Factors affecting COVID-19 outcomes in cancer patients: A first report from Guy’s Cancer Centre in London

Study population

Guy’s Cancer Cohort, a research ethics committee approved research database (Reference number: 18/NW/0297) of all routinely collected clinical data of cancer patients at Guy’s and St Thomas’ NHS Foundation Trust (GSTT), forms the basis of this observational study (25). The database contains routinely collected prospective and retrospective demographic/clinical data on all cancer patients treated at Guy’s Cancer Centre. We have an established clinical database for all cancer patients tested for COVID-19 either in outpatient clinics or ward setting since 29 February 2020. Using the unique hospital number, these databases were merged prior to anonymization for research purposes. We assessed outcomes included in the core outcome sets currently being developed for COVID-19 to ensure all relevant information is collected in our COVID specific database (26).

We have included cancer patients who received a diagnosis of COVID-19, from a positive PCR test, from 29^th February-12^th May 2020. Until 30 April 2020, a COVID- 19 test was ordered for cancer patients if they presented with symptoms necessitating hospitalization or if they were scheduled to undergo a cancer-related treatment. From 1 May 2020, COVID-19 testing was introduced as part of standard care, with about 25% of patients being swabbed daily depending on staff and testing kit availability. A total of 1,507 patients were tested between 29 February and 12 May 2020, of whom 156 had COVID-19 (10%).

Statistical methods

In this analysis of our data, we had three aims:

1. To describe demographic and clinical characteristics of COVID-19 positive cancer patients, in terms of their COVID-19 and cancer diagnoses.

2. To identify which demographic and/or clinical factors were associated with COVID-19 severity in cancer patients.

3. To identify which demographic and/or clinical factors were associated with COVID-19 death in cancer patients.

Descriptive statistics were used to address the first aim. Most variables had several categories for the purpose of these descriptive analyses, but were collapsed for the purpose of regression analyses due to the sample size of our cohort. Socioeconomic status (low, middle, high) was determined based on the English Indices of Multiple Deprivation for postcodes (27). Lymphocyte count (x10⁹) was categorized as ≤0.5, 0.6-0.8, 0.9-1.2, and >1.2 based on the Common Terminology Criteria for Adverse Events v.5 (CTCAE). For the other laboratory variables, we created tertiles instead of clinical cut-offs due to cancer patients already having abnormal values for most of these blood markers (Ferritin, C-reactive protein, and albumin). Radical treatment was defined as those patients with a chance of long-term survival or cure.

For the second aim, we conducted logistic regression analyses. Mild/moderate COVID-19 was defined as pneumonia with or without sepsis (i.e. those patients managed on the ward), whereas severe COVID-19 was defined as acute respiratory distress syndrome (ARDS) or septic shock (i.e. those patients where severity reaches criteria for Intensive Care Unit admission, if deemed clinically appropriate). These definitions were based on the WHO COVID-19 classification (28). The models used to quantify the association between each factor and COVID-19 severity were defined through a directed acyclic graph (DAG) (Figure 1 in Appendix). Each factor was individually set as the main exposure variable in the model in order to determine the minimal adjustments required for each factor (Table 1 in Appendix).

The third research aim was addressed with Cox proportional hazards regression analyses, whereby the models were defined as above (Table 1 in Appendix). Follow- up was defined from date of COVID testing until death or 12 May 2020.

All statistical analyses were conducted with STATA version 15.1

Demographic and clinical characteristics of COVID-19 positive cancer patients

128 patients (82%) presented with mild/moderate COVID-19 and 28 patients (18%) with severe COVID-19 (Table 1). More patients were male (58%) and aged 60+ (68%; median age: 67). However, 14% of the cancer population was aged <50 years (n=21; median age: 41). When stratified by COVID grade, more male cancer patients presented with severe disease (68%). Most patients were from a lower socioeconomic background (81%). With respect to ethnicity, about half were White, 22% were Black (n=32) and 4% were of Asian (n=6) origin. When stratified by COVID grade, a slightly larger proportion of patients from a white ethnic background had severe COVID (57%). Hypertension was the most reported comorbidity (47%), followed by diabetes mellitus (22%), renal impairment (19%) and cardiovascular disease (19%). However, benign lung conditions were more commonly reported for those who presented with severe COVID-19 (29% vs 13% in those with mild COVID-19).

The most frequently reported tumour types were urological/gynaecological (29%), followed by haematological (18%) and breast (15%) (Table 2). The first group (n=45) comprised 21 prostate, 8 renal, 5 bladder and 11 gynaecological cancers. Of the 28 haematological malignancies, 4 (14.3%) were myeloid and 24 (85.7%) were lymphoid. Of all cancer patients tested for COVID-19, 80 were positive after their cancer-related hospital admission (51%), of which 61 were solid tumours (76%) and 19 were haematological cancers (24%). When stratified by COVID-19 severity, the largest proportion of cancers presenting with severe COVID were haematological (36%). A large proportion of patients had advanced cancer (40% stage IV) and were diagnosed with their malignancy in the last 12 months (46%).

Overall, 39% of patients were receiving palliative treatment, 25% were receiving radical treatment and 12% were treatment naive. Treatment distributions were reasonably comparable between COVID-19 severity groups. Of the 81 patients on systemic treatment within the last 2 years, 54 were in a palliative setting, of these 50% were 1^st line, 33% 2^nd line and 13% on ≥3^rd treatment line. However, the majority of severe COVID-19 patients were on third line metastatic treatment. Table 2 provides further details on the cancer characteristics.

46% of the cancer patients diagnosed with COVID-19 in this cohort presented with a cough and 52% had a fever. Most patients were molecularly diagnosed within 7 days of their initial symptoms (58%) (Table 3). More patients in the severe COVID-19 group presented with C-reactive protein (CRP) values in the highest tertile (46 vs 22% for mild/moderate disease). Similarly, they had a lower lymphocyte count (53 vs 21% in the lowest category (≤0.5)) and lower albumin levels (39 vs 22% in the lowest tertile).

Factors associated with COVID-19 severity in cancer patients

The odds ratios (ORs) for the associations between the various demographic and clinical factors and COVID-19 severity status are shown in Table 4. There was a non-statistically significant indication that those patients who were diagnosed with cancer more than 24 months ago were at a higher risk of presenting with severe COVID-19 as compared to those diagnosed during the last 24 months (OR: 1.74 (95%CI:0.71-4.26)). With respect to symptom presentation, those presenting with a fever, dyspnoea, or gastro-intestinal symptoms were at a higher risk of having severe COVID-19 as compared to those without these symptoms (OR: 6.21 (1.7621.99), 2.60 (1.00-6.76), and 7.38 (2.71-20.16), respectively).

Factors associated with COVID-19 death in cancer patients

During a median follow-up of 37 days (IQR:18-49), 34 cancer patients had died of COVID-19 (22%) (Table 5). Several cancer patient characteristics were found to be positively associated with risk of COVID-19 death: Asian ethnicity (as compared to white - HR: 3.73 (95%CI: 1.28-10.91), receiving palliative treatment (as compared to no active treatment - HR: 5.74 (95%CI: 1.15-28.79), time since cancer diagnosis >24 months (as compared to ≤24 months - HR: 2.14 (95%CI: 1.04-4.44), presenting with dyspnoea (as compared to no dyspnoea – HR: 4.94 (95%CI: 1.99-12.25), and high CRP levels (3^rd tertile vs 1^st tertile – HR: 10.35 (95%CI: 1.05-52.21)). In addition, an inverse association with death from COVID-19 was observed with levels normal albumin levels (3^rd tertile vs 1^st tertile – HR: 0.04 (95%CI: 0.01-0.042).

Demographic and cancer characteristics

Several retrospective cohort studies published using data from hospitals situated in Wuhan, China, Northern Italy, Canada and the USA have reported on the clinical characteristics of COVID-19 positive cancer patients with sample sizes varying from 9 to 85, two slightly larger series of >200 patients, and a big data Consortium including over 85 institutions resulting in 900 patients (5-12, 16). The median age reported in these studies was similar to our study, with a range from 63 to 72 years. A larger proportion of male patients has been observed. Lung cancer was the most commonly reported cancer in the Zhang (5), Yu (8), Yang (10) and Stroppa (11) studies, but only accounted for 11 % in our patient population. Zhang et al. (5) estimated that in their cohort 29% of patients tested positive for COVID-19 following hospital admission, whereas this was estimated at 51% in our cohort. Interpretation of this statistic is difficult given the latency between exposure and manifestations of infection, meaning patients diagnosed after admission may have been infected outside hospital. Several studies (6, 10) noted that hypertension, diabetes and coronary heart disease were the commonly reported comorbidities.

Our cancer cohort is similar in distribution of age, sex, and comorbidities to the case series reported to date. The ethnicity and SES of our COVID-19 positive cancer patients are most likely a reflection of the catchment area of our Cancer Centre in South-East London (29), covering more deprived Boroughs (Lambeth and Southwark). Based on the number of cancer patients treated at our Cancer Centre in 2019, about 49% of patients are of a White ethnic background. Variations observed with the published data in terms of cancer type, stage, and treatment may be a reflection of clinical practice (e.g. intensity of treatment and frequency of hospital visits), of relative cancer incidence, or of extent of treatment changes introduced as mitigation in the face of the emerging pandemic. For example, the most recently reported age-standardized lung cancer incidence rates for males and females in Wuhan are 54.1 and 19.1 per 100,000, whereas these are estimated to be 37.5 and 24.3 per 100,000 in London (30). Early modification and prioritization of treatment was introduced at our centre, in accordance with now-published guidance (31).

COVID-19 characteristics and severity

Comparably to our study, both the Zhang and the Du studies also reported fever, cough, shortness of breath and dyspnea as common clinical features (5, 6). As in the Chinese cohort of 85 fatal cases, our severe COVID-19 patients had comparable laboratory findings: decreased lymphocytes, increased CRP, and decreased albumin (6).

Severe events were reported for 54% of the study population and mortality for 29% in the Zhang study, as compared to 18% and 22% in our cohort. Zhang et al. also reported that recent treatment within 14 days was associated with an increased risk of developing severe events (28 days) (5). This difference with our observations may be attributed to different definitions of severe events, as it was not entirely clear how these were defined by Zhang et al. As highlighted by Wynants et al in their assessment of current statistical models published for COVID-19 (19), there is a need for consistent use of outcome definitions. However, our observations of a positive association with CRP levels is in line with most COVID-19 studies published to date (32). Apart from the CCC-19 Consortium (16), no study to date has specifically looked at COVID-19 severity at presentation in COVID-19 positive cancer patients and hence our observation of an association with time since cancer diagnosis and presenting symptoms needs further validation in other large cohorts with homogenous definitions of inclusion criteria, testing strategies, and outcomes measures. However, it is possible that time since cancer diagnosis is also a reflection of extent of disease and progression along the palliative patient pathway from diagnosis to death.

COVID-19 death

The study by Yu et al. reported three deaths (25%) (8). In the larger series from New York, Mehta et al reported an overall case fatality rate of 28%, with 37% for haematological malignancies and 25% for solid tumours (14). The CCC-19 Consortium reported a 30-day mortality rate due to COVID-19 of 13% (16). In our cohort the overall case fatality rate was 22%, with 25% for haematological cancers and 21% for solid tumours. As more than 85 institutions were included in the CCC-19 Consortium (16), it is possible that differences in COVID-19 practice as well as cancer treatments between the numerous centres may explain the slightly lower death rate as compared to reports of single centre study. Moreover, our median follow-up is 37 days as compared to 21 days for the Consortium. The heterogeneity between centres may also explain why performance status was found to be associated with COVID-19 outcomes, an observation not identified in our single centre cohort.

Our observations of Asian ethnicity being associated with increased mortality from COVID-19 in cancer patients is of interest, given the recent speculations about the disproportionate effects of COVID-19 on Black, Asian and minority ethnic communities (33) as well as the confounding factor of vitamin D deficiency (34). However, longer follow-up studies are required to disentangle the association between ethnicity and COVID-19 death in cancer patients

Strengths and limitations

Whilst this is one of the largest single centre COVID-19 positive cancer cohorts to date, our sample size is still relatively modest and hence confidence intervals for some statistically significant observations are still wide. No firm conclusions in terms of prognostic modelling can be drawn as of yet (19). Current analyses were aimed at hypothesis generation about patient or tumour characteristics indicative of severity of or death from COVID-19 in the cancer context. Our data for some of the patient characteristics is limited, for example smoking status was missing for 29% of patients and hence likely underestimates the proportion of ever smokers. COVID testing in the UK has only been implemented gradually during the period of our data collection, and there is selection bias in favour of patients being tested as inpatients. Our analysis is likely to have missed cancer outpatients under our care diagnosed with COVID-19 at other hospitals – however this is most likely to be an even more important issue for global Consortia with many hospitals only adding a few cases to the overall dataset. It is a strength of our study that we used clearly defined definitions of COVID-19 severity, as well as a DAG to develop the different models, as to date very limited knowledge is available regarding the intersection between COVID-19 and cancer (19). Detailed information on our modelling will help comparison with future studies with larger sample sizes and longer follow-up.