To Treat or Not to Treat—Balancing Benefits and Risks of Treatment Delay Among Patients With Cancer During the COVID-19 Pandemic

Decisions regarding the initiation of cancer therapy have become increasingly complex in the era of coronavirus disease 2019 (COVID-19). Health care professionals and patients must balance the benefits and risks of immediate treatment for cancer with the potential subsequent increases in the risk of COVID-19 and its associated complications, including death. Unlike approaches that triage cancer treatment using a tiered approach, Hartman et al¹ have developed a web-based comprehensive model (OncCOVID) to estimate the risk of delaying the initiation of cancer treatment for an individual patient with cancer. This model estimates risk for patients with cancer of varying types and stages, and it incorporates other known risk factors for COVID-19, such as comorbidities, patient age, local COVID-19 prevalence, and reproduction number (which reflects the mean number of new patients with infections by each case in a region). The model is based on the use of epidemiologic data regarding treatment delays and outcomes from patients with cancer to estimate the risk of delayed treatment for a specific patient based on individual risk profile. The model has been developed into a useful application with a simple and intuitive interface. The web-based interface provides an accessible platform that can be used on a smartphone. Many of the necessary inputs for the model have default values that can be modified (for quantities such as relative risk [RR] of infection and other parameters that may vary across cancer care practice or treatment setting), but the user is required to input data on patient-level characteristics.

The utility of this model is clear in several specific settings. First, when the pandemic overwhelmed health care settings in certain areas (eg, New York City and Detroit), health care professionals did not have the capacity to provide treatment for all patients with cancer, and decisions for care rationing had to be made. This scenario may arise again in autumn 2020 with a potential seasonal increase in cases. The OncCOVID model will be useful for cancer care facilities in high-prevalence areas that are required to triage patients because of limited resources. However, at the current time (late summer 2020), the model is unlikely to be used widely given the current infection rates and hospital capacities at levels that appear to allow most cancer care practices to maintain treatment recommendations without delays.

Second, because OncCOVID provides interpretable output regarding risks (eg, the number of weeks or months of life lost owing to a treatment delay of a specific duration), the results from OncCOVID can be used to facilitate treatment timing discussions with patients. The model estimates may help a reluctant patient understand the increased risks of delaying treatment and, vice versa, the estimates may help another patient understand that the risks of cancer treatment delay are minimal, and the patient can safely delay treatment until infection levels in their region have decreased. In other words, the OncCOVID results can provide recommendations to support a clinically appropriate rationale for treatment timing. Such a model could apply to other medical fields. Choosing to receive treatment now vs later is not a substantially different benefit or risk decision than is choosing between 2 potential treatment strategies or between less intense adjuvant treatment immediately vs more intense adjuvant treatment at relapse. In these cases, using data-based inferences to guide clinical recommendations can augment clinical judgment and personal experience.

Despite these strengths, there are still uncertainties when using OncCOVID, as there are with any estimation model. Based on data published earlier this year regarding China’s experience with COVID-19, Hartman et al¹ have set default levels for the RR of COVID-19 at relatively high levels for chemotherapy (RR, 2.5), hospital visits per day (RR, 3.47), and surgery (RR, 5.73). Based on current US health care settings and strategies for mitigating contagion, these levels are likely overestimates. However, it is admittedly a challenge to find more reliable or current estimates, although emerging data suggest that the risks of infection are not higher among cancer patients receiving treatment.² Furthermore, the risk of adverse outcomes associated with COVID-19 (eg, death or requiring ventilation) has not been found to be higher in recent large studies of patients with cancer in US, European, and hematological malignancy cohorts; however, the data are still evolving.^3-6 Another challenge for the comprehensive modeling of RRs in cancer treatment settings is the myriad combinations of treatment modalities, schedules, clinic visit patterns, and associations between specific drugs and infection susceptibility as well as the receipt of multimodality treatment plans among many patients. Hartman et al¹ have considered the risks of several discrete categories of treatment, but the complexities of cancer care make it almost impossible to tailor these assumptions to all cancer patients.

The OncCOVID model does allow risk inputs to be modified. Although a welcome feature of the model, most users will not have valid estimates available and will instead rely on the model defaults. It is recommended that the OncCOVID team regularly update these default values as new information becomes available. Users are encouraged to repeat the model estimation by entering different values for some of the modifiable parameters to assess the sensitivity of the model estimates to various assumptions. The model could also be extended or modified to allow the user to select categorical input for risk levels (eg, low or no risk [RR, 1], moderate risk [RR, 2], or high risk [RR, 4]). Encouraging users to consider multiple outputs to compare results will better characterize the uncertainty of the model estimates based on different assumptions. Another realistic scenario in the pandemic setting is to stop treatment after the patient has received a partial treatment schedule (eg, pause treatment after 2 cycles of chemotherapy instead of completing a planned course of 4 cycles). In this setting, other factors of the patient’s status (eg, adverse effects of chemotherapy, such as lymphopenia) may make patients who are currently receiving treatment more susceptible to infection than those who have yet to start treatment. However, the current OncCOVID model is based on delaying the start of treatment; the data entered in the model are based on days between diagnosis and the initiation of treatment, and they do not address delays in the midst of treatment.

Other sources of uncertainty are future infection levels and reproduction numbers. Based on various inputs, OncCOVID generally estimates lower levels of COVID-19 in approximately 3 months, regardless of the local infection statistics at the current time. Given the potential for a seasonal surge in late 2020; the uncertainty of local, state, and federal guidelines and recommendations; and variable human behavior and adherence to behaviors that limit disease spread, it is challenging to estimate, for any given locale in the US, whether the case numbers will be lower. If the infection rates are higher in a few months, delaying treatment may expose the patient to an increased risk of COVID-19 in later months in addition to the potential disease-based risks of delay.

The OncCOVID model was developed to provide an opportunity for cancer care professionals to triage patient care based on individual risk measures that indicate the potential risks (or benefits) of delaying treatment depending on the individual patient’s risk of developing COVID-19 illness. As described, OncCOVID also provides information to support clinical decision-making and facilitate discussions of the risks and benefits with patients. The authors are to be congratulated for developing a more personalized approach to decision-making that is based on data rather than consensus opinion. Estimates from the OncCOVID model should, however, be used with some degree of caution, and the model’s risk parameters should be updated as new information emerges to reduce uncertainty and increase the accuracy of the model’s estimates.

Corresponding Author: Elizabeth Garrett-Mayer, PhD, Center for Research and Analytics, American Society of Clinical Oncology, 2318 Mill Rd, Ste 800, Alexandria, VA 22314 (liz.garrett-mayer@asco.org).

Published Online: October 29, 2020. doi:10.1001/jamaoncol.2020.4886

Conflict of Interest Disclosures: Dr Rini reported receiving grants and personal fees from AVEO Pharmaceuticals, Bristol Myers Squibb, Genentech, Merck, and Pfizer and personal fees from Alkermes, Aravive, AstraZeneca, and GlaxoSmithKline outside the submitted work. No other disclosures were reported.