Data source
In this retrospective observational cohort study, we used data from the Osaka Cancer Registry (OCR), a population-based cancer registry established in 1962. It includes all patients diagnosed with cancer who reside in Osaka Prefecture at the time of diagnosis. Cancer diagnoses were recorded in the OCR system, along with basic patient information, such as sex, age at diagnosis, residential address at diagnosis, and clinical details based on the diagnosis notice of the medical institution (including cancer sites, histological type, stage, and treatment modality). However, the OCR did not obtain the information on the cause of death. The staging of OCR is based on the Surveillance, Epidemiology, and End Results Summary Staging System15, categorizing cancers into five groups: localized, regional, distant, unknown, and not applicable. The not-applicable category refers to hematological cancers. Information on the vital status of each patient was obtained through the following methods: (1) For patients with a death certificate, the date of death and the confirmation of death were provided by municipalities. (2) For patients without a death certificate, vital status was confirmed by querying the resident register 3, 5, and 10 years after diagnosis, regardless of whether the patient had moved outside Osaka Prefecture16,17. Patients who became untraceable were censored and the last date of confirmed survival was recorded. The most recent date of confirmation from the resident register was recoded as the last date of confirmed vital status in the OCR, and survival days were defined as the time from the date the patient was diagnosed with cancer to the date of death or the last confirmation of vital status, regardless of the cause of death.
Study population
From the anonymized dataset of patients from the OCR database, we extracted patients diagnosed with first primary invasive cancer (International Classification of Disease 10th edition [ICD-10] codes C00–C96) between January 1, 1997, and December 31, 2015. We then excluded the following cases: (a) patients who diagnosed solely on the basis of a death certificate, (b) patients who had missing information on sex or (c) patients who had no recorded residential address at diagnosis. In addition, we excluded patients aged 75 or older because many prefectures, including Osaka Prefecture, are focusing on cancer control for those under 75 years of age, using indicators such as incidence and mortality rates among those under 75 years18,19. Patients under 15 years were also excluded because pediatric cancer differs from adult cancer in terms of tumor site and histological distribution 20. Among the remaining patients, those who received any treatment (surgery, endoscopic therapy, radiotherapy, chemotherapy, or endocrine therapy) were included in this study.
Statistical analysis
The primary outcome of this study was the change in regional variation of 5-year relative survival rate before (1997–2006) and after (2007–2015) the implementation of the BPPCCP. To assess this, we calculated 5-year relative survival rates for each patient’s residential area as well as for Osaka Prefecture as a whole during both diagnostic periods. The patients’ residential areas were categorized into eight divisions (A to H) based on the CMA using their address information14. The names of the CMAs are kept confidential to protect patient identity. In this study, the average survival rate refers to the prefecture-wide average survival rate, and relative survival rates for patients in CMA X was described as “survival for CMA X.” Additionally, regional variation was defined as the difference between the 5-year relative survival rates of each CMA and that for Osaka Prefecture.
All analyses were performed using a flexible parametric Royston–Parmar model. The Royston–Parmar model uses restricted cubic splines to model the baseline hazard, allowing the hazard to vary over time without the proportional hazards assumption required by Cox models21. This flexibility is particularly advantageous when analyzing cancer survival, where hazard rates often change substantially in the years following diagnosis. Given our objective of estimating 5-year relative survival rates, we recognized that the hazard and hazard ratios of patients with cancer could vary over time22; hence, we selected this model to account for potential non-proportional hazards and more accurately capture time-dependent survival patterns.
To calculate the relative survival rate, we used the cohort life table of Osaka Prefecture to obtain the survival expectancy of the general population in this Prefecture23. The outcome measures are reported with their 95% confidence intervals using the delta method. Statistical significance was set at p < 0.05.
The following were considered as confounders and included in the model as covariates: sex, age at diagnosis (natural spline with three degrees of freedom), cancer stage at diagnosis, patients’ socioeconomic deprivation level24, and cancer site, based on the likelihood ratio test. The socioeconomic deprivation level was assessed using the Area Deprivation Index (ADI) of the city or ward where each patient lived at the time of diagnosis. The ADI is a multidimensional measure of socioeconomic status (SES) for each area in which patients live25. As described in previous studies26,27, the ADI for each city or ward in Japan is calculated as the linear sum of census-based factors, such as the proportion of households where all members are aged 65 or older and the total unemployment rate in the area. Since the census is conducted every 5 years, the ADI was based on the census year corresponding to each patient’s diagnosis year: the 1995 census for patients diagnosed in 1997, the 2000 census for those diagnosed between 1998 and 2002, the 2005 census for diagnoses between 2003 and 2007, the 2010 census for diagnoses in 2008–2012, and the 2015 census for diagnoses between 2013 and 2015. The ADIs were categorized into quartiles (Q1–Q4), with Q1 representing the least deprived group and Q4 representing the most deprived group. Cancer sites were divided into 19 categories based on ICD-10 codes as follows: C00–C14 (lip, oral cavity, and pharynx); C15 (esophagus); C16 (stomach); C18–C20 (colorectum); C22 (liver); C23–C24 (gall bladder and bile duct); C25 (pancreas); C32 (larynx); C33–C34 (lung); C43–C44 (skin); C50 (breast); C67 (bladder); C64–C66 and C68 (kidney and urinary tract); C70–C72 (brain and central nervous system); C73 (thyroid); C81–C85 and C96 (malignant lymphoma); C88–C90 (multiple myeloma); C91–C95 (leukemia); and others. All covariates, along with the interaction term for age and sex, were included in the time-dependent effect. The degrees of freedom for the restricted cubic spline in the baseline hazard function for each model were selected based on the Akaike and Bayes information criteria21.
We calculated the 5-year relative survival rates and difference from the average for each CMA without adjusting for confounding factors. Subsequently, we standardized the 5-year relative survival rates for each CMA by adjusting the covariate distribution to that of the entire Osaka prefecture. This allowed for a comparison of 5-year relative survival rates across residential areas28,29. Using this standardized relative survival rate for each CMA and the overall prefecture survival rates, we calculated the difference in relative survival rates from the average. In addition to the all-cancer combined model analysis, subgroup analyses were conducted for five major cancer sites: stomach, colorectum, liver, lung, and breast. These sites were selected due to their prominence in the cancer control measures of Osaka Prefecture30.
All analyses were conducted using Stata/BE Version 18 (StataCorp, TX, USA)31. This manuscript follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines32.
Ethics approval and consent to participate
This study was approved by the Research Ethics Committee of the Osaka International Cancer Institute (Approval No. 21218–2). The Research Ethics Committee of the Osaka International Cancer Institute followed the Japanese government’s “Ethical Guidelines for Medical and Health Research Involving Human Subjects.” The study was conducted in accordance with the ethical standards set forth in the Declaration of Helsinki. All procedures were conducted in accordance with “the Act on the Promotion of Cancer Registries” in Japan, which stipulates that informed consent is not required for the use of cancer registry data. The dataset was provided by the OCR without any personal identifiable information.
AloJapan.com