[1] 9w Evidence for modest familial co-aggregation between dementia and parkinsonism

[1] 250w To investigate the contribution of shared familial risk to the co-occurrence of dementia and parkinsonism by studying familial co-aggregation of Alzheimer's disease (AD) and Parkinson's disease (PD). Using Swedish population-based registers we constructed two cohorts; a first-degree relative cohort of persons born 1932-1960 (n = 2,775,332) and a spouse cohort of persons born 1890-1960 (n = 4,736,006). Study persons were followed up between 1969 and 2009 in the National Patient and Cause of Death Registers. We modeled the association between incidence of disease and having at least one affected relative using Cox proportional hazard regression that estimated hazard ratios (HRs) with 95 % confidence intervals (CIs) adjusted for age, sex and number of relatives. Within each disorder; dementia, AD, parkinsonian disorders and PD, there was a strong association between risk of disease and having at least one affected sibling or parent. There was also a modest shared familial risk between the diseases; risk of parkinsonian disorders was associated with having a sibling with AD (HR 1.35, 95 % CI 1.11-1.65) and risk of dementia was associated with having a sibling with PD (HR 1.20, 95 % CI 1.02-1.41). There were no meaningful familial risks among spouses. The risk of co-occurring dementia in PD was considerably increased (HR 2.83, 95 % CI 2.76-2.89). There is strong familial aggregation within dementia, AD, parkinsonian disorders and PD, and modest familial co-aggregation between dementia and parkinsonism. Thus, co-occurrence of dementia and parkinsonism is not primarily caused by shared familial risk between AD and PD.

[1] 64w Co-occurrence of parkinsonism and dementia is a hallmark of conditions such as Parkinson's disease (PD) dementia, dementia with Lewy bodies [1] and progressive supranuclear palsy [2]. The most common type of dementia, Alzheimer's disease (AD), shares neuropathological characteristics, such as protein aggregation and neurodegeneration, with PD. However, the underlying causes of these disorders and why dementia and parkinsonism co-occur are complex and largely unknown.

[2] 134w There is evidence for familial aggregation within both AD and PD; the relative risk of PD for persons who have a first-degree relative affected by PD has been estimated at 2.9 [3] and the corresponding relative risk for AD has been estimated at 3.5 [4]. Several susceptibility genes and environmental factors have been implicated in the pathologies of AD and PD [5][6][7][8]. In light of the familial predisposition for AD and PD, it has been hypothesized that the co-occurrence of dementia and parkinsonism may be explained by shared familial risk, that is, the same genetic or familial environmental risk factors may contribute to the etiology of both conditions. Several epidemiological studies have investigated the hypothesis of shared familial risk by studying familial co-aggregation of AD and PD, but results have been largely inconclusive [9][10][11][12][13][14][15][16].

[3] 59w The aim of the present study was to investigate familial co-aggregation between dementia and parkinsonism by comparing incidence of disease in persons with and without affected relatives among siblings, parents and offspring and spouses. In addition, we studied familial aggregation within these disorders. We conducted a cohort study of over 5 million Swedish residents using Swedish nationwide population-based data.

[1] 103w The composition of the first-degree relative cohort and the spouse cohort is summarized in Table 1. In the two cohorts combined we identified 265,620 dementia cases, whereof 92,280 (34.7 %) were AD cases, as well as 64,206 parkinsonian disorder cases, whereof 39,504 (61.5 %) were PD cases. The parents of the members of the first-degree relative cohort consisted of 2,749,772 persons (they were not themselves study persons, but were used to ascertain exposure status in parent-offspring analyses). Among the parents, 266,583 (10.4 %) had dementia, including 92,659 who had AD and 55,160 (2.2 %) had a parkinsonian disorder including 31,967 who had PD.

[2] 171w Due to left truncation in 1932 of the first-degree relative cohort, the participants could only be followed to a maximum age of 78 years. As a consequence of the different age structures of the cohorts, the proportion followed until death was 8.6 % of the first-degree relative cohort and 41.0 % of the spouse cohort (73.3 % of the parents of the first-degree relative cohort). The proportion of exposed study persons also differed greatly between relative types, e.g., it was more common to have a parent affected by dementia (18.9 %) than an affected sibling (1.4 %) or spouse (5.8 %). The mean age at detection of diseases was also much lower in the first-degree relative cohort (Table 1). Mean time at risk was 21.4 years (SD 7.4) in the first-degree relative cohort and 24.7 years (SD 10.3) in the spouse cohort. Most cases were identified in the NPR (82.7 % of dementia cases, 78.9 % of AD cases, 93.8 % of parkinsonian disorder cases and 90.0 % of PD cases).

[3] 66w In both cohorts combined, there were 15,048 individuals who had co-occurring dementia and parkinsonian disorders, corresponding to 23.4 % of parkinsonian disorder cases or 5.7 % of dementia cases. 1,750 persons had an AD and a PD diagnosis. There was a strong association between PD status and risk of co-occurring dementia (HR 2.83, 95 % CI 2.76-2.89) or co-occurring AD (HR 1.64, 95 % CI 1.56-1.72).

[4] 67w There was substantial familial aggregation for all disorders, in particular among siblings (Table 2). The familial aggregation was stronger in AD compared to PD, and the difference between familial aggregation of AD and dementia was much larger than the difference in familial aggregation between PD and any parkinsonian disorder. With the exception of AD, there was no association between risk of disease and having an affected spouse.

[5] 93w In the analyses of familial co-aggregation, there was a modest association between risk of dementia and having a sibling affected by any parkinsonian disorder or PD, but there was no association with having a parent affected by a parkinsonian disorder or PD (Table 3). There was also a non-significant association between risk of AD and having a sibling affected by PD (HR 1.22, 95 % CI 0.95-1.57, data not shown). There was a slight inverse association between risk of dementia and having a spouse affected by any parkinsonian disorder or PD (Table 3).

[6] 81w There was a modest association between risk of any parkinsonian disorder and having a sibling or a parent affected by dementia or AD, although the association was stronger in siblings (Table 4). There was also a non-significant association between risk of PD and having a sibling affected by AD (HR 1.15, 95 % CI 0.90-1.47, data not shown). There was a slight inverse association between risk of any parkinsonian disorder and having a spouse affected by dementia or AD (Table 4).

[1] 79w In this population-based study of more than 5 million persons who were followed for up to 41 years, we found evidence of a strong familial aggregation within dementia, AD, parkinsonian disorders and PD and, despite a substantial co-occurrence within individuals, only modest shared familial co-aggregation between dementia and parkinsonism. Thus, the underlying familial causes of dementia are different from those of PD and the cooccurrence of dementia and parkinsonism is primarily caused by factors other than shared familial risk.

[2] 400w These findings have implications for clinicians and persons with first-degree relatives or spouses affected by these disorders. For instance, persons with family members affected by PD, who are aware of the neurodegenerative nature of the disease, may worry about not only their own risk of PD, but also their risk of dementia. As seen in Table 1, 2.7 % of the first-degree relative cohort had a sibling or parent with PD, which is a non-negligible proportion of the population. While we show that the risk of dementia is higher than in families not affected by PD, the increased risk is only modest. Previous studies of familial co-aggregation between AD and PD have been primarily of The number of persons linked to at least one parent is less than the number of persons in the complete cohort because there were n = 13,481 persons who could not be included in parent-offspring comparisons due to their parents being identified but not alive or residing in Sweden during the follow-up period. Thus, in these persons exposure status in parent-offspring analyses could not be determined but the parents could still be used to link to siblings b Refers to the exposure frequency among study persons, i.e., number of study persons with at least one affected first-degree relative, full sibling, parent or spouse case-control [11,16] or reconstructed cohort [9,[12][13][14][15] design and results have been conflicting. The most common type of ascertainment of disease in previous similar studies has been through family history interviews that have been shown to be prone to bias [28]. The present population-based study ascertained disease status in hospital and death register records, reducing the risk of family information bias, but there may still be surveillance bias as persons with family members affected by a disease may be more aware of symptoms and thus more likely to seek care for that disease themselves. However, this possible bias is reduced by our stringent definition of AD and PD: cases were either detected in the NPR where they had to be admitted to hospital with a primary diagnosis listed on the hospital record or in the CDR where the diagnosis had to be listed as the underlying cause of death. There has been one previous register-based cohort study from Sweden which found strong sibling risks for AD and PD in a cohort that was followed up until a maximum age of 69 years [10].

[3] 59w Our finding of strong familial aggregation within AD and PD is in line with previous reports [3,4], and is consistent with finding from studies of heritability [29,30]. There was a greater difference in familial aggregation between AD and any dementia than between any parkinsonian disorders and PD, indicating that dementia may be a more heterogeneous syndrome than parkinsonian disorders.

[4] 116w The spousal risk models allowed us to test the potential contribution of non-genetic familial factors to the etiology of AD and PD. Overall, these results provide further evidence that non-genetic familial factors are of little importance in the etiology of AD and PD. We found no evidence of shared familial risk among spouses with the exception of a small but significant inverse association between dementia and parkinsonian disorders and a slight increased risk of AD in persons with spouses affected by AD. Higher education is a protective factor for AD, but may be a risk factor for PD [31,32]. Thus, a correlation in education among spouses may be a possible explanation for these observed spousal risks.

[5] 85w Consistent with our findings, genome wide association studies find no overlap between risk alleles for AD and PD [33]. Further, there is evidence that other mechanisms, downstream from gene expression, associated with the abundance of alpha-synuclein and amyloid-beta contribute to the pathology in co-occurring dementia and parkinsonism. Alpha-synuclein, the primary component of Lewy bodies in PD, is found as a precursor to the non-amyloid component of Alzheimer's amyloid plaques [34,35]. Interaction between alpha-synuclein and amyloid-beta may promote pathological aggregation of alpha-synuclein into Lewy bodies [36].

[6] 182w Strengths of the present study include the populationbased cohort design, which minimizes selection and survival bias, large size and disease ascertainment method, which ensured that cases were diagnosed independently from the disease status of their relatives. However, the study also had several limitations. Most of our cases were identified through hospital records (NPR) and these diagnoses are of high specificity but up to a quarter of parkinsonian disorder cases and up to half of dementia cases may be missed and the first date of ascertainment may be a poor proxy for date of onset of disease [24]. For the conservative definition of register-based diagnoses of AD and PD that were used, sensitivity is even lower, which may explain why the proportion of AD cases among all dementia cases was lower in the present study than what has been found previously [37]. However, misclassification due to reduced sensitivity was likely non-differential with regards to disease status in relatives. Although registerbased dementia diagnoses in PD patients have not been validated, our estimates of risk of dementia in PD were comparable to previous reports [38,39].

[7] 154w Another limitation was the truncated follow-up in the first-degree relative cohort, resulting in missed late-onset incident cases, especially of dementia. In the sibling comparisons both exposure and outcome were affected by the truncation. However, in the parent-offspring comparisons the exposure was ascertained among the parents who were not affected by the truncation. This may in part explain why the sibling risks were higher than the parent-offspring risks. The risk of parkinsonian disorders was increased among persons with parents affected by dementia, but there was no effect of having a parent with a parkinsonian disorder on the risk of dementia. As the onset of dementia is typically later than the onset of parkinsonian disorders, it is possible that if we had been able to follow the first-degree relative cohort in the 9th and 10th decade of life, there would have been the same modest familial co-aggregation observed for dementia as there was for parkinsonian disorders.

[8] 43w In conclusion, this study provides evidence of strong familial aggregation within dementia, AD, parkinsonian disorders and PD and modest familial co-aggregation between dementia and parkinsonism. Shared familial risk for these conditions plays only a minor role in the cooccurrence of parkinsonism and dementia.

[1] 20w We linked several Swedish individual level populationbased registers using the national personal identification number assigned to all Swedish residents [17].

[2] 37w The Multi-Generation Register (MGR) includes index persons born since 1932 who resided in Sweden in 1961 or later, and links the index persons to their parents, if their parents resided in Sweden in 1947 or later [18,19].

[3] 66w The National Patient Register (NPR) includes hospital inpatient records since 1964 and outpatient records from hospital-based clinics since 2001 (the register does not cover primary care) [20]. Each record lists primary and contributory diagnoses coded according to the Swedish versions of the International Classification of Diseases (ICDs). Sweden has a decentralized unified health care system with universal coverage that ensures equal access to inpatient hospital care.

[4] 34w The Cause of Death Register (CDR) includes death records for all deaths of Swedish residents since 1961 [21]. Each death record lists the underlying and contributory causes of death coded according to the ICD.

[5] 40w In the present study, NPR and CDR records were available from January 1st 1969 to December 31st 2009. Dates of birth and death were retrieved from the Swedish Total Population Register and migration data was retrieved from the Migration Register.

[6] 43w Using the MGR we constructed two study cohorts; one consisting of individuals with at least one identified firstdegree relative (a sibling and/or a parent) and one consisting of individuals with at least one identified spouse. There was substantial overlap between the two cohorts.

[7] 102w We had information from the NPR and CDR for the period between January 1st 1969 and December 31st 2009 and thus this period was chosen as the follow-up during which the study persons were at risk. Study persons were eligible to be included in the cohorts if they were alive and residing in Sweden during the follow-up period, and additionally if they had not died or emigrated before their 40th birthday. In order to be able to assess exposure status, the study person had to have at least one relative who was also alive and residing in Sweden during the follow-up period.

[8] 73w In the first-degree relative cohort we included MGR index persons born between 1932 and 1960 (n = 2,775,332 persons). In the cohort, 93.5 % were linked to both parents, 5.4 % were linked to only their mother and 1.1 % were linked to only their father. Via parents we could link full siblings and in this way we identified 745,143 sibling clusters with on average 3.1 [standard deviation (SD) 1.4] siblings per family.

[9] 54w Through the MGR we identified a cohort of spouses born between 1890 and 1960 (n = 4,736,006 persons). Spouses had to share at least one child born between 1932 and 2012. Most individuals were linked to one spouse but 6.8 % were linked to two spouses and 0.8 % to more than two spouses.

[10] 12w In total, 5,344,042 unique individuals were included in the two cohorts combined.

[11] 12w The study was approved by the Regional Ethical Vetting Board in Stockholm.

[12] 170w We identified cases of any dementia (including AD, vascular dementia, fronto-temporal lobar degeneration and other dementia) or any parkinsonian disorder (including PD, cerebrovascular parkinsonism and other parkinsonism) through the NPR and the CDR (see supplementary table). To reduce misclassification between AD and other dementia or PD and other parkinsonian disorders when identifying AD or PD cases, we restricted the case definition to primary diagnoses in the NPR or underlying causes of death in the CDR. Previous validation studies against gold standards based on established clinical criteria [22,23] show that the positive predictive value (PPV) in NPR was 83.0 % for register-based PD cases [24], 88.0 % for register-based cases with any parkinsonian disorder [24], 72.9 % for register-based AD cases [25] and 81.8 % for registerbased dementia cases [25]. Although diagnostic codes for Parkinson's disease dementia, dementia with Lewy bodies or progressive supranuclear palsy exist, we were not able to distinguish between different types of conditions that include co-occurring dementia and parkinsonism, as their register-based definitions have not been validated.

[13] 53w First date of hospital admission or outpatient visit with a relevant diagnosis was used as date of ascertainment for cases identified in the NPR whereas date of death was used as date of ascertainment for cases identified in the CDR. An individual could be classified as both a parkinsonian disorder and dementia case.

[14] 129w We considered four different incident outcomes in the study persons; any dementia, AD, any parkinsonian disorder and PD. When studying a specific outcome, a person was still at risk if any of the other outcomes occurred. Exposure was having a first-degree relative (sibling and/or parent) or spouse affected by dementia, AD, any parkinsonian disorder or PD (ever or never status), depending on the model. In analyses of familial aggregation within diseases the outcome and exposure were the same condition. In analyses of familial co-aggregation between diseases, when the outcome was dementia or AD, the exposure was having at least one relative affected by any parkinsonian disorder or PD (and vice versa). We were primarily interested in familial co-aggregation of AD with any parkinsonian disorder and PD with any dementia.

[15] 73w In analyses where the exposure was AD status in relatives we excluded study persons with relatives who had dementia other than AD from the reference (unexposed) group to ensure that the unexposed group was comparable to that in analyses where the exposure was any dementia status in relatives. Similarly, when exposure was PD status in relatives we excluded study persons with relatives who had parkinsonian disorders other than PD from the unexposed group.

[16] 26w We studied co-occurrence of dementia and parkinsonism by modeling incidence of any dementia or AD in persons who were ever versus never diagnosed with PD themselves.

[17] 148w The study persons were followed from January 1st 1969, date of immigration or their 40th birthday until December 31st 2009, date of emigration, death or first ascertainment of incident outcome disease, whichever came first. Incidence rates per 100,000 person-years were calculated as number of events divided by total person-time at risk. To quantify the association between disease status in relatives and outcome, we estimated hazard ratios (HRs) with 95 % confidence intervals (CIs) using Cox proportional hazard regression with age as the underlying time-scale. The models were also adjusted for sex and number of informative relatives as a continuous measure (i.e., number of first-degree relatives, full siblings, parents or spouses). To account for dependencies in the data, we adjusted the standard errors using a robust sandwich estimator clustered on family. The proportional hazards assumption was examined using Shoenfeld residuals [26]. Analyses were performed using Stata version 12 [27].