Introduction
The Alzheimer’s Association and the Alzheimer’s Drug Discovery Foundation examined the association between cancer and dementia, and concluded that existing evidence suggested that the risk of dementia was lower among adults with than without cancer (Snyder et al., Reference Snyder2017). A recent systematic review and meta-analysis of 36 studies found that cancer was associated with a reduction of 11% in the risk of dementia compared with non-cancer controls, although there was inconsistency of results and marked heterogeneity across the studies (Zhang et al., Reference Zhang2022). Publication and survivorship bias, and confounding due to exposures such as smoking, diet, and physical inactivity have been offered as possible explanations for the observed association between cancer and dementia (Ganguli, Reference Ganguli2015), but we are not aware of previous studies having investigated how the inception cohort and approach to the analysis of the data may have affected the results of these reports.
We used data from the Health In Men Study (HIMS) to determine if incident cancer affects the risk of incident dementia (Norman et al., Reference Norman, Flicker, Almeida, Hankey, Hyde and Jamrozik2009). We excluded from the cohort prevalent cases of cancer (past or current) and dementia, so as to minimize the potential bias associated with the selection of cancer survivors. This approach allowed us to investigate the contribution of novel cases of cancer to the onset of dementia over 22 years.
Methods
Study design, setting, and participants
We used the electoral roll (voting is mandatory in Australia) to select 38,173 men who were living in the Perth metropolitan region and were aged 65–85 years in 1996. We excluded 342 participants who had received a diagnosis of dementia prior to enrollment and another 6750 who had had cancer. One other participant was excluded because he died before the invitation date. Hence, the study sample consisted of 31,080 older participants who were followed until they received the diagnosis of dementia, died, or reached the end of the follow-up period (31 December 2018). The University of Western Australia Human Research Ethics Committee and the Ethics Committee of the Department of Health of Western Australia approved and oversaw the conduction of this study.
Outcome of interest: dementia
We used the Western Australian Data Linkage System (WADLS) to retrieve information about health contacts associated with a recorded diagnosis of dementia. WADLS brings together hospital morbidity data (inpatient and outpatient), emergency department, and mental health contacts, as well as the cancer and death registries (Holman et al., Reference Holman2008). The causes leading to health contacts are recorded, and the coding of diseases follows the guidelines of the International Classification of Diseases (ICD): ICD-9 from 1 January 1970 to 30 June 1999, and ICD-10 from 1 July 1999. The death registry provides information about the date and cause of death. The following ICD codes were used to ascertain the presence of dementia (all causes): 290, 294.1, 294.2, 331.0, 331.1, 331.82, and 331.2 (ICD-9), and F00, F01, F02, F03, G30, G31.0, G31.1, and G31.83 (ICD-10). The date of the first recorded diagnosis of dementia was considered the date of onset.
Cancer
The reporting of cancers is mandatory in Australia and diagnoses were retrieved using the WADLS and the following ICD codes: 140-9, 150-7, 159, 160-5, 170, 172-3, 158, 171, 176, 175, 185-9, 190-9, 200-8 and 238.0 (ICD-9), and C00-9, C10-9, C20-6, C30-9, C40-1, C43-9, C50, C60-9, C70-9, C80-9, C90-7. These codes record malignant neoplasms of the lip or oral cavity or pharynx, digestive organs, respiratory and intrathoracic organs, bone and articular cartilage, skin, mesothelial and soft tissue, breast, genital organs, urinary tract, eye of brain or other parts of the central nervous system (CNS), thyroid and other endocrine glands, ill-defined or secondary or unspecified sites, lymphoid or hematopoietic or related tissue, and independent multiple sites.
Statistical Analyses
We used the statistical software Stata 17.0 (StataCorp LLC, 2022) to manage and analyze the data, descriptive statistics to summarize categorical measures as count and proportions (%) and continuous measures as mean, range, and standard deviation of the mean (SD). We split time-span records according to the date of the recorded diagnosis of cancer, so that participants would contribute information as non-cancer cases up to the date of diagnosis of cancer. We then used Cox regression analyses according to the Breslow method to estimate the hazard ratio (HR) and respective 95% confidence interval (95% CI) of dementia associated with the diagnosis of any cancer, with age being used as the time scale. We subsequently repeated this approach to each specific cancer site. Sensitivity analyses were used to exclude participants who received a diagnosis of dementia within 1 and 2 years of the diagnosis of cancer, as more frequent contact with health services could increase the opportunity for a diagnosis of dementia to be made. Finally, a post hoc analysis examined the possibility of reverse causality, that is dementia affecting the risk of cancer. We used log-log plots to inspect the proportional hazards assumption and tested for the distribution of Schoenfeld residuals after the fitting of the regression models. Alpha was set at 5%, and all statistical tests were two-tailed.
Results
Cancer and risk of dementia
The age of participants ranged from 65.0 to 85.7 years (mean ± SD = 72.2 ± 4.5 years), and none had a recorded diagnosis of either cancer or dementia at the start of the study. They were followed for up to 22.7 years (mean = 13.2 ± 6.7 years). During follow-up, 18,693 (60.1%) men had a recorded diagnosis of cancer and 6897 (22.2%) of dementia. Of 18,693 who developed cancer during follow-up, 1930 (10.3%) received a diagnosis of dementia after the diagnosis of cancer and 1860 (10.0%) before. Among participants who had cancer and later developed dementia, the time elapsed between the two diagnoses was 5.8 ± 4.4 years. The risk of dementia was higher among older participants who developed cancer (HR = 1.13, 95% CI = 1.07, 1.20), but declined after the exclusion of those diagnosed with dementia within 1 (HR = 0.97, 95% CI = 0.91, 1.02) and 2 years (HR = 0.85, 95% CI = 0.80, 0.91) of the diagnosis of cancer (Figure 1). Table 1 shows the number of participants affected by specific types of cancer, as well as those who developed dementia.
Figure 1. Age-adjusted proportion of men with and without diagnosis of cancer who remained free of dementia during follow-up. The shady bands represent the 95% confidence interval (95% CI) of the proportion. The left panel shows data for the entire sample of 31,080 participants: the hazard ratio (HR) of dementia-associated cancer was 1.13 (95% CI = 1.07, 1.20). The right panel excludes 449 participants who developed dementia within the first 2 years after the diagnosis of cancer: the HR of dementia associated with cancer was 0.85 (95% CI = 0.80, 0.91).
Table 1. Age-adjusted hazard of dementia associated with specific types of cancer a and after the exclusion of 449 participants with dementia diagnosed within 2 years of the diagnosis of cancer b
Participants were considered to be "free of cancer" until the date of the diagnosis, after which they were counted as "exposed cases" for the calculation of the hazard ratio. Cases of dementia arising before the diagnosis of cancer were censored at the time the diagnosis of dementia was recorded. 95% CI: 95% confidence interval of the ratio. CNS: central nervous system.
a All participants were free of cancer and dementia at study entry.
b Excludes 449 participants who developed dementia during the first 2 years following the diagnosis of cancer.
Bold print used to highlight associations that were statistically significant (p < 0.05).
Post hoc analyses: reverse causality
Among participants who develop cancer during follow-up, the diagnosis of dementia preceded that of cancer in 1860 (9.9%). The diagnosis of dementia increased the age-adjusted risk of incident of cancer (HR = 2.32, 95% CI = 2.20, 2.44), but this association was no longer significant when 1200 men who developed cancer within 2 years of a diagnosis of dementia were dropped from the analyses (HR = 0.96, 95% CI = 0.89, 1.04). During follow-up, 24,931 (80.2%) participants died, and of these, 11,849 (47.5%) died of cancer. Among the men who died of cancer, 2248 (19.0%) had a recorded diagnosis of dementia.
Discussion
The design of this study sought to minimize inception bias and bias associated with frequent health contacts triggered by the diagnosis of cancer. We found that the direction of the association between cancer and dementia varied according to the approach used to analyze the data: increased hazard when we used time-to-event analyses and decreased hazard when participants who developed dementia within 2 years of the diagnosis of cancer were excluded from the analyses.
Strengths of the study include the use of a large community-representative sample of older men, access to health record linkage, long duration of follow-up that allowed for the accrual of thousands of cases of cancer and dementia, and an analytical approach that minimized the risk of bias (the association between cancer and decreased risk of dementia has been reported by others, but the approach we used to examine this association represents a novel contribution, as does the reporting of the association between different types of cancer and dementia). This study also generated data about specific cancer groups and showed that the association between cancer and dementia is not specific to one type of cancer (except for skin cancer). Limitations include the non-availability of women in the sample, confounding associated with unmeasured factors, residual confounding, and the absence of detailed clinical information to confirm the diagnosis of dementia, although recent evidence suggests that the use of health administrative data can provide robust information about the rates of dementia in the community (Flicker et al., Reference Flicker2022). Survivorship bias could have also affected the results by maintaining in the sample cancer survivors who are at lower risk of dementia than those who die early. Moreover, we cannot be certain about the time lag between the onset of the disease and the diagnosis of cancer, so that delayed diagnosis could conceivably bias the results by mislabeling "cases" as "non-cases" during part of the follow-up period. Our sensitivity analyses should have contributed to circumvent, at least in part, this limitation. Nonetheless, it is possible that the diagnosis of cancer may have contributed to decrease the focus on the need for dementia screening in later life (or vice versa), in which case the inverse association between cancer and dementia could have arisen as a result of bias. Again, the results of our sensitivity analyses suggest this is an unlikely explanation for our findings, particularly if one accepts that cognitive screening is more likely to be neglected during the acute phase following the diagnosis of cancer. Finally, we acknowledge that the observed effect size of the association between cancer and dementia was small and most likely of limited clinical significance to an older person with cancer. However, these results imply that a physiological mechanism mediating such an association exists and, once unveiled, could potentially lead to the development of novel and clinically meaningful approaches to reducing the risk of dementia in the population.
A recent neuropathological study has confirmed that cancer is associated with lower burden of Alzheimer’s disease pathology (Karanth et al., Reference Karanth2022), although the mechanisms mediating this association remain unclear (Ganguli, Reference Ganguli2015; Snyder et al., Reference Snyder2017). Future studies should seek to clarify if the inverse relationship between cancer and dementia is a consequence of current cancer treatments, or is due to environmental or genetic factors. For example, shorter telomeres have been associated with increased risk of dementia (Honig et al., Reference Honig, Kang, Schupf, Lee and Mayeux2012), whereas longer telomeres seem to increase the risk of cancer (Telomeres Mendelian Randomization Collaboration, 2017). The systematic evaluation of the relationship between cancer and dementia may generate valuable information to address the health needs of an aging population (Ahmad and Anderson, Reference Ahmad and Anderson2021).
Data availability
The data reported in this study are available to other researchers upon request to Prof. Leon Flicker (leon.flicker@uwa.edu.au).
Conflict of interest
None.
Description of authors’ roles
Obtaining funding for the study: all investigators.
Data collection: all investigators.
Supervision of research staff: LF.
Data Analysis: OPA.
Drafting of paper: OPA.
Critical review of paper for important intellectual content: all investigators.
Consent of submission for publication: all investigators.
Acknowledgements
The authors thank research staff, colleagues from the Western Australian Data Linkage Unit, and participants for their generous contribution. This research was funded by the National Health and Medical Research Council of Australia (NHMRC) grant numbers 1128083, 1045710, 0634492, and 0403963.
Open access
