Metastatic Patterns: Insights from Japanese Pathological Autopsy Registry Analysis

Malignant neoplasms often cause lethal organ dysfunction at the primary or metastatic site. Metastasis is the leading cause of cancer-related mortality in approximately 90% of cancer deaths [1]. Thus, understanding its mechanisms and patterns is crucial [2, 3]. However, the definitive mechanism underlying metastasis remains unclear. Current conceptual knowledge suggests that metastasis involves complex mechanical mechanisms influenced by anatomical and hemodynamic factors, as well as seed-and-soil mechanisms due to various tumor and tissue-related factors [4‒8]. These factors include organ blood flow, arteriovenous shunts within organs, blood-organ barriers, sentinel lymphatic pathways, adjacent spaces (e.g., cerebrospinal and body cavities), tissue temperature, tissue oxygenation, organ microenvironment, and genetic and/or epigenetic evolution of tumor clonal cells [3, 4, 9, 10].

Additionally, understanding the characteristics of metastasis patterns in individual primary organ malignancies is crucial for developing effective diagnosis and treatment strategies. While metastatic patterns of malignancies have been extensively studied through autopsy series [10‒13] or healthcare database analyses [14, 15], the frequency and target sites of diagnostic imaging are influenced by the healthcare environment [16], which may result in the inability to detect certain metastases. Autopsy remains the gold standard for examining metastatic sites and provides comprehensive insight. However, previous studies have predominantly focused on major malignancies and common metastatic organs, leaving rare primary malignancies, uncommon metastatic sites, and unique organ-specific metastasis patterns underexplored.

The observation that metastasis patterns are similar across certain primary malignancies suggests the possibility of a shared metastatic mechanism among these malignancies. This insight offers a novel perspective for further elucidating the mechanisms of metastatic progression specific to each primary malignancy. There is a lack of comprehensive studies or established frameworks for classifying or analyzing the similarity of metastatic patterns by cancer type. Few reports provide detailed classifications or similarity analyses of metastatic patterns across all malignancies, including rare primary malignancies.

The Annual of the Pathological Autopsy Cases in Japan is a national registry that aggregates data from 1,173 Japanese facilities, amounting to approximately 1.2 million visceral autopsy cases. Of these, approximately 20% also include brain autopsies, and the registry captures over 95% of all autopsies conducted nationwide. Despite a decline in the autopsy rate in Japan from approximately 20% in 1993 to 1% by 2020, this registry remains an invaluable resource. Leveraging data from the Japanese Society of Pathology’s study on primary malignancy and metastatic sites based on International Classification of Diseases-Tenth Revision (ICD-10) codes, we constructed a comprehensive database of previously reported metastatic forms of rare genitourinary malignancies [17]. This pathology database contains detailed observations of rare malignancies and a comprehensive range of metastatic organs.

A multivariate cluster analysis was used to classify participants based on similarities in their characteristics [18]. This technique was previously used to analyze distant metastatic patterns [19, 20]. The current analysis method and analysis of a large amount of autopsy data can classify various malignancies, including rare malignancies, using a more scientific classification method. Therefore, we conducted this study to elucidate the metastasis patterns of primary malignancies by organ, including rare malignancies, using a comprehensive autopsy database. Additionally, we aimed to classify primary malignancies based on these metastasis patterns by multivariate cluster analyses.

This study was designed based on epidemiological assessments derived from a multiple-case series conducted using a nationwide registry. The Annual of Pathological Autopsy Cases in Japan is a nationwide registry that has collected data from 1,173 of the 1,230 facilities that perform autopsies throughout Japan [21]. The decision to perform an autopsy depends on the consent of the bereaved family and can be based on various reasons, such as determining the cause of death, investigating complications or unexpected deaths, respecting the wishes of the deceased or their family, and contributing to research. The bereaved family has consented to the registration of the anonymized results of the pathological autopsy in the Japanese Autopsy Database for study purposes. The original database of pathology diagnoses includes autopsy cases of all-cause mortality, not limited to clinical cancer. The primary malignancy site was defined as confirmed and suspected primary malignancy, or nonfatal primary malignancy in patients with multiple malignancies, while the metastatic organ was defined as an invasion or distant metastasis according to the Annual of Pathological Autopsy Cases. Data for multiple malignancies were tabulated by treating each malignancy separately. The number of metastases was counted as a one-organ metastasis, even if multiple metastases were identified in a single organ. The original database contained personal information such as age, sex, and occupation, as well as information about clinical diagnosis, treatment history, pathological diagnosis, and pathological findings such as metastasis.

The dataset used in this study, including paper media, was last updated on April 10, 2024. The dataset was created as previously described [17]. We extracted a cross-tabulation table of the primary malignancy and organ metastasis sites from the Annual of Pathological Autopsy Cases in Japan, 1993–2021, compiled according to ICD-10 codes. Data for which organ identification was impossible were excluded from the analysis. Data on metastatic sites with unclear organ identification were included in the overall dataset but excluded from the metastatic pattern analysis. The corresponding table of the correlated ICD-10 tumor classifications used in this study is shown in online supplementary Table 1 (for all online suppl. material, see https://doi.org/10.1159/000542684). To obtain the number of actual cases analyzed, the number of early stage and latent cancer cases was subtracted from the number of autopsy cases. The metastatic frequency (number of metastases per autopsy) and the proportion (percentage of organs with metastases out of the total in a primary malignancy) for 48 organ metastasis sites across 76 primary malignancies were calculated.

Open-source R software (version 4.2.2) was used for the data processing, general statistical analysis, and graph generation. The metastatic frequency for each primary malignancy and the proportion (percentage) of organ metastases based on each primary malignancy were calculated. Heatmaps illustrating the proportion of organ metastases per primary malignancy were created using the R software “ggplot2” and “ComplexHeatmap” packages [22]. Standardized metastatic proportion data were employed for cluster classification, using hierarchical and nonhierarchical clustering. For hierarchical clustering, the default method of “hierarchical clustering” with “Euclidean distance” was used, while for nonhierarchical clustering the K-mean method with Euclidean distance was employed. Determination of the metastatic category in the hierarchical cluster classification was confirmed using forward selection. Cohen’s kappa and weighted kappa [23] using the R software “psych” package were used to find the agreement of two categories of hierarchical and nonhierarchical clustering. The 95% confidence interval (CI) for kappa was calculated from its standard error.

A total of 332,195 autopsy cases and 810,206 organ metastases were included in the analysis. Seventy-six primary organ malignancies (shown in Table 1) were included in the analysis to determine metastatic frequency and proportion. Online suppl. Table 1 presents the number of autopsy cases (median, 1,204; interquartile range [IQR], 574–4,638), total number of metastatic organs (median, 16,234; IQR, 1,048–10,486), and metastasis frequency (median, 2.44; IQR, 1.76–3.34) of primary malignancies. The most frequently reported primary malignant autopsy cases were the bronchus and lungs with 56,026 occurrences, followed by the liver and intrahepatic bile ducts, stomach, malignant neoplasms of hematopoietic origin, pancreas, malignant neoplasms of lymphoid origin, esophagus, prostate, rectosigmoid junction and rectum, and breast. Placental malignancies were the least often reported (11 cases), followed by craniopharyngeal duct, aortic body and other paraganglia, parathyroid, spinal meninges, ear and external auricular canal, overlapping lesions of the biliary tract, lip, peripheral and autonomic nervous system, and urethra and paraurethral gland.

The overall metastasis frequency was 2.44. The top three primary malignancies (metastasis frequency) with a tendency toward multiple organ metastases were the placenta (5.09), eye and adnexa (4.24), and ovaries (3.94). In contrast, primary malignancies that tended to metastasize to fewer organs at the time of death, with a metastasis frequency of less than one, were mostly central nervous system (CNS) tumors: craniopharyngeal duct (0.60), brain stem (0.69), cerebrum (0.72), larynx (0.96), and cerebellum (0.98).

The most frequently reported metastasized organs were the lung (113,678 cases), liver (107,849 cases), bone (58,698 cases), adrenal gland (51,324 cases), and peritoneum (49,382 cases). The number of metastases is influenced by morbidity and autopsy rates. Therefore, we compared the proportion of metastatic sites by primary malignancy because the proportion was not affected by morbidity or autopsy rate. Parathyroid malignancy had the highest proportion of numbers with lung metastases (41.9%), followed by thyroid (24.1%) and liver and intrahepatic bile duct (23.7%) malignancies. For liver metastases, overlapping biliary tract malignancy had the highest percentage (24.0%), followed by the sigmoid colon (23.7%) and gallbladder (21.6%) malignancies; for bone metastases, prostate malignancy topped the list (22.3%), followed by spleen (16.9%) and aortic body and paraganglia (14.7%) malignancies; for adrenal metastases, lung (9.3%), kidney (8.9%), and renal pelvis (8.0%) malignancies; for peritoneal metastases, appendix (17.5%), ovary (16.1%), and cecum (12.5%) malignancies. For brain metastases, the brain stem was the most common site (44.7%), followed by the craniopharynx (41.7%) and cerebrum (33.8%) malignancies.

Standardized metastatic proportions for each primary malignancy were grouped using hierarchical clustering. In this approach, the route or organs important for metastatic pattern classification were the lung, liver, bone, peritoneum, brain, and hematolymphoid. We classified the patients into seven malignancy groups based on their clinical significance of the classified groups. We considered separating malignancies with similar organogenesis and anatomic proximity to be overclassifying. The results of the seven stratified cluster analyses are shown in Figure 1. Similarly, the classification into seven groups by nonhierarchical cluster classification using the K-mean method closely matched that of the hierarchical cluster classification, confirming the robustness of the classification (shown in Fig. 2). The analysis of 76 primary malignancies comparing the two classification methods demonstrated substantial agreement, with an unweighted kappa value of 0.84 (95% CI: 0.75–0.93) and a weighted kappa value of 0.77 (95% CI: 0.60–0.93).

Table 2 displays the seven subpopulations classified by hierarchical and nonhierarchical cluster analyses and the affiliation of primary malignancies commonly classified by both classifications. Many malignancies of the gastrointestinal system, urogynecological and pelvic organ, hemolymphatics, and CNS are categorized into the same group. Lung metastases are common in patients with thoracic organ malignancies. In addition to prostate and breast malignancies, head and neck malignancies of the maxillary sinus, nasal cavity and middle ear, nasopharynx, and parotid gland show a high frequency of bone metastases. Ovarian malignancy is characterized by peritoneal metastasis and grouped with the gastrointestinal organs, rather than other gynecological malignancies. Tonsil malignancy, similar to other hemolymphatic tumors, is characterized by hematogenous metastasis.

Table 3 shows the proportions of characteristic metastatic sites in major primary malignancies, representing the seven subpopulations. Table 4 shows the proportions of characteristic metastatic sites in rare primary malignancies. Online supplementary Table 2 presents the patterns and proportions of representative metastatic sites in all primary malignancies in the metastatic group. The common sites of metastatic organs is influenced by anatomic location (e.g., brain metastasis of intracranial malignancies, lung metastases of thoracic organ malignancies, and peritoneal and liver metastases of abdominal organ malignancies) and/or organ specificity (e.g., bone metastasis of prostate and breast cancer) of the primary malignancy. Supplementary comma-separated values data show the numbers, proportions, and metastatic frequencies of all 48 metastatic sites across all 76 primary malignancies in the metastatic group.

The association between the seven classified subpopulations and standardized proportions of hematogenous metastases in the brain, lung, liver, bone, peritoneum, adrenal gland, and spleen is shown in Figure 3. The lung metastasis (group A), and lung and liver metastasis groups (group B) differed in the proportion of lung metastasis. Bones with the visceral metastasis group (group C) consisted predominantly of bone metastasis. The peritoneal with visceral metastasis group (group D) consisted predominantly of intraperitoneal organs dominated by portal blood flow, with a high frequency of liver metastasis. The hematolymphoid metastasis group (group E) showed hematogenous metastases to the spleen. In the brain metastasis group (group F), brain metastasis accounted for most of the metastases, while the brain with visceral metastasis group (group G) showed brain metastases accompanied by lung, liver, and bone metastases.

Here we analyzed a large amount of Japanese autopsy data from the Asian region and demonstrated a metastatic relationship between primary malignancies and the metastatic organs. Our finding showed that a large number of malignancies could be classified into seven patterns, deemed to possess potential clinical significance. Furthermore, our metastasis classification demonstrated high consistency between classification methods, as evidenced by an unweighted kappa value of 0.84, exceeding the 0.8 threshold, indicating substantial agreement. Although many metastatic studies have been conducted, few have analyzed all primary malignancies in an attempt to comprehensively classify them by organ [5, 10, 13, 15]. To our knowledge, this seven-category classification is the first of its kind. Our findings will be valuable to clinicians because they provide insight into the metastatic patterns of rare malignancies and enable the prediction of metastases that have not been extensively studied. Our study of the major cancer types is also important, particularly for Asians, since it recognizes ethnic differences in metastatic patterns. Our findings provide important information for discussing the characteristics of metastasis of primary malignancies based on the similarity of metastasis patterns and may be useful in clinical practice for planning monitoring of organ dysfunction, considering preventive interventions for bone fractures and brain metastasis, and selecting drug administration routes.

When organ metastasis patterns for the major cancer types were compared with those reported in Europe and the USA [13‒15], most malignancies were similar and no major ethnic differences in metastasis patterns were observed. However, differences in the frequency of liver metastasis have been observed in some malignancies, particularly hepatic and esophageal. Liver metastasis from liver malignancies is probably due to assessment bias caused by differences in definitions, and liver metastases from esophageal malignancies are more common in overseas reports, a difference that is probably due to the fact that esophageal cancer is predominantly squamous cell carcinoma in Japan [24], whereas adenocarcinoma from Barrett’s esophagus is predominant in Western countries.

In our analysis, brain, lung, liver, bone, peritoneum, and hematolymphoid metastases were considered medically meaningful classifiers. At these sites, the anatomical and biological features are associated with metastasis. Malignancies of the digestive system often metastasize to the liver due to portal blood flow [25]. Ovarian [26], peritoneal [27], and gastrointestinal [28, 29] cancers are prone to intraperitoneal dissemination [30]. Bone metastasis of prostate cancer can result from the spread to the vertebral venous plexus [31] and adaptation to the bone environment [5, 6, 32]. Metastasis to the lungs occurs frequently via hematogenous venous drainage, the lymphatic spread of antegrade and retrograde lymphatic spread, or direct metastasis via the pleura [33, 34]. The brain’s microenvironment, anatomical structures, metabolic constraints, and immune environment differ drastically from those of extracranial lesions, imposing distinct and profound selective pressure on tumor cells [35]. Splenic metastasis is characterized by mechanical factors such as a high blood flow that prevents the implantation of cancer cells, a lack of afferent lymphatic vessels, the inhibitory effect of the splenic microenvironment, and splenic extension by peritoneal dissemination [36].

Malignant cells not only randomly metastasize to organs, but they can also have nonrandom metastatic patterns that tend to metastasize systematically to some metastatic sites [37]. There are few reports on the incidence of distant metastases in rare parathyroid carcinoma; however, the most common metastasis site is the lungs [38], which is supported by our report. The previously reported incidences of bone metastases in prostate and breast cancers are 65–84.4% [5, 39, 40] and 60–80% [41, 42], respectively. Our results showed a similar trend in the proportion of bone metastases to total organ metastases, although there was some reporting bias due to the inherently favorable natural history. CNS tumors, such as those of the brain stem, cerebellum, cerebrum, pineal gland, and pituitary gland, have a low overall metastatic frequency and a low proportion of lung and liver metastases, suggesting the rarity of hematolymphoid metastases.

This study identified several malignancies with high and low frequencies of organ metastases. These results are consistent with those of many previous similar studies, such as the high metastatic frequency of eye, ovarian, and pancreatic cancers and the low frequency of laryngeal and lip cancers. However, there are some differences, such as in testicular tumors [13]. These differences may be attributed to internal or external ethnic differences, changes in treatment systems over time, or sample bias. The number of distant metastatic organs does not necessarily represent the malignant grade of the cancer. The metastatic frequency was likely influenced by multiple distant organ metastases due to peritoneal dissemination (e.g., ovary, retroperitoneum, peritoneum, and appendix) with active histopathologic metastatic potential (e.g., placenta and pancreas), a long natural history (e.g., eye, adnexa, nasal cavity, middle ear, vagina, anus, and prostate cancer) [40], or a primary site that could itself be lethal [43] (e.g., craniopharyngeal duct, brain stem, cerebrum, cerebellum, pineal gland, and pituitary gland).

The strength of this study is that a large number of organ malignancies and rare organ metastases were included and the characteristics of the metastatic patterns were analyzed using an epidemiologically and statistically robust database in a multivariate technique [17]. This study is the first large-scale analysis conducted in Asia. However, it has several limitations. First, detailed associations with pathological tissue type, gender, age, etc., could not be presented because the secondary analysis was made of the database rather than individual data analysis. Moreover, limited histopathological information was available. Second, because only autopsy cases were considered, selection bias could not be ruled out. The autopsy database exhibits a high coverage rate of over 95% in terms of the number of facilities. However, the autopsy rate in Japan has declined from approximately 20% in 1993 to 1% in 2020, suggesting the presence of selection bias [21]. Third, the analysis of metastatic patterns was based on autopsy information from fatal cases, without a time series. Fourth, the number of brain and bone metastases is underestimated in this study. In the current Japanese database, only approximately 20% of cases underwent brain autopsy. Concerning the search for bone and brain metastases, not all asymptomatic autopsy cases are evaluated, and not all bone sites are pathologically evaluated at autopsy. Therefore, the implications and significance of our findings, particularly those concerning the seven metastatic groups, require further clarification.

Here we conducted an extensive analysis of autopsy data from Japanese patients to characterize the metastatic organ patterns across various primary malignancies, including rare malignancies. An analysis of metastatic sites using these large autopsy datasets and clustering methods revealed the propensity for metastasis to certain sites as well as the natural history of metastasis. The metastatic patterns of primary malignancies include the brain, lung, liver, bone, peritoneum, and hematolymphoid metastases, leading to seven classifications. Knowledge of cancer metastatic patterns, including common and rare malignancies, provides valuable insight into imaging finding and metastatic-site detection methods and encourages further investigations into the mechanisms of metastasis.

We express our deepest gratitude to the bereaved families who provided consent for the autopsies, the pathologists throughout Japan who worked hard to populate the vast database, and the Pathological Society of Japan. An English-speaking medical editor proofread the manuscript and verified its accuracy.

This study was approved by the Research Ethics Review Board of Toranomon Hospital (Approval No. 2451), which waived the requirement for informed consent due to its retrospective nature, and conducted in accordance with the Declaration of Helsinki. All data cited in this paper were derived from published information. The “Deep L Pro” artificial intelligence tool (DeepL SE) was used to draft the first version of the manuscript before expert English editing and data checking were performed; however, it was not used for the data analysis.

The views expressed in this article are those of the authors and do not necessarily reflect the official views of the Pharmaceuticals and Medical Devices Agency.

Suguru Oka has a grant from the Okinaka Memorial Institute for Medical Research. Shinji Urakami had research funding support from Boston Scientific Japan (No. 2310014002). This study was supported by these grants that facilitated the extraction and synthesis of data, as well as the refinement of English language in the text.

Tomohiko Hara: conceptualization, data curation, investigation, formal analysis, validation, software, methodology, visualization, writing – original draft, and writing – review and editing. Suguru Oka: data curation, investigation, formal analysis, validation, software, writing – original draft, and writing – review and editing. Shinji Ito: methodology, supervision, and writing – review and editing. Takeshi Yamaguchi and Kazushige Sakaguchi: supervision and writing – review and editing. Michikata Hayashida: writing – review and editing. Shinji Urakami: project administration, supervision, funding acquisition, and writing – review and editing.

Tomohiko Hara and Suguru Oka contributed equally to this work.

All data cited in this paper were derived from published information and are presented in the manuscript and/or supporting information files. The original data underlying the results presented in the study are available from the Annual of Pathological Autopsy Cases in Japan (https://pathology.or.jp/kankoubutu/JSP-hyou.html).