Neurodegenerative diseases represent one of the main public health issues in Western societies. In 2019, about 57 million individuals were living with dementia worldwide.¹ Alzheimer's disease (AD) accounts for 60% to 70% of dementia cases (29 million) and represents the first cause of dementia, whereas dementia related to synucleinopathies (Parkinson's disease [PD] dementia and dementia with Lewy bodies [DLB]) are the second most common neurodegenerative causes of dementia.² PD is the second most common cause of neurodegenerative disease, estimated worldwide to affect 6.1 million individuals in 2016.³ All these neurodegenerative diseases are known to have a very long preclinical stage. For AD, accumulation of amyloid can precede a clinical diagnosis of dementia by up to 20 years.⁴ Synucleinopathies, including PD and DLB, also have a long prodromal stage, as demonstrated by isolated rapid eye movement (REM) sleep behavioral disorders, which can precede symptomatic PD or DLB by 10 years or more.⁵ The first dysautonomic signs have been estimated to start as early as 15 years before motor symptoms in PD.^{6, 7}

Neurological disorders are often underdiagnosed in the primary care setting.⁸ Analyses on general practitioners’ (GPs) databases have already led to the identification of disease-specific prodromal symptoms of PD and AD as well as risk factors in the primary care data.^{9, 10} However, it is unknown how the prodromal features and risk factors of these disorders differ. Analyses on the prodrome of DLB have so far been limited to core symptoms of DLB.^{11, 12} Overall, cross-disorder analyses focusing on shared and specific associations with neurodegenerative diseases are rare.¹³ We undertook cross-disorder analyses in order to identify prediagnostic associations and risk factors of the most common neurodegenerative conditions (AD, PD, and DLB), using data collected by GPs over a long-time window. We sought to identify disorders and drug prescriptions associated with AD, PD, and DLB, as well as those that are specific to each condition. Predictors of dementia in PD observable in primary care are also missing.¹⁴ In a secondary analysis, we aimed at understanding which drug prescriptions are associated with the presence of dementia in the 5 years following a first PD diagnosis. We relied on the Health Improvement Network UK primary care for discovery and the UK Biobank (UKB) for selected replication.

Methods

Results

In the THIN database, our full initial analytical cohort included 20,214 patients with a final diagnosis of AD, 28,222 patients with a diagnosis of PD, 4,682 patients with a diagnosis of DLB, and 20,214 healthy control patients. Table 1 describes the characteristics of the different populations. The median age at diagnosis was younger for PD (77 [70–83] years) than for DLB (80 [74–85] years) and patients with AD (82 [76–87] years) and for healthy controls (82 [76–87] years). There were more men than women in the PD and DLB cohorts (sex ratio of 57%) and the opposite in the AD and controls cohorts (sex ratio of 31%). There were 13,625 patients with AD, 14,539 patients with PD, 2,279 patients with DLB, and 10,205 healthy controls who had data available for more than 10 years before diagnosis. Demographic characteristics were similar between the different retrospective follow-up intervals (0–5, 10 or 15 years before diagnosis, Table S1). Only 5% of patients with AD had one prescription of antiparkinsonian drugs, whereas 76% of patients with PD had at least 2 prescriptions of antiparkinsonian drugs (Table S2).

All conditions were more frequent than in controls in all 3 neurodegenerative diseases during the period 0 to 5 years before diagnosis (Fig 1). The prevalence of tremor, sleep disorders, constipation, anxiety, and urinary tract disorders was higher in patients with PD and patients with DLB than in controls up to 15 years prior diagnosis, and anxiety and depression only for patients with DLB (Table S3). The pre-diagnosis profile of the 12 health conditions investigated in the THIN study in AD, PD, DLB, and controls over time are shown in Figure 2 (Table S4). Treatments against depression, anxiety, and constipation were all more frequently prescribed in individuals with a later diagnosis of AD, PD, and DLB (see Fig 1). A higher frequency of benzodiazepines and serotonin reuptake inhibitors prescription was observed in the years preceding the diagnosis of all 3 diseases (Table 2). Lower antidiabetic medications use, including insulins, biguanides, sulfonylureas, and DDP4 inhibitors, was observed before the diagnosis of AD and PD (see Table 2). A lower prescription of agents acting on the renin–angiotensin–aldosterone system and selective beta-2-adrenoreceptor agonists was significantly observed before the diagnosis of the 3 diseases when compared to the healthy controls (see Table 2).

Associations with the Risk of Developing Dementia after a Parkinson's Disease Diagnosis

To investigate whether the observed associations we describe above were specific to the disease or to the syndrome (dementia), we ran a subsequent analysis in the subpopulation of patients with PD, comparing patients with PD with a subsequent diagnosis of dementia and those who remained free of dementia 5 years after PD diagnosis. The median age at diagnosis of PD was younger for patients with PD who did not develop dementia in the 5 years after the first PD diagnosis (73 [65–79] years) than for patients with PD who developed a dementia (79 [73–83] years; see Table S7, p < 0.0001). There were more men than women in both groups (56% in the PD group without dementia and 59% in the PD group with dementia). When comparing the 2 groups, after correcting for sex and age at diagnosis, the prescriptions of drugs against hypercholesterolemia, constipation, depression, and psychosis were individually associated with the development of a dementia in the 5 years after PD diagnosis (Fig 3). They all remained significantly associated with dementia in the multivariable model (Table 3). By contrast to what was found in the whole population, drugs against hypertension and anxiety were not individually associated with dementia after a PD diagnosis.

TABLE 3. ORs and 95% CIs for a Logistic Model Assessing the Association of at Least One Prescription in the Considered Class of Treatment in the 5 Years Preceding a First Parkinson's Disease Diagnosis With a Diagnosis of Dementia in the 5 Years Following a First Parkinson's Disease Diagnosis

	Prevalence of at least one prescription in the 5 years before PD diagnosis		Univariable analysis		Multivariable analysis
	PD patients with dementia in the 5 years after PD diagnosis (n = 1,682)	PD patients without dementia in the 5 years after PD diagnosis (n = 6,924)	OR (95% CI)	p	OR (95% CI)	p
Drugs used against hypertension	1,074 (63.85%)	4,126 (59.59%)	1.01 (0.9, 1.14)	0.805 (5.633)	0.85 (0.75, 0.96)	0.01
Drugs used against hypercholesterolemia	579 (34.42%)	1814 (26.2%)	1.5 (1.33, 1.69)	<0.001 (<0.001)	1.49 (1.31, 1.69)	<0.001
Drugs used against diabetes	140 (8.32%)	535 (7.73%)	1.07 (0.87, 1.31)	0.517 (3.62)	0.9 (0.73, 1.11)	0.334
Drugs used against constipation	671 (39.89%)	1,685 (24.34%)	1.56 (1.39, 1.76)	<0.001 (<0.001)	1.42 (1.25, 1.6)	<0.001
Drugs used against depression	653 (38.82%)	2,212 (31.95%)	1.56 (1.38, 1.75)	<0.001 (<0.001)	1.44 (1.27, 1.63)	<0.001
Drugs used against anxiety	255 (15.16%)	1,112 (16.06%)	1.05 (0.9, 1.23)	0.504 (3.53)	0.85 (0.72, 1.0)	0.044
Drugs used against psychosis	430 (25.56%)	1,262 (18.23%)	1.37 (1.2, 1.57)	<0.001 (<0.001)	1.23 (1.08, 1.42)	0.003

• Note: In the multivariable model, we included all 7 health conditions. The treatment has to be prescribed in 0 to 5 years before Parkinson's disease diagnosis. Analyses are corrected for age at PD diagnosis and sex. The p values are corrected for multiple comparisons in the univariable analysis.
• Abbreviations: CI = confidence interval; OR = odds ratio; PD = Parkinson's disease.

Discussion

In this large cross-disorder analysis, we compared the prodromal symptoms and drug prescriptions before the diagnosis of three of the main neurodegenerative diseases (AD, PD, and DLB). Our results complement previous approaches which tested separately a range of prodromal features on PD,⁹ AD,²⁰ or DLB,¹¹ observable at the GP level. By contrasting AD with PD and DLB, we found cardinal motor or non-motor symptoms of each disease to be more frequent than in controls at least 5 years before diagnosis: memory problems for patients with AD, tremor for patients with PD, and hallucinations, confusion, movement disorders, and memory problems for patients with DLB. We also observed that GPs reported more falls in future patients with DLB than in patients with AD and patients with PD in keeping with the McKeith 2017 criteria for DLB.²¹ We also found a robust association among all 3 neurological disorders and constipation in accordance with previous studies,²² even though the association was stronger for PD and DLB than for AD. More generally, we showed that all neurological disorders present more autonomic dysfunction in primary care compared to healthy controls, including constipation, urinary disorders, and hypotension several years before diagnosis, with a stronger association for PD and DLB than for AD. These autonomic dysfunctions might be used by primary care physicians in the future to anticipate a neurodegenerative disease. This result may support the hypothesis that mechanisms of neurodegenerative diseases pathological affect the periphery before spreading into the brain.²³ These autonomic dysfunctions might alert the GP to a potential evolution toward a neurodegenerative disease.

We also investigated drug use and found a lower prescription of all the main antihypertensive families, including agents acting on the renin–angiotensin–aldosterone system (for all 3 diseases), calcium channel blockers (for AD and PD), and diuretics (for AD and DLB), in accordance with previous epidemiological studies showing an inverse association between antihypertensive drugs and AD, PD, or DLB.²⁴ These results may support the hypothesis that vascular diseases are a risk factor of neurodegenerative diseases,²⁴ a frequent comorbid condition of these diseases, and that decreasing cardiovascular risks by prescribing antihypertensive drugs would help to decrease the risk of developing all neurodegenerative diseases. Alternatively, some of these drugs may have unspecific neuroprotective effects,²⁵ as antidiabetics drugs that were less prescribed in the years preceding all 3 diseases.²⁶ The design of this study did not allow us to distinguish between these 2 possibilities. The difference observed on beta-blocker agents, which were more prescribed in individuals who later developed PD and DLB compared to AD or healthy controls, confirms previous similar results and may be explained by their use to relieve tremor.²⁷ The lower risk of developing PD with the use of Beta-2-adrenoreceptors and the higher risk with using beta-blocker agents may also be consistent with the current assumption that β2-adrenoreceptor is a regulator of the α-synuclein gene (SNCA).²⁸ Benzodiazepine's usage have been repeatedly linked to a higher risk of dementia,²⁹ and benzodiazepines were indeed more prescribed in the years preceding with all 3 diseases. Of note, all 3 diseases were also associated with increased rates of anxiety and sleep disorders. As above, the design of the study did not allow us to distinguish between a direct effect of benzodiazepines on the risk of dementia or the consequence of the need to treat prodromal anxiety. Our observed association of benzodiazepines with PD and DLB suggests that this association is not specific to AD.

More surprisingly, we found a lower prescription of statins in the years before the diagnosis of PD compared to dementia-related disorders (AD and DLB). These differences remained significant when adjusting for potential confounding variables, such as APOE status, body mass index, education level, and Townsend deprivation status, and in the replication (UKB) dataset. This observation could have several explanations. First, because statins are prescribed in secondary prevention of cardiovascular diseases, it may be the result of higher vascular comorbidities in the AD and DLB groups.³⁰ However, lipid lowering therapy intake was similar between AD and healthy controls. Moreover, we observed that antihypertensives were similarly prescribed in the 3 diseases, which suggests that the 3 populations may have similar vascular risk factors. Second, AD participants were differentiated from vascular dementia in the UKB. Another explanation would be that subjects treated with lipid lowering therapies are at lower risk to develop PD. Indeed, higher cholesterol levels have been associated with a lower risk of PD,^{31, 32} including a recent mendelian randomization analysis suggesting that this association could be causal.³³ Finally, lipid lowering agents may have a protective effect on PD, as supported by a study showing that the Mediterranean diet adherence is associated with a reduced risk of prodromal PD,³⁴ even though similar studies also exist in AD.³⁵ A recent randomized controlled trial of simvastatin for delay of disease progression in PD was negative,³⁶ indicating that the association might not be causal. Another assumption would be that PD induces a global hypermetabolism and thus fat burning, which could indirectly lower cholesterol levels and thus a lower use of statins. The mechanisms by which lipid lowering therapies may be protective against PD remains to be understood.

Interestingly, the increased usage of statins prior the diagnosis of dementia in the whole population (i.e., AD or DLB) was also found when restricting the analysis in the PD population. Indeed, we found that patients with PD who developed dementia within 5 years used more anti-cholesterol drugs at the pre-PD stage than patients not developing dementia. It may argue toward a specific role of statins in the development of dementia.

Finally, we also identify other prescriptions associated with future development of dementia in the PD population (drugs against constipation and depression). Using past prescriptions of the GP may be useful in the future to design approaches to distinguish patients with PD who will develop dementia shortly after their first initial PD diagnosis.

This study is, to our knowledge, the largest addressing on the same database as the prodromal phase of the 3 neurodegenerative diseases allowing to compare their clinical profile and treatment prescriptions as observed by GPs. The observations of increased non-motor, non-cognitive prodrome in patients with DLB might help GPs who encounter these symptoms in their clinics. Indeed, many are facing challenges of knowing when to refer for subspecialist evaluations, especially in underserved areas. Another notable strength of this study is the replication on the UKB database, with consistent results obtained for the 2 analyses.

Our work has also several limitations. First, in the THIN database, the diagnosis of neurodegenerative diseases was made by GPs, which may result in a delayed diagnosis, explaining why we can observe disease-specific symptoms before their diagnosis. However, most patients have likely been seen by a specialist because previous studies showed that it was the case for 78% of patients with a PD diagnosis.³⁷ Initial diagnosis of probable PD was later confirmed in 83% cases.³⁷ In addition, we used ICD codes, which cannot capture disease severity, which might under-detect soft clinical signs. However, the size of the database offers the opportunity to start progressing our understanding of prodromal risk factors of DLB in France and the United Kingdom. Conversely, in the UKB database, diseases’ records came from hospitalization or death registers, which may be bias toward more severe disease forms. Second, the duration of the study is long. It represents one of the major strengths of this work, because it enables study of possible associations several years before the first diagnosis. However, over that time, diagnostic and prescribing behaviors of generalists have evolved, making the overall cohort heterogeneous in their ascertainment. Third, the frontiers of each complex neurodegenerative disease may not be defined perfectly, and their distinction is a simplification with respect to disorders heterogeneity³⁸ and distinction between developing a DLB or co-existence of PD and AD might be difficult to do in clinical practice. However, the differences observed in the prodromal stage of patients with DLB, show that DLB is not simply the addition of the symptoms of AD and PD at the GP level. Finally, in the THIN analysis, controls were matched with the AD population, which was older and had a higher proportion of female patients than the PD and DLB cohorts, and we may therefore have underestimated associations particularly in the younger PD population. Patients with AD in the UKB might also not be representative of the general patient population as they are still relatively young.

Conclusion

Our results highlight the long-term clinical prodromal phase of neurodegenerative diseases, and therefore could enable more efficient primary and secondary prevention measures. This work also represents one of the first steps to understand early prodromal symptoms of DLB, helping GPs to better detect, and refer to specialists, from the first symptoms of the disease. To the best of our knowledge, this is the first study showing that there may be a differential role of lipid lowering therapies in the prodromal phase of neurodegenerative diseases. Further studies are needed to better understand the relationship among cholesterol levels, the risk of neurodegenerative diseases, and the risk of developing dementia after a PD diagnosis.

Author Contributions

T.N., B.C.-D., A.S., J.-C.C., and S.D. contributed to the conception and design of the study. T.N., B.C.-D., F.M., L.G., A.D.-B., A.S., J.-C.C., and S.D. contributed to the acquisition and analysis of data. T.N., B.C.-D., A.D.-B., F.M., R.C., L.G., B.L., N.V., A.S., S.D., and J.-C.C. contributed to drafting the text or preparing the figures.

Potential Conflicts of Interest

F.M., L.G., and B.E. are full time employees of Cegedim.