Objective To undertake a systematic review and meta-analysis of studies that investigated prognostic factors and survival in patients with progressive supranuclear palsy (PSP) and multiple system atrophy (MSA).
Methods Publications of at least 10 patients with a likely or confirmed diagnosis of PSP or MSA were eligible for inclusion. Methodological quality was rated using a modified version of the Quality in Prognostic Studies tool. For frequently examined prognostic factors, HRs derived by univariate and multivariate analysis were pooled in separate subgroups; other results were synthesised narratively and HRs could not be reported here.
Results Thirty-seven studies presenting findings on 6193 patients (1911 PSP, 4282 MSA) fulfilled the inclusion criteria. We identified the following variables as unfavourable predictors of survival. In PSP, PSP-Richardson’s phenotype (univariate HR 2.53; 95% CI 1.69 to 3.78), early dysphagia and early cognitive symptoms. In MSA, severe dysautonomia and early development of combined autonomic and motor features but not MSA phenotype (multivariate HR 1.22; 95% CI 0.83 to 1.80).
In PSP and MSA, survival was predicted by early falls (multivariate HR 2.32; 95% CI 1.94 to 2.77), the Neuroprotection and Natural History in Parkinson Plus Syndromes Parkinson Plus Score and the Clinical Global Impression Disease Severity Score but not sex (multivariate HR 0.93; 95% CI 0.67 to 1.28). There was conflicting evidence regarding the prognostic effect of age at onset and stridor.
Conclusion Several clinical variables were strongly associated with shorter survival in PSP and MSA. Results on most prognostic factors were consistent across methodologically diverse studies; however, the lack of commonality of prognostic factors investigated is a significant limitation.
- Progressive supranuclear palsy
- multiple system atrophy
- prognostic factors
- systematic review
Statistics from Altmetric.com
- Progressive supranuclear palsy
- multiple system atrophy
- prognostic factors
- systematic review
Progressive supranuclear palsy (PSP) and multiple system atrophy (MSA) are rare, debilitating adult-onset neurodegenerative conditions. Their estimated prevalence is 2–7 per 100 000.1–4 PSP is characterised by the accumulation of abnormally phosphorylated tau protein in the basal ganglia, frontal lobe and brainstem, whereas MSA is characterised by the accumulation of alpha synuclein-positive glial cytoplasmic inclusion bodies5 affecting the basal ganglia, cerebellum, pons and substantia nigra.6 PSP can be divided into PSP-parkinsonism (PSP-P) and PSP-Richardson’s syndrome (PSP-RS)7 and MSA can be classified into MSA-cerebellar (MSA-C) and MSA-parkinsonism (MSA-P) subtypes.8 Diagnostic criteria have been proposed for the clinical diagnosis of PSP and MSA9–11; however, pathological brain examination at autopsy remains the gold standard for diagnostic classification.
Prognosis of PSP and MSA is poor. The mean survival after diagnosis is only 3–4 years.12 13 To date, there are no narrative or systematic reviews on prognostic factors in PSP and MSA. Predictors of survival were investigated in numerous studies, including several prospective multicentre studies,14–16 single centre clinical studies,17 18 postmortem studies19 20 and a meta-analysis of pathologically confirmed case studies.21 Researching prognosis of PSP and MSA is challenging in view of their rarity, insidious onset and heterogeneity of clinical presentation. Therefore, studies are frequently underpowered and at risk of diagnostic ascertainment bias or selection bias depending on whether participants were diagnosed clinically or recruited postmortem from a brain bank. A systematic review and meta-analysis is therefore desirable to increase power and to investigate and quantify the methodological and statistical heterogeneity across studies. Here, we sought to summarise, synthesise and contrast published studies that investigated the association between potential prognostic factors and survival.
Study registration and inclusion criteria
This work adheres to the Preferred Reporting Items for Systematic Reviews and Meta-analyses standards (PRISMA).22 A protocol was registered with PROSPERO, registration number: CRD42016032968. Studies of any design, reporting on one or more potential prognostic factors predicting survival in patients with PSP and/or MSA were eligible for inclusion. Participants were patients identified by study authors as having a likely or confirmed diagnosis of PSP, MSA, Steele-Richardson-Olszewski syndrome, striatonigral degeneration, olivopontocerebellar atrophy or Shy-Drager syndrome. The following prognostic factors were considered: demographic variables, symptoms, clinical signs, disease phenotypes, rating scales, biomarkers and imaging findings. Exclusion criteria were as follows: (1) meeting abstracts or conference proceedings; (2) articles in languages other than English, German or French; (3) case studies or studies including fewer than 10 participants; and (4) statistical analysis other than time-to-event analysis. Studies were also eligible if they investigated the difference in survival between PSP and MSA.
Identification of studies
The electronic databases of Medline, Embase, CINAHL and The Cochrane Library were searched from inception until December 2015/January 2016 for studies conducted on humans; no language restrictions were employed. The search strategy included a combination of keywords and MeSH terms (see online PROSPERO protocol). For studies retrieved in full text, hand searching of reference lists and citation searches were performed. Grey literature searches were performed in several grey literature databases.
Methodological quality assessment and data extraction
Studies were individually graded for risk of bias by one reviewer (SAG), using a modified version of the Quality in Prognostic Studies tool.23 A random sample of five papers was reviewed by a second reviewer (RAS) to check that there was no systematic bias in the method followed by the primary reviewer. The study’s risk of bias was rated as low, moderate or high on 14 items according to prespecified criteria (online supplementary table S1). A summary score was not calculated in keeping with recommendations by the Cochrane Collaboration.24 One reviewer (SAG) extracted data from all included studies using a prespecified data extraction sheet and a second reviewer independently extracted results reported by studies included in the main text; there were no disagreements.
Supplementary file 1
For three studies, corresponding authors were contacted to provide clarification on missing data; however, none could provide these.
HRs derived by univariate and multivariate analysis were pooled in separate subgroups in generic inverse variance meta-analyses in Review Manager 5.3 using a random effects model. HRs adjusted for age and/or sex were included in the subgroup containing HRs derived by multivariate analysis. The meta-analytic summary includes the pooled estimate for each subgroup and its 95% CIs. Statistical heterogeneity was quantified using the I2 test; I2 represents the percentage of heterogeneity that cannot be attributed to chance. Where only Kaplan-Meier curves were reported by studies, methods by Tierney and colleagues25 were used to extract HRs from published Kaplan-Meier curves (12 prognostic factors, 8 studies). Kaplan-Meier curves were inspected for severe violations of the proportional hazards assumption prior to data extraction. Anonymised patient data published by Birdi and colleagues26 were analysed in SPSS version 23.0 with univariate Cox regression. Sensitivity analysis was undertaken to gauge the effect of including studies at high risk of diagnostic ascertainment bias in meta-analysis. Results of such studies were graphically displayed in forest plots but were not weighted in the pooled effect estimate. Publication bias was examined for by inspecting funnel plots for the two most commonly reported prognostic factors. Results were synthesised narratively where heterogeneity across study definitions of a prognostic factor did not allow meta-analysis. A p value <0.05 was considered statistically significant.
The search yielded the following results: Medline: 934 results, Embase: 1420 results, CINAHL: 155 results and The Cochrane Library: 38 results. The selection process is illustrated in figure 1. A total of 40 publications describing 37 studies investigating 34 distinct patient cohorts were therefore included in narrative synthesis, of which 22 studies were also included in one or more meta-analyses. Meta-analysis was possible for seven prognostic factors: sex, age at disease onset, MSA phenotype, PSP phenotype, onset of falls, levodopa response and orthostatic hypotension.
Eleven studies recruited patients with PSP,12 19 26–34 20 studies recruited patients with MSA15–18 21 35–51 and six studies recruited both patients with PSP and MSA.3 9 13 14 20 52 53 Seven studies reported pathological confirmation of diagnosis in 100% of patients. Studies reported on a total of 6193 patients including 4282 patients with MSA of whom 759 had a pathologically confirmed diagnosis and 614 were included in postmortem studies. There were 1911 patients with PSP of whom 415 had a pathologically confirmed diagnosis and 250 were included in postmortem studies. The mean age at disease onset ranged from 61.0 to 67.2 years in PSP and from 54.0 to 64.5 years in MSA, and the proportion of male and female patients was approximately equal. The study characteristics can be found in online supplementary table S2 and the risk of bias ratings in online supplementary table S3; differences in symptom profile between clinical and postmortem studies are reported in online supplementary table S4.
Supplementary file 2
Supplementary file 3
Supplementary file 4
The cumulative survival in PSP and MSA is shown in figures 2 and 3, respectively (online supplementary figures S1 and S2 for colour images). Ten studies reported on patients that survived beyond 15 years after disease onset, of which three studies reported pathological confirmation of diagnosis in 100% of patients. Median survival from disease onset ranged from 5.3 to 10.2 years. One study43 reported a median survival of 13 years but was judged to be at high risk of diagnostic ascertainment bias. Studies on patients with MSA generally reported higher median survival than studies on patients with PSP; however, studies including both patients with MSA and PSP found similar survival between PSP and MSA. In the Neuroprotection and Natural History in Parkinson Plus Syndromes (NNIPPS) study, PSP was associated with longer survival only when adjusted for dysautonomia, disease severity, disease duration and the Schwab and England Activities of Daily Living Scale in multivariate analysis.14
Table 1 summarises the evidence on frequently examined prognostic factors. Ten studies (n=2728) comparing survival between male and female patients were combined in meta-analysis (figure 4). The pooled HRs in the subgroup containing HRs derived by univariate analysis was 1.05 (0.95, 1.18), I2=32% and n=2102, whereas the pooled HRs in the subgroup containing HRs derived by multivariate analysis was 0.93 (0.67, 1.28), I2=34% and n=675. There were no significant differences between subgroups (p=0.45). Five studies only reported p values but not HRs and could not be included in meta-analysis,17 28 30 38 53 one of which reported a significant effect of sex on survival in multivariate analysis28
Fourteen studies (n=2954) investigating the effect of age at disease onset on survival were combined in meta-analysis (online supplementary figure S3). Nine studies compared the survival of patients aged above and below the median age at disease onset reported by each study. The pooled HR in the subgroup containing HRs derived by univariate analysis was 1.92 (1.50, 2.44), I2=0% and n=288, and the pooled HR in the subgroup containing HRs derived by multivariate analysis was 1.75 (1.32, 2.32), I2=63% and n=973. Subgroups were not significantly different (p=0.74). Six studies reported the prognostic effect of age at disease onset per year increase, the pooled HR in the subgroup containing HRs derived by univariate analysis was 1.01 (1.0, 1.02), I2=23% and n=1669, and the pooled HR in the subgroup containing HRs derived by multivariate analysis was 1.10 (0.98, 1.23), I2=68% and n=260. Six studies only reported p values but not HRs and could not be included in meta-analysis,10 17 28 32 50 53 of which two studies reported a significant effect of age at disease onset on survival.17 28
Eight studies (n=1572) comparing survival of the MSA-P and MSA-C phenotypes were combined in meta-analysis (figure 5). The pooled HR in the subgroup containing HRs derived by univariate analysis was 1.12 (0.97, 1.29), I2=9% and n=1310, and the pooled HR in the subgroup containing HRs derived by multivariate analysis was 1.22 (0.83, 1.80), I2=19% and n=311. Subgroups were not significantly different (p=0.67). Four studies only reported p values but not HRs and could not be included in meta-analysis, none of which reported a significant effect of MSA phenotype on survival.17 41 50 53
Four studies (n=268) comparing survival of the PSP-RS phenotype with the PSP-P phenotype were combined in meta-analysis (figure 6). The pooled HR in the subgroup containing HRs derived by univariate analysis was 2.53 (1.69, 3.78), I2=18% and n=268; one study (n=110) reported a HR derived by multivariate analysis of 2.37 (1.21, 4.64). One study only reported the difference in median survival (PSP-RS 6.8 years vs PSP-P 10.9 years).27
Onset of falls
Six studies (n=1116) investigating the effect of falls within 1–3 years of disease onset on survival were combined in meta-analysis (figure 7). The pooled HR in the subgroup containing HRs derived by univariate analysis was 1.62 (0.55, 4.77), I2=87% and n=777. The pooled HR in the subgroup containing HRs derived by multivariate analysis was 2.32 (1.94, 2.77), I2=0% and n=1073. The high statistical heterogeneity in the univariate subgroup is almost exclusively due to Figueroa and colleagues,35 who found early falls to predict shorter survival in multivariate but not univariate analysis. Subgroups were not significantly different (p=0.52). One study only reported the difference in median survival (early falls 5.2 years vs late falls 6.8 years).30
Four studies (n=1147) investigating the effect of levodopa response on survival were combined in meta-analysis (online supplementary figure S4). The pooled HR in the subgroup containing HRs derived by univariate analysis was 0.87 (0.73, 1.02), I2=1% and n=960. One study (n=187) reported a HR derived by multivariate analysis of 0.60 (0.31, 1.14).
Symptoms and signs in MSA
Severe dysautonomia was found to predict shorter survival in MSA, as two studies comparing survival between patients with mild or absent and severe dysautonomia15 18 and three studies evaluating the Clinical Global Impression Dysautonomia Score and/or Composite Autonomic Scoring Scale14 18 35 reported a significant effect on survival. Development of dysautonomia within 1, 2 or 2.5 years of MSA onset and the development of combined autonomic and motor features within 3 years of MSA onset predicted shorter survival in three studies18 20 46 and two studies,17 41 respectively. However, the presence of dysautonomia at MSA onset did not predict survival in five studies.17 18 37 41 43 Orthostatic hypotension within 1 year of MSA onset predicted shorter survival in a meta-analysis of two studies,18 35 while there was substantial heterogeneity regarding the effect of bladder symptoms on survival in five studies.15 16 18 35 39 Two studies reported significantly reduced survival in the presence of stridor,45 49 while two studies reported reduced survival in univariate but not multivariate analysis18 39 and two small studies failed to reach statistical significance40 46 (online supplementary table S5).
Supplementary file 5
Symptoms and signs in PSP
Four studies found early development of dysphagia to predict survival.12 29 30 33 There is conflicting evidence regarding the prognostic effect of vertical supranuclear gaze palsy examined in five studies: two studies with overlapping patient cohorts found vertical supranuclear gaze palsy to predict survival,27 29 two studies failed to show a predictive effect on survival30 33 and one study reported a predictive effect on survival in univariate but not multivariate analysis.12 Early development of cognitive symptoms or dementia was a predictor of shorter survival in two studies12 29 but not in two lower quality studies30 32 (online supplementary table S6).
Supplementary file 6
Disease severity, progression and rating scales
There is evidence to support the use of the NNIPPS Parkinson Plus Score and Clinical Global Impression Disease Severity Score for prognostic stratification from one study.9 14 Stage 5 of the Hoehn and Yahr Staging Scale likely predicts shorter survival as shown by one study39 but otherwise the Hoehn and Yahr Staging Scale did not predict survival in two studies.14 16 Two studies found the PSP Rating Scale to predict survival,27 28 while the Northwestern University Disability Score did not predict survival in one study.47 The Unified MSA Rating Scale did not predict survival in one study15 although its Activities of Daily Living Subscore predicted survival in one study.16 A shorter interval between disease onset and first clinical milestone also predicted shorter survival in two studies12 20 (online supplementary table S7).
Supplementary file 7
Three studies34 42 43 were excluded from meta-analysis due to high risk of bias ratings in the diagnostic certainty domain. Inclusion of the study by Sakushima and colleagues43 in the meta-analysis of multivariate HRs comparing survival in MSA-P and MSA-C resulted in MSA-P becoming a significant predictor of shorter survival. Otherwise, inclusion of the above studies did not affect the overall effect estimates (online supplementary table S8).
Supplementary file 8
Funnel plots are shown in online supplementary figures S5 and S6.
This is the first systematic review and meta-analysis of studies investigating prognostic factors in PSP and MSA. The following variables were identified as unfavourable predictors of survival: PSP-RS phenotype, early dysphagia and early cognitive symptoms in PSP; severe dysautonomia and early development of combined autonomic and motor features in MSA; and early falls, the NNIPPS Parkinson Plus Score and the Clinical Global Impression Disease Severity Score in PSP and MSA. Conversely, MSA phenotype and sex did not predict survival. Survival was similar between patients with PSP and MSA and between pathologically confirmed cases and clinically probable cases.
Many studies were too small to reach statistical significance; therefore, the increased power gained by combining studies in statistical and narrative syntheses yields more conclusive evidence on the validity of the above prognostic factors. This review examined the statistical heterogeneity (I2) of prognostic results between clinically and methodologically diverse studies. This includes studies on patients with PSP and MSA and studies on pathologically confirmed cases as well as clinically probable cases. I2 values were low (<40%) in all meta-analyses except age at disease onset, indicating that differences between the results on prognostic factors across studies were mainly due to chance. However, the lack of commonality of prognostic factors investigated is a significant limitation.
The PSP-RS phenotype compared with the PSP-P phenotype was highly predictive of shorter survival in a meta-analysis of two clinical and two postmortem studies. Early falls and early development of cognitive symptoms, which are characteristics of the PSP-RS phenotype,7 also predicted shorter survival. The latter symptom predicted shorter survival in larger studies of higher quality, whereas smaller studies failed to show a prognostic effect. The evidence regarding the prognostic effect of supranuclear gaze palsy remains inconclusive. Levodopa response, which is characteristic of the PSP-P phenotype, did not predict longer survival in meta-analysis. This may be due to variation in the definition of levodopa response across studies. Early onset of dysphagia, which is common in both phenotypes, was predictive of shorter survival.
MSA phenotypes did not predict survival in a meta-analysis of mainly Western studies; this finding should be confirmed in Asian countries.54 Interestingly, symptoms that indicate shorter survival in MSA, namely severe dysautonomia and early falls, occur with a similar prevalence in both phenotypes.18
There has been conflicting evidence on the prognostic effect of dysautonomia in previous studies, which may be due to heterogeneous definitions of dysautonomia. This review has identified trends in the literature and offers potential explanations to guide future research. Severity of dysautonomia, measured with rating scales or subjective impression, predicted survival. Similarly, urinary catheterisation, indicating severe impairment, was highly predictive of survival in two studies, whereas conflicting evidence was found regarding the effect of incomplete bladder emptying, which may range from mild to severe, on survival.
The development of dysautonomia within a specified period of MSA onset and the time until the development of combined motor and autonomic features, but not the presence of dysautonomia at MSA onset predicted survival. It can be hypothesised that the first two prognostic factors represent the same entity: studies investigating the prognostic effect of development of dysautonomia within a specified time of MSA onset may have included many patients presenting with motor onset who developed dysautonomia soon after. This hypothesis would be in keeping with a series of five long-term MSA survivors.55 Similarly, a shorter interval between disease onset and first clinical milestone predicted shorter survival. This review therefore emphasises the importance of rate of progression and severity of MSA symptoms as prognostic indicators.
There was conflicting evidence regarding the prognostic value of age at disease onset and stridor in MSA. The high statistical heterogeneity in the meta-analysis of age at disease onset may be due to variation in the median age of onset. Stridor was an independent predictor of survival in multivariate analysis in one study, which was the only study to control for the effect of stridor treatment on survival. The beneficial effect of stridor treatment on survival should be confirmed in a larger cohort of patients.
Few studies investigated rating scales; thus, further research is warranted. The NNIPPS Parkinson Plus Score and the Clinical Global Impression Disease Severity Score were found to predict survival; the PSP Rating Scale may predict survival but effect sizes were not reported.
Limitations of primary studies
The small number of prospective studies and low proportion of pathologically confirmed cases are limitations of this review. The risk of diagnostic ascertainment bias was especially high in five studies, where PSP or MSA diagnosis was based on the subjective assessment of the treating neurologist or on diagnostic criteria that are not internationally accepted. Studies of the latter type were not weighted in the meta-analytic summary effect estimates.
Several studies did not report HRs or Kaplan-Meier curves and were therefore not included in meta-analysis. However, most p values reported by studies excluded for this reason were in keeping with the respective meta-analytic estimates. Some studies did not report on insignificant results, which may have led to publication bias. Failure to quantify follow-up and missing data has not allowed us to adequately assess the risk of attrition bias and sampling bias in some studies. Many studies excluded erectile dysfunction from the definition of MSA onset while some studies excluded all autonomic symptoms, which may have overestimated mean age at onset and thus underestimated disease duration. Similarly, several prognostic factors such as severe dysautonomia were not sufficiently well defined. Nonetheless, in view of the low statistical heterogeneity in most meta-analyses it is likely that bias arising from retrospective data collection, attrition bias and variations in the definition of MSA onset has not substantially influenced the review outcomes.
Limitations of the review methodology include its restriction to articles published in English, French or German and the low specificity of the search strategy, which increases the risk of missing relevant studies. The presence of reporting bias cannot be excluded given the limitations of funnel plots where several prognostic factors are reported by individual studies.
The present review is the first to summarise, synthesise and contrast evidence on prognostic factors in patients with PSP and MSA, and it identified several clinical variables that were strongly associated with survival (table 1). Results on most prognostic factors were consistent across methodologically diverse studies, including both clinical and postmortem studies. However, the lack of commonality of prognostic factors investigated is a significant limitation.
Supplementary file 9
Supplementary file 10
Supplementary file 11
Supplementary file 12
Supplementary file 13
Supplementary file 14
The authors would like to thank Yasmin Abdi Tubai for independently extracting prognostic data reported in the main text.
Contributor SAG performed the study selection, risk of bias assessment, data extraction, statistical analysis and drafted the manuscript. RAS reviewed a random sample of five papers. All authors participated in study design, revised the protocol, contributed to interpretation of the results, critically revised the manuscript for important intellectual content, read and approved the final version of this manuscript.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.