The study was approved by the local ethics committee of Osaka International Cancer Institute (Approval number: 1707105108) and the study protocol was in accordance with the principles set out in the 1964 Declaration of Helsinki. The requirement for informed consent was waived due to the retrospective nature of the study.
Data sources
This was a multicentre retrospective cohort study using the Osaka Cancer Registry (OCR) and administrative data [14-18]. The OCR is a population-based cancer registry that compiles information on cancer diagnoses and outcomes in patients residing in Osaka Prefecture, Japan. OCR data include age, sex, history of smoking, type of cancer, date of cancer diagnosis, date of the last follow-up, date of any cause of death, cancer stage (i.e., localised, regional to lymph nodes, regional by direct extension, and metastatic) according to SEER (surveillance, epidemiology, and end results) [19]). OCR also includes treatment information (i.e., curative surgery/endoscopic treatment, chemotherapy, hormonal therapy, and radiation therapy). Cancer types are defined according to the International Classification of Diseases for Oncology, Third Edition (ICD-O-3). Furthermore, administrative data from Japan’s Diagnosis Procedure Combination Per-diem Payment System (DPC) were collected from 36 designated cancer care hospitals in Osaka Prefecture. The DPC data include medication and history of PCI. In addition, upon hospital admission, patient data on activities of daily living (ADL, Barthel Index score), smoking habits, and International Classification of Diseases, Tenth Revision (ICD-10) diagnoses are recorded. OCR data are linked to administrative data at the patient level, using each hospital’s patient identification number.
Study population
Study investigators identified gastric (ICD-O-3 topographical codes: C16.x), colorectal (C18.x-C20.x), prostate (C61.x), and lung (C34.x) cancer patients who were diagnosed between 2010 and 2015. This decision was based on data that patients with these cancers underwent PCI most frequently (See Supplementary Figure 1, Additional File 1). Exclusion criteria included a number of items: having undergone coronary artery bypass grafting (CABG), history of myocardial infarction, history of PCI, and missing data (including vital status, DPC, and/or other baseline characteristics) at index admission for PCI or IHD for primary analysis (described below) and at cancer diagnosis for secondary analysis (described further on). The patient selection flowchart can be seen in Figure 1. The presence of IHD, including angina pectoris, asymptomatic myocardial ischemia, and acute myocardial infarction, was determined as a patient receiving IHD as the main diagnosis, having IHD as comorbidity upon admission, or having IHD as an in-hospital complication of index admission based on ICD-10 in DPC data (See Supplementary Table 1, Additional File 1).
Exposure
Patients were categorised into 3 groups: 1) those diagnosed with IHD who underwent PCI (the PCI+ group); 2) those diagnosed with IHD who did not undergo PCI (the PCI-/IHD+ group); and 3) those without a diagnosis of IHD (PCI-/IHD- group). To assess its effects on long-term prognosis, only patients who underwent PCI within 3 years of their cancer diagnosis were included in the PCI+ group. The 3-year threshold was chosen because it includes 90% of patients undergoing PCI after the diagnosis of cancer. Among patients with IHD not undergoing PCI, only those who had been diagnosed with IHD within 3 years of their cancer diagnosis were included in the PCI-/IHD+ group. As a sensitivity analysis, all-cause mortality was also assessed for patients who had undergone PCI or had received a diagnosis of IHD within 1.5 years of their cancer diagnosis.
Potential confounders
Data on medications (statins, b-blockers, angiotensin-converting enzyme [ACE] inhibitors, angiotensin II receptor blockers, and oral anticoagulants [warfarin and direct oral anticoagulants]), coronary risk factors (hypertension, dyslipidaemia, diabetes mellitus, and overweight); and other confounders (atrial fibrillation, congestive heart failure, and chronic kidney disease) were retrieved from the DPC database according to ICD-10 codes (See Supplementary Table 1, Additional File 1). The medication was considered in the analysis if it had been introduced before discharge from the index hospitalisation (See Supplementary Table 2, Additional File 1). Overweight status was defined as a body mass index >25 kg/m2. The Barthel Index was used to measure ADL, and patients were divided into 3 groups based on their scores: 0–39, 40–59, and 60–100.
Statistical analysis
In the primary analysis, we analysed the effect of PCI on long-term all-cause mortality in cancer patients with IHD by comparing the PCI+ and PCI-/IHD+ groups. Survival was calculated from the index admission for PCI or IHD. Subgroup analysis by cancer type was also performed. In the secondary analysis, the PCI+ and PCI-/IHD- groups were compared to examine the combined impact of PCI and IHD on cancer prognosis. Survival was calculated from the index cancer diagnosis. As a sensitivity analysis, the difference in all-cause mortality between the PCI+ group, excluding those with the acute coronary syndrome (ACS), and the PCI-/IHD- group was assessed.
Propensity score-matched survival analyses were performed in both primary and secondary analyses. The propensity score for PCI treatment was calculated using all 22 covariates described in Table 1. For the subgroup analysis of lung cancer patients, small cell carcinoma (ICD-O-3 morphological codes: 8041-8045) was also included as a factor. After 1:1 matching, 5-year all-cause mortality was assessed using Kaplan-Meier analysis. Caliper width was set as 0.2 times the standard deviation of the propensity scores. The balance of each factor was assessed using the standardised difference. Since the time interval between cancer diagnosis and admission for PCI varied, it was considered to represent an immortal time bias in the secondary analysis. Consequently, we used extended Kaplan-Meier analysis by adjusting for immortal time bias [22,23] after propensity score matching. In the PCI+ group, the number at risk during the interval between cancer diagnosis and admission for PCI was 0. Therefore, PCI+ group patients were grouped with PCI-/IHD- group patients during no-risk periods, with survival analysis in both groups starting at the date of cancer diagnosis [22].
Cox proportional hazard analysis with inverse probability of treatment weighting (IPTW) for 5 years from PCI or IHD admission was also performed to confirm the robustness of the results. The entire cohort was weighted by stabilised average treatment effect weight [23]. Proportional hazards assumptions were confirmed by Schoenfeld residuals. For further confirmation, multivariable Cox proportional hazard analysis of the propensity score-matched sample was performed with a history of PCI, age (continuous variable), sex, cancer type, cancer stage, Barthel Index, ACS, and interval from cancer diagnosis to index admission for PCI or IHD as covariates for the primary analysis. Each of these variables, except ACS and interval from cancer diagnosis to index admission for PCI or IHD, was used for the secondary analysis.
JMP (version 11.0; SAS Inc., Tokyo, Japan) was used for data organisation and propensity score matching while graphing and all other analyses were performed using STATA (version 15; STATA Corporation, College Station, TX). Results meeting a 2-tailed P<0.05 were considered statistically significant, and P<0.1 was used to indicate a trend towards significance.