Cause of Death among Long-Term Cancer Survivors: The NANDE Study

Survival information for Japanese patients with cancer is based only on survival status and the cause of death among these patients remains unclear. In this study, Osaka Cancer Registry data (1985–2014) and vital statistics data (1985–2016) were linked to create a database, permitting the extraction of data on the causes of death. In total, 522,566 subjects diagnosed with cancer between 1995 and 2011 were analyzed. Follow-up for vital status was conducted 5 and 10 years after cancer diagnosis. To evaluate the three causes of death (index cancer, non-index cancer, and non-cancer death), cause-specific hazard and cumulative incidence functions were estimated using a life table and Gray’s methods. The number of deaths owing to any of the causes in the observation period (median: 3.51 years, mean: 4.90 years) was 394,146. The 5- and 10-year cancer-specific survival rate was 48.56% and 39.92%, respectively. Immediately after cancer onset, the hazard of index cancer death was high. The proportion of non-index cancer deaths was high in patients with mouth and pharynx cancers. The hazard of index cancer death remained constant for breast and liver cancers. In prostate, breast, and laryngeal cancers with good prognosis, the hazard of non-index cancer and non-cancer death constantly increased.


Introduction
Cancer accounts for a significant proportion of deaths worldwide, accounting for 5,659,700 deaths in 1990 and over 9 million deaths in 2017, as the number of deaths increases worldwide [1,2]. In Japan, the reported number of deaths from cancer in 2021 was 381,505 (26.5%), and cancer was the leading cause of death. The second leading cause of death in 2021 was heart disease, with 214,710 (14.91%) reported deaths, followed by cerebrovascular disease with 104,595 deaths (7.26%) [3].
Cancer incidence rates continue to increase [4][5][6], although cancer mortality is declining [7,8]. The number of patients with cancer with long-term survival has increased worldwide [9,10]. In addition, with the aging world population, the age of patients with cancer is also increasing. Therefore, it is postulated that more patients will die from causes of death other than primary cancer [5,7]. Few studies have examined the cause of death by cancer site; however, studies examining causes of death limited to breast cancer patients and others that examined survival rates only, without limiting the cause of death by cancer site, have indicated that the incidence of non-cancer death has increased among patients with breast and prostate cancer [11,12]. The Japanese cancer registries only present information on survival status, with information regarding the cause of death remaining unknown. In addition, data such as the proportion of deaths by cause and time trends are lacking. The increase in the number of cancer survivors is a major public health challenge. For cancer survivors to be active in society, it is important to optimize cancer care and provide support

Linking Cancer Infection Information and Death Information
The present study was one of the derived studies conducted by the collaborative study group, NANDE (Neoplasms and other Causes of Death), that identifies and characterizes the cause of death in patients with cancer. First, we created a database to examine the causes of death in patients with cancer. No unique identifier (e.g., social security number) was available in the OCR or VS to allow for deterministic linkage of the corresponding records from the same person. Therefore, OCR data and VS were probabilistically linked using nine indicators: prefectures at the time of death, municipalities at the time of death, sex, birth year, birth month, birthday, year of death, month of death, and the date of death. In the OCR, 716,873 patients with cancer were confirmed to have died between 1985 and 2014. Eventually, 692,069 people from the registry data (96.5% of all deceased) were probabilistically linked with the corresponding data from VS using nine indicators (680,261 people) or eight indicators, except for any of the following: birthday, date of death, or municipality at the time of death (11,808 people). Furthermore, 371,918 people, in whom death could not be confirmed ten years after cancer diagnosis, were included. The final database developed for the NANDE study included 1,063,987 patients.

Selection of Participants
To examine the cause of death in patients with cancer, we extracted only cases with accurate information on morbidity and death derived from the full NANDE dataset. Participants diagnosed with cancer from 1995 to 2011 were included in the analysis. In Japan, ICD-10 coding for causes of death has been used since 1995. Therefore, only deaths from 1995 to 2016 were analyzed in this study, because the ICD-9 and ICD-10 codes do not match entirely. The following cases were excluded from the survival analysis: patients diagnosed after 2012 with less than five years of follow-up (165,416 cases, 15.55% of the total), death certificate only (110,471 cases, 10.38% of the total), carcinoma in situ with good prognosis (46,077 cases, 4.33% of the total), patients with a survival time of 0 days, and patients in whom the number of months survived was unknown (133,298 cases, 12.53% of total). Excluding these cases, 522,566 subjects were included in the last analysis from the 1,063,987 patients in the NANDE dataset, corresponding to 49.11% of the total number of cancer cases.

Ethics Approval
In this study, all patient data were de-identified, and the requirement for informed consent requirement was waived. This study was approved by the ethics committee of the Osaka International Cancer Institute (approval number 1707105103).

Statistical Analysis
The 5-and 10-year crude observed survival rate was calculated using the Kaplan-Meier method [13], and corrected survival rates were calculated by treating the other cause of deaths occurring with five or ten years of follow-up as withdrawals. For example, for calculating the adjusted survival rate for index cancer deaths, non-index cancer deaths, and non-cancer deaths were considered censored. The corrected survival rate of 50% for index cancer death indicates that 50% of index cancer patients are spared the risk of death from the disease within 10 years of diagnosis [14]. The proportion of deaths at the last follow-up was calculated for each cancer site and sex. We created a Sankey diagram to show the correlation between the site of cancer incidence and the ultimate cause of death in relation to gender. Hazard rates were calculated for the following three causes of death using the life table method every half year: index cancer death, non-index cancer death, and non-cancer death [15]. In addition, the hazard ratio was calculated using the cancer-specific hazard rate as the denominator, and the non-index cancer-specific non-cancer-specific hazard rate as the numerator. To estimate the cumulative hazard function by competing for the cause of death, we used the Fine and Gray method, with a proportional hazards model for the sub-distribution of a competing risk [16,17]. The significance level was 5% on both sides, and all statistical analyses were performed using SAS (version 9.4; SAS Institute, Cary, NC, USA). Table 1. Correspondence between ICD-O-3 T and ICD-O-3 M codes on cancer incidence, and IDC-10 code on the cause of death.

Results
A total of 327,954 patients died by any cause of death in the observation period (median: 3.51 years, mean: 4.90 years). The proportion of index cancer deaths was 51.83% for male patients and 42.96% for female patients, while the proportion of non-index cancer deaths was 7.88% for male patients and 6.35% for female patients, and the proportion of non-cancer deaths was 8.58% for male patients and 5.91% for female patients.   Tables 4 and 5 summarize the 5-year crude and corrected survival rates by the cause of death and cancer site. These survival rates are estimated at 5 years after cancer diagnosis and show a similar trend to the 10-year survival rates, with all crude observed survival rates being higher at 5 years. In corrected survival rates, there is a difference of only a few percentage points in index cancer death between the 5-year and 10-year estimates. However, there is a nearly 10% difference in non-index cancer death and non-cancer death estimates.  Pancreatic cancer had the highest proportion of index cancer deaths in male patients (pancreatic cancer death, 89.21%; other cancer death, 2.81%; and non-cancer death, 2.83%). The next highest was lung cancer (lung cancer death, 76.26%; other cancer death, 3.46%; and non-cancer death, 5.58%). In female patients, the cancer with the highest proportion of index cancer deaths was the pancreas (pancreatic cancer death, 89.30%; other cancer death, 2.59%; and non-cancer death, 2.28%). The next highest was the gall bladder (gall bladder cancer death 78.34%, other cancer death 4.71%, and non-cancer death 3.98%). Patients with a poor prognosis had a low proportion of other cancers and non-cancers. Our results revealed that index cancer deaths of the pancreas, lung, gall bladder, liver, esophagus, multiple myeloma, leukemia, and malignant lymphoma were high in male patients, and the index cancer deaths of the pancreas, gall bladder, liver, lung, multiple myeloma, esophagus, leukemia, ovary, and malignant lymphoma were high in female patients. In male patients with mouth and pharynx cancer, the proportion of non-cancer death was 11.47%, which was higher than that of other cancers. This tendency was not observed in female patients. Figures A1 and A2 summarize the relationship between the site of cancer incidence and the ultimate cause of death in a diagram. The composition of causes of death shows that deaths due to cancer at the incidence site are common, and among causes other than cancer, deaths due to cardiovascular disease and respiratory disease are common. Figures 1 and 2 show the temporal trend of the cumulative hazard function of the three causes and the hazard ratio for 6 months. The cumulative hazard function curve shows the cumulative proportion of the causes of death. The red area shows the index cancer death, the blue area shows the other cancer deaths, the green area shows the non-cancer deaths, and the other white area shows the cumulative survival rate.         This tendency was different for each cancer site, and four characteristic types were found. In the first type, the hazard rate of the index cancer was not stable and the index cancer mortality was observed over a long period. This type includes liver cancer, breast cancer in female patients, ovarian cancer, multiple myeloma, and leukemia. With respect to these sites, the hazard of index cancer death does not decrease with secular change. Even after ten years, there is a significant difference between the index cancer death, and non-index cancer deaths and non-cancer deaths. In the second type, most patients die from the index cancer immediately after the onset of cancer. This type included pancreatic, gallbladder, and lung cancer, with over 75% of patients dying from the index cancer. In the third type, including laryngeal cancer in female patients, and skin cancer, and breast cancer in male patients, with relatively low cancer incidence, the confidence intervals of the three causes of death overlap immediately after the onset of cancer. In the fourth type, including prostate, skin, and laryngeal cancers in male patients, index cancer mortality tends to decrease over time.
The hazard ratio shows whether non-index or non-cancer mortality is higher than index cancer mortality. In the first and second types, hazard ratios of both non-index cancer death and no-cancer death were always below 1. In the other cancer sites, the hazard rate of death from the index cancer increased immediately after the onset of cancer; the hazard rate became constant for approximately five years and was almost the same as that of other cancer deaths and non-cancer deaths in 8 to 10 years.

Discussion
This study clarified the long-term characteristics of the cause of death other than cancer in Japanese patients with cancer. While few studies have examined the cause of death by cancer site, previous studies that were limited to breast cancer patients or those that examined survival rates only, without limiting the cause of death by cancer site, have reported results that were similar to those of this study [11,12,[18][19][20][21][22]. In Japan, there is scarce research on the cause of death among patients with cancer and, to the best of our knowledge, the present study is the first large-scale study to examine the characteristics of the cause of death of patients with cancer in Japan. In addition, this research is novel because it can characterize cancer deaths among patients with cancer by linking the OCR, which is the total registration of cancer incidence, and VS, which represent a complete registry of causes of death in the Japanese population. By dint of its population-based cancer registry, Japan was the only country in Asia that could provide data in the CONCORD study, which estimated the relative survival rate of cancer worldwide [9,23]. The OCR was one of the three prefectures participating in the CONCORD study, and this registry is highly regarded worldwide for its scale and accuracy.

Cancer Sites with Consistently High Index Cancer Deaths over the Long Term
There are few studies on the hazards by cause of death for all cancer sites in the 10-year follow-up period after cancer incidence. To the best of our knowledge, few studies have performed an analysis that considers competing risks. These four characteristic types of trends are consistent with those of previous studies. Mortality of patients with breast and liver cancer remains high more than 5 years after cancer, similar to previous studies [12]. In this study, similar trends were observed for liver, ovarian, multiple myeloma, and leukemia; however, there are no previous studies on sites other than breast cancer [11]. In breast cancer, hazards for non-index cancer deaths and non-cancer deaths showed a constant increase in the decade after diagnosis, although this was not observed for liver cancer, ovarian cancer, multiple myeloma, and leukemia. The difference in prognosis between breast cancer and cancer at other sites may explain these results.
In pancreatic, gallbladder, and lung cancer, which are generally considered to have a poor prognosis, the hazard of index cancer death is particularly high, and the hazard declines depending on age after diagnosis. However, it was higher than that of non-index cancer and non-cancer deaths. The 10-year relative survival rate for these cancer sites is very low [8]. In this study, the death proportion of index cancer deaths was 75% or above within 5 years of a cancer diagnosis. Because lung or liver cancers are sites that are likely to be an organ to which primary cancer metastasizes from other organs, if the primary cancer is not listed in the death certificate, it may become the cause of death.
However, if a death certificate states multiple cancer sites, other cancers may be selected as the cause of death, in accordance with the original death cause selection rule. For this reason, the hazard of death from index lung and liver cancer in this study may have been underestimated [24]. It is not possible to directly confirm the contents of the death certificate. However, we consider that this study uses the most robust data currently available.

Cancer Sites with Consistently High Rates of Both Index Cancer Deaths and Non-Index Cancer Deaths
Among male patients, cancer with high mortality from original cancer and high mortality from other cancers are mouth cancer and pharynx cancer, which is different from the trends for the other cancers studied herein. Mouth and pharynx cancers are affected by field cancerization, a phenomenon widely spread across multiple areas due to long-term exposure to common cancer-inducing factors. In previous studies, mouth and pharynx cancers were often affected by esophageal cancer and secondary lung cancer [25]. In this study, male patients with mouth and pharynx cancer had a high risk of death from other cancers, mainly esophageal cancer (15.35%), followed by lung cancer (14.89%); these results are consistent with existing literature.

Cancer Sites with a Relative Decrease in Index Cancer Deaths over the Long Term
Prostate cancer, laryngeal cancer, and breast cancer in female patients showed constant increases in the hazards of other cancer deaths and non-cancer deaths when evaluating the elapsed years after morbidity. Previous prostate and breast cancer studies have shown that the proportion of non-cancer-related our results findings of the present study [11,[26][27][28]. In this study, similar trends are also seen in laryngeal cancer, although there is no previously reported research on this point. These three cancer sites are known to have relatively good prognoses. However, similar trends were not observed in thyroid cancer, which is considered to have a good prognosis. In prostate cancer, the age at diagnosis was the highest, and a high frequency of deaths within 5 to 10 years was observed. Although the number of laryngeal cancers was small, similar trends were also observed. Conversely, as patients with thyroid cancer were diagnosed at a young age, it was considered that other cancer deaths and non-cancer deaths cannot be observed in the observation period of 10 years. In this study, prostate cancer had the highest risk of other cancer and noncancer mortality. Previous studies have found that men diagnosed with prostate cancer are less likely to die from prostate cancer, which is consistent with the results of high other cancer and noncancer deaths [26]. However, deaths due to prostate cancer in this study were not consistent with those reported by previous studies because the mortality rate was constant for 10 years after cancer diagnosis [26]. Potential reasons for this discrepancy include the characteristics of the Japanese medical insurance system, wherein even nonfatal cancers are treated aggressively, and the characteristics that make cancer more likely to be selected as the underlying disease if the cause of death selection rules are followed; however, it is difficult to examine this issue in detail from the present data [24]. In prostate cancer, the age at diagnosis is typically high, and it is necessary to further understand the incidence of second cancers during the observation period of 10 years after treatment. In Japan, a national health insurance system is in place, and even elderly patients attending hospitals and diagnosed with cancer are actively treated. In other countries, aggressive treatment for elderly patients with cancer is not undertaken. According to our data on cancer care in the Japanese medical insurance system, all people of all ages are equally admitted to a hospital if diagnosed with cancer, and have the opportunity to receive treatment; selection bias in treatment is considered to be strong. This study has several limitations. First, it is difficult to analyze patient characteristics and treatment because the registry and vital statistics data do not include clinical information. In addition, there is no adjustment for age at diagnosis and classification by year of birth (for cohort effects). Second, the detailed contents of the death certificate could not be obtained. The wording of the death certificate is primarily influenced by the discretion of the doctor who conducted the death diagnosis, so we could not verify whether the cause of death on the death obtained was the correct cause of death for each patient.
We believe that this study will help people affected by cancer to learn about complications and other diseases that they should be aware of and to review their lifestyle with medical care providers and other stakeholders. The strength of this research is that it is the first study in Japan simultaneously investigating information on cancer incidence and information on the cause of death among patients with cancer. We focused on three causes of death without limiting the cancer sites. In the future, we believe that our study methods will lead to analysis according to personal characteristics such as age at cancer diagnosis, aging, treatment method, degree of clinical progress, and multiple cancer morbidity. In addition, we believe this indicates the need for a study to distinguish the causes of death in more detail and to evaluate the relationship between affected cancer sites and cause of death.

Conclusions
This study clarified secular trends in the three causes of death after cancer diagnosis and the characteristics of each cancer site. These results indicate that many fatalities caused by index cancers occur immediately after cancer incidence, hazards decrease with time, and the difference between non-index cancer deaths and non-cancer deaths disappears in approximately five to eight years. The results also suggest that in cases of breast and liver cancer with many recurrences, the hazard of the original cancer death does not decrease within 10 years. Furthermore, in prostate, breast, and laryngeal cancers with a relatively good prognosis, non-index cancer deaths and non-cancer deaths increase. These findings are useful for developing strategies against cancer and providing important information to patients. We consider that these data will contribute to the optimization of cancer care and support the provision of appropriate information for cancer survivors.

Institutional Review Board Statement:
The study was conducted in accordance with the Declaration of Helsinki and was approved by the ethics committee of the Osaka International Cancer Institute (approval number 1707105103).

Data Availability Statement:
The data that support the findings of this study are not publicly available due to privacy and ethical restrictions. The data are available from the corresponding author on reasonable request. Rectosigmoid junction and rectum (2105) Other malignant neoplasms, stated or presumed to be primary, of lymphoid, hematopoietic and related tissue (2120)

Appendix A. Cause of Death Classification (Code) As Defined by the Ministry of Health, Labour and Welfare
Diseases of the respiratory system (10,000) Liver and intrahepatic bile ducts (2106) Other malignant neoplasms (2121) Diseases of the digestive system (11,000) Gallbladder and other biliary tracts (2107) In situ neoplasms and benign neoplasms and neoplasms of uncertain or unknown behavior (2200) Diseases of the genitourinary system (14,000) Pancreas (2108) Diseases of the blood and blood-forming organs and certain disorders involving the immune mechanism (3000) Accidents (      The diagram above illustrates the correlation between the site of cancer incidence and the specific cause of death. The left side of the diagram represents the site of cancer incidence, while the right side represents the classification of causes of death. The number before the cause of death is the code for the cause of death classification as defined by the Japanese Ministry of Health, Labour and Welfare. The height (width of the bands) represents the number of individuals, with thicker bands indicating higher numbers. It is evident that deaths due to index cancer are prevalent, but for causes other than cancer, heart disease and respiratory system diseases are more common.