IntroductionEpidemiological studies have identified an inverse association between cancer and dementia. Underlying methodological biases such as confounding, diagnostic bias, and selective survival have been postulated, yet no studies have systematically investigated the potential for each source of bias within a single dataset. We used the UK Biobank to generate and compare estimates for the cancer-dementia association using different analytical specifications designed to address different sources of bias.MethodsWe included 140,959 UK Biobank participants aged ≥ 55 without dementia before enrollment, and with linked primary care data. We used cancer registry data to identify cases of prevalent cancer before UK Biobank enrollment and incident cancer after enrollment. We used multivariable-adjusted Cox models to evaluate the associations of prevalent and incident cancer with incidence of all-cause dementia, Alzheimer’s disease (AD), and vascular dementia. And we systematically evaluated each potential source of bias.ResultsThe cohort accumulated 3,310 incident dementia diagnoses over a median of 12.3 years of follow-up. All-site incident cancer was positively associated with all-cause dementia risk (hazard ratio [HR]=1.14, 95% CI: 1.02-1.29). The adjusted HR for prevalent cancer was 1.03 (95% CI: 0.92-1.17). Results were similar for vascular dementia. AD dementia was not associated with prevalent or incident cancer. Dementia diagnosis was substantially elevated in the first year after cancer diagnosis (HR=1.83, 95% CI: 1.48-2.36), suggesting diagnostic bias.ConclusionIncident cancer diagnoses were associated with a higher risk of subsequent dementia diagnoses. Increased chance of dementia diagnosis associated with increased health care utilization after a cancer diagnosis may be a source of bias in electronic health records-based studies.

Background and Objectives: Cancer survivors are less likely than comparably-aged individuals without a cancer history to develop Alzheimer's disease and related dementias (ADRD). We investigated the association between cancer history and structural magnetic resonance imaging (MRI) markers for ADRD risk, using linear mixed-effects models to assess differences at the mean values of MRI markers and quantile regression to examine whether the association varies across the distribution of MRI markers of brain aging. Methods: Among UK Biobank participants with ≥1 brain MRI, we considered total gray matter volume, total brain volume, hippocampal volume, white matter hyperintensity volume, and mean cortical thickness in the Alzheimer's disease (AD) signature region. Cancer history was ascertained from national registry and self-report. We first specified linear mixed models with random intercepts to assess mean differences in MRI markers according to cancer history. Next, to examine whether effects of cancer history on these markers varies across the ADRD risk distribution, we specified quantile regression models to assess differences in quantile cut-points of the distribution of MRI markers according to cancer history. Models adjusted for demographics, APOE-ε4 status, and health behaviors. Results: The sample included 42,242 MRIs on 37,588 participants with no cancer history (mean age 64.1 years), and 6,073 MRIs on 5,514 participants with a cancer diagnosis prior to MRI (mean age 66.7 years). Cancer history was associated with smaller mean hippocampal volume (b=-19 mm3, 95% confidence interval [CI]=-36, -1) and lower mean cortical thickness in the AD signature region (b=-0.004 mm, 95% CI=-0.007, -0.000). Quantile regressions indicated cancer history had larger effects on high quantiles of white matter hyperintensities (10th percentile b=-49 mm3, 95% CI=-112, 19; 90th percentile b=552 mm3, 95% CI= 250, 1002) and low quantiles of cortical thickness (10th percentile b=-0.006 mm, 95% CI=-0.011, -0.000; 90th percentile b=0.003 mm3, 95% CI=-0.003, 0.007), indicating individuals most vulnerable to ADRD were more affected by cancer history. Discussion: We found no evidence that cancer history was associated with less ADRD-related neurodegeneration. To the contrary, adults with cancer history had worse MRI indicators of dementia risk. Adverse associations were largest in the highest-risk quantiles of neuroimaging markers.

BackgroundGenetic risk variants for late‐onset Alzheimer’s disease (LOAD) predict brain atrophy, cognitive changes, and dementia onset in late‐life. Little is known about the association between LOAD genetic variants and regional brain volumes in middle aged adults. Further understanding of when associations emerge across mid‐to‐late‐life may point to the earliest manifestations of LOAD.MethodWe studied 41,435 UK Biobank participants aged 45‐80 who enrolled 2007‐2010, had no dementia at baseline, and subsequently underwent a brain MRI (3‐Tesla scanner). A genetic risk score for LOAD (LOAD‐GRS) was calculated as a weighted sum of 27 variants previously confirmed to be genome‐wide significant predictors of LOAD dementia including APOE. MR images underwent standardized quality control and data processing by UK Biobank. Volumes (both hemispheres combined) for 38 regions of interest (ROIs) were derived from Freesurfer (Version 6.0) packages for subcortical segmentation and cortical parcellation with the Desikan‐Killiany‐Tourville atlas. We examined associations of LOAD‐GRS with each ROI tested whether effects differed across age using linear regression with an interaction between age and LOAD‐GRS. A cross‐validated model determined the earliest age at which trends in MRI volume diverged between those with low compared to high LOAD‐GRS. All models adjusted for sex, genetic ancestry, and intracranial volume.ResultUsing a Bonferonni correction (p<0.05/38=0.0013]), 4 of 38 regional volumes had significant interactions between age and LOAD‐GRS, all indicating that for people with 1 SD higher genetic risk of LOAD, older age was more strongly associated with smaller brain volumes than among people with lower LOAD genetic risk. Significant regions included the amygdala (‐172.6 mm3/SD; 95%CI=‐176.7,‐168.5); hippocampus (‐351.9 mm3/SD; 95%CI=[‐359.9,‐343.9], nucleus accumbens (‐87.7 mm3/SD; 95%CI=‐89.4,‐86.0], and thalamus (‐526.4 mm3/SD; 95%CI=[‐538.2,‐514.6]. Best fitting model‐based age‐curves for people with high versus low LOAD‐GRS scores began to diverge around age 45 for the nucleus accumbens, hippocampus, and thalamus and after age 51 for the amygdala.ConclusionGenetic factors that increase risk of LOAD begin to predict lower brain volumes as early as middle age, decades prior to average age of dementia onset. Our findings prioritize particular brain regions that may help predict LOAD onset, including the hippocampus, amygdala, nucleus accumbens, and thalamus.