain and Odds Ratio as a multi-criteria for pattern selection. Additionally, two versions of IGSD are proposed to evaluate the dynamic adjustment of the search optimization thresholds during subgroup space exploration. Thus, a comparison is performed among FSSD, SSD++, and the proposed IGSD versions on the considered clinical dataset. Obtained patterns have been validated through clinical guidelines recommendations and domain experts ratings in order to evaluate the performance of the different algorithms compared. Evaluation results show that IGSD outperforms the other two methods in terms of clinicians acceptance rates and also in terms of SD quality metrics. Also, among SD metrics proposed in the literature, we show that confidence and ORR (Odd ratio rate) values seem to be better related to a higher acceptance rate for the considered dataset.

Note:
Funding declaration: We state that this paper has been supported by Fundación AECC and Instituto de Salud Carlos III (grant AC19/00034), under the frame of ERA-NET PerMed.

Conflict of Interests: We confirm that there are no known conflicts of interest associated with this publication and there has been no financial support for this work that could have influenced its outcome.

Ethics: The research that led to the results presented in this work was approved by the Ethics Committee of Universidad Politécnica de Madrid (Spain).

Keywords: Subgroup Discovery, Lung cancer, Pattern mining