Alzheimer’s and Parkinson’s Diseases and the Risk of Cancer: A Cohort Study

Setting and data sources

The Danish National Health Service provides tax-supported health care to all residents. Nationwide registries track vital status, diagnoses, and procedures for the entire population. Data can be linked accurately among the registries using the unique civil registration number assigned to all Danish residents at birth or upon immigration. Emigration and vital status are tracked by the Civil Registration System (CRS) [21]. The Danish National Patient Registry (DNPR) covering all Danish hospitals has recorded all clinical inpatient discharge diagnoses given to patients since 1977 and diagnoses made at outpatient clinics visits since 1995 [22]. The Psychiatric Central Research Registry (PCRR) has similarly recorded inpatient psychiatric diagnoses since 1969 and diagnoses conferred at psychiatric outpatient visits since 1995 [23]. In these registries, diagnoses were coded according to the Eighth Revision of the International Classification of Diseases (ICD-8) through 1993 and according to the Tenth Revision (ICD-10) since 1994 [24]. The Danish Cancer Registry (DCR) maintains information on all incident cancers diagnosed in Denmark since 1943, including information on morphology, histology, and stage at diagnosis [25]. Diagnostic codes used in this study are provided in Supplementary Table 1.

Design and study population

We used the DNPR and PCRR to identify a cohort of patients diagnosed with a first-time inpatient or out-patient hospital-based diagnosis of any type of dementia or PD without dementia in the period 1980–2012. The dementia diagnoses used in our study reflect the clinical diagnosis given to patients based on the ICD and thus recorded in the DNPR or the PCRR. Patients who had prevalent cancer recorded up to and including the date of dementia or PD diagnosis (n = 28,571) were excluded from the cohort. Subcohorts were established based on type of dementia diagnosis (AD, VaD, and all-cause dementia), and PD. The date of the first incident diagnosis of one of these conditions was defined as the index date. We obtained information from the DNPR on inpatient and outpatient diagnoses of comorbidities included in the Charlson Comorbidity Index (CCI) as of the index date [26].

Endpoints and follow-up

Patients with dementia or PD were followed from the index date for 10 years since patients with dementia rarely survive for an extended period of time [18], until a diagnosis of incident cancer recorded in the DCR, death, emigration from Denmark, or 31 December 2013 (end of follow-up period), whichever occurred first. Benign neoplasms were generally not defined as cancer, but benign neoplasms of the brain and meninges were included for the purpose of this study.

Analytic variables

Age on the index date was categorized as≤49, 50–59, 60–69, 70–79, and≥80 years. Comorbidity burden was classified using a version of the CCI [26], that excluded malignancy and dementia and categorized as total scores of 0, 1, 2-3,≥4. Comorbid conditions also were considered individually.

Cancer after dementia was examined as a composite endpoint (any cancer) and as specific types of cancer as shown in detail in Supplementary Table 1 [27]. This approach was based on previous studies indicating that the association between dementia and cancer may differ for smoking-related cancer and non-smoking-related cancer [1, 28]. Cancer stage at diagnosis of solid tumors was classified as localized, regional, distant, and unknown.

Statistical analysis

We first characterized the subcohorts according to descriptive characteristics. The absolute reduction in cancer risk attributable to dementia was computed as the expected minus the observed number of cancer cases divided by the person time to 10 years of follow-up per 10,000 person years. Finally, we computed standardized incidence rate ratios (SIRs) with 95% confidence intervals (CIs) by comparing observed cancer incidence among dementia or PD patients with that expected, based on cancer incidence in the entire general population, using national incidence rates by person-days of cancer by age, sex, and single year of diagnosis, and stratified by baseline characteristics. Associated 95% CIs were derived using Byar’s approximation, assuming that the observed number of cases in a specific category followed a Poisson distribution. We used exact 95% CIs when the observed number of cancers was less than ten. Analyses were then stratified for dementia survivors by 0–1 year of follow-up, >1–5 years of follow-up, and > 5-10 years of follow-up.

To examine whether there was evidence for diagnostic biases such that patients with dementia had a higher prevalence of distant metastases or unknown cancer stage, we stratified SIRs by cancer stage categories within selected solid tumor cancer sites that have been examined in previous research: colon cancer, pancreatic cancer, lung cancer, breast cancer, prostate cancer, liver cancer, melanoma, kidney cancer, and bladder cancer.

Sensitivity analysis

To account for differential diagnostic accuracy in the classification of dementia patients in each subcohort, all analyses of AD as the exposure were repeated with exclusion of the ICD-8 code for senile dementia in the AD definition. Diagnostic imaging has been recorded in the DNPR since 2004. We therefore repeated analyses for dementia patients with a record of brain imaging (CT, MR, or imaging with PET) within one month before or after the dementia diagnosis.

Analyses were conducted using SAS software version 9.4 (SAS Institute, Cary, NC, USA).

This study was approved by the Danish Data Protection Agency (record number 2007-58-0010). Danish law does not require an ethical approval or an informed consent from patients for studies based on routinely collected registry data [29].

Risk of cancer

During 10 years of follow-up, the observed to expected number of cancer cases were 4,204/6,178 for AD; 1,408/1,992 for VaD; 10,981/14,295 for all-cause dementia; and 3,004/3,057 for PD (Supplementary Tables 2 to 5). The corresponding absolute reduction in risk for cancer attributable to dementia were 79.9 per 10,000 person-years for AD [SIR = 0.68 (95% CI: 0.66, 0.70)], 74.4 [SIR = 0.71 (95% CI: 0.67, 0.74)] for VaD, 55.8 [SIR = 0.77 (95% CI: 0.75, 0.78)] for all-cause dementia, and 4.0 per 10,000 person-years [SIR = 0.98 (95% CI: 0.95, 1.02) for PD (Table 2).

At time of dementia or PD diagnosis, cancer risk during the entire follow-up period was increased or null in most age groups (<60 years of age for AD and PD and < 70 years of age for VaD and all-cause dementia) (Table 2). SIRs were reduced for most individual comorbidities in those patients classified as AD and VaD. When we categorized follow-up periods, we observed consistently reduced SIRs during all follow-up intervals for patients with AD and VaD. For all-cause dementia [SIR = 1.06 (95% CI: 1.03, 1.10)] and PD [SIR = 1.40 (95% CI: 1.30, 1.50)], a slightly increased association was observed in the first year of cancer diagnosis, but an inverse association was observed for the > 5–10 year follow-up period [SIR = 0.67 (95% CI: 0.64, 0.70) for all-cause dementia and SIR = 0.86 (95% CI: 0.80, 0.93) for PD] (Table 2). Risk of lung cancer was reduced and risk of malignant melanoma was increased in patients with PD.

The stage distributions for colon, pancreatic, lung, breast, prostate, liver, melanoma, kidney, and bladder cancer are presented in Table 2. Associations were nearly similar for solid tumors diagnosed at localized, regional, or distant stages, with few exceptions. For breast cancer and melanoma, SIRs were increased for missing/unknown cancer stage at diagnosis for all patients. Patients with PD had increased SIRs for breast cancer, melanoma, and kidney cancer.

Estimates of association remained largely unchanged in sensitivity analyses that (1) re-categorized ICD-8 codes for senile dementia and (2) that only included dementia cases assessed with neuroimaging (data not shown).

Strengths and limitations

Strengths of this population-based cohort study included its large size, which allowed for meaningful informative stratification by several patient characteristics. All Danish hospitals report data on diagnoses, including dementia, to the national registries DNPR and the PCRR [23, 24], and DCR data are nearly complete and valid due to compulsory reporting [25]. Most tumors are histologically confirmed after 2004 [25]. Free access to healthcare within the Danish health care system and tracking of patients through the CRS permitted nearly complete follow-up of all study participants [21, 24]. The positive predictive value of the dementia diagnosis in the DNPR overall has been shown to be 89% and is higher for AD than for other dementia types [30]. However, it has been demonstrated in the United States that some patients diagnosed with AD had mixed pathologies on autopsy [31]. To account for such potential inaccurate coding, we conducted sensitivity analyses that re-categorized ICD-codes for dementia subtypes, with largely similar results.

Several weaknesses of the study should be considered. Because dementia, PD, and cancer are associated with a long latency period prior to diagnosis, the temporal indeterminacy of disease onset is a concern, particularly for shorter-term associations. Our results did not vary substantially by calendar period despite the expected improvement in dementia recording with the transition to ICD-10 from ICD-8 in 1994. These potential biases may have resulted in a bias toward unity.

Previous literature

Our findings largely accord with previous studies, although our SIRs were somewhat higher. These studies showed an inverse association between AD and cancer risk overall, as well as for cancers at specific sites [1, 28]. Studies that distinguished AD from other dementias also observed a gradient toward a stronger protective effect associated with AD [1, 19]. For example, Roe et al. reported a hazard ratio of 0.31 (95% CI: 0.12, 0.86) for any cancer in patients with AD alone, and hazard ratios ranging from 0.41 to 0.89 for mixed dementia pathologies in patients aged 65 years or older from four communities in the United States [19]. In their nested case-control study, Driver et al. found a hazard ratio of 0.38 (95% CI: 0.25, 0.56) for possible AD and 0.44 (95% CI: 0.32, 0.61) for any dementia [1]. Some studies suggested a particularly reduced risk of smoking- and alcohol-related cancers in association with dementia [1], while others did not [28]. Attner et al.’s Swedish registry-based case-control study, conducted in a setting similar to ours, found an odds ratio of 0.66 (95% CI: 0.52, 0.69) for any dementia diagnosed before a cancer diagnosis. By cancer site, the odds ratios were 0.60 (95% CI: 0.40, 0.91) for colon cancer, 0.38 (95% CI: 0.09, 1.59) for pancreatic cancer, and 0.53 (95% CI: 0.31, 0.90) for lung cancer [2]. We reported nearly parallel SIR estimates: the 10-year SIR for colon cancer was 0.61 (95% CI: 0.55, 0.67); 0.52 (95% CI: 0.42, 0.63) for pancreatic cancer; and 0.54 (95% CI: 0.54, 0.61) for lung cancer, which is nearly similar to the results by Attner et al. Though there was no overall association with cancer after PD, our results broadly confirm those previously observed for some specific cancer sites [11, 17].

Mechanisms

A potential explanation for an association between AD and cancer is the possibility that cancer may be diagnosed less often in cognitively impaired patients with a progressive, essentially fatal neurologic disorder. Detection bias may be less strong in PD than AD as the age of onset of PD is earlier, life expectancy is longer, and early disability is often minor. However, several of our findings challenge this explanation for dementia. First, though we did not have any information on diagnostic procedures, we did report associations by cancer stages. Since cancer stage was nearly similar across cancer sites, this does not suggest that differential diagnostic or lead time bias explain our results. Second, because both dementia and cancer are associated with a high level of alcohol consumption and smoking, and since we found similar associations with both smoking-and non-smoking related cancer, this potential bias by lifestyle and socioeconomic status seems unlikely. PD patients had increased SIRs for breast cancer, melanoma, and kidney cancer, which are related to lifestyle factors, such as high socioeconomic status (breast cancer), sun exposure (melanoma), and obesity (kidney cancer) [5], which further supports a causal inverse association between AD and cancer. Because we found inverse associations for dementia and cancer with cancer stage and across lifestyle related cancer sites, this may suggest that lead time bias and a diagnostic bias is negligible. Neurodegenerative diseases and carcinogenesis may share several molecular signaling pathways and factors involved in cell cycle dysfunction such as Pin1 [32], tumor suppressor gene p53 [33], the WNT signaling pathway [34], and neurotransmitter signaling [35–37]. It was recently shown in a mouse model that human cyclophilin 40, which is upregulated in at least some cancer types [38, 39], may exert potential neuroprotective effects in the brain by breaking down tau protein aggregates [40].

Conclusions and implications

Evidence from this study adds to the existing body of research showing an inverse association between dementia and cancer risk. The association was robust for 10-year survivors of AD within cancer stages and within lifestyle related cancer sites, suggesting that underlying pathological processes of some central nervous system diseases may play a role in the development or detection of subsequent cancer. The risk of cancer varied by type of dementia, with a gradient toward a stronger protective effect associated with AD and VaD. However, we cannot fully exclude the possibility that cancer may be diagnosed less often in patients with dementia. We did not find a consistent inverse association between PD and cancer risk, suggesting that an inverse association is not characteristic of all neurodegenerative disorders.