Dying from cancer with COVID-19: age, sex, socio-economic status, and comorbidities

Abstract

Background

The COVID-19 pandemic has caused excess deaths (all causes) and has disproportionately affected the elderly with certain characteristics.

Objectives

To study how COVID-19 affected cancer deaths regarding age, sex, socio-economic status, comorbidities, and access to palliative care. An additional objective was to study changes in place of care and death.

Material and methods

A descriptive, retrospective study of all cancer patients who died during March–May 2020 in the Stockholm region, n = 1467 of which 278 died with a COVID-19 diagnosis, compared with deaths in 2016–2019. The Stockholm Regional Council’s central data warehouse was used. T-tests, 95% CI, Wilcoxon and chi-squared tests were used for comparisons.

Results

There were excess cancer deaths compared with 2016–2019 (p < 0.001) and patients dying with a COVID-19 diagnosis were older (79.7 vs. 75.9 years, p < 0.0001), more often male (67% vs. 55%, p < 0.0001), and had more comorbidities (Charlson Comorbidity Index 1.6 vs. 1.1, p < 0.0001). Patients with COVID-19 more seldom had access to palliative care (34% vs. 59%, p = 0.008), had more changes in place of care during the last two weeks of life (p < 0.0001) and died more often in acute hospitals (34% vs. 14%, p < 0.0001). For the subgroup with access to palliative care, the hospital deaths for individuals with and without a COVID-19 diagnosis were 11% and 4%, respectively (p = 0.008).

Conclusion

Cancer patients dying with a COVID-19 diagnosis were older, more often male, and had more comorbidities. A COVID-19 diagnosis negatively affected the probability of being admitted to specialized palliative care and increased the likelihood of dying in an acute hospital.

Keywords:

• Cancer
• COVID-19
• mortality
• age
• palliative care services

Background

The excess deaths caused by the COVID-19 pandemic have been extensively investigated [1–5]. Early studies have shown that older age, sex (male), frailty, and having several comorbidities are associated with a higher risk of death [6–9]. Additionally, ethnicity and socio-economic factors, e.g., living in a less affluent area [4,5,10,11], and residing in a nursing home [4,5,12] have also been established as risk factors.

Early in the course of the pandemic, it was established that comorbidities in the form of hypertension, cardiovascular disease, diabetes, chronic respiratory disease, and chronic kidney disease were of importance [3,6,7,13,14]. In addition, active cancer and ongoing cancer treatments have been suggested as risk factors for mortality [15–20].

In a recent study on excess deaths in the county of Stockholm, Sweden, age and socio-economic status were strong predictors of mortality [5]. However, the median age of those who died was 83 years which is much higher than the expected median age of death for patients dying from cancer, as life expectancy for cancer patients is lower than for the general population [21]. As cancer patients die earlier, we do not know to what extent age is a risk factor in this patient group. Socio-economic factors at large constitute risk factors for COVID-19 deaths [4,5,10,11] but, so far, socio-economic groups have not been specifically studied in cancer patients dying with a COVID-19 diagnosis. Another knowledge gap concerns the impact of residing in a nursing home. In general, nursing home residents constitute a disproportionately higher proportion of those who have died from COVID-19 [5,12], but as most cancer patients in Sweden reside elsewhere than in nursing homes, we cannot extrapolate general nursing home data on COVID-19 mortality to cancer patients.

Cancer patients who are approaching death often have many and complex symptoms and therefore need skillful symptom control and support. In the county of Stockholm, a considerable number of patients are admitted to advanced palliative home care (ASIH “avancerad sjukvård i hemmet”) and such care is often integrated with oncological care provided by acute hospitals. The palliative home care services are successful in terms of pain and symptom control and are highly appreciated both by the patients and their families [22]. In addition, designated palliative wards are available for inpatient care when needed. Access to such care was obviously desirable during the first wave of the pandemic, but we do not know to what degree this was possible. Additionally, at the end of life (EOL), patients normally wish to avoid acute visits to emergency departments and often have their own preferences regarding their place of care and, eventually, place of death. We do not know, however, whether the pandemic has actually changed these patterns of care.

What we do know, however, is that the COVID-19 pandemic has changed many of the normal conditions for care. Therefore, it is of interest to study predictors of COVID-19-related cancer deaths, as well as care consumption. Such studies are made possible by using Region Stockholm’s central data warehouse which covers most of the health care use in the county of Stockholm, which incorporates about 2.3 million inhabitants.

Objectives

To study how COVID-19 affected cancer deaths with reference to age, sex, socio-economic status, comorbidities, access to palliative care, and changes in place of care and death, compared with a reference population from 2016 to 2019.

Patients and methods

The methods and the results sections are, whenever possible, reported based on the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) criteria [23].

Study design

The study is based on information retrieved from administrative data bases. In the Stockholm region all appointments, hospital visits, diagnoses and major costs are registered and stored in VAL, the Stockholm Regional Council’s central data warehouse. The data in VAL registers are based on Swedish personal numbers but encrypted, meaning that an individual’s health care consumption can be followed without revealing their personal identity.

The monthly deaths for March to May 2020, i.e., the most affected months of the first wave (the first verified COVID-19 death in Sweden was on March 11, 2020), were identified and compared with the corresponding months over four consecutive years, 2016–2019. The data were further analyzed in relation to age, sex, living arrangements (residents in nursing homes versus all others), and socio-economic status by means of Mosaic [24].

Populations

Study population

All monthly deaths registered in the VAL databases during March to May 2020 (data retrieved June 29, 2020), which are the three months with known excess death rates [5]. In accordance with the guidelines issued by the Public Health Agency of Sweden (Folkhälsomyndigheten), any death with a COVID-19 diagnosis according to ICD-10 should be considered as a death from COVID-19 [25]. As there was an extreme shortage of tests during the first weeks of the pandemic, both PCR-verified cases (U07.1) and clinically verified cases, but without tests (U07.2) were included.

Reference population

All monthly deaths registered in the VAL databases during March to May 2016–2019 (four-year-cohorts). Based on these data, mean values with 95% confidence intervals (CI) were calculated. The reason for choosing a mean from the four previous years (instead of just data from the previous year) was to smooth out any short-term spikes, e.g., due to an influenza outbreak.

Variables

Deaths (all causes, versus cancer deaths) as well as deaths with a COVID-19 diagnosis were used as outcome measures. Age, sex, living arrangements (nursing homes versus all others) as well as socio-economic status in the form of Mosaic groups were used as explanatory variables. The Mosaic methodology is based on the assumption that people tend to live in neighborhoods where others are quite similar to themselves and the final Mosaic groups are based on iterative cluster-analyses, which themselves are based on more than 40 socio-economic variables, and are used in scientific studies [24,26,27]. Mosaic, therefore, provides socio-economic information that makes it possible to define and allocate different areas of residence to three different socio-economic classes (Mosaic 1-3). The information is mainly based on income and education but also, for example, on family situation (single/cohabiting/children, etc.) and living arrangements (owned or rented housing etc.), phase of life, origin and ethnicity, and degree of urbanization. The county of Stockholm is itself divided into 1300 small areas and each area is classified as either Mosaic 1, 2, or 3, the three groups being approximately equal in size. Mosaic group 1 refers to persons living in the most affluent areas and 3 the least affluent. Thus, the Mosaic groups characterize areas, rather than individuals.

Comorbidity was calculated using the Charlson Comorbidity Index (CCI) where different comorbidities are weighted depending on their probability of predicting mortality [28,29]. As we exclusively studied individuals who died from cancer, we excluded cancer from the CCI.

For the studied variables we used two types of comparisons: (a) cancer deaths in March–May 2020 compared with deaths during the corresponding months for the period 2016–2019 (comparisons with 95% CI); (b) a comparison of those who died in March–May 2020 with or without a COVID-19 diagnosis (t-test for age, Wilcoxon Rank Sum test for CCI, other tests with chi-squared tests).

Selection bias

Dropouts: s reporting data to VAL is an obligation for each clinic/care unit, the data are complete with very few missing values. Immediacy: The VAL databases are updated every month, thus, it is possible to retrieve even very recent data.

Nursing home residents: Nursing home residents were identified through the registrations of medical interventions by physicians as such care use is exclusive to nursing home residents and they have a unique, identifiable code. It is most unlikely that a nursing home resident would not have a single registration. If so, they were not included in the analysis.

Study size

The study covers total cohorts, i.e., all deaths (all causes and, specifically, cancer deaths) as well as all reported COVID-19-related deaths during the months of March to May 2020, with the data being compared with the data for four similar year-cohorts (2016–2019). Therefore, no power calculations were made.

Statistical methods and missing data

Deaths with a COVID-19 diagnosis during March–May 2020 (the most affected months of the first wave) were compared with means and 95% confidence intervals (95%CI) for those who died in 2016–2019, during the corresponding months. T-tests for independent means were used for the comparison of age groups, Wilcoxon Rank Sum test was calculated for CCI, and chi-square tests were applied for comparisons of proportions.

The few missing data were not substituted. The SAS version 9.4 and SPSS version 25 software programs were used for statistical analysis.

Ethics: The study was approved by the Swedish Ethical Review Authority (Etikprövningsmyndigheten, Dnr 2020-02186).

Results

Comparison 1: between 2020 and 2016–2019 (reference years)

Excess deaths. Compared with 2016–2019, there were excess deaths during the period of March to May 2020, both for cancer and non-cancer deaths (for each comparison the actual value was higher than the upper 95% CI limit), for details, see Table 1. The proportion of excess cancer deaths was lower than excess non-cancer deaths for each month and this was especially evident for April 2020 when there were 38% excess cancer deaths and 144% excess non-cancer deaths.

Table 1. Excess deaths.

Month	Cancer deaths			Non-cancer deaths
	2016–2019 N (95% CI)	2020	Excess deaths	2016–2019 N (95% CI)	2020	Excess deaths
March	416 (373–458)	461	11%	1059 (937–1180)	1358	28%
April	402 (385–421)	560	38%	974 (908–1038)	2374	144%
May	375 (334–416)	446	19%	912 (878–946)	1414	55%

There were more deaths in 2020, than the average death rates in 2016–2019: For each month, the number of deaths in 2020 was higher than the upper limit for the 95% confidence interval.

Cancer deaths. During the period of March to May 2020, there were 1467 cancer deaths, which was higher than the corresponding mean value of 1194 (95% CI, 1179-1208) for the years 2016–2019, Table 2. In addition, when stratifying for sex, higher values were found: during March to May 2020, 626 women (95% CI, 537–594) and 841 men (95% CI, 598–658) died.

Table 2. Characteristics 2020 for all deaths, with or without COVID-19, compared to 2016–2019 (with 95% CI).

Characteristics	2016–2019 (March–May)	2020 (March–May)	p Value
Deaths, mean (95%CI) or n^a	1194 (1179–1208)	1467^b
Women, N (95%CI)	565 (537–594)	626^b
Men, N (95%CI)	628 (598–658)	841^b
Age, all years (SD)^c	75.3 (13.0)	76.6 (13.5)	0.0009
Women^c	75.2 (13.6)	76.0 (14.3)	0.21 (n.s.)
Men^c	75.4 (12.5)	77.1 (12.8)	0.0008
Living in nursing homes
% of all cancer deaths^d	17	19	0.03
Age, only cancer deaths^c	84.7	86.7	0.008

^aMean and 95% CI refer to the years 2016–2019, whereas n (actual number) refers to 2020.

^bThe values for 2020 are significantly higher than the upper 95% CI limit for 2016–2019.

^cT-test was used for comparison of age.

^dChi-2 test was used for comparison of proportions.

Age. The mean age of all cancer deaths in the period of March to May 2020 was 76.6 years, which was 1.3 years higher than for the same months in 2016–2019 (p < 0.001). The difference was also statistically significant when comparing men (p < 0.001) but not women, for details, see Table 2.

Cancer deaths in nursing homes. In the period 2016–2019, 17% of all cancer deaths were in nursing homes, with the figure being somewhat higher in 2020, 19% (p = 0.03). Patients dying in nursing homes in 2020 were older than those who died during the period 2016–2019, 86.3 versus 84.7 years (p = 0.008), Table 2.

Comparison 2: Deaths in 2020, with or without a COVID-19 diagnosis

Age. Among the 1467 cancer deaths in the period March–May 2020, 278 were with and 1189 without a COVID-19 diagnosis, for details, see Table 3. Those cancer patients who died with a COVID-19 diagnosis were, on average, 3.8 years older (79.7 vs. 75.9 years, p < 0.0001), for details, see Table 3. The proportion of patients aged 80 years or older was 53% for those dying with a COVID-19 diagnosis during March to May and 44% for non-COVID cancer deaths (χ² = 8.5, p = 0.004), Table 3.

Table 3. Deceased cancer patients March–May 2020 without or with a COVID-19 diagnosis, (t-test for age, Wilcoxon Rank Sum test for CCI, other tests with Chi-squared tests).

Characteristics	2020 no COVID-19 diagnosis	2020 COVID-19 diagnosis	p Value
Deaths, n total March–May	1189	278	N/A N/A N/A N/A N/A
March 2020	446	15
April 2020	401	159
May 2020	342	104
Age, all (SD)	75.9 (13.8)	79.7 (11.4)	<0.0001
Women	75.1 (14.5)	81.3 (12.0)	<0.0001
Men	76.5 (13.2)	79.0 (11.0)	0.01
Proportion ≥80 years (%)	44%	53%	0.004
Sex
Women, n (%)	535 (45)	91 (33)	0.0002^a
Men, n (%)	654 (55)	187 (67)
Mosaic (%)
Group 1	29	27	0.36 (ns)
Group 2	39	38	0.21 (ns)
Group 3	32	35	0.29 (ns)
CCI^b	1.1	1.6	<0.0001
Palliative care (access) n (%)	698/1189 (59)	94/278 (34)	<0.0001
Change of care
Once, last 2 weeks	487/1189 (41)	172/278 (62)	<0.0001
2 or more times	261/1189 (22)	138/278 (50)	<0.0001
Hospital death	161/1189 (14)	95/278 (34)	<0.0001
Hospital death, despite palliative care^c	30/698 (4)	10/94 (11)	0.008

^aThe significance level (p < 0.0001) refers to the proportions of women/men without COVID-19 (45%/55%) compared with patients with a COVID-diagnosis (33%/67%).

^bCCI, Charlson Comorbidity Index.

^cHospital death despite palliative care = enrolled in palliative care but died in an acute hospital.

Sex. Of the 278 who died with a COVID-19 diagnosis, 33% were women and 67% were men, which was different from non-COVID deaths with 45% women and 55% men (χ² = 13.8, p = 0.0002), for details, see Table 3.

Mosaic groups. The distribution of patients belonging to Mosaic groups 1 and 3, respectively, was similar regardless of COVID-19 status without statistically significant differences in any comparison, Table 3.

Comorbidities. The mean modified values for the CCI were 1.6 for patients with a COVID-19 diagnosis and 1.1 for the others (p < 0.0001), Table 3.

Access to specialized palliative care

During March to May 2020, 752 patients (52%) had access to palliative care. There was a difference regarding access to specialized palliative care: among those dying from cancer with a COVID-19 infection, only 34% had access to palliative care compared with 59% for those without a COVID-19 diagnosis (χ² = 56.2, p < 0.0001), for details, see Table 3.

The proportion of patients infected with COVID-19 was lower than for those enrolled in palliative care: 12% of those dying in palliative care services had a COVID-19 diagnosis compared with 27% for those dying elsewhere (p < 0.0001, data not shown in tables).

Change in place of care during the last two weeks of life

The proportion of patients who changed their place of care was higher for COVID-19 patients, regardless of cutoff point (one or more changes, or two or more changes). During the last two weeks of life, 50% of those with COVID-19 had at least two changes in place of care compared with 22% for the others (p < 0.0001), for details, see Table 3.

Place of death

More patients with a COVID-19 diagnosis, 34% versus 14%, died in acute hospitals (p < 0.0001). Hospital deaths were few among those patients who were enrolled in palliative care services, but still with more deaths for those with a COVID-19 diagnosis, 11% versus 4% (p = 0.008), for details, see Table 3.

Cancer deaths in nursing homes. During the period 2016 to 2019, 17% of all cancer deaths were in nursing homes and the figure was somewhat higher in 2020, 19% (p = 0.03). Those patients dying in nursing homes in 2020, however, were older than those who died in 2016–2019, 86.3 versus 84.7 years (p = 0.008), see Table 2.

Discussion

The study revealed that there were excess cancer deaths during March to May 2020, but the excess death rate was lower than for non-cancer patients. Those dying with a COVID-19 diagnosis were older, more often male, and had more comorbidities but the deaths were not related to socio-economic factors. Cancer patients with COVID-19 had less access to palliative care services and more changes in place of care. Access to palliative care was associated with fewer emergency room visits and fewer hospital deaths.

When studying all the diagnoses, the findings have both similarities and dissimilarities compared with the overall situation during the COVID-19 pandemic. There were excess cancer deaths during the first wave (March–May) but, for each month, the excess was lower than for non-cancer diagnoses. This finding was most prominent for April 2020, with 38% excess cancer deaths and 144% excess non-cancer deaths. The reasons for this remain unclear, although it is likely that a considerable portion of the non-cancer deaths occurred in nursing homes in much older individuals with pronounced frailty and many comorbidities [5,8,9,12].

Cancer patients dying with a COVID-19 diagnosis were more often male and older than other cancer patients, 79.7 years (median 81 years), but somewhat younger than other patients dying with a COVID-19 diagnosis in comparable studies [5]. Age and male sex as risk factors are well in line with studies of all COVID-19 deaths [5–7]. However, whereas socio-economic factors and ethnicity have been associated with a higher risk for COVID-19-related deaths [4,10,11], such associations were not seen for the cancer patients. We cannot satisfactorily explain this finding, but as excess death rates were much higher for non-cancer patients, a group where socio-economic factors are important [5], it cannot be ruled out that certain socio-economic factors were more prevalent in the non-cancer population.

Comorbidities, as measured by the CCI, were more often seen in patients dying with a COVID-19 diagnosis. This is an interesting finding, as prognosis in cancer is mainly discussed in relation to cancer characteristics, whereas comorbidity is seldom included in a systematic way when discussing prognosis, although data indicate that the burden of comorbidities may vary between different cancers [30]. Comorbidities are especially important as a prognostic factor in elderly cancer patients [31].

Nursing homes have been at the center of the debate during the pandemic, with high death rates related to all underlying diagnoses [12]. However, when specifically studying cancer deaths in nursing homes, the differences between 2020 and 2016–2019 were surprisingly small: the percentage of cancer patients was 19% in 2020, compared with 17% in 2016–2019. Most probably, cancer patients residing in nursing homes are not representative of all nursing home residents.

Access to palliative care is desirable for cancer patients in EOL situations, however, this was not always achieved during the pandemic: a much lower proportion of cancer patients with a COVID-19 diagnosis was admitted to palliative care services compared with other cancer patients. The reason for this cannot be determined from the study design but a possible explanation is that cancer patients infected with COVID-19 living in their own homes were acutely admitted to hospital, and when there was a deterioration in their general condition there was no time or routines available to admit an infected, dying patient to palliative care facilities.

The above explanation is strengthened by the observation that cancer patients with a COVID-19 diagnosis had more changes in place of care during the last two weeks of life, including emergency room visits, hospital admissions, and changes of hospital wards (e.g., oncology departments, infection clinics, ICUs). For the same reason, hospital deaths were much more prevalent among cancer patients with a COVID-19 diagnosis.

Not being referred to a palliative care service has an impact on care during EOL [32]. As seen in this study, those patients who were referred to palliative care, mainly in the form of palliative home care, had fewer acute admissions and hospital deaths than those who were not referred. This was seen for both groups, cancer patients dying with a COVID-19 diagnosis as well as the other cancer patients.

Strengths and limitations

The current data have high reliability with very few missing values. This is related to the fact that almost all health care in the Stockholm Region is financed by taxes and reporting to the VAL databases is a mandatory basis for remuneration. For patients, the out-of-pocket expenses are low in Sweden: the maximum fee for all outpatient visits is 1150 SEK (approximately 110 Euros) per year whereas the cost for hospital stays is 100 SEK (approximately 10 Euros) per day. The maximum fee for all pharmaceutical drug prescriptions is 2350 SEK (approximately 230 Euros) for a 12-month period.

A limitation of the study is that we cannot judge to what extent the patients died from COVID-19 or primarily from cancer with COVID-19 as a secondary diagnosis. Death certificates would be required for such an analysis. However, for study reasons we used the definition proposed by the Public Health Agency of Sweden, namely that any death with an ICD-10 code of COVID-19 should currently be counted as a death from COVID-19 [25]. Moreover, we had only access to a specific register of deceased persons. A study including both the deceased and those still alive would probably have given a more nuanced picture.

Conclusions

Cancer patients dying with a COVID-19 diagnosis were older, more often male, and had more comorbidities. A COVID-19 diagnosis negatively affected the probability of being admitted to specialized palliative care and increased the likelihood of dying in acute hospitals.

Acknowledgments

The authors thank Region Stockholm for generously providing us with the data for the study. The Stockholms Sjukhem Foundation is acknowledged for providing excellent facilities in their research and development unit. David Boniface is acknowledged for linguistic revision.

Disclosure statement

Dr. Strang reports receiving grants from the RCC, Regional Cancer Center Stockholm-Gotland (Dnr VKN 2019-0070), from The Cancer Research Funds of Radiumhemmet (no. 201241) and from Stockholms Sjukhem Foundation’s Jubilee Fund during the conduct of the study. Christel Hedman and Torbjörn Schultz declare that they have nothing to disclose.