Introduction

The world was caught off guard by the coronavirus disease 2019 (COVID-19) pandemic. As the pandemic spread and serious infections mounted, initial responses focused on flattening the infection curve and preserving health care resources. Guidelines encouraging masking, social distancing, and stay-at-home were promulgated by multiple agencies, and health care systems limited in-person patient visits,1, 2 access to diagnostic and screening procedures,3 and elective surgery due to a lack of capacity and concerns over patient and caregiver safety.4 It has been estimated that 41% of US adults delayed or avoided medical care during the early phases of the pandemic as a result of COVID-19-related concerns.5 The course of the pandemic in the United States has evolved over time having been impacted by recurrent infection waves,6 increasing availability of vaccines,7, 8 and an overall trend toward gradual reopening.

Cancer is the second leading cause of the death in the United States, and timely diagnosis and intervention are critical to ensure optimal outcomes. With this is in mind, we undertook a retrospective analysis to identify trends in new cancer cases within a large, community-based health care system following the onset of the COVID-19 pandemic. We report on data through the first 15 months of the pandemic to evaluate for both early and late effects, which to date have not been adequately explored.

Materials and Methods Study Cohort and Measures

This study includes all patients with an electronic medical record encounter between January 1, 2019, and May 31, 2021, linked to a new principal diagnosis of cancer at any disease site at any of 40 Providence St. Joseph Health (PSJH) system inpatient or outpatient facilities located in 5 western US states (Alaska, Washington, Montana, Oregon, and California). These 40 sites share an electronic medical record system that deposits data into a cloud data warehouse (CDW) on a daily basis. The CDW is readily searchable for a broad range of discrete and nondiscrete data fields including International Classification of Diseases, Tenth Revision codes using in-house, validated query logic. Encounter-related International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes were used to identify and define the population, whereas Surveillance, Epidemiology, and End Results (SEER) Program benign classification was used to exclude noninvasive cancers. Anatomically related cancers were combined into disease site groups for analysis purposes (Supporting Table 1). Primary study outcomes include the weekly number of new cancer cases and trends in weekly case numbers. Secondary outcomes include weekly numbers of new cancer cases by primary disease site, patient characteristics, and positive COVID-19 test rates. This study was approved by the PSJH institutional review board, which waived the informed consent requirement because of the minimal risk, retrospective nature of the study.

Statistical Analysis

A time series analysis with an autoregressive integrated moving average (ARIMA) model built on data from the time interval spanning January 1, 2019, through March 3, 2020 (pre-COVID-19) was performed using R package “forecast” to estimate the expected weekly post-pandemic cases and compared to observed post-COVID-19 cases that occurred between March 4, 2020 and May 31, 2021. The week of March 4, 2020, corresponds to initial reports of increasing COVID-19 infection rates and the onset of the pandemic in the United States. Segmented regression analysis was performed using R package “segmented” to identify the slope change in weekly cancer cases, with consideration of time dependence in the model. Using 3 identified break points (March 4, April 1, and June 10), we grouped cases into 1 of 4 periods for analysis: pre-COVID-19 (January 1, 2019, to March 3, 2020), early-COVID-19 (March 4, 2020, to March 31, 2020), mid-COVID-19 (April 1, 2020, to June 9, 2020), and late-COVID-19 (June 10, 2020, to May 31, 2021). The identified breakpoints correspond to the onset of abrupt changes in the observed trends in weekly cancer cases including a steep decline (March 4) followed by an intermediate (April 1) and gradual (June 10) recovery. Patient demographics and disease characteristics are summarized as descriptive statistics. Continuous variables are presented as means or medians, and categorical variables are presented as frequencies. Comparisons between groups were performed by ANVOA tests for continuous variables and χ2 or Fisher exact tests for categorical variables. All tests were 2-sided and statistical significance was set at P < .05. All statistical analyses were performed using R software (version 3.6.3).9

Results Population Demographics and Trends in Weekly Cancer Cases

The overall study cohort comprised 80,138 patients with a new cancer diagnosis between January 1, 2019, and May 31, 2021, of which 39,593 patients were diagnosed before March 4, 2020, and 40,545 patients were diagnosed afterward. No patient experienced more than 1 new cancer diagnosis during the study. Overall patients had a mean age of 64.9 years (SD, 15.4) at diagnosis. A majority of patients were Caucasian/White (89.8%), lived in an urban setting (95.4%), were insured (84.2%), and were female (53.9%). Among the Caucasian/White group, 2.7% identify as Hispanic/Latino (data not shown).

Weekly new cancer cases are presented in Figure 1 as line graphs. Ljung-Box test for residuals from the ARIMA (3,1,1) model show there is insufficient evidence to conclude a lack of fit (P value 0.551). The median (interquartile range) weekly cancer cases during the pre-COVID-19 period was 661 (627-685) cases. Post-COVID-19 new cancer cases followed a triphasic pattern declining precipitously (−110.0 cases per week [95% confidence interval (CI), −190.2 to −29.8]) during the early-COVID-19 period, rising modestly (+23.7 cases per week [9.1 to 38.4]) from a low baseline during the mid-COVID-19 period and trending slowly back toward pre-COVID-19 baseline levels during the late-COVID-19 period. Subgroup analyses by state showed a similar tri-phasic variation pattern in post-COVID-19 case numbers in Washington, Oregon, and California with stable case numbers in Alaska and Montana (Fig. 2A and Supporting Table 2). Subgroup analyses by disease site showed that among the 10 most common cancers, breast, prostate, and nonmelanoma skin cancer demonstrated similar trends, whereas others were impacted to a lesser extent (Fig. 2B; Tables S3 and S4.

Weekly cancer cases over time. The onset of each of the post–coronavirus disease 2019 (COVID-19) study periods identified by segmented regression analyses is indicated by vertical lines (early-COVID-19: March 4-31, 2020; mid-COVID-19: April 1, 2020; and late-COVID-19: June 10, 2020). The solid blue line represents the observed weekly cancer cases. The dashed black line corresponds to the fitted regression line for each of the 4 study intervals. The orange solid line represents expected cancer cases based on the autoregressive integrated moving average time series model with 95% confidence intervals shaded in gray.

Weekly cancer cases (solid dots) over time by (A) state and (B) disease site. Fitted segmented regression solid lines with associated 95% confidence intervals are shown (gray-shaded region).

Significant differences in patient demographics including region, age, gender, and race were observed across periods (Table 1). However, these differences were small and not clinically meaningful. The proportion of patients having health insurance increased substantially during the late-COVID-19 period (87.1% vs 82%, 81.8%, and 82.9% for the pre-, early-, and mid-COVID-19 periods, respectively; P < .01 for the trend). There was no difference in the proportion of patients living in urban versus rural group settings across periods (P = .26). A total of 6761 (16.7%) patients had a clinically indicated COVID-19 test, among which, 359 (5.3%) patients tested positive. COVID-19 testing rates and the frequency of positive tests increased over time.

TABLE 1. Patient Demographics Pre COVID-19 (n = 39,593) Early COVID-19 (n = 2121) Mid COVID-19 (n = 4734) Late COVID-19 (n = 33,690) P Region, No. (%) <.01 Alaska 1747 (4.4) 94 (4.4) 243 (5.1) 1446 (4.3) California 12,041 (30.4) 681 (32.1) 1384 (29.2) 10,055 (29.8) Montana 1158 (2.9) 68 (3.2) 160 (3.4) 912 (2.7) Oregon 9446 (23.9) 462 (21.8) 1063 (22.5) 7954 (23.6) Washington 15,201 (38.4) 816 (38.5) 1884 (39.8) 13,323 (39.5) Age at diagnosis, mean (SD), y 65.1 (14.6) 64.9 (14.6) 65 (14.7) 64.7 (16.4) <.01 Age at diagnosis, No. (%) <.01 <18 y 185 (0.5) 11 (0.5) 20 (0.4) 184 (0.5) 18-60 y 12,696 (32.1) 697 (32.9) 1508 (31.9) 11,212 (33.3) >60 y 26,712 (67.5) 1413 (66.6) 3206 (67.7) 22,292 (66.2) Gender, No. (%) <.01 Female 21,313 (53.8) 1146 (54) 2353 (49.7) 18,373 (54.5) Male 18,273 (46.2) 974 (45.9) 2380 (50.3) 15,300 (45.4) Race, No. (%) <.01 White or Caucasian 31,245 (90.3) 1608 (89.4) 3589 (89.4) 25,302 (89.2) American Indian or Alaska Native 321 (0.9) 22 (1.2) 47 (1.2) 278 (1) Black or African American 885 (2.6) 48 (2.7) 125 (3.1) 826 (2.9) Asian/Pacific Islander 2160 (6.2) 121 (6.7) 255 (6.3) 1953 (6.9) Missing 4982 322 718 5331 Urban/rural group, No. (%) .26 Urban 36,804 (95.5) 1973 (94.8) 4376 (95) 31,249 (95.4) Rural 1751 (4.5) 108 (5.2) 232 (5) 1494 (4.6) Missing 1038 40 126 947 Insurance, No. (%) <.01 Not insured 7129 (18) 386 (18.2) 809 (17.1) 4352 (12.9) Insured 32,464 (82) 1735 (81.8) 3925 (82.9) 29,338 (87.1) COVID-19 test results, No. (%) Patients tested 246 (11.6) 940 (19.9) 5575 (16.5) <.01 Negative test 240 (97.6) 919 (97.8) 5243 (94) Positive test 6 (2.4) 21 (2.2) 332 (6) <.01 Abbreviations: COVID-19, coronavirus disease 2019; SD, standard deviation. Year-Over-Year Comparison in Weekly Cancer Cases

An analysis comparing year-over-year differences in observed weekly cancer cases post-COVID-19 relative to corresponding weeks in the calendar year immediately before the pandemic confirmed an overall reduction in the number of cases that slowly returned to baseline over a one year period (Fig. 3). Specifically, beginning March 4, 2020, except for a large spike in cases during the last week of August 2020, the number of cancer cases per week was either less or similar when compared to the same week in the prior year. The week of April 8 to 15, 2020, displayed the largest decrease (−41.9%). Overall, during the post-COVID-19 time interval from March 4, 2020, through March 3, 2021, there was a cumulative decline in new cancer cases of 7.3% relative to same time interval (total diagnoses of 31,348 vs 33,822) (Table 2). The observed decline in case numbers varied by region (range, −11.5% to −4.2%) and disease site (range, 20.2% to 5.0%). Breast, prostate, and nonmelanoma skin were the 3 cancers most significantly affected by the pandemic and represented 43.9% of cases overall. The observed reduction in case numbers for these cancers was 14.3% for breast, 12.8% for prostate, and 20.2% for nonmelanoma skin.

Year-over-year comparison in weekly cancer cases.

TABLE 2. Cumulative Year-Over-Year % Change in Cancer Cases Total Cancer Cases Cumulative % Change in Cancer Cases March 4, 2019-March 3, 2020 March 4, 2020-March 3, 2021 All 33,822 31,348 −7.3 Subgroup by region Washington 13,059 12,512 −4.2 California 10,225 9054 −11.5 Oregon 8071 7490 −7.2 Alaska 1499 1383 −7.7 Montana 968 909 −6.1 Subgroup by top 10 primary disease site Breast 6135 5257 −14.3 Prostate 5371 4683 −12.8 Nonmelanoma skin 4120 3286 −20.2 Colorectal 2090 1927 −7.8 Lung and bronchus 1991 1861 −6.5 Heme 1847 1754 −5.0 Uterus 1568 1444 −7.9 Thyroid 1298 1126 −13.3 Melanoma 1185 953 −19.6 Bladder 978 877 −10.3 Discussion

This observational study of patients in a large health care network spanning a geographically, ethnically, and economically diverse part of the western United States is the first to report 1-year trends in new cancer cases following the onset of COVID-19 in a clinical practice setting. These data demonstrate that the number of new cases declined precipitously in the early days of the pandemic, but it took approximately 1 year for cases to return to expected pre-COVID-19 volumes. The importance of the study lies in clarifying the cadence at which patients (ie, new cases) returned to the health care system and poses challenging and unanswered questions about the effect of delayed diagnosis on stage at diagnosis and, thus, clinical outcomes. We observed a 7.3% decline in the expected number of cancer cases compared to the year prior, which is not explained by demographic, geographic, diagnostic, or insurance-related concerns. Cancers experiencing the greatest decline included those that are typically detected by screening or as part of a routine clinical examination including breast, prostate, melanoma and nonmelanoma skin, and thyroid. These findings suggest that alternative mitigation strategies should be considered in the future to avoid diagnostic delay.

In addition to being community-based, important strengths of this study include its size, scope (inclusion of cancers from all disease sites), and extended follow-up. Data collection was facilitated by a shared electronic medical record system linking 40 participating inpatient and outpatient facilities spanning 5 states in the western United States and Alaska. Overall, 39,593 new cancer cases were detected during the 14-month interval from January 1, 2019, through March 3, 2020, and this served as a baseline pre-COVID19 study cohort. Demographics characteristics of this cohort are largely in line with the 2018 US cancer population overall as summarized by the Center for Disease Control (CDC) (https://gis.cdc.gov/Cancer/USCS/#/Demographics/) and based on data from the National Program of Cancer Registries and National Cancer Institute–SEER Registry.10 The mean age of cancer cases in our cohort was 64.9 years, and 89.8% of cases were Caucasian/White. Our cohort includes a greater proportion of cases of Asian/Pacific Islander (6.5% vs 3.3%) and American Indian (1.0% vs 0.5%) and lower proportion of cases of Black (2.7% vs 10%) and Hispanic/Latino (2.7% vs 7.9%) compared to CDC US cancer statistics. We observed a slightly higher proportion of cases among women (53.9% vs 49.5%). The distribution of cases by disease site was consistent with US population statistics, with the exception of slightly fewer lung and more skin cancer cases. Nearly 16% of patients in our baseline pre-COVID-19 cohort lacked health insurance, which is higher than reported rates of 12.4% and 7.7% of uninsured cancer patients at diagnosis before and after the 2018 Affordable Care Act Medicaid in a recent meta-analysis.11

We noted a tri-phasic pattern in cancer cases following the onset of the pandemic. Compared to the pre-COVID-19 period, cancer cases declined dramatically for 4 weeks beginning on March 4, 2020, and then increased from a low baseline, first at a moderate and then a slow rate. A subanalysis by state revealed the observed pattern was largely driven by trends within the more populous states (Washington, California, and Oregon) with the less populous states (Montana and Alaska) reporting stable case numbers across the study period. The observed stable case numbers in Montana and Alaska may be a result of lower numbers of COVID-19 infections and/or less intense behavioral and health system restrictions in these areas.12 Our findings are in-line with prior reports on trends in cancer screening visits, diagnosis, and clinical trial participation during the early phases of the pandemic.13-16 These studies demonstrate a consistent pattern of an initial, brief but steep decline in visits and/or cases followed within weeks to months by a moderate rebound. This pattern is not surprising considering the early measures that were implemented at the onset of the pandemic to address the extreme burden of COVID-19 on the health care system and concern for the health and safety of patients and health care staff. Our study, which provides additional follow-up spanning 15 months following the onset of the pandemic, identified a third phase beginning roughly 3 months after the onset of pandemic that was characterized by a slow, persistent increase in weekly case numbers that was just starting to approach pre-pandemic levels by the end of the study interval. Interestingly, this third phase appeared to be largely unaffected by ongoing dynamics, such as second and third infection waves associated with viral variants, the availability and uptake of vaccines, and gradual re-openings in some regions. Assuming there were no significant changes in the population served by the PSJH network, we observed a year-over-year cumulative decline in cancer cases of 7.3% during the interval from March 4, 2020 (onset of the pandemic) through March 3, 2021 (n = 31,348), compared to the same period before the pandemic (n = 33,822). Among the 10 most common cancers, melanoma, nonmelanoma skin, breast, prostate, and thyroid cancer experienced the greatest reduction in case numbers. These findings are likely attributable to the interruption in screening services and/or access to routine clinical examination and procedures typically used to detect these cancers. We noted reductions in other screen-detected cancers including colorectal and lung cancer, however, these cancers were impacted to a lesser extent. The prolonged and significant reduction in cancer cases that were observed is of concern and suggests that despite progress in control of the pandemic, patients remained reluctant to engage the health care system. Targeted educational and recruitment strategies to encourage participation in recommended cancer screening and routine preventive care visits may be necessary.

The observed changes in case volume were accompanied by significant demographic changes in the study population. Post-COVID-19, cancer cases tended to be of slightly younger age and more often non-White/Caucasian. However, these changes were minor and not likely clinically meaningful. We noted a 5% increase in proportion of cancer patients with health insurance during the late-COVID-19 period relative to baseline (87% vs 82%), suggesting that pandemic-related barriers to accessing cancer care services disproportionately impact the uninsured.

There is evidence that cancer patients are at elevated risk for COVID infection, and concern about infection may have contributed to a reluctance to participate in cancer screening and detection services. A large cohort study that included review of electronic health records from over 73 million patients including 2,523,920 patients with 1 of the 13 most common cancers and 16,570 patients with a COVID-19 diagnosis reported that relative to noncancer patients, cancer patients were at greater risk of COVID 19 transmission, with an adjusted odds ratio (OR) for COVID-19 infection of 1.46 (95% CI, 1.42-1.50). Risks were higher for patients with a cancer diagnosis within a year before study entry (OR, 7.14; 95% CI, 6.91-7.39) and a diagnosis of leukemia (OR, 12.6).17 The rate of COVID-19 infections among our cohort of cancer patients was modest (5.9% among 6761 participants tested) and roughly double the 2.3% severe acute respiratory syndrome coronavirus 2 immunoglobulin G seropositivity rate among asymptomatic health care providers from one of our major participating institutions as previously reported by members of the team.18 Systematic testing was not performed in the current study and instead limited to clinical indications including symptom evaluation and/or routine testing before diagnostic procedures. Consequently, these results may under- or overestimate the true prevalence of COVID-19 infection. Nonetheless, our data are largely in line with reported frequencies of COVID-19 infection among other cancer cohorts (0.79% and 4.2%), including patients receiving anticancer therapies.19-21

Limitations

This study has several limitations. First, because the cohort was defined by coding data, it is possible that the primary diagnosis was misclassified, potentially leading to the inclusion of patients without cancer. Second, review of pathology reports was not feasible and information available from the Electronic Medical Record was not sufficient to classify the site of origin for roughly 1.8% of cases identified. Third, the ARIMA model used to predict expected cancer cases post-COVID-19 relied on historical data. Although case rates remained reasonably stable over the pre-COVID-19 period, variation in cancer incidence or in the number of patients seeking care within the PSJH system may have contributed to differences between observed versus expected cases during the post-COVID-19 interval. Finally, our analysis focused on trends in cancer diagnosis and was not designed to evaluate the impact of COVID-19 on patient outcomes, which will require additional studies.

The results of this study demonstrate a significant and persistent delay in cancer diagnoses following the onset of COVID-19 pandemic. As of Spring 2021, case volumes in our hospital system appear to be returning to expected norms. As the pandemic continues, strategies to mitigate the potential adverse effects of diagnostic delay, especially those targeting patients lacking health insurance, need to be considered. The rate of COVID-19 transmission among newly diagnosed patients does not appear to be substantially elevated and should not be a barrier to participating in cancer screening and diagnostic services.

Funding Support

No specific funding was disclosed.

Conflict of Interest Disclosures

Richard Bryan Bell received support for the present manuscript from Providence Portland Medical Foundation, grants or contracts from the National Institutes of Health (1R01CA234343-01A1) and Bristol-Myers Squibb, and payments or honoraria from Bristol-Myers Squibb, Merck, Macrogenics, and Regeneron; he participated on a data safety monitoring board or advisory board for Macrogenics and had a leadership or fiduciary role with the American Cancer Society. The other authors made no disclosures.

Author Contributions

Charles W. Drescher: Conceptualization, overall supervision, data analysis, and conclusions. R. Bryan Bell: Conceptualization, overall supervision, data analysis, and conclusions. Adam J. Bograd: Conceptualization, overall supervision, data analysis, and conclusions. Shu-Ching Chang: Data curation, data analysis, and statistical analysis. Roshanthi K. Weerasinghe: Data curation and data analysis. Ann Vita: Data curation and data analysis. All authors have contributed to writing, review, and editing.