Adult‐onset idiopathic dystonia: A national data‐linkage study to determine epidemiological, social deprivation, and mortality characteristics

Abstract Background and purpose Accurate epidemiological information is essential for the improved understanding of dystonia syndromes, as well as better provisioning of clinical services and providing context for diagnostic decision‐making. Here, we determine epidemiological, social deprivation, and mortality characteristics of adult‐onset idiopathic dystonia in the Welsh population. Methods A retrospective population‐based cohort study using anonymized electronic health care data in Wales was conducted to identify individuals with dystonia between 1 January 1994 and 31 December 2017. We developed a case‐ascertainment algorithm to determine dystonia incidence and prevalence, as well as characterization of the dystonia cohort, based on social deprivation and mortality. Results The case‐ascertainment algorithm (79% sensitivity) identified 54,966 cases; of these cases, 41,660 had adult‐onset idiopathic dystonia (≥20 years). Amongst the adult‐onset form, the median age at diagnosis was 41 years, with males significantly older at time of diagnosis compared to females. Prevalence rates ranged from 0.02% in 1994 to 1.2% in 2017. The average annual incidence was 87.7/100,000/year, increasing from 49.9/100,000/year (1994) to 96.21/100,000/year (2017). In 2017, people with dystonia had a similar life expectancy to the Welsh population. Conclusions We have developed a case‐ascertainment algorithm, supported by the introduction of a neurologist‐reviewed validation cohort, providing a platform for future population‐based dystonia studies. We have established robust population‐level prevalence and incidence values for adult‐onset idiopathic forms of dystonia, with this reflecting increasing clinical recognition and identification of causal genes. Underlying causes of death mirrored those of the general population, including circulatory disorders, respiratory disorders, cancers, and dementia.


INTRODUC TI ON
Dystonia is a hyperkinetic movement disorder characterized by sustained muscle contractions, producing repetitive movements and abnormal postures, impacting both physical and social functioning [1,2]. Accurate epidemiological information is essential for the improved understanding of dystonia syndromes, as well as better provisioning of clinical services and providing context for diagnostic decision-making. However, the true prevalence of dystonia remains largely unknown, with notable variation in reported rates worldwide, depending on disorder subtype, cohort ethnicity, and study design [3].
A recent meta-analysis estimated the global prevalence of dystonia to be 16.43 per 100,000 [4], although work to date suggests that this may vary geographically. Northern European countries, the USA, and Colombia have higher rates of prevalence than those observed in Asia and Southern Europe [5][6][7][8][9][10][11]. Similar variation between ethnicities has also been observed in rates of incidence.
Higher rates of adult-onset idiopathic, isolated focal cervical dystonia (AOIFCD) have been identified amongst Caucasian populations [12,13]. Higher rates of early-onset (≤28 years old) idiopathic torsion dystonia (5 per 100,000) were identified amongst Ashkenazi Jewish populations, in part due to the founder effects of genetic forms of dystonia [14]. Prevalence estimates are also dependent on age. The prevalence of early-onset dystonia (<20 years) is estimated to be between 0.3 and 5 cases per 100,000 [15,16], whereas the prevalence of adult-onset (>20 years) dystonia is between 0.3 and 732 cases per 100,000 [3,15,17]. Table 1 summarizes dystonia prevalence studies.
Different study types have been used to determine the epidemiological characteristics of dystonia. These have included servicebased, linkage record-based, and population-based studies. Studies also vary in their reliance upon patient-reported diagnoses and clinical confirmation (Table 1), possibly explaining the variation in observed prevalence rates. Current estimates of dystonia prevalence are also likely to be underestimates, due to a combination of lack of recognition, underdiagnosis, and the limited number of patients seeking treatment [4,17].
Recent advances in record-linked population data repositories have led to increased opportunities for population-based studies, providing more detailed epidemiological data across a range of neurological disorders [18][19][20]. There is a specific need to apply this population-based approach to dystonia syndromes, particularly in relation to adult-onset focal syndromes [21]. Use of data from several sources may address conflicting epidemiological estimates, and also has the potential to generate detailed information over several years and across diverse geographic areas, which is currently lacking in the field. The Secure Anonymised Information Linkage (SAIL) Databank contains anonymized, routinely collected data for the Welsh population (3.15 million [2019]) [22]. It is one of the most accessible and broadest sources of anonymized data in the world, covering an extensive percentage of the Welsh population (100% of Welsh hospitals and ~80% of general practitioners). SAIL also provides complete data linkage of social, education, and health care data, enabling a more comprehensive understanding of patient care, and allowing for analysis of multifactorial relationships. In comparison, another UK primary dataset (Clinical Practice Research Datalink) has linkage to only around 10% of English general practices (GPs) [23]. Other linked population data include those in Scandinavia and Australia.
However, those in Australia are managed at a state level, including data collections, individual data linkage units, and ethics committees. The Lumos programme, the first state-wide database linked to GP records throughout New South Wales (NSW), was recently established and covers 16% of the NSW population [24].
Using SAIL, we aimed to derive an algorithm to identify individuals diagnosed with dystonia, and to establish key epidemiological characteristics, including links to deprivation and causes of mortality.

Reference population and diagnostic validation
Anonymized records from 90 patients with a confirmed diagnosis of AOIFCD were linked to records held in SAIL. Each participant was re-

Sensitivity algorithm
A list of Read codes (version 2) and International Classification of Diseases, 10th Revision (ICD-10) codes was selected and used to identify individuals with a primary and/or secondary dystonia diagnosis. All codes relevant to dystonia including diagnosis, symptoms, and therapy were reviewed by a clinical neurologist with movement disorder expertise. Code lists were created to maximize the positive predictive value, while maintaining a reasonable sensitivity. In a stepwise manner, Read codes that did not contribute to the identification of the reference cohort were removed. All ICD-10 codes, excluding drug-induced dystonia, were used. Sensitivity analyses of Read codes are available in Table S1. A full list of the final Read and ICD-10 codes is provided in Table 2.

Dystonia diagnosis
An individual was defined as having a diagnosis of dystonia if their GP or hospital record contained a Read version 2 or ICD-10 code from the list of codes (Table 2). Dystonia subtypes were not mutually exclusive; an individual could have more than one dystonia subtype.
Results from primary and secondary care extracts were combined to create a joined dystonia cohort. Individuals diagnosed with a potential secondary cause of dystonia were excluded (Tables S2-S4) [29,30]. Our stringent exclusion criteria removed diagnostic codes linked with tremor (apart from dystonic tremor) and diagnoses that may include tremor as part of the phenotype. Medicine codes for all forms of dopaminergic therapy were also included in this exclusion algorithm, providing an additional mechanism to exclude degenerative forms of tremor. Inclusion of the tremor code was shown to increase the sensitivity of our patient identification algorithm by 6% (73%-79%) in our validation cohort, allowing for recognition of dystonic tremor as a form of primary dystonia, while limiting the possibility of including other forms of tremor.
Additionally, individuals were required to be resident in Wales Population-based approaches identify cases from the general community to represent the population, for example, door-to-door surveys or electronic questionnaires. b Service-based studies ascertain cases from inpatient clinics, for example, service users.
c Record-linkage studies combine data from health and other services, for example, medical records, death records, and socioeconomic status. Abbreviations: BSP, Blepharospasm.  We identified all dystonia cases in primary and secondary care datasets; however, further analysis including epidemiological, deprivation, and mortality characterization focuses on individuals with adult-onset idiopathic dystonia (≥20 years of age).

Incidence and prevalence
Estimates for prevalence and incidence were calculated annually for the study period (January 1994-December 2017

Mortality
Date and underlying causes of death during the study period were determined from the ADDE. Underlying causes of deaths coded using ICD-9 were converted to ICD-10.

Statistical analysis
Data was analysed using R software (version 4.0.4). Where applicable, a t-test or Mann-Whitney U-test was used to determine any TA B L E 2 ICD-10 and read codes used to identify dystonia patients in hospital and general practice electronic records, respectively

Ethics
This study design uses anonymized, routinely collected data and therefore does not require ethical approval and written informed consent. The SAIL independent Information Governance Review Panel, experts in information governance and members of the public, approved this study (Reference: 0768).

Validating the dystonia diagnosis
The optimized case-ascertainment algorithm ( Figure 1) had a diagnostic sensitivity of 79%, identifying 70 of 89 of the clinically confirmed reference population (one patient could not be identified within SAIL).  Figure 2a and

Deprivation
At the time of diagnosis, adult-onset cases were equally distributed across deprivation quintiles (Table 3 and Figure 3b). There were no significant changes in deprivation quintile throughout the study between entry into the dataset (during the study period) and diagnosis (median = 0, p < 0.01; Figure 3a), diagnosis and follow-up

Mortality
There were 4,315 deaths (10%) in the adult-onset dystonia cohort ( Table 3). Distribution of mortality data by dystonia syndromes, including the 10 most common causes of death, is shown in Figure 5.

DISCUSS ION
Our longitudinal, nationwide population-based cohort study has combined anonymized electronic data derived from GP (family doctor/community health care) and hospital records for more than Previous estimates of dystonia prevalence, from service-or population-based studies, have varied between 6.1 and 70.1 per 100,000 in the former [31,32], to between 7.91 and 732 per 100,000 in the latter [15,17]. However, these approaches are likely to introduce bias, identifying only those seeking out or receiving clinical care, as well as underascertaining cases, suggesting that rates of dystonia within the population are higher than previously suggested [21]. In support of this, our results suggest an overall prevalence of  (31) 1092 (43) Median age at death, years 80 (17) 65 (10) 70 (7) 79 (19) 76 (12) 83 (13) 70.5 (12.5) 69 (26) 78 (19) 81 (16) Female median age at death, years   over the same period [10,16,[33][34][35]. The UK study also used a linkage-based approach, ascertaining cases by reviewing hospital records, as well as recruiting individuals registered as members to the Dystonia Society, and via a postal survey [36]. Our study has expanded upon this using detailed linked data available across the whole of Wales. More recent epidemiological studies of dystonia show a trend toward elevated estimates compared to the literature prior to 2012 [21], which may be in part due to an ageing population, increased clinical recognition of dystonia, and identification of an ever-expanding number of causative genes [37].
Three US-based studies have estimated the incidence of cervical dystonia between 0.8 and 1.18 per 100,000 person-years [5,12,38], and incidence rates of other types of dystonia varied between 0.2 and 0.4/100,000 person-years [5]. As with the multiple factors Elevated rates of incidence are consistent with increased recognition, awareness, and improved availability of neurological services.

F I G U R E 3 (a) Change in Welsh Index of Multiple Deprivation (WIMD) quintile and (b) number of cases per WIMD quintile at entry into dataset (during study period) and diagnosis by dystonia subtype in adult
Our study demonstrated that deprivation scores did not change following a diagnosis of dystonia, arguably an unexpected finding given the potential for dystonia to impact employment and social interaction. To our knowledge, there has been no research that directly investigates changes in deprivation, with previous studies that have included measures of this nature identifying conflicting findings. A population-based survey in India found no significant differences in dystonia prevalence estimates when comparing slum dwellers and non-slum dwellers, despite their socioeconomic differences [9]. Other studies have also noted that dystonia affects employment, with pain and severity of symptoms impacting work status in up to 40% of cases [40]. However, given that many of the treatments typically used to manage motor symptoms, notably botulinum neurotoxin and deep brain stimulation (DBS), have been shown to improve quality of life, and reduce pain and motor symptom severity, it is plausible that this may contribute to maintained working status [41,42]. Interestingly, increased rates of DBS surgery have been observed in the USA between 2002 (n = 2372) and 2014 (n = 5260), with a similar trend likely in the UK [43].
To our knowledge, ours is the first study to assess mortality in patients with dystonia. Of the cohort, 10% died during the study pe- Dystonia is primarily diagnosed in secondary care by a specialist, and subsequently coded in primary care by general practitioners.
Our algorithm was 79% sensitive, with 74% of the reference population being identified from primary records alone. Although there is little incentive for GPs to subtype dystonia, in general, there is evidence to suggest that primary care coding is relatively accurate in the UK [46]. Several previous studies, using the SAIL database, have validated the accuracy of using these diagnosis codes for case ascertainment of other neurological conditions [20,47,48]. However, coding practices may vary between GPs as well as having the potential There is always the potential for diagnostic misclassification through coding. For example, we did not identify any cases of idiopathic familial dystonia. This likely reflects the focus on achieving the most accurate clinical diagnosis for the individual patient, with information relating to other affected family members not always being known to the patient and/or the treating clinician.
Including hospital inpatient and outpatient data, in addition to primary care data, provided only a small increase in case ascertainment sensitivity (0.9%). This may be because hospital coding is primarily related to recording the admission or attendance, with the reasons for being admitted and comorbidities being less frequently documented [20].
We were unable to obtain a measure of specificity for our case ascertainment algorithm in general and for dystonia subtypes, as we were unable to upload relevant negative control groups. It therefore remains possible that we included a proportion of cases without dystonia. However, we sought to account for this by applying stringent exclusion criteria that would account for most known causes of acquired dystonia or misdiagnosis of dystonia. Lastly, deprivation is difficult to measure, and although the WIMD is one of the more comprehensive deprivation scores available, it does not cover every aspect of deprivation. It is also an area-based rather than individualbased measure of deprivation. It is therefore possible that an increase in deprivation for an individual would not change their WIMD score if they did not change address. We may therefore lack recognition of social change in deprivation.
Our results indicate that prevalence and incidence rates of dystonia are higher than previously estimated, 1,220/100,000/year (1.2%) and 96/100,000/year, respectively, potentially indicating more accurate population estimates given the unbiased nature of patient identification. We have shown that a diagnosis of dystonia does not appear to have a detrimental impact on socioeconomic status, with no changes in deprivation observed at follow-up. In addition, there was no evidence of decreased life expectancy and causes of underlying death, mirroring the Welsh leading causes of death.
The findings from this work have important implications for patients, carers, health care providers, third sector organizations, and health care policy, with our case-ascertainment algorithm providing a platform for application in future population-based dystonia studies.

ACK N OWLED G EM ENTS
This study makes use of anonymized data held in the SAIL system, which is part of the national e-health records research infrastructure for Wales. We would like to acknowledge all the data providers who make anonymized data available for research.

CO N FLI C T O F I NTE R E S T
The authors report no conflicts of interest.

RO LE O F TH E FU N D ER /S P O N SO R
The funders had no role in the design or conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data may be obtained from a third party and are not publicly available.
The data used in this study are stored within the SAIL Databank at the Health Information Research Unit (HIRU) at Swansea University.
All proposals to use SAIL datasets must comply with HIRU's information governance policy and are subject to review by an independent Information Governance Review Panel (IGRP). Before data can be accessed, approval must be given by the IGRP. SAIL has established an application process to be followed by anyone who would like to access data via SAIL: https://www.saild ataba nk.com/appli catio n-process.