Year: 2024
Type: editorial
Abstract: Treatment decision-making involves balancing the efficacy of an intervention against the harms. Although the efficacy measurement has been well established (eg, disease-free survival, progression-free survival, overall survival) - tools to measure the harms have been more challenging. Traditionally, toxicities have been reported as adverse events, which can be short term or long-lasting, and often do not reflect the patient’s experience, as studies have found that providers under report symptomatic toxicities (1). Patient-reported outcomes (PROs) provide valuable insights into patients’ perspectives on their health, quality of life, and the tolerability of the treatments and have been increasingly incorporated into both clinical trials and clinical practice. Collection of these data can be used to provide more tailored guidance on persistent, sudden, and severe adverse symptoms; to assist with clinical decision-making; to improve drug management and dose optimization; and to help providers mitigate adverse events by promoting effective and timely supportive care. Despite the increasing incorporation of PROs into clinical trials, challenges exist with regard to determining the optimal statistical methods for PRO analysis and data presentation. As a result, international standards and recommendations have been developed to guide investigators (2,3). In addition to methodologic challenges in designing the trial with regard to selection of appropriate PRO measures, the timing and frequency of data collection, patient compliance, and cultural and language differences in measures and regulatory standards are key considerations. Challenges also exist reporting reliability with regard to establishing appropriate and interpretable endpoints. For example, it is unclear whether studies should be designed to determine a minimally clinically relevant difference between groups at a specific time point or over time, or whether they should report the proportion of patients with a predefined deterioration in PRO score. In the current issue of the Journal, Dent et al. (4) report the prespecified PRO analysis from KEYNOTE-522, a phase 3 trial evaluating the addition of neoadjuvant and adjuvant pembrolizumab to standard chemotherapy for triple-negative early-stage breast cancer. The efficacy of this regimen has been previously reported, with an improvement of both event-free survival and, more recently, overall survival (5,6). Briefly, participants completed several PRO measures including the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire–Core 30 (EORTC QLQ-30), the EORTC Breast Cancer-Specific 12 Quality of Life Questionnaire (EORTC QLQ-BR23, and the EQ5D at 3 timepoints during the neoadjuvant phase; at 3 timepoints during the adjuvant phase; and then every 12 months for 2 years. Between-group differences in least-squares mean change from baseline (neoadjuvant baseline in neoadjuvant phase, adjuvant baseline in adjuvant phase) in PROs at the primary analysis timepoint were assessed using a constrained longitudinal data analysis model. The model implicitly treated missing data as missing at random. During the neoadjuvant phase, a substantial decline in global quality of life, physical function scores, and breast cancer specific scores in both groups were reported, with a trend to lower scores in those who received pembrolizumab, but without significance between groups. In the adjuvant phase, there was a modest improvement among these measures in both groups over time, with a similar nonsignificant trend of better improvement in the placebo arm. The authors concluded that no substantial differences in PRO assessments were observed between neoadjuvant pembrolizumab + chemotherapy followed by adjuvant pembrolizumab vs neoadjuvant placebo + chemotherapy. The study has some notable strengths. First, despite enrolling from 181 sites in 21 countries, PRO completion rates ranged between 80% and 90% at each timepoint, which is higher than PRO completion in other clinical trials (7,8). Standard survey instruments were also available in multiple languages, which enabled targeting of a demographically diverse population of patients receiving treatments. The authors also use the QLQ-BR23, a breast-specific survey asking patients a range of questions about their body image, sexual functioning, systemic therapy side effects, breast symptoms, arm symptoms, and hair loss. This gives us an overall understanding of symptoms both short term and long term that are important to women with breast cancer undergoing neoadjuvant therapy; however, most of these side effects are a reflection of the chemotherapy that was identical in both arms, and reassuring that the addition of pembrolizumab did not appear to influence these side effects beyond that of chemotherapy. As noted above, the authors reported that no substantial differences were observed with the addition of pembrolizumab to neoadjuvant chemotherapy. It should be noted that the primary article of KEYNOTE-522 reported treatment-related adverse events of 99% in the pembrolizumab arm (76.8% ≥grade 3) and 99.7% in the placebo arm (72.2% ≥grade 3) (5). The fact that everyone did so poorly overshadows any between-group differences. This high toxicity rate is reflected in significant declines in PROs in both treatment groups. This highlights that our efforts going forward should focus on reducing the harms from the chemotherapy backbone used in this regimen. One way to do this may be to de-escalate treatment early for those who appear to have a complete response. This is a strategy that is being evaluated by in clinical trials such as I-SPY 2 (Investigation of Serial studies to Predict Your Therapeutic Response with Imaging and Molecular AnaLysis 2), a neoadjuvant breast cancer trial for early-stage, high-risk patients. A key advancement in the trial is the notion of a targeted de-escalation or escalation strategy on the basis of response to therapy. This adaptive design may be an effective strategy to maximize benefits and minimize toxicity. Pembrolizumab has unique toxicities as a result of immune-related adverse events (irAEs). In the primary article, adverse events of interest (irAEs) differed between the groups (38.9% with 12.9% ≥grade 3 in pembrolizumab vs 18% and 1.8% ≥grade 3 in placebo). One may wonder why the PROs would be similar between groups despite these differences. This may be related to the choice to present overall mean change from baseline. Comparing QOL in patients who developed specific toxicities such as irAEs with those who did not may show differences that assist with decision making and toxicity mitigation strategies. Another option is to include the patient-reported outcomes version of the Common Terminology Criteria for Adverse Events (PRO-CTCAE) (9). These PROs were developed to evaluate symptomatic toxicity in patients on cancer clinical trials as a companion to CTCAE; a proportion analysis of each symptomatic adverse event across the 2 groups may highlight individual symptoms of high and low prevalence. Several other options for presenting PRO data can highlight differences between groups. These include line plots depicting individual-level data for a single domain across varying timepoints (10). This approach is most effective when the sample size is small. Challenges arise when multiple PROs are reported that use different scales with differing directions of favorability. To enhance interpretability, the use of a standardized effect size graph, which shows all PROs from a study on the same figure on the same scale has been shown to be effective, particularly when depicting changes from baseline within a population (11). Other presentation strategies may include visualizing symptom clusters associated with outcomes such as worsening QOL (12) or development of a grade 3 or higher adverse event. With many symptoms and toxicities reported within a single trial, there has been an increased interest in reporting summary measures to capture the overall tolerability or symptom burden from treatment. Number, duration, and severity of toxicity over time can have a profound impact on a patient’s quality of life. One measure increasingly being incorporated into clinical trials is the FACT-GP5 (13). The single item asks participants to score “I am bothered by side effects of treatment.” In recent studies of phase 3 trials, mean GP5 scores significantly differed by maximum adverse event grade (P <.001), with a clear trend toward increasing GP5 scores with level of increasing adverse event grade (13). The mean GP5 score has also been shown to be associated with early drug discontinuation (14) and trends with worsening QOL and increased symptom burden in neoadjuvant treated triple-negative breast cancer patients (15). In addition to comparing PRO in randomized trials, there is also increased interest in the use of PROs in early phase studies, where maximum tolerated dose (MTD) is used to define the dosing for subsequent efficacy trials (16). For example, there have been efforts in developing novel methods to estimate the MTD using both clinician and patient-reported toxicity data (17). Additionally, tailored PRO tools are being developed specifically for assessing tolerability in phase I trials using a condensed set of PRO-CTCAE items and can be used to identify dose-limiting toxicities (18). In summary, we congratulate the research team on their rigorous collection of PROs in KEYNOTE-522. Studies such as these underscore the need for PRO collection and modernized reporting guidelines to help patients and providers understand the added burden of any new treatment. Moving forward, real-time assessment and mitigation strategies for patients on new therapeutics may help guide toxicity management as well as provide guidance to practitioners and patients once drugs are approved and used among patients who may not be represented well in clinical trials. No new data were generated or analyzed for this editorial. Amrita Basu, PhD (Conceptualization; Writing—original draft; Writing—review & editing), Dawn L. Hershman, MD, MS (Writing—original draft; Writing—review & editing). No funding was use for this editorial. The authors have no disclosures. None declared.
DOI: 10.1093/jnci/djae157
Authors: Amrita Basu, Dawn L Hershman
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4401200630