Study design and data source
We conducted a multicentre retrospective cohort study of cancer patients residing in Osaka Prefecture, Japan. The study was performed using a database comprising hospital-based cancer registry data, administrative data, and population-based cancer registry data. First, clinical data from a hospital-based cancer registry were linked with administrative data produced by hospitals for reimbursements under Japan’s Diagnosis Procedure Combination/Per-Diem Payment System. The details of this record-linked database have been reported previously [19,20,21,22,23,24,25,26]. Briefly, the hospital-based cancer registry collects patient demographic information, as well as information on the diagnosis and treatment of newly diagnosed cancer cases; this includes the date of diagnosis, topographical and morphological codes based on the International Classification of Diseases for Oncology Third Edition (ICD-O-3), cancer stage at the time of diagnosis based on the Seventh Edition of the Union for International Cancer Control staging system, and cancer treatment modality. Cancer treatment refers to the initially planned course of treatment that occurred within four months of cancer diagnosis. The administrative data contained inpatient clinical summaries for hospitalisation episodes. In addition to the hospital-based cancer registry data and administrative data, our study database also incorporated data from the Osaka Cancer Registry, a population-based cancer registry that collects information on vital statuses of Osaka Prefecture residents using death certificates and official resident registrations. Using these data, we identified residents diagnosed with cancer in 2014 and 2015; their vital statuses were tracked in May 2018 and May 2019, respectively.
The study database, which was organisationally supported by the Council for Coordination of Accredited Cancer Hospitals, comprised data that were voluntarily provided by 31 hospitals in Osaka Prefecture. These hospitals are accredited as cancer hospitals by the national or prefectural government, and treat approximately half of all newly diagnosed cancer patients residing in the study region.
Study population
Using the hospital-based cancer registry data, we first identified 18,018 eligible subjects who (a) were diagnosed with gastric (ICD-O-3 topographical codes: C16.x), colorectal (C18.x–C20.x), or lung cancer (C33.x–C34.x) between April 1, 2014 and December 31, 2015; (b) were aged 65–99 years at the time of cancer diagnosis; (c) received cancer treatment at any of the 31 study hospitals; and (d) had at least one cancer-specific hospitalisation episode within 90 days before or after a cancer diagnosis as identified in the administrative data. The three cancer sites were chosen due to their relatively high prevalence in older adults residing in the study region. When a single patient had two or more cancer records for one site, we selected his/her earliest record of the most advanced-stage cancer. Cancer treatment included best supportive care as well as treatment with curative intent. We excluded tumours of the three sites if they were diagnosed as sarcoma (ICD-O-3 morphological codes: 8800–9044, 9120–9262, or 9540–9581; n = 147), haematological tumour (9590–9989; n = 91), or melanoma (8720–8790; n = 3). In addition, patients were excluded if they had missing data for dementia status at the time of cancer diagnosis (n = 2). One of the study hospitals had no eligible patients, and the final study population comprised 17,775 patients from 30 hospitals.
Preexisting dementia
Dementia status at the time of cancer diagnosis was analysed as the exposure of interest. Japanese hospitals are required to include dementia status in clinical summaries for inpatients aged 65 years or older discharged on or after April 1, 2014. Dementia status is evaluated upon hospital admission using a scale described below. For each patient identified in the hospital-based cancer registry data, we searched the clinical summaries in the administrative data for the cancer-specific hospitalisation episode closest to the cancer diagnosis (designated the index hospitalisation) to determine dementia status.
Dementia status was evaluated using a dementia scale that assigns ranks to persons based on their symptoms and degree of independence in activities of daily living: no dementia (rank 0); having symptoms of dementia, but can live independently in one’s home and community without assistance (rank I); having symptoms of dementia, but can live independently in one’s home and community with assistance (rank II); having symptoms of dementia that sometimes affect daily life such that occasional caregiving is required (rank III); having symptoms of dementia that frequently affect daily life such that full-time caregiving is required (rank IV); and having severe symptoms of dementia that require specialised medical treatments (rank M) [27]. This scale has been shown to have good reliability and validity [27]. The ranks are aggregated into three categories (rank 0, ranks I to II, ranks III to IV or M); for this study, we considered these categories to indicate the severity of dementia (no dementia, mild dementia, and moderate-to-severe dementia, respectively).
Cancer staging
Cancer stage at the time of diagnosis was determined using pathological staging. However, clinical staging was used for patients who did not undergo surgical resection for their cancer or had received neoadjuvant therapy prior to surgical resection. Patients with missing information on cancer stage in the hospital-based cancer registry data were classified as having ‘unstaged’ cancer. To assess if there were differences in staging between patients with and without dementia, we performed multivariable logistic regression analyses where the outcome was the receipt of an unstaged cancer diagnosis (vs. staged cancer). Dementia status was examined as the explanatory variable of interest, and the covariates included age (65–69, 70–74, 75–79, 80–84, ≥ 85 years), sex, and comorbidities at the time of diagnosis. Information on age and sex were obtained from the hospital-based cancer registry data, and information on comorbidities was acquired from the clinical summaries in the administrative data corresponding to the index hospitalisation. Comorbidities were measured using updated Charlson Comorbidity Index (CCI) scores based on International Classification of Diseases, Tenth Revision codes [28, 29]. These scores were calculated as the sum of the individual component scores (ranging from 1 to 4) for 10 major diseases (e.g., heart failure and renal disease) associated with increased mortality. Dementia was excluded from CCI scoring. In addition, metastatic cancer was also excluded from CCI scoring because it may be associated with cancer stage in the target cancers. Patients were categorised as follows: no comorbidity (CCI score: 0), moderate comorbidities (1–2), and severe comorbidities (≥ 3). The measurement method is described in further detail in our previous study [19].
Next, patients with unstaged cancer (n = 233) were excluded, and cancer stages were categorised into early stage (0 and I) and advanced stage (II, III, and IV). We performed multivariable logistic regression analyses where the outcome was the receipt of an advanced-stage cancer diagnosis (vs. early-stage cancer). Dementia status was examined as the explanatory variable of interest, and the covariates included age, sex, and comorbidities.
Cancer treatment
We examined cancer treatment after excluding patients with unstaged cancer (n = 233), patients who were diagnosed with small cell lung cancer (ICD-O-3 morphological codes: 8041–8045; n = 461), and patients who were diagnosed with stage 0 lung cancer (n = 8). These patients were excluded because standard treatments could not be determined according to cancer stage, because of major differences with non-small cell lung cancer (NSCLC) patients, and because of the small sample size, respectively. Cancer treatment included endoscopic resection, open surgical resection, laparoscopic resection, thoracoscopic resection, pharmacotherapy, and radiotherapy; information on these treatments was obtained from the hospital-based cancer registry data. The first four modalities were collectively categorised as ‘tumour resection’. Debulking surgery and radical resection were also included in tumour resection. On the other hand, palliative procedures that relieved symptoms but did not reduce tumour mass (e.g., bypass surgery and endoscopic stent placement) were not included in cancer treatment.
First, we performed multivariable logistic regression analyses where the outcome variable was the receipt of any cancer treatment modality (vs. no treatment) included in the initially planned course [14]. Next, we performed multivariable logistic regression analyses where the outcome variable was the receipt of a standard treatment modality (vs. no standard treatment). To identify the standard treatment modalities (tumour resection, pharmacotherapy, and radiotherapy) for each cancer site and stage among older patients in current real-world settings, we performed preliminary analyses to investigate the most common treatment modality for each stage of cancer using our dataset. We found that tumour resection was most common for stage I (91%), II (53%), and III (32%) gastric cancer; pharmacotherapy was most common for stage IV (42%) gastric cancer; tumour resection was most common for stage 0 (99%), I (85%), II (84%), III (42%), and IV (27%) colorectal cancer; tumour resection was most common for stage I (74%) and II (48%) NSCLC; and pharmacotherapy was most common for stage III (23%) and IV (46%) NSCLC. Therefore, tumour resection was regarded as the standard treatment modality for stage I, II, and III gastric cancer, all stages of colorectal cancer, and stage I and II NSCLC. Pharmacotherapy was regarded as the standard treatment modality for stage IV gastric cancer and stage III and IV NSCLC. In these multivariable logistic regression models, dementia status was examined as the explanatory variable of interest, and the covariates included age, sex, comorbidities, and stage. In addition, we constructed stage-stratified multivariable logistic regression models where the outcome was the receipt of a standard treatment modality (vs. no standard treatment) for each stage. Dementia status was examined as the explanatory variable of interest, and the covariates included age, sex, and comorbidities.
Mortality
We constructed Cox proportional hazards regression models where the outcome of interest was overall survival time for a maximum follow-up period of three years. The duration of follow-up was defined as the period between the date of cancer diagnosis and the date of death from any cause. Patients were censored at the date of the last follow-up with an ‘alive’ status from the registry data or administrative data, whichever was later. Dementia status was examined as the explanatory variable of interest, and the covariates included age, sex, comorbidities, stage, and histology. ‘Unstaged’ was included in cancer stage because we considered it to have potential prognostic value. Histology was only included as a covariate for lung cancer, and was classified into small cell lung cancer (ICD-O-3 morphological codes: 8041–8045) and NSCLC (all other codes). Patients with no follow-up for vital status (n = 7) and patients with stage 0 lung cancer (n = 8) were excluded from the survival analyses.
Other statistical procedures
Crude percentages were used to compare the distribution of demographic and tumour characteristics among the three dementia status groups (no dementia, mild dementia, and moderate-to-severe dementia). Adjusted odds ratios and adjusted hazard ratios were reported with their 95% confidence intervals for the logistic regression models and Cox proportional hazards regression models, respectively. All analyses were performed separately for each of the three cancer sites. All P values were two-sided, and P < 0.05 was considered statistically significant. Survival analyses were performed using SAS 9.4 (SAS Institute Inc., Cary, NC, USA), and all other analyses were performed using IBM SPSS Statistics version 22 (IBM Corp., Armonk, NY, USA).