The prevalence and incidence of progressive supranuclear palsy and corticobasal syndrome: a systematic review and meta-analysis

A systematic review was conducted according to a pre-determined protocol based on the PRISMA statement for systematic review and meta-analysis (see Supplementary Table 3). A search of PubMed and EMBASE from their date of inception to 13th July 2021 was performed using the search strategies detailed below (Table 1). The search was restricted to studies in English. The study protocol was registered with the PROSPERO registry before the initial review of the titles and abstracts (crd.york.ac.uk/PROSPERO/; registration number: CRD 42021266193) [9].

Studies were screened by the first author (SL). Where it was unclear whether to include a study, the senior author (SOD) acted as arbiter. No automated tools were used in the selection of studies. For the purposes of the systematic review portion of the study, we adopted broad inclusion criteria defined using the CoCoPop model (Supplementary Table 1) [10]. We considered any study which attempted to enumerate neurodegenerative disease in a general population and which reported cases of PSP or CBS as part of that enumeration. Studies were excluded if cases of PSP or CBS were not reported or could not be calculated from reported data. We considered a prevalence study to consist of a study of a general population which attempted to identify cases of PSP/CBS in a comprehensive way (for example by multiple source referrals within a defined population, screening of medical records within a region, or a standardised system of notification) and reported prevalence or reported data such that prevalence could be calculated. We considered an incidence study to consist of a study of a general population in which a comprehensive attempt was made to ascertain all cases of PSP and CBS within a defined time-period, and incidence was reported, or data reported such that incidence could be calculated. We included in the systematic review studies that identified PSP and CBS in a variety of ways: (i) in person examination of patients with identification based on published guidelines or diagnosis by expert clinicians, (ii) identification from notes and letters detailing patient diagnosis and (iii) notification to registries by treating physicians, (iv) association of patients with an appropriate code or entry in database and registry studies. Papers and conference abstracts meeting these criteria were selected for full-text review. The references of included articles and relevant review articles were hand searched for additional articles. All additional articles were evaluated in the same manner as those identified in the initial search. When studies were identified in both abstract and paper format, the paper format was included as representing a more complete data set.

Data was extracted from articles using a standardised data collection form by the first authors (SL) in cases where data the relevant data was unclear or questionable, these cases were reviewed with the senior author (SOD). Studies were recorded as reporting data on prevalence and/or incidence of PSP, CBS, or both. If multiple articles reported data on the same study population the most comprehensive data were utilised. In cases where multiple studies reported prevalence or incidence data for the same geographic area at different time points all studies were included as separate estimates. Demographic data extracted included age, sex, and study location. Diagnostic criteria used, data on phenotypic subtypes, and methods of case recruitment and identification were recorded. The number of cases of PSP or CBS, and the size of the population studied, and/or person-years used to calculate incidence rates were extracted. Incidence and/or prevalence estimates from each study were extracted and reported per 100,000 population or 100,000 person-years as applicable. Study quality was assessed using the Joanna Briggs Inventory (Supplementary Table 4) [11].

Meta-analysis was limited to a subset of studies with the following characteristics: (1) full articles, excluding abstracts, (2) were designed to investigate PSP/CBS, parkinsonism, Parkinson’s disease, or FTLD. Studies of dementia were excluded as they may fail to identify movement-predominant presentations, (3) data on individual cases accessed either by review of clinical notes or examination of participants (as opposed to databases or registries) (4) did not restrict recruitment by age beyond that required by diagnostic criteria (i.e. 40 years of age) (5) specified established diagnostic criteria/guidelines, excluding the 2017 MDS Diagnostic Criteria (these were used only in one study and the expansion of the phenotype recognised by these criteria makes meta-analysis unreliable), (6) studies based on sampling were excluded.

Number of cases (numerator) and population and/or person-years (denominator) in included studies were used to calculate pooled prevalence and incidence of PSP and CBS using a random effects model. Meta-analyses were conducted using the metafor package [12]. Heterogeneity was assessed by the X² test on Cochrane’s Q statistic, which was quantified by I², assuming I² values of 25%, 50%, and 75% respectively representing low, medium, and high heterogeneity [13]. Outliers were identified using studentised residuals. We conducted a leave-one-out sensitivity analysis to explore the sensitivity of the pooled prevalence estimates to the exclusion of individual studies.

The search strategy yielded 1,650 citations (1,020 from EMBASE, 630 from PubMed) (Fig. 1) carried out between 1988 and 2021. Of the studies reviewed, 32 reporting the incidence and/or prevalence of PSP and/or CBS were included in our systematic review [14‐45]. Nine studies reported data on both conditions. A variety of ascertainment methods, often overlapping within a single study, were used (see Table 2 for summary details of study contents and Supplementary Table 2 and 3 for more details, Supplementary Table 6 contains a complete list of included studies). Seven studies assessed populations of fewer than 20,000 people (Supplementary Table 1 and 2).

Prevalence data on PSP and/or CBS was reported in 23 studies. This included 20 papers and three conference abstracts. 15 studies reported prevalence data on PSP only, three reported on CBS only, and five reported data on both PSP and CBS. One paper reported prevalence figures for PSP in three nested populations using a “Russian doll” design, thus, there were 22 estimates of PSP prevalence. Studies identified PSP/CBS in the context of a range of conditions as follows: PSP/CBS specifically (n = 7), parkinsonism (n = 7), Parkinson’s disease (n = 5), dementia (n = 3), frontotemporal lobar degeneration (n = 1).

17 articles and three abstracts provided information on PSP prevalence (see Fig. 2(A) and Supplementary Table 1). Reported estimates of population prevalence ranged from 1.00 (0.9–1.1) [22] to 18 (8–28) [32] per 100,000. Age-adjusted prevalence was reported in 10 populations with estimates ranging from 1.39 to 17.26 per 100,000 [16, 39]. In a study of a large US database prevalence peaked in the age range 75–79, with a prevalence of < 2 per 100,000 before the age of 60 [45]. A UK study demonstrated a peak prevalence in the 70–74 age range [38], however, a large Swiss study reported a peak prevalence of 72.3 per 100,000 in the oldest age group (80–89 years) [40]. Prevalence data by sex was available for seven studies [16, 21, 22, 27, 32, 39, 40]. Reported prevalence figures for men ranged from 1.00 (0.9–1.1) to 25 (8–43), and from 1.0 (0.8–1.1) to 17.76 (10.38–30.39) per 100,000 for women. One study reported data on phenotypic subtypes. Richardson’s syndrome was the most common phenotype (16/25, 64%), followed by PSP-parkinsonism (3/25, 12%), and PSP-progressive gait freezing (2/25, 8%) [14]. Two studies of PSP prevalence in Yonago City, Japan conducted 11 years apart demonstrated an increase in the recorded prevalence of PSP in this area during that time (from 5.82 to 17.9 per 100,000) [27, 39]. Meta-analysis of nine studies yielded a pooled prevalence estimate of 6.92 (4.33–11.06) per 100,000, with a high degree of heterogeneity (I² = 89%, τ² = 0.3907, P < 0.01) (Fig. 4(A)). One study [38] was a significant outlier (studentised residual =  – 2.00), exclusion of this study resulted in a prevalence estimate of 8.29 (5.50–12.49). Pooled prevalence figures of relevant subgroups are available in Fig. 5(A). Pooled prevalence for three studies from Japan yielded a prevalence of 12.85 (10.97–19.68) (I² = 76%, τ² = 0.2969, P = 0.02) while six studies carried out in Europe yielded a pooled prevalence of 4.95 (3.19–7.68) (I² = 81%, τ² = 0.1789, P < 0.01).

Eight studies (seven papers and one abstract) reported the prevalence of CBS (Table 2 and Supplementary Table 2). Reported prevalence figures ranged from 0.77 (0.02–4.32) to 25 (0–59) per 100,000. Age adjusted prevalence was available for two studies of CBS with rates of 3.2 (1.5–4.8) and 6 (0–12) per 100,000 [32, 40]. Prevalence by sex were available for two studies, with male vs female prevalence figures of 2.6 vs 3.3[40], 16 vs 3 per 100,000 [32]. Meta-analysis of three studies of the prevalence of CBS yielded a pooled prevalence 3.91 (2.03- 7.51, I² = 72%, τ² = 0.2573, P = 0.03).

The incidence of PSP/CBS was reported in 12 studies (PSP only n = 5, PSP and CBS n = 7, no studies with CBS alone), including 11 papers and one abstract. Two additional studies reported an indirectly calculated incidence rate, using prevalence and survival data and were not included in the analysis of incidence [16, 21]. Included studies reported on the incidence of parkinsonism (n = 6), PSP and CBS specifically (n = 2), dementia (n = 1), FTLD (n = 2), and a single study reported on a number of uncommon neurological diseases. Age-adjusted rates were available for four studies of PSP [14, 31, 40, 45] and two for CBS [14, 40]. Three of these studies used direct standardisation to a national population with details of standardisation not reported in one study.

Reported incidence rates for PSP ranged from 0.3 to 2.6 per 100,000 person-years (see Fig. 2(C) and Supplementary Table 2), while age-adjusted rates ranged from 0.12 to 2.0 per 100,000 person-years. Incidence rates in populations > 60 years ranged from 0.84 (0.27–1.96) [14] to 9.7 per 100,000 person [18] years reported. Incidence of PSP was reported in comparable age increments in three studies (Fig. 3). PSP incidence was reported by sex in eight studies. Incidence among men ranged from 0.17–3.9 per 100,000 person-years and from 0.1–0.8 per 100,000 person-years among women. Incidence of PSP was reported in studies of the population in Olmsted County, Minnesota over two time periods, the first study reported on the period 1976–1990 produced an incidence of 0.9 per 100,000 person-years [18], while a study of the 1991–2005 population using the 2017 MDS criteria returned an incidence rate of 2.6 per 100,000 person-years [43]. Meta-analysis of eight studies yielded a pooled incidence of 0.81 per 100,000 person-years (0.48–1.37, I2 = 86%, τ2 = 0.4847, p < 0.01, Fig. 4(C)). One study [14] was a significant outlier (studentised residual =  – 2.57); when this was excluded from the analysis, a pooled incidence of 0.99 (0.65–1.51) was found.

The incidence of CBS ranged from 0.03 [14] to 0.8 per 100,000 person-years [18] (Fig. 2(D) and Supplementary Table 2). Age-adjusted rates were available for two studies of CBS with rates of 0.02 (0.01–0.12) [14] and 1.4 (0.3–2.4) [40] per 100,000 person-years reported. Meta-analysis of six studies yielded an incidence of 0.28 per 100,000 person-years (0.15–0.52, I² = 70%, τ² = 0.3896, p < 0.01, Fig. 4(D)). Pooled incidence figures for relevant subgroups are reported in Fig. 5(B) and (C).

Registry and database-based studies have generally reported lower prevalence and incidence numbers than studies which have sought referral from a comprehensive array of sources [44, 45]. The 2001 study by Nath et al. demonstrates the limitations of a database and registry-based approach, with a Russian-doll model demonstrating higher prevalence figures with more granular and comprehensive methods of ascertainment [22]. While database and registry studies provide the sample size necessary to study rare diseases, they may lack the ability to accurately capture cases in disease where reliable biomarkers are lacking, such as PSP and CBS. However, studies with more complete ascertainment may be limited in size, identifying relatively few cases in smaller populations and generating estimates of incidence and prevalence with very broad confidence intervals.

The degree of heterogeneity between studies makes inferences regarding longitudinal trajectories of prevalence difficult, however, two studies conducting in Yonago City, Japan 11 years apart, demonstrated an increase in prevalence over that time, from 5.82 (1.78–9.86) to 17.9 (12.12–26.42) per 100,000 [27, 39]. This increase remained statistically significant following adjustment for age and sex. However, although PSP-RS was numerically increased this did not reach statistical significance, suggesting that increased recognition of non-RS subtypes might have played a role in this apparent trend. This may be driven by several factors which are likely to be replicated in other settings, such as an aging population and increased recognition of PSP subtypes, and some which are unique to the local context, such as the inclusion of PSP in the Japanese Specified Treatment Research Program in 2003. However, the later of these studies is a relative outlier in terms of reported prevalence and, as geographical clusters of PSP are known to occur, may not be generalisable. Studies of the population of Olmsted County, Minnesota demonstrated an incidence rate which was essentially unchanged in the periods 1976–1990 and 1991–2005. Two studies published since the conclusion of our data collection two studies have added more information on secular trends in the prevalence and incidence of PSP and CBS. Swallow et al. have demonstrated stable incidences of PSP and CBS over a 20 year period in a Northern European population [47]. Logroscino et al., whose earlier study is included in our analysis have published updated data on the incidence of a range of fronto-temporal lobar degeneration syndromes in nine countries with combined estimated incidence of PSP and CBS of 0.51 (0.22–1.20) per 100,000 person-years [48].

Evolution in diagnostic criteria may explain a significant proportion of apparent change over time. Many existing studies used the NINDS-SPSP criteria for PSP which, despite excellent specificity, has poor sensitivity for Richardson’s syndrome early in the disease course and does not account well for variant presentations, which may be the initial presentation in up to 60–75% of PSP cases [49, 50]. Therefore, these studies may underestimate the true prevalence and incidence of PSP. In one study, only 25% of cases in which PSP was demonstrated pathologically received a PSP diagnosis at initial evaluation [50]. Two studies conducted in Olmsted County, Minnesota provide insight into the impact of the 2017 MDS criteria on PSP incidence: in a 2013 study using the NINDS-SPSP criteria the incidence of PSP was 0.9, in a 2020 study of the same population using the 2017 criteria, an incidence rate of 2.6 was reported [34, 43]. The most common criteria used in the diagnosis of CBS were the 2013 Armstrong criteria which despite usefully expanding the range of phenotypes seen in CBS may have limited specificity [51].

In common with many other neurological conditions the epidemiology of PSP/CBS is much more extensively studied in high-income countries with Europe, America, and Japan being particular areas of activity with low-and-middle-income-countries (LMICs) underrepresented. As the prevalence of neurodegenerative disease increases with age, and the populations of many LMICs are aging[52], there is an urgent need to expand the study of PSP/CBS to these areas. In addition, of particular interest is the relatively high prevalence of PSP reported in studies from Japan, and further studies from that region may provide clues to genetic and environmental risk factors [32, 39].

This study comprehensively collates and analyses the existing literature on the incidence and prevalence of PSP/CBS. A recent systematic review and meta-analysis of the prevalence of PSP and CBS by Swallow et al. identified similar issues with heterogeneity of study design and a wide range of reported prevalence figures [8]. Our study complements this work, including a discussion on data from existing registry based studies and data and meta-analysis on incidence of PSP and CBS. Our study has several limitations. There is a high degree of heterogeneity between the identified studies, which limits the ability to generalise from recorded observations, especially when the possibility of geographic or genetic clustering may occur. Many of the studies are small and underpowered to estimate the prevalence of rare conditions such as PSP and CBS. Nevertheless, significant heterogeneity remains even when such studies are excluded. Several studies restricted their recruitment by age [17, 26, 33] and these studies reported prevalence of PSP/CBS higher than unrestricted studies, consistent with an increasing prevalence of neurodegenerative disease in older age. The relatively small number of studies makes identifying trends problematic. All studies of PSP and CBS are limited by the degree of phenotypic heterogeneity and the absence of reliable antemortem biomarkers of the underlying neuropathology, and any inference derived from meta-analysis will be susceptible to similar limitations. PSP has also demonstrated a tendency to cluster for environmental reasons [53, 54] and genetic reasons [55] which may further confound attempts to generalise from existing studies, especially from small sample sizes. Lastly, because of the recognition of the wider phenotype associated with PSP and CBD pathology it is difficult to extrapolate epidemiological estimates from studies which use older diagnostic criteria.

Future epidemiological studies of PSP and CBS face several challenges. Studies of PSP/CBS require relatively large populations to generate accurate estimates. Accurate detection of variant presentations of PSP and CBS will require collaboration of multiple specialties, including psychiatry of older age, neurology, medicine for the elderly, and family physicians. Given the possibility of evolution of clinical syndromes and consequent diagnostic instability studies should include a range of neurodegenerative diagnoses and longitudinal follow-up. Large, prospective registries of neurodegenerative diseases with comprehensive reporting requirements and including detailed clinical information might offer the best hope for accurate population level data, although the logistic and data management challenges would be formidable. In the future, with the development of fluid and imaging based biomarkers, epidemiology informed by in-vivo molecular profiles may improve the reliability of epidemiological data [56].