Increased risk of death from pneumonia among cancer survivors: A propensity score‐matched cohort analysis

Abstract Background The repeated global pandemic of the new virus has led to interest in the possibility of severe pneumonia among cancer patients and survivors. Here, we aimed to assess the association between incident cancer and risk of death from pneumonia in Japanese in a large population‐based cohort study. Methods We used the data from The Japan Public Health Center‐based Prospective Study (JPHC Study), which enrolled subjects aged 40 to 69 between 1990 and 1994 and followed their cancer incidence and mortality until 2013. After identifying 103,757 eligible subjects for analysis and imputing missing data on covariates by the chained equations approach, we conducted propensity score‐matched analysis for 1:4 matching, leaving 14,520 cases diagnosed with cancer and 48,947 controls without cancer during the study period for final analysis. A Cox proportional hazards regression model was used to estimate the hazard ratio (HR) and corresponding confidence interval (CI) for the risk of death from pneumonia with comparison of cancer cases and cancer‐free controls. Results Compared to cancer‐free individuals, risk of death from pneumonia was significantly higher among those who had any diagnosed cancer (HR, 1.41; 95%CI, 1.08–1.84); those within 1 year of diagnosis (HR, 23.0; 95% CI, 2.98–177.3); within 1 to <2 years (HR, 3.66; 95% CI, 1.04–12.9); and those with regional spread or distant metastatic cancer at initial diagnosis (HR, 2.01; 95% CI, 1.26–3.21). A history of lung, oesophageal, and head and neck cancer conferred the higher risk among site‐specific cancers. Conclusion We found a positive association between incident cancer and risk of death from pneumonia in this study. These results imply the possibility that the immunocompromised status and respiratory failure due to antitumor treatment may have resulted in a more severe outcome from pneumonia among cancer survivors than the general population.


| INTRODUCTION
Globally, pneumonia is the leading cause of death from infection, accounting for 2.5 million deaths in 2019. 1 Although the disease burden of pneumonia is primarily considered to affect older people worldwide and children under age 5 in developing countries, the repeated emergence of new viruses over the last few decades has re-emphasised pneumonia as a public health risk. 2 In Japan, approximately 95,000 deaths were due to pneumonia in 2019, accounting for 6.9% of total deaths, and of which 97% were in patients over 65 years old. 3 Interest is growing in the risk of viral pneumonia among cancer survivors. [4][5][6] Cancer patients are more susceptible to infection than the general population because various cancer treatments place them in an immunocompromised state. 7,8 Previous studies have reported a higher risk of hospitalisation and mortality from pneumonia among patients with haematological malignancies. 9,10 Patients who undergo treatment for haematological malignancies are more susceptible to infection due to a high degree of defects involving the innate and adaptive immune systems. 11 Surgical procedures for lung, oesophageal and head and neck (HN) cancers are highly invasive and can have several severe postoperative complications, including pneumonia. 12 Furthermore, cancer patients often suffer from conditions partially caused by antitumour treatment and various comorbidities, including cardiac disease, diabetes, dyslipidaemia, hypertension and obesity, which may lead to severe pneumonia. 9,11,13 The risk factors for various cancers are closely similar to those for pneumonia prognosis and other chronic diseases, including older age, smoking, poor diet, obesity and alcohol intake. 14 Cancer diagnosis and antitumor treatment may be possible risk factors for severe pneumonia, [15][16][17][18][19] although evidence for this putative association is scarce. Given the above background, the risk of death from pneumonia might be high for cancer survivors, especially lung, oesophageal and HN cancers and haematological malignancies.
Here, we investigated the association between incident cancer and risk of death from pneumonia using a largescale, population-based longitudinal survey in Japan.

| Study setting
The study used data from the Japan Public Health Centrebased Prospective Study (JPHC Study). The JPHC Study consisted of two cohorts: Cohort I (baseline; launched in 1990) and Cohort II (in 1993). In all, 140,420 participants in 11 public health centre areas (PHCs) were enrolled. Participant age ranged from 40-59 years in Cohort I and 40-69 years in Cohort II. Details of the JPHC Study have been described elsewhere. 20,21 Figure 1 shows the participants included in the present study. Participants residing in the Tokyo area were excluded because of the incomplete availability of data on incident cancer. The other exclusion criteria were non-Japanese nationality, incorrect birth data, multiple registrations, pre-commencement loss or death, no response to the baseline survey, history of cancer before the baseline survey identified by self-report or from registries, cancers identified on death certificates only and missing date of death. Finally, the analysis included 103,757 participants. Incorrect birth data was identified through a registration system administered by the local government. In our cancer registry system, very few cancer cases were ascertained only by death certificates (1.2%) and were thus excluded.

| Study population
The study was approved by the Institutional Review Board of the National Cancer Centre of Japan (approval number: 2001-021) in accordance with the relevant ethics guidelines for medical research in Japan. When each participant completed the baseline questionnaire, the study's purpose and follow-up methods were fully explained, and informed consent was implicitly obtained.

| Assessment of exposure
The baseline survey was conducted from 1990 to 1992 for Cohort I and from 1993 to 1994 for Cohort II. We may have resulted in a more severe outcome from pneumonia among cancer survivors than the general population.

K E Y W O R D S
cancer survivors, epidemiology, mortality, pneumonia, propensity score, prospective cohort study used any incident cancer that occurred between the baseline survey and 31 December 2013 (or 30 December 2012 for the Suita area). Cancer cases were identified through active patient notifications from major local hospitals and linkages with population-based cancer registries. Cancer diagnoses were coded using The International Classification of Diseases for Oncology, Third Edition. Discrete periods after cancer diagnoses and clinical stages at diagnoses were also used. Based on the Surveillance, Epidemiology, and End Results (SEER) classification, clinical stage at diagnosis was categorised as localised, regional and distant. We used the following site-specific cancers: lung (C34), oesophagus (C15), head and neck (HN) (C00-C14, C32) and hematologic malignancies (morphology codes 959-972, 974-975, 973-976 and 980-994, respectively) according to previous studies as described in the Introduction section. In cases where a participant was diagnosed with multiple cancers during follow-up, we used only the first incidence case.

| Ascertainment of outcome
Death certificates and permission to identify the cause of death for deceased participants were provided by the Ministry of Health, Labour and Welfare. Causes of death were collected from the primary cause of deaths and classified based on the International Classification of Diseases, 10th Revision (ICD-10). Pneumonia death was defined as  ICD-10 codes J10-J18. Information on the incidence of pneumonia was not available in this study.

| Statistical analysis
We conducted an analysis based on propensity scores in accordance with the methodology of Saito et al. 22 First, we conducted multiple imputation by chained equations with 20 iterations to address missing data under the assumption that they were missing at random. The covariates included into the propensity score analysis were age at baseline (40-44, 45- Second, we conducted logistic regression using the covariates described above to estimate the probability of developing cancer based on the baseline variables in each imputed dataset. 23 Then, we used a pooled propensity score to match, with replacement, subjects with and without incident cancer in a 1:4 ratio. We established a calliper width of 0.25 of the propensity score's standard deviation (SD), resulting in a better balance of subjects with incident cancer and those without as matched samples (C-statistic = 0.67). We assessed the distributions of basic characteristics among groups with standardised mean differences (SMD). Third, we defined an index date for cases as the date of cancer diagnosis. Accordingly, cancer-free controls were assigned the index date corresponding to the date of cancer diagnosis of their matched cases. All subjects were followed until the date of death, migration out of the study area or end of follow-up (31 December 2013), whichever occurred first. We excluded controls who were censored before the index date. If all four matched controls were excluded, their matched case was also excluded. Finally, Cox proportional hazards regression models using attained age as the time scale were employed to estimate the hazard ratio (HR) of mortality from pneumonia with comparison of cancer cases to controls stratified by the propensity score-matched pairs (Model 1). A history of asthma at the baseline survey was further adjusted for as a risk factor for pneumonia (no or yes) (Model 2). We also analysed the association between any incident cancer and pneumonia death using years after cancer diagnosis (cancer-free, 5+ years, 4-<5 years, 3-4 years, 2-<3 years, 1-2 years, or 0-1 years) and clinical stage at cancer diagnosis (localised, regional or distant).
We further conducted the same approach for each cancer site as secondary analyses. Considering the possibility of misclassifying a death from cancer as from pneumonia and vice versa, 24 we also analysed the models following exclusion of cases diagnosed within 1 year after diagnosis as a sensitivity analysis. All p values were two-sided and had values smaller than 0.05, indicating statistical significance. All analyses were conducted using STATA version 16.0 software (StataCorp LP).

| RESULTS
The propensity scores matching analysis included 103,757 subjects from the JPHC Study during a mean follow-up time of 17.8 years. The mean age at study entry (±standard deviation) was 53.0 (±7.8) years, and the proportion of female was 52.3%. After matching, controls who were censored before the index date (n = 9425) and those who lost all matched controls (n = 73) were excluded, leaving 14,520 cases and 48,947 controls as final analytic samples. Table 1 shows the baseline characteristics of the included participants with and without incident cancer during follow-up. Comparison of SMD, age, sex, study area, smoking status and alcohol intake improved the balance between cases and controls after matching. During 496,590 personyears of follow-up, we identified 762 deaths from pneumonia after 1:4 propensity score matching with replacement. The number of site-specific cancers was 1885 for lung, 411 for oesophagus, 426 for HN, 747 for haematologic and 11,088 for other cancers. Table 2 presents the HR with corresponding 95% confidence intervals (CI) for the association between incident cancer, years since diagnosis and clinical stage at diagnosis for any cancer and the risk of death from pneumonia. Compared to cancer-free individuals, those who had any diagnosed cancer (HR, 1.41; 95% CI, 1.08-1.84), those within 1 year of diagnosis (HR, 23.0; 95% CI, 2.98-177.3) or within 1 to <2 years (HR, 3.66; 95% CI, 1.04-12.9) and those whose cancer was regionally spread or distantly metastatic at diagnosis (HR, 2.01; 95%CI, 1.26-3.21) all had a significantly higher risk of dying from pneumonia. In contrast, more than 2 years after diagnosis, risk in those with cancers that were localised at diagnosis was no different from that of cancer-free individuals. Table 3 shows the association between the incidence of site-specific cancers and risk of dying from pneumonia. Compared to cancer-free controls, those with a history of lung cancer, lung cancer diagnosed less than 5 years before and lung cancer with regional spread or distant metastasis at diagnosis were significantly associated with a higher risk of death from pneumonia. Similarly, a history of oesophageal or HN cancer, oesophageal cancer diagnosed more than 5 years before, localised oesophageal cancer at diagnosis, and HN cancer with regional spread or distant metastasis were associated with an increased risk of death from pneumonia. No associations were found for a history of haematological malignancy or other cancers. Among cases with other cancers, the increased risk was found only in those with a diagnosis within the last 5 years. Risk was diminished for the model which excluded cases within 1 year of diagnosis, but was significantly higher among those with regional spread or distant metastases (Table 4).

| DISCUSSION
This study found a positive association between incident cancer and higher risk of death from pneumonia compared to cancer-free controls in the Japanese population, including variations by cancer site, time since diagnosis and clinical stage at diagnosis. In particular, risk was higher in those less than 2 years from diagnosis and those whose cancers were regionally spread or distantly metastasised at diagnosis. Secondary analyses for site-specific cancers revealed that patients with lung, oesophageal, and HN cancers had a higher risk of dying from pneumonia than cancer-free controls. In contrast, no associations were found with haematological malignancies and other cancers. To our knowledge, this is the first time a largescale prospective cohort study has been used in a general population in Japan to quantify the relative risk of death from pneumonia in cancer survivors. Although a systematic review found that incident cancer was not a definitive risk factor for pneumonia, 14 an excess risk of pneumonia deaths among cancer patients has been consistently reported in recent years. Previous reports have shown that cancer patients were 2 to 21 times more likely to die from pneumonia than the general population. [15][16][17] These results had greater impact than our estimate because they focused on deaths during the short term. Helena et al. reported higher risks of hospitalisation and mortality from influenza among those who survived cancer for more than 1 year than among cancer-free controls. 18 Risk diminished with time since diagnosis, and no difference was observed at 10 years after diagnosis. 18 Surveys of childhood cancer survivors have reported that antitumour treatment causes late pulmonary toxicity and chronic pulmonary complications, limits daily activity, increases oxygen need and causes recurrent pneumonia and subsequent excess pneumonia death. 19 Those who survive cancer may be affected by acute, iatrogenic or treatment-induced infections. 25 Both the cancer itself and various antitumour treatments can impair functional status and drastically alter the host's immune system, placing them in an immunosuppressed condition. 19 Immune system suppression using corticosteroids and other immunosuppressive agents or bone marrow suppression caused by cytotoxic agents render patients more prone to infection. 8 Direct irradiation of the chest and lungs can cause treatment-induced pneumonitis and interstitial lung injury. 26 Postoperative pneumonia, although not a focus of this study, is one of the most common complications in patients with lung, oesophagus and HN cancers undergoing surgical resection. 27 The occurrence of postoperative pneumonia can lead to prolonged hospital stays and respiratory failure. 28 Advanced cancer weakens the patient's immune system and requires multimodality therapies, resulting in immunocompromise and increasing the risk of infectious disease. 29,30 Increased exposure to resistant bacteria during frequent treatment may lead to treatment failure and severe pneumonia. 31 In the present study, the increased risk of pneumonia within one or 2 year(s) after cancer diagnosis may be reflective of cases under active T A B L E 4 Hazard ratios (HR) and 95% confidence intervals (CIs) of death from mortality during follow-up by any incident cancer excluding those with a follow-up time less than 1 year since diagnosis in the JPHC study The number was calculated after matching by pooled propensity score for subjects with and without incident cancer during follow-up in a 1:4 ratio with replacement, in which propensity scores were estimated by age at baseline (40-44, 45-49, 50-54, 55-59, 60-64, or 65+ years old), public health centre area (10 areas), sex, smoking status (never, former, or current), history of second-hand smoke at home (no or yes), history of second-hand smoke at the workplace (never, sometimes, or always), alcohol intake (never/former, <1 time/week), regular (ethanol converted g/day) [<23, 23-<46, 46-<69, 69-<92, or ≥92], BMI (in kg/m 2 ; <18.5, 18.5-<25, 25-<30, or ≥30), weekly leisure-time sports or physical exercise (<1 time, 1-4 time(s), or ≥5 times), coffee intake (almost never, <1 cup/d, or ≥1 cup/d), green tea intake (almost never, <1 cup/d, or ≥1 cup/d), and history of diabetes (no or yes). b Cox proportional hazards model using attained age as time scale stratified on the propensity-score matched pairs. treatment, as the entire course of some anticancer regimens extends over more than a year. 32,33 Higher risk of death from pneumonia among cancer survivors may also be explained by the high overlap between the risk factors for developing cancer and for pneumonia death (e.g., tobacco exposure). 10,34 Other risk factors for pneumonia death may also include common cancer comorbidities or/ and comorbidities subsequent to antitumour treatment (e.g., cardiac disease, diabetes, dyslipidaemia, hypertension and obesity). 9,11,13 The high risk of pneumonia among lung cancer patients is consistently attributable to bronchial obstructions caused by tumours, bronchoscopy, surgery and coexisting structural lung diseases. 15 In particular, advanced lung cancer is an independent risk factor associated with pneumonia, mainly as a result of the long-term anatomical abnormalities caused by pneumonectomy. 35,36 This study showed an increased risk in localised lung cancer but not advanced cases, likely because of a high rate of cancer death in advanced lung cancer cases. Surgery for oesophageal cancer frequently results in postoperative complications, which may be one reason for the low pneumonia survival rates among those with this cancer type. 37 HN cancer patients commonly experience a long-term dysphagia, various adverse respiratory outcomes and pneumonia after curative surgery and chemoradiotherapy. [38][39][40] In the present study, regional or distant metastases of HN cancer were associated with pneumonia death. Pulmonary metastases frequently occur in HN cancers and require pulmonary metastasectomies, 41 which may explain the higher pneumonia risk among patients with advanced HN cancer. These findings suggest that a diagnosis of cancer, especially cancer of the lung, oesophagus and HN, should receive higher attention in efforts aimed at preventing pneumonia.
Although some studies have shown a higher risk of pneumonia among haematological malignancies, others have reached inconsistent conclusions. 8,42 Data from the present study showed no similar associations, possibly because the classification of haematological malignancies differed among some studies and was not determined in others.
For other types of cancer, the present study found no association between incident cancer and death from pneumonia. The mix of various cancer types might have diluted the association between incident cancer and pneumonia deaths in other types of cancer. Analysis for the most common types of cancers, including stomach, colon, liver, pancreas, prostate, and breast cancer showed null associations. Given that gastric and liver cancers are mostly caused by infectious agents, and breast and prostate cancers are often caused by sex hormones, we speculate that people with these cancers were less affected by risk factors of pneumonia deaths than those with lung, HN and oesophagus cancers, resulting in the lack of association with pneumonia deaths. The association was not prominent even in non-aggressive cancers, which might be ascribable to less competition from primary cancer deaths, such as prostate cancer. This null association may be partially because cancer survivors appear more likely than the general population to receive vaccination against pneumococcal and influenza viruses due to guideline recommendations. 43 Lifestyle improvements and increased opportunities to access healthcare after diagnosis may also guard against infection. 11 Primary cancer deaths might compete with pneumonia deaths. 25 Competing risk of death from primary cancer may explain the null association for aggressive cancers such as pancreas cancer. Nevertheless, we decided that competing risk assessment was not appropriate in an association study, 44 as an analysis accounting for competing risk may underestimate the relative risk or result in protective pneumonia deaths, leading to erroneous interpretation.

| Strengths and limitations
The study has several noteworthy advantages, including its large sample size, high response rate (81%) and low loss to follow-up. Our analysis using propensity score matching provides robust evidence to mitigate the critical differences in risk factors between cancer cases and controls. Detailed information on the date of cancer diagnosis, clinical stage and site-specific cancers enabled us to assess the gradient of risk among cancer survivors. The study's subjects reflected Japan's general population, making our findings applicable to the nation as a whole.
This study has some limitations. First, because we had no detailed data on cancer treatments, we could not analyse by treatment type or status (whether patients were under active treatment or not). Different regimens may have different effects on pneumonia outcomes. 15,45 Therefore, the inability to assess the impact of cancer treatment may have led to overestimating or underestimating results. Second, although the prospective nature of our study enabled robust adjustment of confounders, we could not rule out the possibility of residual confounding. The study did not obtain data on pneumococcal vaccination, which may have a positive effect on pneumonia outcomes. Without adjustment for vaccination history, our study may have underestimated the effects. Third, we could not assess the impact of lifestyle change or the status of comorbidities after a cancer diagnosis. Fourth, our exclusion of subjects living in Tokyo, an urban area, could have resulted in selection bias because the different lifestyle and better access to medical care of urban compared to rural areas may influence cancer incidence and pneumonia death. Finally, physicians may have made errors in coding the cause of death because of competing effects in comorbidity-associated mortality, 24 especially in cases where the patient was under active treatment based on an expert's opinion. We addressed this issue by excluding those diagnosed within 1 year as a treatment period. We further confirmed that the rate of pneumonia death among cancer cases did not substantially differ by study area, indicating that cause of death was accurately or at least consistently identified nationwide.
Allowing for these limitations, our study nevertheless provides insightful evidence regarding pneumonia in cancer survivors. As mechanisms of tissue injury in pneumonia induced by viral agents seem to share some aspects with severe acute respiratory syndrome coronavirus 2 (SARV-CoV-2), 46 it is intriguing to speculate that our results are extrapolatable to coronavirus disease 2019 (COVID-19)-induced pneumonia, which is caused by SARS-CoV-2. Further study of the association between cancer history and COVID-19 outcomes in the current global pandemic is warranted.

| CONCLUSIONS
This prospective cohort study found a higher risk of death from pneumonia among cancer survivors compared to cancer-free controls. The increased risk was pronounced for cases within 2 years after diagnosis and those with advanced cancer at diagnosis. By site, lung, oesophageal, and HN cancer survivors had a higher risk of death from pneumonia. The results may suggest that the possible immunocompromised status and respiratory failure of patients following antitumor treatment impacted the greater severity of outcomes from pneumonia in cancer survivors than in the general population. 47 These vulnerable people may need to be given special precautions to prevent infection, including optimization of vaccination and patient education.