Leveraging electronic patient-reported outcomes collected at six cancer centers to identify symptom clusters across the first four weeks of chemotherapy.

Abstract

366 Background: Symptom clusters are groups of two or more related concurrent symptoms. Interventions that target symptom clusters may be more efficient than those which address individual symptoms, yet few studies have identified symptom clusters using patient-reported outcomes collected in routine practice. The purpose of this study was to identify symptom clusters de novo using data reported in eSyM, an electronic health record-integrated cancer symptom management program deployed across six large and diverse health systems. Methods: Adults starting chemotherapy to treat thoracic, gastrointestinal, and gynecologic cancers used eSyM to report symptoms twice weekly for up to six months. We analyzed patient-reported symptom data from cycle 1, day 2 (T0), one week (T1), and two (T2), three (T3), and four (T4) weeks later. Patients reported attributes (i.e., frequency, severity, and interference) of 12 symptoms using items derived from the PRO-CTCAE. We calculated a composite grade (0, 1, 2, or 3) for each symptom and conducted an exploratory factor analysis using polychoric correlations and unweighted least squares estimation to identify symptom clusters at each time point. We considered rotated factor loadings ≥|0.30| meaningful and assessed the Eigenvalues, structure, and clinical relevance of two- through four-factor (i.e., cluster) models. Finally, we labeled each cluster according to its predominant symptoms using labels adopted from prior studies of symptom clusters in patients undergoing chemotherapy. Results: Our analytic sample at each time point included 730 (T0), 1605 (T1), 1571 (T2), 1617 (T3), and 1456 (T4) symptom reports. Of the two-, three-, and four-factor models, the two-factor model had the least cross-loading and was the most clinically relevant across time points. The two-factor model accounted for 100% of the total variance in symptoms. In this model, a gastrointestinal symptom cluster was comprised of vomiting, nausea, decreased appetite, trouble drinking, and diarrhea at all time points, while a psychoneurological symptom cluster was comprised of fatigue, pain, anxiety, shortness of breath, and constipation at all time points. Rash was present in the psychoneurological cluster at T0 but was not present in any cluster at T1-T4. Conclusions: Patients experience gastrointestinal and psychoneurological symptom clusters in the first four weeks of chemotherapy. Our findings suggest it is feasible to use routinely collected data to identify symptom clusters using composite PRO-CTCAE grades. Future research will investigate the prevalence of symptom clusters, clinical and demographic characteristics of patients at high risk for severe symptoms, and how to better manage symptom clusters.