Fatigue as a Driver of Overall Quality of Life in Cancer Patients

Abstract

Background: This manuscript describes an approach for analyzing large amounts of disparate clinical data to elucidate the most impactful factor(s) that relate to a meaningful clinical outcome, in this case, the quality of life of cancer patients. The relationships between clinical and quality of life variables were evaluated using the EORTC QLQ-C30 global health domain--a validated surrogate variable for overall cancer patient well-being.

Methods: A cross-sectional study design was used to evaluate the determinants of global health in cancer patients who initiated treatment at two regional medical centers between January 2001 and December 2009. Variables analyzed included 15 EORTC QLQ-C30 scales, age at diagnosis, gender, newly diagnosed/ recurrent disease status, and stage. The decision tree algorithm, perhaps unfamiliar to practicing clinicians, evaluates the relative contribution of individual parameters in classifying a clinically meaningful functional endpoint, such as the global health of a patient.

Findings: Multiple patient characteristics were identified as important contributors. Fatigue, in particular, emerged as the most prevalent indicator of cancer patients' quality of life in 16/23 clinically relevant subsets. This analysis allowed results to be stated in a clinically-intuitive, rule set format using the language and quantities of the Quality of Life (QoL) tool itself.

Interpretation: By applying the classification algorithms to a large data set, identification of fatigue as a root factor in driving global health and overall QoL was revealed. The ability to practice mining of clinical data sets to uncover critical clinical insights that are immediately applicable to patient care practices is illustrated.

Fig 1. A pathway model of patient quality of life adapted from Wilson & Cleary, 1995.

The pathway generally progresses from left to right, starting with the construct of disease state, symptom status, functional status, overall quality of life and patient satisfaction with quality of life. Each construct is composed of multiple patient attributes and is also affected by individual and environmental characteristics.

Fig 2. An example decision tree generated from newly diagnosed patients.

To predict a patient’s global health level, start at the root node (top oval), traverse the branches–depending on the specific values of Individual patient data–and come to a leaf node (rectangle). The leaf node is the prediction of Low, Medium or High global health for that patient. Paths travelled from the root node to each leaf node can be restated as a conditional rule set listing the drivers of global health levels.

Fig 3. A second example of decision tree generated from newly diagnosed stage 4 patients.

This tree has role function as the root node (first split) and fatigue and pain as next splits. ‘N’ in each node represents the number of patients.