Are There More Accurate Ways to Measure What Patients Are Telling Us?

Abstract

We who specialize in oncology care know that children, adolescents, and adults with cancer experience multiple objective and subjective symptoms during disease-directed treatments. Increasingly, we are asking patients to report their subjective symptoms and adverse events during treatment to help us more accurately and completely determine the total impact of the therapies we are administering. This knowledge of treatment impact derived from patient-reported outcomes (PROs) and our efforts to understand such information could guide us in giving more individualized supportive care to our patients and their families. Our efforts to use PROs to tailor supportive care align well with this era of precision cancer treatments. However, our traditional approaches to analyzing PRO data are not at the level of each individual person but instead are at the level of “group” data analyses. Perhaps, the analytic approaches that keep us furthest away from using PROs to individualize care are the measures of central tendency. Certainly, these measures provide valuable information about a group of patients average symptom score, or the central value of 1 symptom compared with that of all other measured symptoms, or the most frequently reported symptom score from a certain set of patients, but these are about the “whole” of the studied group—not about individuals within the whole. In addition to being insufficiently informative for our purposes of using PROs to tailor care, central tendency values can mask important item toxicity findings for individuals or subgroups. Masking symptom, function, or quality-of-life scores that merit our clinical intervention prevents us from providing individualized supportive care. Using only central tendency measures with PRO data conveys an assumption of population homogeneity— that individuals with cancer are alike or essentially the same in their experience with cancer-related symptoms, functioning, and quality of life. We know that this assumption is very far from the reality of the cancer experience. Although important commonalities may exist, each person experiences cancer uniquely. For both clinical care and research reasons, we need analytic strategies that advance our PRO data analyses much closer to being able to use PROs to improve care of individuals affected by cancer. I believe this means that we need a methodological shift in how we are analyzing PROs. There are signs that a methodological shift is underway. Patient-reported outcome data are now being analyzed for subgroups, signaling that we recognize clinically important differences exist between persons experiencing cancer —although they may have the same diagnosis and be at the same time point of receiving the same cancer therapies. This methodological shift is reflected in our increasing use of analytic strategies such as latent class analyses (grouping patients by their symptom, function, or quality-of-life scores). This type of analysis places patients into similar subgroups matched by their degree of treatment impact at a specific treatment milestone or across time. This type of analysis could teach us whether symptoms or subjective adverse events vary at the subgroup level in ways that we could then anticipate and work to reduce or prevent the negative impact. This analytic strategy directly challenges and even defies the assumption of homogeneity by instead assuming heterogeneity of the cancer experience. The heterogeneity assumption moves us somewhat closer to our clinical reality of person-specific cancer experiences. While acknowledging the possibility of measurement error in using a single item for clinical care situations, we are using individual symptom item scores to predict the profile membership of pediatric oncology patients in terms of symptom suffering (high, moderate, or low) during cancer treatment. Single item scores are also being used to predict symptom bother, symptom distress, or symptom interference with function. This willingness to examine outcomes predicted by a single symptom item score adds to our clinical and research efforts to provide more evidence-based, tailored supportive care and thus a more precise use of PROs. Methodological shifts occur when findings from current research are a misfit with clinical realities. We are at that point now with needing PROs to be used in clinical care at the level of each person experiencing cancer. Most certainly, time is needed tomove the outcomes of a methodological shift into benefiting patient care, but we need to up the speed with which we are currently making this shift. Please do examine the symptom, function, and qualityof-life data bases within your reach for heterogeneity at the subgroup and item levels—advances in supportive care will very likely result from our collective efforts. With much regard for each of you,