In April 5, 2020, we launched an internet-based survey study examining the impact of the COVID-19 pandemic on mental health and well-being. Participants were invited to take part in a study that focused on “how the pandemic is affecting you” through social media (ie, Twitter and Facebook), listservs, social networks, websites (eg, Buzzfeed), and the research match website at Columbia University. Information about this study was included in the beginning of the survey; participants indicated consent by proceeding to the survey. The survey comprised multiple validated instruments as well as items that were developed to help us understand experiences that were specific to the COVID-19 pandemic. The mean completion time for the full survey was approximately 35 minutes. As this study was not funded, participants were not offered compensation. At the culmination of the survey, participants were provided with a list of resources (eg, crisis hotlines) in case they experienced distress. Participants were also asked to provide their email addresses if they consented to being recontacted for future follow-ups. Participants who provided their email addresses were surveyed again between April 26 and May 5, 2020. All study procedures were reviewed and approved by the Columbia University Institutional Review Board.

Participants were asked about any interruptions they experienced to their health care in the following item, which was developed specifically for this study: “Have you experienced any delays or interruptions in your healthcare (e.g. cancelled or delayed appointments, tests, procedures) during the coronavirus outbreak?” Response options for this item were dichotomous (“yes” or “no”), and participants who responded affirmatively were asked to elaborate with an open-ended, text-based answer. Responses ranged from no delayed care to multiple types of delayed care. Some participants included more detailed responses.

To characterize and code the types of delays, we used Microsoft Excel. Three researchers (ELP, DR, and HG) coded participants’ responses to the open-ended item about the types of health care delays and interruptions they experienced [15,16]. The coding process was as follows. First, one researcher (DR) reviewed a set of 25 randomly selected responses and developed a draft codebook, which primarily focused on the types of care affected. Second, the three researchers collaboratively reviewed the codebook draft and planned the logistics of the coding process. Each researcher coded the same 50 responses, identified coding uncertainties, and noted potential codebook edits. To establish interrater reliability, the three researchers discussed coding discrepancies until they were resolved with 100% agreement among all three researchers. The codebook was edited to clarify codes, add additional codes, and remove codes. Third, each researcher coded a third of the remaining responses. Any uncertainties were resolved by communication and discussion with the other two researchers. Finally, upon completion, researchers debriefed to further refine the codebook (eg, adding a dermatology code based on its frequency of appearance in the data set). All codes (n=2776) were integrated into a single spreadsheet.

Basic descriptive statistics were used to describe the research sample. We conducted two sets of inferential procedures to determine which factors were associated with self-reported health care delays, with a focus on three outcome variables. First, we examined the determinants of experiencing any health care delays due to the pandemic (dichotomous variable; “yes” or “no”). Among those who experienced delays, we were particularly interested in the following three types of delays: dental, preventive, and diagnostic care delays. For each dependent variable, bivariate analyses were conducted to describe the associations of the outcomes with demographic characteristics or self-reported levels of worry about their general health and COVID-19. Independent samples t tests were used for continuous independent variables, whereas chi-square tests were used for categorical variables. All independent variables were then entered into a multiple logistic regression model; R (The R Foundation) [17] was used for the statistical inference procedures.

Table 3 shows the associations between the dependent variable “any health care delays” and independent variables as a series of separate bivariate tests. Several demographic factors were associated with reporting any health care delays resulting from the pandemic. On average, older age, cisgender women, White individuals, higher levels of education, higher levels of self-reported worries about general health, and higher levels of self-reported worries about COVID-19 were all associated with experiencing health care delays. We then entered all independent variables, regardless of their significance at the bivariate level, into a multiple logistic regression model. The results from the logistic regression models indicated that age (P<.001), gender identity (P<.001), race (P=.005), education (P=.007), and the degree of worry about COVID-19 (P<.001) and general health (P<.001) were all significantly associated with experiencing health care delays due to the pandemic. The largest effect was observed for gender; the odds of patient-reported health care delays for cisgender men were 0.60 (95% CI 0.46-0.78) times that of cisgender women.

We also examined delays in specific types of health care—dental, preventive, and diagnostics care. Both the bivariate tests (Table 4) and the multiple logistic regression model (Table 5) showed that the only significant independent variable that was associated with dental delays was self-reported worries about general health. Increased levels of worry were associated with fewer delays in dental care (P=.002). Our bivariate analyses identified two factors that were associated with experiencing delays in diagnostics (Table 4)—age and self-reported worry about general health. In the multiple logistic regression model, both factors remained significant at the .05 level of confidence. Age had a larger effect on experiencing delays in diagnostics. Every 10-year increase in age increased the odds of experiencing delays in diagnostics by a multiple of 1.36. The bivariate tests of association for delays in preventive care showed that age (P=.03) and race and ethnicity (P=.02) were significant. Neither of these factors remained significant in the multiple logistic regression model, and although there was trend between race and delays (people of color experienced fewer delays than White participants), it was not significant (P=.06).

In June 2020, Dr Lasic, an interventional cardiologist at Jamaica Hospital Medical Center and Lenox Hill Hospital in New York, made the following prediction: “I think the toll on non-COVID patients will be much greater than COVID deaths” [18]. To our knowledge, ours is the only study that aimed at understanding the toll on patients without COVID-19 by describing rates of multiple types of health care delays as well as characterizing people who were the most affected by these delays during the beginning of the pandemic. In June 2020, under dynamic circumstances, researchers found that 32% and 12% of adults deliberately delayed or avoided routine care and emergency care, respectively [19]. Our findings suggest that a not insignificant proportion of the US populace may be experiencing delayed or interrupted care as a result of the pandemic, which may lead to decrements in health over the long term. Across all types of delays, older people, people with higher levels of education, and people who were more worried about their health were more likely to report delays or interruptions in their care. Men and people of color were less likely to report delays or interruptions.

Our findings highlight that dental care appears to be the most impacted (reported by over one-third of participants; 351/921, 38.1%). This is unsurprising, given that many dental offices closed at the outset of the pandemic due to concerns of patient and provider safety. However, studies have shown a relationship between dental health and heart disease, between dental health and diabetes, and between dental health and prenatal outcomes [20-22]. This is suggestive of the potential, downstream, deleterious health implications of delayed or interrupted dental care. Dentists are often also on the frontlines of identifying child abuse [23] and intimate partner violence [24], of which both have reportedly increased in incidence during the pandemic [25-27]. Further, a recent study conducted in Qatar has suggested that routine oral care is associated with a reduction in the risk of COVID-19 complications (eg, hospitalization and ventilation) [28]. Several months into the pandemic (July 2020), the American Dental Association released a policy declaring that dentistry was essential care (ie, care that is integral to systemic health) [29]. Of great concern are our findings associated with the frequency of reported delays in preventive and diagnostic care, particularly delays in such care for older adults. The implications of this finding are described in the Introduction section.

The lessons learned from the initial surge of COVID-19 cases can help mitigate potential future disruptions. Mitigation strategies for future disruptions include effective patient prioritization and triage [30-34]. For instance, Medically Necessary Time-Sensitive Prioritization is a scoring system for surgical triage that integrates multiple factors associated with the environment, patient risk, and ethics that are intended to be generalizable across hospital settings [35]. Additionally, the Centers for Medicare & Medicaid Services have released guidelines for the provision of non–COVID-19 care [36-38], However, considerations should also include patient life factors (eg, family support and transportation) as well as systems factors at the hospital (eg, visitation policy), local and state (eg, infection control mandates), or national levels (eg, policies). Using systems thinking perspectives is necessary for addressing complex problems [39]. Cancer care providers have also recommended balancing delays against COVID-19 risk, practicing effective social distancing, and managing the appropriate allocation of resources [32]. Key to reducing delays among patients is the relaying of accessible, clear, and actionable messaging by health care systems about how and when to safely access health care.

More research on health care delays and interruptions during the pandemic is needed. Further investigation is needed into the differences between the care of chronic illness and emergent conditions. For instance, the first pandemic surge resulted in a decline in hospitalization rates and lengths of hospital stay associated with acute cardiovascular conditions as well as other common conditions, such as acute appendicitis, bone fractures, cancer, and live births. Given the evidence on differences based on sex [40], race and ethnicity [41], sexual and gender identity [42,43], and other key factors, we need to investigate sociodemographic differences in delayed or interrupted care. Further, the significant economic effects of the COVID-19 pandemic as well as the high rates of job loss may result in the loss of insurance. Thus, more work is needed to understand how these factors affect health care. The potential impacts on patient-level outcomes such as mental health, quality of life, and pain must also be considered.

There are some key limitations that need to be considered. First, this study relied on a convenience sample that is not representative in terms of race, ethnicity, education, age, and gender. Convenience samples are limited in terms of the generalizability of prevalence rates. However, they have the advantage of efficiency, which is critical in a dynamic situation, and can be useful for understanding patterns and relationships associated with a phenomenon of interest. Recent research has shown that internet-based convenience samples correlate with truly random probability samples to a surprising extent. For example, Mullinix et al [44] found a correlation of 0.75 when they conducted 20 experiments that replicated results based on national probability samples on Amazon Mechanical Turk (an internet-based survey platform). More recently, Coppock [45] conducted 15 replication experiments and similarly found that convenience and national probability samples provided similar estimates of treatment and moderator effects. However, despite the limitation of using a convenience sample, our study is unique and critical to documenting experiences at the outset of the pandemic, given that the conditions under which the data were collected are not replicable. In addition, a study conducted over the internet does not reach individuals who do not have access to the internet. However, this is perhaps the only mechanism for capturing data from a fleeting moment in time that is characterized by dynamic and uncertain circumstances.

The pandemic has amplified people’s attention to health disparities and inequities in the United States and has created a need for inclusive and equitable research. The limitations of this study highlight pervasive gaps and challenges associated with conducting such research (and doing so efficiently, given the dynamic nature of the COVID-19 pandemic). There is an urgent need for sharing lessons learned, disseminating effective strategies for reaching more diverse populations (eg, engaging leaders of marginalized communities, understanding and addressing research hesitancy, etc), and encouraging the research community to use and improve upon these strategies in future research. Although the conditions for this study cannot be replicated, the methodological lessons learned can serve as a sort of pilot study for future crises, thereby creating more diverse and inclusive bodies of research that drive health equity forward. Without explicit discussions of research limitations, the research community cannot make progress in collecting data to inform the design of effective programs for addressing health inequities that have existed long before the COVID-19 health crisis. The impact of health care disruptions resulting from the COVID-19 pandemic may be difficult to measure in the short term. However, characterizing these disruptions and improving research methods for such characterizations is critical to informing systemic and equitable planning, mitigation, and recovery strategies for the long term.