Determinants of Severity in Cancer Patients with COVID-19 Illness

Abstract

New York State had 180,458 cases of SARS-CoV-2 and 9385 reported deaths as of April 10^th, 2020. Patients with cancer comprised 8.4% of deceased individuals¹. Population-based studies from China and Italy suggested a higher COVID-19 death rate in patients with cancer^2,3, although there is a knowledge gap as to which aspects of cancer and its treatment confer risk of severe COVID-19 disease⁴. This information is critical to balance the competing safety considerations of reducing SARS-CoV-2 exposure and cancer treatment continuation. From March 10^th to April 7^th, 2020, 423 cases of symptomatic COVID-19 illness were diagnosed at Memorial Sloan Kettering Cancer Center (from a total of 2035 cancer patients tested). Of these, 40% were hospitalized for COVID-19 illness, 20% developed severe respiratory illness, (including 9% who required mechanical ventilation), and 12% died within 30 days. Age >65 years and treatment with immune checkpoint inhibitors (ICI) were predictors for hospitalization and severe disease, while receipt of chemotherapy and major surgery were not. Overall, COVID-19 illness in cancer patients is marked by substantial rates of hospitalization and severe outcomes. The association observed between ICI and COVID-19 outcomes in our study will need further interrogation in tumor-specific cohorts.

MAIN

The characterization of COVID-19 in patients with cancer remains limited in published studies and nationwide surveillance analyses. Reports from outside the US raise the possibility that patients with cancer on active therapy have higher risk of COVID-19 related severe events ^4–7. In the current study, we report on the epidemiology of COVID-19 illness experienced at our tertiary-care cancer center during the height of incident cases in New York City, and offer an analysis of risk factors for severe infection that is pertinent to cancer patient populations. From March 10, 2020, until May 7, 2020, SARS-CoV-2 was detected in 946 patients at Memorial Sloan Kettering Cancer Center. In comparison with New York City COVID-19 data, the age-stratified rates of hospitalization and death among 946 lab-confirmed cases from March 10 until May 7, 2020 are shown in Supplemental Figure 1. Clinical characteristics were abstracted for a final study population consisting of 423 case patients diagnosed from March 10 until April 7, 2020 with a follow up period of at least 30 days or death (Figure 1).

Figure 1:

Number of SARS CoV-2 positive cases, hospitalizations, ICU admissions and deaths from March 10, 2020 to May 7, 2020. Case abstraction study period is from March 10, to April 7, 2020. Follow up of abstracted cases until May 7, 2020.

Table 1 shows the demographic and clinical characteristics of the 423 cases. Most patients were adults over the age of 60 years (234, 56%). The most frequent cancer types included solid tumors such as breast (86, 20%), colorectal (37, 9%) and, lung cancer (35, 8%). Lymphoma was the most common hematologic malignancy (48, 11%). Over half of the cases were metastatic solid tumors (238, 56%). At least one of the specified co-morbid conditions were present in 248 (59%) individuals: diabetes, hypertension, chronic kidney disease, cardiac disease. Of the presenting symptoms examined, fever (78%) and cough (82%) were the most common, whereas shortness of breath (44%) and diarrhea (26%) were less common but not rare. Chest radiographic findings were varied and are summarized in Supplemental Table 1. In the cohort, 168 (40%) out of 423 patients were hospitalized and 87 (20%) developed severe respiratory illness, including 47 (11%) who required high-flow oxygen and 40 (9%) who required mechanical ventilation. (Figure 1B). In the absence of approved therapy for COVID-19 illness during this period, a number of investigational treatments were administered, including hydroxychloroquine, azithromycin, remdesivir, tocilizumab, convalescent plasma, and corticosteroids (Supplemental Table 2). Illness in seven pediatric cases was mild and without complications. The overall case-fatality rate was 12% (51 out of 423). Case fatality for hospital and ICU admittance were 24% (41 of 168) and 35% (17 of 48), respectively.

Table 1:

Patient demographics and clinical characteristics (N=423)

Characteristic	No. (%)
Age (years)
<18	7 (2)
18–29	11 (3)
30–39	19 (4)
40–49	51 (12)
50–59	101 (24)
60–69	134 (32)
≥70	100 (24)
Sex
Male	212 (50)
Female	211 (50)
Race
White	263 (62)
Black	68 (16)
Asian	36 (9)
Other	56 (13)
Body mass index (BMI)
Below 18.5 (underweight)	13 (3)
18.5 to 24.9 (normal)	135 (32)
25.0 to 29.9 (overweight)	146 (35)
30.0 to 39.9 (obesity)	109 (26)
40 or higher (severe obesity)	20 (5)
Smoking status
Current	10 (2)
Former	157 (37)
Never	249 (59)
Unknown	7 (2)
Underlying Cancera
Hematologic
Leukemia	32 (8)
Lymphoma	48 (11)
Myeloma	22 (5)
Solid Tumor
Breast	86 (20)
Colorectal	37 (9)
Lung	35 (8)
Prostate	26 (6)
Other	137 (32)
Metastatic disease	238 (56)
Major surgery (30d)b	31 (7)
Asthma	43 (10)
COPD	29 (7)
Diabetes	84 (20)
Cardiac dysfunctionc	84 (20)
Chronic kidney disease	36 (9)
Hypertension	214 (51)
Systemic chemotherapy (within 30d)d	191 (45)
Chronic corticosteroide	66 (16)
Chronic lymphopeniaf	9 (2)
Immune checkpoint inhibitor (ICI)g	31 (7)
Symptoms at Onset
Fever	331 (78)
Shortness of breath	186 (44)
Cough	347 (82)
Diarrhea	109 (26)
Outcomes (30d)
Hospitalization status
Not admitted	243 (57)
Admitted	168 (40)
Already hospitalized	12 (3)
Oxygen requirementsh
Nonei	279 (66)
Low-flow oxygen	57 (13)
High-flow oxygenj	47 (11)
Mechanical ventilation	40 (9)
Death	51 (12)

COPD chronic obstructive pulmonary disease

^aIn patients with multiple malignancies, the most active and recently treated was used to define underlying cancer.

^bDefined as any surgical procedure requiring general anesthesia

^cHeart failure, myocardial infarction, valvular replacement, or cardiomyopathy

^dSystemic, parenteral chemotherapy

^eCorticosteroids (equivalent of prednisone 20 mg or higher) for at least 10 days.

^fAbsolute lymphocyte count <500 per microliter over five previous consecutive measurements

^gImmune checkpoint inhibitor therapy consisted of the following, given within 90 days: pembrolizumab (18), nivolumab (5), atezolizumab (3), avelumab (1), durvalumab (1), ipilimumab (1), nivolumab + ipilimumab (1), pembrolizumab followed by nivolumab (1)

^hHighest oxygen requirement over 30 days is shown.

ⁱNote, 2 patients had chronic tracheostomy because of prior medical issues, but did not require supplemental oxygen during COVID-19 infection.

^jIncludes the following oxygen delivery routes: non-rebreather mask, high-flow nasal oxygen, bilevel positive airway pressure (BiPAP)

Next, we assessed risk factors for hospitalization and severe respiratory illness, the latter defined as the requirement for high-flow oxygen supplementation or mechanical ventilation. In the multivariate analysis, the following risk factors were independently associated with hospitalization: non-white race, hematologic malignancy, a composite measure of chronic lymphopenia and/or corticosteroid use, and treatment with ICI therapy (Table 2). Age >65 years, former or current smoker, hypertension and/or chronic kidney disease, and history of cardiac disorder were significant predictors in univariate, but not in multivariate analysis. The risk factors for severe respiratory illness due to COVID-19 were similar to those for hospitalization, but not identical (Table 2). Severe respiratory illness was significantly more common with age > 65 years. Of note, treatment with ICI also remained an independent predictor of severe respiratory illness. Age and ICI are illustrated by the Kaplan-Meier estimator for this endpoint in Supplemental Figure 2. Notably, metastatic disease, recent chemotherapy, or major surgery within the previous 30 days did not show a significant association with either hospitalization or severe respiratory illness. Given the apparent association of ICI with COVID-19 severity, we explored this further by calculating stratum-specific rates of hospitalization and severe respiratory illness (Table 3). Since PD-1 blockade, a type of ICI, is commonly used to treat lung cancer, we examined the occurrence of outcomes by ICI use and underlying cancer (lung vs non-lung) and observed higher frequencies of hospitalization and severe respiratory illness in both ICI-treated groups. Further, we incorporated lung cancer as a covariate with immune checkpoint blockade in a post-hoc analysis and found separate, distinct effects of lung cancer and ICI treatment on hospital admission and severe respiratory illness (Supplemental Table 3).

Table 2:

Predictors of hospitalization and severe respiratory illness for COVID-19

Predictors of hospitalization, by logistic (N=411a)
Variable	Univariate		Multivariate
	OR (95% CI)	P-value	OR (95% CI)	P-value
Age (>65 years)	1.81 (1.20 – 2.72)	0.004	1.53 (0.96 – 2.43)	0.072
Sex (female)	0.89 (0.60 – 1.32)	0.575
Race (non-white)	1.36 (0.91 – 2.04)	0.135	1.62 (1.05 – 2.51)	0.029
BMI (≥30)	0.89 (0.58 – 1.36)	0.585
Smoking (Current/Former)	1.60 (1.07 – 2.40)	0.022	1.37 (0.88 – 2.13)	0.169
Asthma/COPD	1.39 (0.81 – 2.37)	0.226	1.07 (0.59 – 1.92)	0.828
Cancer (non-met solid)	1.00 (Ref)	-	1.00 (Ref)
Cancer (met solid)	0.89 (0.53 – 1.50)	0.647	0.76 (0.43 – 1.34)	0.338
Cancer (hematologic)	2.24 (1.25 – 4.06)	0.007	2.49 (1.35 – 4.67)	0.003
Major Surgery (within 30d)	1.24 (0.53 – 2.84)	0.612
Diabetes	1.20 (0.73 – 1.96)	0.467
Cardiac Disorder	1.86 (1.13 – 3.07)	0.015	1.35 (0.77 – 2.36)	0.297
HTN/chronic kidney disease	1.84 (1.24 – 2.75)	0.003	1.51 (0.96 – 2.39)	0.077
Systemic chemotherapy (within 30d)	1.04 (0.70 – 1.54)	0.845
Chronic lymphopenia or corticosteroids	1.86 (1.11 – 3.15)	0.019	1.85 (1.06 – 3.24)	0.030
Immune checkpoint inhibitor	2.53 (1.18 – 5.67)	0.017	2.84 (1.24 – 6.72)	0.013
Predictors of severe respiratory illness, by Cox proportional hazard (N=423)
Variable	Univariate		Multivariate
	HR (95% CI)	P-value	HR (95% CI)	P-value
Age (>65 years)	2.02 (1.33 – 3.08)	0.001	1.67 (1.07 – 2.60)	0.024
Sex (female)	1.04 (0.68 – 1.58)	0.859
Race (non-white)	1.20 (0.79 – 1.84)	0.394
BMI (≥30)	1.01 (0.64 – 1.59)	0.965
Smoking (Current/Former)	1.78 (1.17 – 2.72)	0.007	1.39 (0.89 – 2.17)	0.148
Asthma/COPD	1.63 (0.98 – 2.71)	0.059	1.24 (0.72 – 2.13)	0.436
Cancer (non-met solid)	1.00 (Ref)	-	1.00 (Ref)	-
Cancer (met solid)	0.87 (0.48 – 1.59)	0.658	0.75 (0.40 – 1.41)	0.371
Cancer (hematologic)	1.69 (0.92 – 3.10)	0.092	1.79 (0.97 – 3.32)	0.063
Major Surgery (within 30d)	1.31 (0.63 – 2.71)	0.464
Diabetes	1.09 (0.65 – 1.83)	0.745
Cardiac Disorder	2.02 (1.28 – 3.19)	0.002	1.44 (0.88 – 2.37)	0.147
HTN/chronic kidney disease	1.68 (1.09 – 2.58)	0.020	1.18 (0.73 – 1.89)	0.505
Systemic chemotherapy (within 30d)	1.19 (0.78 – 1.82)	0.407
Chronic lymphopenia or corticosteroids	1.59 (0.97 – 2.59)	0.066	1.42 (0.86 – 2.34)	0.165
Immune checkpoint inhibitor	2.38 (1.29 – 4.38)	0.005	2.74 (1.37 – 5.46)	0.004

COPD chronic obstructive pulmonary disease; HTN hypertension

^a12 subjects were excluded from the hospitalization endpoint; they were already hospitalized for other reasons, prior to testing positive SARS-CoV-2 RNA PCR testing.

Table 3:

Stratum-specific point estimates of outcomes, by immune checkpoint inhibitor (ICI) treatment and underlying solid cancer (lung vs. non-lung) (N=275)

Cancer	Endpoint	Non-ICI no./total no. (%)	ICIc no./total no. (%)
Lung cancer—	Hospitalizationb	12/23 (52)	10/12 (83)
	Severe respiratory illness	8/23 (35)	7/12 (58)
Other solid cancersa	Hospitalizationb	82/216 (38)	8/17 (47)
	Severe respiratory illness	34/221 (15)	5/19 (26)

^aIncludes the following non-lung malignancies where ICI was given as therapy: breast (86), lymphoma (48), colorectal (37), prostate (26), kidney (11), genitourinary (9), skin (9), brain (6), uterus (5), cervix (3)

^bFor hospitalization endpoint, 7 patients already admitted to the hospital at time of COVID-19 diagnosis are excluded from calculation

^cWithin 90 days

In two separate models, we evaluated the symptoms and laboratory markers as predictors of clinical outcomes of interest. Fever, cough, new onset dyspnea, and diarrhea at the time of clinical presentation were associated with a higher risk of hospitalization and severe respiratory illness (Supplemental Table 4), but only dyspnea and diarrhea were independently predictive of severe outcomes. For laboratory biomarkers, evaluated longitudinally and coded as time-dependent predictors, procalcitonin, lymphopenia, interleukin-6, D-dimer, and lactate dehydrogenase correlated with subsequent severe respiratory illness (Supplemental Table 5). Finally, we assessed the distribution of first threshold cycle (Ct) values at the time of laboratory diagnosis for both analytic targets of the SARS CoV-2 PCR assay and found no significant association between Ct values and severe respiratory illness (Supplemental Figure 3).

Patients with cancer are among those most vulnerable to severe illness from respiratory viral infections⁸. Our early experience with COVID-19 at a large tertiary care cancer center demonstrated severe disease in 20% of patients diagnosed with COVID-19 illness, with an overall case fatality rate of 12%. Similar to other studies in the general population, we found that age, non- white race, cardiac disease, hypertension, and chronic kidney disease correlated with severe outcomes^9,10. Contrary to early reports, receipt of chemotherapy within 30 days before COVID-19 diagnosis was not associated with a higher risk of complications⁶. Recent major surgery and metastatic disease also did not confer a significant risk of severe COVID-19 disease. Treatment with ICI predicted both hospitalization and severe disease, although there was considerable heterogeneity in ICI treated tumor types, and disease-specific factors could not be individually addressed. COVID-19 illness among children with cancer exhibited a milder course, consistent with early reports in children without cancer, but represented a small portion of the evaluated population (7, 2%).

A raw comparison with NYC cases during the same time period shows that MSK cancer patients in the age range of 0–64 years were hospitalized at higher rates than the general NYC COVID-19 population (Supplemental Figure 1). Crude death rates were similar across age groups at MSK and citywide, except the elderly (≥75 years of age), in which we observed lower rates at MSK. Beyond cancer, it is plausible that competing mortality risks, functional status, and socioeconomic disparities contribute to the age-specific differences observed between the MSK and citywide population.

Very recently, other groups have reported their observations of COVID-19 in cancer-wide patient populations; these include CCC19, a multi-institution registry including centers across the United States, Canada, and Spain, ¹¹ UKCCMP, a registry spanning the United Kingdom,¹² as well as two studies from large healthcare systems in New York City, one Montefiore,¹³ and another Mount Sinai.¹⁴

Our reported case fatality of 12% was similar to those reported by CCC19 (13%) and Mount Sinai (11%). Interestingly, Montefiore and UKCCMP both reported a 28% case fatality in their cancer population. The reasons for this finding are unclear, but these two studies may have had patients who were both older (both reported a median age of 69) and sicker. In the Montefiore study, most cancer patients who died were nursing home or shelter residents (36%), not on active cancer therapy and had prominent co-morbid conditions (69% of the deceased had at least one other severe co-morbid disease). It is possible that differences in utilization of critical care resources may have contributed to variation in outcomes in patients with cancer. Socioeconomic and racial disparities described in study populations add further complexity to the interpretation of the relationship between cancer and COVID-19 outcomes. Similar to our findings, these studies found no increased risk with chemotherapy treatment or with metastatic disease.

A notable finding of our study is the association of checkpoint inhibitor immunotherapy as a risk factor for severe outcomes in ICI-treated patients, which was independent of age, cancer type, and other co-morbid conditions. While we observed more severe COVID-19 illness in ICI recipients with underlying lung cancer, non-lung cancer patients treated with ICI also demonstrated severe outcomes (Table 3). A possible explanation for this observation is an exacerbation of ICI-related lung injury, or ICI-triggered immune dysregulation by T-cell hyperactivation, that in turn may facilitate acute respiratory distress syndrome, a dreaded COVID-19 disease complication^15,16. The association of ICI treatment with other severe infections is influenced by the use of corticosteroids for control of immune mediated adverse events¹⁷. In this study, only one of 31 ICI-treated patients received corticosteroid therapy prior to the severe illness endpoint, and none had immune-mediated pneumonitis at the time of COVID-19 diagnosis. Thus, corticosteroids and immune-mediated adverse events were highly unlikely to influence the association of ICI treatment with COVID-19 disease severity reported in this study. The findings in our current study should be interpreted with caution due to the limitations of our ICI dataset in evaluating tumor-specific risk.

Other studies have examined cancer immunotherapy as a risk factor, and unlike our study, did not find an association with poor outcomes.^12,13,18 However, these studies had few patients on immunotherapy and examined death as an endpoint. In this work, we examined an endpoint based on significant oxygen needs, which was more common than death. A disease-specific analysis of PD-1 blockade in lung cancer patients from our institution with COVID-19, but treated at multiple hospitals, did not suggest a discernible association¹⁸. It is unclear if that study had the statistical power to uncover the effect size reported in this manuscript. It is important to note that these two studies had distinct endpoints, and small overlapping study populations. It is possible that patients with lung cancer or other malignancies have confounding effects from other factors that were not fully evaluable in our population. Specifically, in lung cancer, global consortium efforts are underway to understand the observed impact of COVID-19¹⁹. Until further evidence is available, it is prudent not to alter treatment decisions, but to consider increased vigilance with SARS CoV-2 testing in patients initiating or continuing treatment with ICIs, irrespective of symptoms.

There are several other limitations to our study. First, we describe a single center retrospective analysis in a heterogeneous group of patients with cancer. Second, the effectiveness of experimental therapeutics used for the management of COVID-19 was not explicitly evaluated. With the postponement of all non-essential cancer care during the study period, COVID-19 testing practices were targeted towards symptomatic patients that needed medical evaluation, potentially overestimating the overall severity of COVID-19. Finally, with our study population and design, we are unable to provide a reliable comparison of COVID-19 related outcomes between cancer and non-cancer populations. Such an analysis should be conducted in homogenous cohorts with adequate adjustment for comorbidities, inclusion of cancer patients on active therapy, a similar testing strategy, and the ability to measure the effects of interrupting oncologic care.

Our study has the distinct strength of reporting the most extensive single-institution experience in cancer patients from the epicenter of the US outbreak. Although we had a high number of COVID-19 related hospitalizations, critical care resources were never in short supply. Further, we describe outcomes on COVID-19 associated respiratory compromise and include a full 30 day follow up for reporting primary study outcomes and death rates.

In summary, the outcome of COVID-19 illness is worse among those with underlying conditions, including cancer. Our group of 423 cancer patients cancer had substantial rates of severe respiratory outcomes (20%) and death (12%) with COVID-19 illness^3,13,20–22 In addition, as was seen with the SARS epidemic in 2003, the ongoing risk of contracting the illness and indirect consequences of treatment disruptions are expected to have a lasting effect on the health and safety of patients undergoing treatment for cancer.²³ Continuous preparedness is paramount as routine cancer care is resumed in the coming weeks and months amidst the unpredictable threat posed by COVID-19. Informed approaches with universal screening, aggressive testing, and rigorous control measures will be essential for the safe ongoing delivery of oncologic care.

ONLINE METHODS

Memorial Sloan Kettering Cancer Center (MSKCC) is a 514-bed tertiary cancer center in New York City with approximately 25,000 admissions and 173,000 patient days annually. MSKCC maintains 19 ambulatory sites across New York State and New Jersey, with in excess of a million combined yearly outpatient visits. Approximately 23,000 individuals per year are under active treatment. During February and March 2020, a total of 14,067 individuals received parenteral chemotherapy at MSKCC. Critical care resource allocation was determined by patients expressed advanced directives and not subject to shortage during the study period.

Study population

From March 10, 2020, until April 7, 2020, all consecutive adult and pediatric cases of symptomatic and laboratory-confirmed SARS-CoV-2 infection were included. The only exceptions were asymptomatic subjects tested before surgery or prior to receipt of select myeloablative chemotherapy regimens (n=30). Thirty-three patients included in this study cohort were also included in Luo, et al. Identification of case-patients, their medical background, and clinical course during COVID-19 illness, were abstracted from electronic medical records. The MSKCC Institutional Review Board granted a HIPAA waiver of authorization to conduct the study.

Laboratory methods

Nasopharyngeal swab samples were collected using flocked swabs (Copan Diagnostics) and placed in viral transport media. SARS-CoV-2 RNA was detected using the Centers for Disease Control and Prevention (CDC) protocol targeting two regions of the nucleocapsid gene (N1 and N2), with the following modifications: Nucleic acids were extracted from specimens using the NUCLISENS EasyMag (bioMérieux, Durham, NC) following an off-board, pre-lysis step.^24,25 Real-time reverse transcription PCR was performed on the ABI 7500 Fast (Applied Biosystems Foster City, CA) in a final reaction volume of 20-μL, including 5 μL of extracted nucleic acids. Samples were reported as positive if both the N1 and N2 targets were detected.

Statistical analysis

We first assessed patient risk factors for hospitalization as part of the management for COVID-19 illness, using logistic regression. Patients with nosocomial infection (n=12) were excluded from this analysis. Next, we assessed risk factors for severe respiratory illness, defined as the requirement for high-flow oxygen supplementation or mechanical ventilation. Cause-specific Cox proportional hazard modeling was applied for this. Analysis time began at the time of COVID-19 diagnosis and was censored at 30 days after diagnosis or death, in the absence of the endpoint beforehand. The proportional hazards assumption was verified by examining Schoenfeld residuals for all predictors.

For the outcomes described above, the following clinical variables were assessed: age, sex, race, diabetes, hypertension cardiovascular disease (myocardial infarction, heart failure, heart valve replacement, or cardiomyopathy), chronic obstructive pulmonary disease, asthma, chronic kidney disease, obesity (body mass index >30), smoking status, underlying cancer, major surgery (surgery requiring general anesthesia), chronic lymphopenia (absolute lymphocyte count less than 500 per microliter for at least five measurements, immediately preceding positive COVID-19 PCR test), chronic corticosteroid use (prednisone of 20 mg per day or equivalent, for at least ten days), systemic parenteral chemotherapy within 30 days, and major surgery within 30 days, treatment with an immune checkpoint inhibitors (ICI) within 90 days. In addition to past underlying conditions, new symptoms at the time of testing were assessed: fever, cough, shortness of breath, and diarrhea.

For comparison with New York City (NYC) data we derived cases, hospitalization rates and deaths for specified age strata from publicly available data sources maintained by the NYC Department of Health (https://www1.nyc.gov/site/doh/covid/covid-19-data.page) and compared frequencies with MSK counts for the same age groups. Additional evaluation of risk factors was done for symptoms present at the time of diagnosis (fever, cough, dyspnea, diarrhea), and monitoring of clinical laboratory biomarkers (procalcitonin, absolute lymphopenia, interleukin-6, D-dimer, lactate dehydrogenase. The laboratory biomarkers were monitored over the course of analysis time and encoded as a time-dependent predictor in the time-to-event model.

For both outcomes, predictors were first analyzed separately in a univariate analysis. Predictors with a univariate P-value of less than 0.25 were incorporated into a multivariate model.^26,27 The Kaplan-Meier estimator was calculated and shown for the cumulative probability of severe respiratory illness, for independent predictors. Details of sample population and statistical tests are provided in the Life Sciences Reporting Summary. All study analyses were performed using R version 3.5 (R Development Core Team, Vienna, Austria).

Data availability

All data generated or analyzed during this study are included in this published article (and its supplementary information files).