Physical frailty, depressive symptoms, and cognitive decline before and after Parkinson’s disease diagnosis: a population-based prospective study of the SHARE and ELSA cohorts Authors: Ziyang Ren, Zuliyaer Talifu, Xinyue Lin, Yuchun Sun, Yihao Zhao, Jing Chen, Danhua Zhao, Jufen Liu, Junliang Yuan, Xiaoying Zheng Published: 2026 ABSTRACT AbstractParkinson’s disease (PD) is increasingly common, yet its multidomain prodromal evolution remains incompletely characterised. Among 49,039 participants aged ≥50 years from the Survey of Health, Ageing and Retirement in Europe (SHARE), 624 incident PD cases occurred for up to 9 years. Compared with non-frailty, physical frailty was associated with a higher PD risk (HR = 2.31, 95% CI: 1.82–2.94); corresponding HRs were 1.95 (95% CI: 1.55–2.46) for severe depressive symptoms and 1.66 (95% CI: 1.34–2.06) for severe cognitive impairment. Risks were greatest when deficits co-occurred across all three domains (HR = 3.20, 95% CI: 2.42–4.24). Short-term worsening in multidomain health was associated with higher PD risk, whereas recovery to a more favourable state was associated with lower risk. The joint trajectory characterised by high physical frailty, high depressive symptoms, and low cognitive performance showed the highest risk of PD (HR = 4.21, 95% CI: 2.36–7.51). PD was also associated with progression to less favourable health states and reduced recovery, while future PD cases in both the SHARE and English Longitudinal Study of Ageing (ELSA) showed faster pre-diagnostic deterioration, particularly in depressive symptoms and verbal fluency, with abrupt worsening around diagnosis. These findings support repeated multidomain monitoring to improve PD risk stratification in older adults.Subject terms: Diseases, Health care, Neurology, Risk factors Introduction Parkinson’s disease (PD) is one of the fastest-growing neurological disorders worldwide1,2. With population ageing, and possibly exacerbated by longevity and environmental exposures, the number living with PD is projected to reach 25.2 million by 2050, a 112% increase from 20213,4. PD is a multisystem neurodegenerative disease with a prolonged prodrome5. Prior studies have identified several early features associated with future PD6. Physical frailty, in particular, can comprise a set of autonomic changes and predict PD regardless of genetic susceptibility7,8. Depression is among the most consistent, with prevalence rising progressively from ~10 years before diagnosis through 10 years after9. Likewise, subtle cognitive deficits predate diagnosis; population research reports poorer global cognition up to 6 years before Parkinsonism6,10. Yet most work targets single systems, and evidence on combined multidomain deficits is limited. The Movement Disorder Society proposed PD diagnostic criteria, but relied primarily on specialised assessments not commonly used in general ageing populations11,12. In contrast, routine physical, psychological, and cognitive measures, such as physical frailty, depressive symptoms, and cognitive function, can help characterise integrated prodromes and multidomain decline preceding PD. Furthermore, physical frailty often coexists with depressive symptoms and cognitive impairment in older adults13, raising the question of whether multidomain deficits confer cumulative risks. To date, however, no large prospective study has comprehensively evaluated how concurrent deficits in physical, psychological, and cognitive health together influence PD incidence in community-dwelling populations. Similarly, the dynamic evolution of these domains before and after PD diagnosis remains underexplored. To address these gaps, we conducted this prospective cohort study to examine the associations of baseline status, short-term changes, and long-term joint trajectories of physical frailty, depressive symptoms, and cognitive function with incident PD; to characterise trajectories of these multidomain indicators before and after PD diagnosis; and to quantify the associations of PD with multidomain health-state transitions. Results Baseline characteristics of study participants Among 49,039 participants included in the incident PD analysis, 624 developed incident PD for up to 9 years (Table S1). Compared with those who remained free of PD, participants who developed PD were older (72.3 ± 8.4 vs. 66.4 ± 9.4 years), more often men and urban-dwelling, and less likely to smoke or drink at baseline. Multidomain deficits were also more common among future PD cases than among non-cases; the prevalence of all three impaired domains was 15.5% versus 6.4%. Baseline multidomain health and incident PD In the primary analysis, physical frailty, depressive symptoms, and cognitive impairment were each associated with a higher subsequent risk of PD (Table 1). Compared with non-frail, frail was associated with an HR of 2.31 (95% CI 1.82–2.94); compared with no depressive symptoms, severe depressive symptoms were associated with an HR of 1.95 (1.55–2.46); and compared with no cognitive impairment, severe cognitive impairment was associated with an HR of 1.66 (1.34–2.06). In the combined multidomain analysis, the highest risk was observed among participants with deficits in all three domains (HR 3.20, 95% CI 2.42–4.24), followed by those with physical frailty plus one additional impaired domain (HR 2.14, 95% CI 1.68–2.74). The findings were broadly robust in complete-case analyses, after excluding incident PD within 4 years after baseline, in time-dependent Cox models, and in competing-risk analyses (Table S2). There was no evidence of effect modification by sex (Table S3). Table 1Physical frailty, depressive symptoms, cognitive impairment, and risk of incident Parkinson’s diseaseEvents/participants (%)HR (95% CI)Physical frailty Non-frail104 (0.7)1.00 (Ref) Pre-frail203 (1.1)1.32 (1.04–1.68) Frail317 (2.1)2.31 (1.82–2.94)Depressive symptoms No419 (1.1)1.00 (Ref) Mild109 (1.5)1.29 (1.04–1.59) Severe96 (2.1)1.95 (1.55–2.46)Cognitive impairment No364 (1.0)1.00 (Ref) Mild116 (1.8)1.35 (1.08–1.68) Severe144 (2.2)1.66 (1.34–2.06)Combined multidomain health status Healthy182 (0.8)1.00 (Ref) Physical93 (1.8)1.84 (1.42–2.38) Psychological28 (0.7)1.13 (0.76–1.69) Cognitive78 (1.4)1.42 (1.08–1.87) Psychological-Cognitive19 (1.4)1.68 (1.04–2.72) Physical-Psychological/Cognitive127 (2.0)2.14 (1.68–2.74) Physical-Psychological-Cognitive97 (3.0)3.20 (2.42–4.24)Models were adjusted for age, sex, residence, education, household assets per capita, marital status, smoking, and drinking. In the combined multidomain analysis, Healthy indicates no physical frailty, no depressive symptoms, and no cognitive impairment; Physical, Psychological, and Cognitive indicate impairment in that domain only; Psychological-Cognitive indicates depressive symptoms plus cognitive impairment; Physical-Psychological/Cognitive indicates physical frailty plus one additional impaired domain; and Physical-Psychological-Cognitive indicates impairment in all three domains.CI confidence interval, HR hazard ratio, PD Parkinson’s disease. Physical frailty, depressive symptoms, cognitive impairment, and risk of incident Parkinson’s disease Models were adjusted for age, sex, residence, education, household assets per capita, marital status, smoking, and drinking. In the combined multidomain analysis, Healthy indicates no physical frailty, no depressive symptoms, and no cognitive impairment; Physical, Psychological, and Cognitive indicate impairment in that domain only; Psychological-Cognitive indicates depressive symptoms plus cognitive impairment; Physical-Psychological/Cognitive indicates physical frailty plus one additional impaired domain; and Physical-Psychological-Cognitive indicates impairment in all three domains. CI confidence interval, HR hazard ratio, PD Parkinson’s disease. Short-term changes in multidomain health and incident PD Short-term changes from Wave 4 to Wave 5 were also informative (Fig. 1). In the continuous analyses, worsening PFI and EURO-D scores were associated with higher PD risk, whereas declines in cognitive z score were associated with higher PD risk; there was no strong evidence of non-linearity for these associations (P for non-linearity = 0.097 for PFI, 0.438 for EURO-D, and 0.566 for cognitive z score). In the categorical multidomain analysis, transition from Healthy to Combined was associated with a twofold higher risk of incident PD compared with persistent Healthy status (HR 2.01, 95% CI 1.11–3.64), and transition from Single to Combined was likewise associated with increased risk compared with persistent Single status (HR 1.78, 95% CI 1.21–2.60). In contrast, recovery from Combined to Healthy was associated with a lower subsequent PD risk relative to persistent Combined status (HR 0.39, 95% CI 0.17–0.88). Short-term changes in multidomain health in relation to incident Parkinson’s disease. a Adjusted HRs and 95% CIs for incident PD according to continuous change in PFI. b Adjusted HRs and 95% CIs for incident PD according to continuous change in EURO-D score. c Adjusted HRs and 95% CIs for incident PD according to continuous change in cognitive z score. In panels a) to c), associations were estimated using Cox proportional hazards models with restricted cubic splines. Change was defined as the Wave 5 value minus the Wave 4 value. The orange, blue, and pink solid lines indicate adjusted HRs, the shaded areas indicate 95% CIs, and the grey dashed horizontal lines indicate HR = 1.0. Higher changes in PFI and EURO-D score indicate worsening physical frailty and depressive symptoms, respectively, whereas lower changes in cognitive z score indicate worsening cognitive function. Models were adjusted for age, sex, residence, education, household assets per capita, marital status, smoking, drinking, and the corresponding Wave 4 measure. d Adjusted HRs and 95% CIs for incident PD according to categorical changes in combined multidomain health status from Wave 4 to Wave 5. Healthy indicates no physical frailty, no depressive symptoms, and no cognitive impairment; Single indicates impairment in exactly one domain; and Combined indicates impairment in at least two domains. Black squares indicate point estimates, horizontal black lines indicate 95% CIs, and the solid vertical black line indicates the null value on the log(HR) scale. Percentages indicate the proportion of incident PD cases within each category. Models in (d) were adjusted for age, sex, residence, education, household assets per capita, marital status, smoking, and drinking. CI confidence interval, EURO-D European Depression Scale, HR hazard ratio, PD Parkinson’s disease, PFI physical frailty index. Joint trajectories of multidomain health and incident PD Across Waves 1 to 5, five joint trajectory groups of physical frailty, depressive symptoms, and cognitive function were identified (Fig. 2). Compared with the most favourable trajectory group characterised by low PFI, low EURO-D, and high cognitive performance, the group characterised by moderate PFI, moderate EURO-D, and low cognitive performance had a higher PD risk (HR 2.61, 95% CI 1.53–4.45), and the group with high PFI, high EURO-D, and low cognitive performance had the highest risk (HR 4.21, 95% CI 2.36–7.51). Joint trajectories of multidomain health in relation to incident Parkinson’s disease. a Five joint trajectory groups of PFI, EURO-D score, and cognitive z score across Waves 1, 2, 4, and 5, identified using group-based trajectory modelling. The blue line indicates the trajectory group with low PFI, low EURO-D, and high cognitive function; the orange line indicates low PFI, low EURO-D, and moderate cognitive function; the green line indicates moderate PFI, moderate EURO-D, and high cognitive function; the pink line indicates moderate PFI, moderate EURO-D, and low cognitive function; and the black line indicates high PFI, high EURO-D, and low cognitive function. Lower PFI and EURO-D scores and higher cognitive z scores indicate a more favourable health status. b Incident PD events, participants, and adjusted HRs with 95% CIs for each joint trajectory group. The reference group was the trajectory characterised by low PFI, low EURO-D, and high cognitive function. Models were adjusted for age, sex, residence, education, household assets per capita, marital status, smoking, and drinking. CI confidence interval, EURO-D European Depression Scale, HR hazard ratio, PD Parkinson’s disease, PFI physical frailty index, Ref reference. Multistate transitions in multidomain health according to PD status In the multi-state analyses, PD was associated with accelerated progression to worse multidomain states and impaired recovery (Table 2). For the combined multidomain status, PD increased the hazards of transition from Healthy to Single (HR 2.18, 95% CI 1.64–2.90), from Single to Combined (HR 1.72, 95% CI 1.42–2.08), and from Combined to Death (HR 1.53, 95% CI 1.32–1.76). Conversely, PD was associated with lower recovery hazards from Combined to Single (HR 0.34, 95% CI 0.26–0.46) and from Single to Healthy (HR 0.61, 95% CI 0.40–0.91). Domain-specific analyses showed a similar pattern. For example, PD was associated with higher progression hazards from non-frail to pre-frail (HR 3.43, 95% CI 1.97–5.98), from no depressive symptoms to mild symptoms (HR 2.42, 95% CI 2.00–2.92), and from no cognitive impairment to mild impairment (HR 1.86, 95% CI 1.53–2.26), while several recovery transitions were attenuated. Table 2Parkinson’s disease and the risk of transitions in physical frailty, depressive symptoms, cognitive impairment, and combined multidomain health statusTransitions during follow-upNo. of records contributing to the transitionHR (95% CI)Start stageEnd stagePhysical frailty Non-frailPre-frail97733.43 (1.97–5.98) Pre-frailFrail10,8962.03 (1.65–2.50) FrailDeath38931.86 (1.61–2.14) FrailPre-frail55910.38 (0.27–0.53) Pre-frailNon-frail50810.66 (0.32–1.40)Depressive symptoms NoMild96012.42 (2.00–2.92) MildSevere28701.22 (0.84–1.77) SevereDeath10361.37 (1.16–1.62) SevereMild26460.71 (0.46–1.07) MildNo84030.78 (0.60–1.02)Cognitive impairment NoMild73081.86 (1.53–2.26) MildSevere34991.37 (0.96–1.95) SevereDeath22551.77 (1.53–2.06) SevereMild25690.80 (0.50–1.28) MildNo60990.58 (0.42–0.78)Combined multidomain health status HealthySingle11,8022.18 (1.64–2.90) SingleCombined87141.72 (1.42–2.08) CombinedDeath32581.53 (1.32–1.76) CombinedSingle58850.34 (0.26–0.46) SingleHealthy88460.61 (0.40–0.91)Models were adjusted for age, sex, residence, education, household assets per capita, marital status, smoking, and drinking. In the combined multidomain model, Healthy indicates no physical frailty, no depressive symptoms, and no cognitive impairment; Single indicates impairment in exactly one domain; and Combined indicates impairment in at least two domains.CI confidence interval, HR hazard ratio, PD Parkinson’s disease. Parkinson’s disease and the risk of transitions in physical frailty, depressive symptoms, cognitive impairment, and combined multidomain health status Models were adjusted for age, sex, residence, education, household assets per capita, marital status, smoking, and drinking. In the combined multidomain model, Healthy indicates no physical frailty, no depressive symptoms, and no cognitive impairment; Single indicates impairment in exactly one domain; and Combined indicates impairment in at least two domains. CI confidence interval, HR hazard ratio, PD Parkinson’s disease. Pre- and post-diagnosis trajectories of multidomain health in SHARE and ELSA Participants included in the pre- and post-diagnosis trajectory analyses from SHARE and ELSA were similar in age and sex, although education, smoking, and alcohol-use distributions differed between cohorts (Table S4). In SHARE, participants who developed PD showed an abrupt worsening around diagnosis, with a sudden increase in PFI of 0.066 (95% CI 0.055–0.077) and in EURO-D score of 0.479 (95% CI 0.239–0.718), together with sudden declines in global cognitive z score (−0.172, 95% CI −0.261 to −0.082), immediate recall (−0.154, 95% CI −0.259 to −0.050), delayed recall (−0.205, 95% CI −0.310 to −0.099), and verbal fluency (−0.100, 95% CI −0.192 to −0.008) (Fig. 3). Before diagnosis, future PD cases already showed steeper deterioration than participants without incident PD, particularly for depressive symptoms (0.084 vs. 0.034 points/year) and verbal fluency (−0.035 vs. −0.017 z-score units/year). After diagnosis, adverse trajectories persisted, with post-diagnosis slopes of 0.025/year for PFI, 0.061/year for EURO-D, −0.056/year for global cognition, and −0.042/year for verbal fluency. Predicted trajectories of multidomain health before and after incident Parkinson’s disease. a Predicted trajectories of PFI. b Predicted trajectories of EURO-D scores. c Predicted trajectories of cognitive z scores. d Predicted trajectories of immediate recall z scores. e Predicted trajectories of delayed recall z scores. f Predicted trajectories of verbal fluency z scores. In (a–f), trajectories were estimated using fully adjusted mixed-effects models and are illustrated for PD diagnosis at 8 years after baseline. Red solid lines indicate participants with incident PD, black solid lines indicate those without incident PD, the lighter surrounding bands indicate 95% CIs, and the vertical grey dotted lines indicate the illustrative diagnosis time. “No-PD slope” denotes the annual change in the no-PD group; “Sudden change” denotes the estimated level change at PD diagnosis; “Pre-PD slope” denotes the annual change before PD diagnosis; and “Post-PD slope” denotes the annual change after PD diagnosis. Models were adjusted for age, sex, residence, education, household assets per capita, marital status, smoking, and drinking. Predicted values were generated for a reference woman aged 65 years, urban dwelling, with upper secondary/vocational education, median household assets per capita, married/partnered, never smoking, and none or less than monthly alcohol use. CI confidence interval; EURO-D European Depression Scale; PD Parkinson’s disease; PFI physical frailty index. ELSA showed similar peri-diagnostic and longitudinal patterns (Fig. S1). Future PD cases had abrupt worsening at diagnosis, including increases in PFI and CES-D score and declines in global cognition, immediate recall, and verbal fluency. Pre-diagnosis deterioration was again steeper among those who later developed PD, particularly for depressive symptoms and verbal fluency, and post-diagnosis decline remained pronounced. The delayed-recall step change in ELSA was directionally less consistent and not statistically significant, likely reflecting the smaller number of incident PD cases in the validation cohort. Discussion In this prospective cohort study, we found that physical frailty, depressive symptoms, and cognitive impairment were each associated with a higher subsequent risk of incident PD, with the strongest associations observed when deficits co-occurred across domains. Short-term worsening in these measures and long-term adverse joint multidomain trajectories were associated with additional risk, whereas recovery from poorer multidomain states was associated with lower risk. We also found that future PD cases showed faster deterioration before diagnosis, most notably in depressive symptoms and verbal fluency, together with abrupt worsening around diagnosis. After diagnosis, PD was associated with accelerated transitions toward more adverse multidomain states and reduced recovery to more favourable states. Our findings corroborate previous evidence on individual prodromal features of PD6,8,14 and extend it by showing that multidomain deficits convey cumulative prognostic information in community-dwelling older adults. This pattern is consistent with the concept that prodromal PD may involve a broader multisystem process rather than isolated dysfunction in a single domain. Recent research has indicated a “diffuse-malignant” PD subtype characterised by early multidomain involvement, including motor dysfunction plus cognitive and autonomic impairments, with more rapid progression and poorer prognosis, which supports our findings15, and our findings are compatible with the possibility that some individuals who later develop PD follow a similarly diffuse prodromal pathway. Widespread α-synuclein-related pathology affecting multiple neural systems may underlie the co-occurrence of physical frailty, depressive symptoms, and cognitive impairment before diagnosis16,17. More broadly, our results support the view that PD is not solely a motor disorder with a late onset, but may emerge against a background of progressive loss of resilience across several functional domains18. The short-term change analyses further reinforce this interpretation. Worsening PFI and EURO-D scores and declining cognitive performance over ~2 years were associated with higher subsequent PD risk, while recovery from poorer multidomain states was associated with lower risk. These patterns should be interpreted cautiously. On the one hand, they may indicate that worsening multidomain health captures an evolving prodromal process, whereas apparent recovery may reflect transient, non-neurodegenerative fluctuations. On the other hand, these findings should not be interpreted as evidence that modifying these domains will necessarily prevent PD. In this context, the most defensible implication is one of risk stratification rather than causality: emerging or persistent deterioration in physical frailty, depressive symptoms, or cognitive function may help identify older adults in whom prodromal PD warrants closer clinical attention. The pre-/post-diagnosis trajectory analysis provides further granularity on how PD unfolds over time. We observed that in the years leading up to diagnosis, individuals who eventually developed PD experienced significantly faster deterioration in physical, cognitive, and especially psychological health compared to those with non-PD. The slope of depressive symptom increase was more than twice as steep in future PD patients, followed by verbal fluency declined about ~2 times faster in the prodromal phase of PD than in normal ageing, which is similar to previous evidence9,19. It is likely not a coincidence that the fastest pre-diagnostic changes we detected were in depressive symptoms and fluency, as both are related to frontal-subcortical circuits modulated by monoaminergic systems (serotonergic, noradrenergic, and dopaminergic dysfunction), which are known to be affected in the early stage of PD pathology20–23. We also documented a step-change in trajectory at PD diagnosis, which was evident across domains. This could reflect a threshold effect where burgeoning pathology crosses the clinical diagnostic boundary, coupled with factors like peri-diagnostic heightened detection or stress that can acutely worsen function. Importantly, we replicated these trajectory patterns in an independent cohort (ELSA), lending robustness to our findings. These findings strengthen the evidence for a prodromal acceleration phase in PD, wherein multidomain health indicators deteriorate faster than normal ageing before diagnosis, followed by an inflexion at diagnosis and a persistent or even steeper post-diagnostic decline. Our multi-state analyses complement these findings by showing that, once PD is present, the disease is associated not only with progression to worse physical, depressive symptom, cognitive, and combined multidomain states, but also with a reduced likelihood of recovery. In the combined multidomain model, PD was associated with higher hazards of transition from Healthy to Single and from Single to Combined, as well as a higher hazard of transition from Combined to Death, while recovery from Combined to Single and from Single to Healthy was less likely. The model-estimated 5-year transition probabilities illustrated the same pattern. Together, these findings suggest that PD is linked to a sustained shift toward more adverse health states across multiple domains, rather than to a decline in a single isolated dimension. This study has several strengths. We used large, population-based longitudinal cohorts with repeated measurements over extended follow-up, enabling evaluation of baseline status, short-term changes, joint trajectories, and diagnosis-anchored pre-/post-diagnosis trajectories within a coherent framework. We also externally validated the trajectory findings in ELSA. In addition, the multi-state models allowed us to quantify how PD was associated with progression and recovery across multidomain health states, and the main findings were broadly robust across complete-case, lagged-event, time-updated, and competing-risk analyses. Our findings also carry important public health and clinical implications. In community and primary-care settings, repeated information on physical frailty, depressive symptoms, and cognitive function is often more feasible to obtain than specialised biomarker or imaging assessments. Our results suggest that concurrent deficits across these domains, as well as their short-term worsening, may help flag individuals who merit closer follow-up for possible prodromal PD. At the same time, these measures should be viewed as complementary risk indicators rather than diagnostic tools, and they are unlikely to substitute for neurological evaluation when PD is clinically suspected. Some limitations should be acknowledged. First, PD cases were identified via self-reported doctor diagnoses, which could miss subclinical or undiagnosed cases and cannot distinguish PD from other Parkinsonian syndromes, although we further excluded individuals with baseline dementia to avoid confusion. In addition, because diagnosis dates were interval-censored between interviews and imputed at the midpoint between the first PD report and the preceding wave, the estimated peri-diagnostic changes may be somewhat blurred. Second, cognitive function was assessed using brief population-based tasks rather than a full neuropsychological battery, and the psychological domain was represented by depressive symptoms rather than broader measures of psychological well-being. Finally, although we observed that worsening multidomain health predicted subsequent PD and that recovery was associated with lower risk, these patterns likely reflect a mixture of prodromal manifestations, general vulnerability, and potentially modifiable factors; they should not be interpreted as establishing causal effects. In conclusion, PD is preceded by a prolonged multidomain prodromal phase characterised by worsening physical frailty, depressive symptoms, and cognitive function, with accelerated deterioration before diagnosis and abrupt worsening around diagnosis. Concurrent deficits across these domains, as well as their short-term worsening and adverse joint trajectories, were associated with higher subsequent PD risk. After diagnosis, PD was associated with progression toward more adverse multidomain health states and reduced recovery to more favourable states. These findings support the value of repeated monitoring of physical frailty, depressive symptoms, and cognitive function in older adults for PD risk stratification and for informing future multidomain early-detection strategies. Methods Study population The primary analysis used the Survey of Health, Ageing and Retirement in Europe (SHARE), a longitudinal study of adults 50 years and older across Europe and Israel24. Validation analyses used the similarly designed English Longitudinal Study of Ageing (ELSA), a nationally representative cohort in England25. All procedures involving human participants were performed in accordance with the Declaration of Helsinki. SHARE received ethics approval from the Ethics Committee of the University of Mannheim for earlier waves and from the Ethics Council of the Max Planck Society for later waves. ELSA received ethical approval from the NHS Research Ethics Committees. All participants provided written informed consent at each interview. For the SHARE-based incident PD analyses, participant selection proceeded sequentially according to the requirements of each analysis (Fig. S2). Among respondents interviewed in SHARE Wave 5 (2013), 49,956 individuals aged 50 years and older with available baseline information and at least one post-baseline follow-up record were included in the multi-state analysis. After further excluding those with prevalent PD or dementia at baseline, 49,039 participants were included in the baseline incident PD analysis. Of these, 30,119 had the additional Wave 4 data required for short-term change analyses, and 11,805 had repeated measurements across Waves 1, 2, 4, and 5 for long-term trajectory analyses. For the pre- and post-diagnosis trajectory analyses, 23,848 SHARE participants interviewed in Wave 2 (2006–2010) and 7770 ELSA participants interviewed in Wave 3 (2006/2007) were included (Fig. S2). In both cohorts, participants were required to be aged 50 years and older, free of PD and dementia at baseline, have complete baseline data on the relevant physical, depressive symptom, and cognitive measures, and have at least one follow-up interview. Participants were followed through SHARE Wave 9 (2021/2022) and ELSA Wave 10 (2021–2023), respectively, to align baseline and follow-up windows. Ascertainment of Parkinson’s disease and death In both cohorts, PD was determined based on self-reported physician diagnoses at each wave. For participants with incident PD, the diagnosis date was imputed as the midpoint between the first wave reporting PD and the preceding interview. In SHARE, vital status was additionally ascertained through end-of-life proxy interviews. Construction of physical frailty index (PFI) in SHARE SHARE-based frailty tools, such as the five-item phenotype screening instrument (SHARE-FI) and the 40-item deficit-accumulation index (SHARE-FIx) constructed by Romero-Ortuno et al.,26,27 were originally derived using Wave 1 data. Additionally, Rockwood et al.28 developed a 70-item deficit-accumulation index based on SHARE Waves 1 and 2. However, their application in later waves can be limited because some items are not consistently available across all waves. Although deficit-accumulation FIs have been computed longitudinally across multiple SHARE waves to map frailty prevalence in European populations29, to our knowledge, no index has been built exclusively from physical-domain deficits. Following Rockwood and colleagues’ recommendations on FI construction30,31, we developed a PFI based on the deficit-accumulation model that can be applied consistently across SHARE waves (excluding SHARELIFE), in order to meet our study demands. We first enumerated health-related variables that were present in all regular SHARE waves (excluding SHARELIFE), and recoded each to the [0,1] interval so that 1 always denoted a deficit (Table S5). PD was not included as it was the study outcome. For IADL/mobility limitations, earlier SHARE-based frailty tools often used a subset of functional items to keep domain balance and avoid redundancy26,27. To preserve broad physical-domain coverage and enable consistent computation across waves, we initially considered the full set of functional items and then excluded four IADL items (making telephone calls, taking medications, managing money, and using a map to figure out how to get around) because these tasks may rely more heavily on cognitive abilities than on physical capacity alone. Low physical activity was operationalised from SHARE’s behavioural frequency item. Although behaviour variables are generally discouraged when they do not reflect a health problem, low activity is a core physical phenotype component (used in SHARE-FI)26 and is widely interpreted as a manifestation of limited physical capacity in older adults; we therefore retained it. Table S6 displayed the item-level missingness and prevalence (score = 1). Two items were dropped due to extreme prevalence, consistent with recommended thresholds (<1% or >80% deficit)30: Nursing home stay in the last 12 months (0.35%) and Doctor visits in the last 12 months (89.49%). Low handgrip strength (HGS), defined using sex- and BMI-specific cut-offs32, was excluded given >5% missing. We further graphed each item’s deficit prevalence by age (Table S7). Most items rose with age, as expected; however, low body mass index did not increase strictly monotonically and was thus excluded. For redundancy screening, all 42 remaining items were dichotomised as deficit present versus not, and then examined using the phi coefficient. The largest phi value was 0.57, observed between shopping for groceries and doing work around the house or garden (Fig. S3), which was well below the r > 0.95 threshold recommended for excluding near-duplicates when building an FI30; therefore, no items were removed at this step. In summary, we finally determined 42 items when constructing the PFI (see Table S8). The distribution of PFI was right-skewed and closely approximated by a gamma-like distribution (Fig. S4); the 99th percentile was 0.65 ( < 0.7), consistent with the submaximal frailty limit reported across FI studies31,33. The PFI increased with age (Spearman correlation coefficient = 0.39; log-PFI regressed on age yielded standardised R² = 0.13). PFI alone achieved Harrell’s C = 0.718 for all-cause death (Table S9). Adding age and sex increased discrimination to 0.799. In age- and sex-adjusted Cox models, categorised PFI showed a strong, graded association with death: pre-frail hazard ratio (HR) and 95% confidence interval (CI) = 1.55 (1.43–1.68) and frail HR = 3.25 (3.01–3.51) versus non-frail for all-cause death; corresponding 5-year HRs were 1.53 (1.36–1.71) and 3.61 (3.25–4.01) (Table S10). For each participant at each wave, we computed PFI as (sum of item deficits) / (number of items observed), provided the individual had ≤20% missing items (i.e., at least 80% of the 42 items observed). Physical frail status was classified as non-frail (PFI ≤ 0.10), pre-frail (0.10 < PFI ≤ 0.21), and frail (PFI > 0.21)34,35. Depressive symptoms and cognitive function in SHARE Depressive symptoms were used to represent the psychological domain. In SHARE, depressive symptoms were assessed using the 12-item European Depression Scale (EURO-D), a validated instrument covering mood, pessimism, hopelessness, loneliness, sleep, and related symptoms (score range 0–12)36. Depressive symptom status was classified as no (0–3), mild (4-5), and severe (≥6)37. Cognitive function was assessed using immediate and delayed 10-word recall and verbal fluency (animals named in 1 min). These brief cognitive tasks are consistently available in both SHARE and ELSA and have been widely used in population-based ageing studies to capture memory and executive function. To place the three tests on a common scale and facilitate comparisons across waves and between cohorts, each cognitive component was standardised using the cohort-specific baseline distribution, and the standardised component scores were averaged to derive a composite cognitive z score38. Thus, the cognitive z score was used as a relative measure of cognitive performance within each cohort rather than as a clinical normative score. Cognitive status was classified as no (>−1), mild (−1.5 to −1.0), and severe (≤−1.5). Multidomain health status in SHARE Multidomain health status was defined as mutually exclusive categories: Healthy; Physical (physically frail only); Psychological (mild or severe depressive symptoms only); Cognitive (mild or severe cognitive impairment only); Psychological-Cognitive; Physical-Psychological/Cognitive (physically frail plus one additional impaired domain); and Physical-Psychological-Cognitive (all three domains impaired). For the transition analyses, these categories were collapsed into Healthy, Single (one impaired domain), and Combined (at least two impaired domains). Physical frailty, depressive symptoms, and cognitive function in ELSA In ELSA, physical frailty was measured using a 38-item PFI derived from a validated longitudinal frailty-index framework, using the same coding principles and denominator rule (≥80% non-missing; Table S11)39. Depressive symptoms were measured using the 8-item Centre for Epidemiologic Studies Depression scale (CESD-8)40. Cognitive function was assessed using the same word-recall and animal naming tasks as in SHARE. To enhance cross-cohort comparability, cognitive measures in ELSA were standardised to the cohort-specific baseline distribution, and the standardised component scores were averaged to derive a composite cognitive z score, using the same approach as in SHARE. Covariates Covariates were obtained from structured interviewer-administered questionnaires at baseline. In SHARE, these included age, sex, residence (urban/rural), education (less than upper secondary, upper secondary/vocational, or tertiary), household assets per capita (tertiles), marital status (married/partnered, separated/divorced/widowed, or never married), smoking (current, former, or never), and alcohol use (none or less than monthly, monthly, weekly, or daily). In ELSA, residence was unavailable, and alcohol use was coded as never or ever. Statistical analysis We used Cox proportional hazards models to estimate HRs and 95% CIs for incident PD from Wave 5 to Wave 9 in relation to baseline physical frailty, depressive symptoms, cognitive impairment, and combined multidomain health status. Models were adjusted for age, sex, residence, education, household assets per capita, marital status, smoking, and drinking; missing covariate values were retained as separate categories. For short-term changes from Wave 4 to Wave 5, we examined both continuous changes in PFI, EURO-D score, and cognitive z score and categorical changes in combined multidomain health status. Continuous changes were defined as the Wave 5 value minus the Wave 4 value and were analysed using Cox proportional hazards models with restricted cubic splines (four knots) to allow for potential non-linearity; these models were additionally adjusted for the corresponding Wave 4 measure. Categorical changes were analysed using Cox proportional hazards models with the same covariate adjustment as in the primary analysis. To characterise long-term multidomain patterns, we applied group-based trajectory modelling (GBTM)41,42 to identify joint trajectories of PFI, EURO-D score, and cognitive z score across Waves 1, 2, 4, and 5. The number of trajectory groups was selected on the basis of a lower Bayesian Information Criterion together with clinical interpretability. Associations between trajectory-group membership and incident PD were then estimated using Cox proportional hazards models adjusted for the same covariates. To examine how PD was associated with changes in health-state transitions, we fitted continuous-time multi-state Markov models for physical frailty, depressive symptoms, cognitive impairment, and the combined multidomain status43. These models are appropriate for panel-observed longitudinal data in which health states are recorded at intermittent interviews and the exact transition times are not directly observed. The models assumed that transition intensities depended on the participant’s current state. State definitions are shown in Fig. S5. We allowed bidirectional transitions between living states and transitions from any living state to death; death was treated as an absorbing state. PD was modelled as a time-dependent exposure. Transition-specific HRs were estimated with full covariate adjustment, and model-estimated 5-year transition probabilities were derived from the fitted transition intensity matrices for illustrative presentation at age 65 (Fig. S6). To characterise trajectories before and after PD diagnosis, we fitted mixed-effects models in SHARE and, for validation, in ELSA38,44. Outcomes included PFI, depressive symptom score, global cognitive z score, and cognitive subtests. Fixed effects included time since baseline, an indicator for incident PD during follow-up, the interaction between time and incident PD, an indicator for the post-diagnosis period, time since diagnosis, and covariates. Random effects included a participant-specific random intercept and random slopes for time since baseline and time since diagnosis. Models in SHARE were adjusted for age, sex, residence, education, household assets per capita, marital status, smoking, and drinking; those in ELSA were adjusted for the corresponding covariates except residence. Predicted trajectories were generated for a reference profile of a 65-year-old woman with upper secondary/vocational education, median household assets per capita, married/partnered, never smoking, and no or less than monthly alcohol use in SHARE and never drinking in ELSA; residence was set to urban in SHARE. We performed four sensitivity analyses: complete-case analysis excluding participants with missing covariate data; exclusion of incident PD occurring within 4 years after baseline to reduce potential reverse causation; time-dependent Cox models updating physical frailty, depressive symptoms, cognitive impairment, and combined multidomain health status during follow-up; and Fine-Gray subdistribution hazards models treating all-cause death as a competing event. This study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. Analyses were performed using SAS version 9.4 (SAS Institute) and R version 4.4.3. All tests were two-sided, and P values < 0.05 were considered statistically significant. Supplementary information Supplementary Materials STROBE-checklist Acknowledgements We appreciate the efforts made by the original data creators, depositors, copyright holders, and funders of the data collections and their contributions to accessing data from SHARE and ELSA. This study is funded by the National Natural Science Foundation of China (22376006, 82071552) and Demand and Utilization of Health Services for the Elderly Population in China (2023P253QG041). The funding source had no involvement in the collection, analysis, or interpretation of data; in the writing of the report; or in the decision to submit the paper for publication. Author contributions Z.R., J.L., and X.Z. conceptualised and designed the study. Z.R. managed and analysed the data with the assistance of Z.T. and X.L. Z.R., X.L., Y.S. and Y.Z. visualised the data. Z.R. prepared the first draft and was responsible for revising the manuscript. J.L., X.Z., J.Y., J.C. and D.Z. contributed significantly to data interpretation and commented on the initial and subsequent revisions of the manuscript. J.L., Z.R., Z.T., X.L. and Y.S. had full access to the data in the study. All authors read and approved the final version of the manuscript and accepted responsibility for the decision to submit for publication. Data availability Data for the study came from the SHARE and ELSA project and are available to all researchers for purely scientific purposes upon request on the website: https://share-eric.eu/data/ and https://www.elsa-project.ac.uk/. Code availability The underlying code for this study is available on GitHub and can be accessed via https://github.com/RenzySleeping/Physical-psychological-and-cognitive-decline-before-and-after-Parkinson-s-disease. Detailed codes are available from the corresponding author upon reasonable request. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Jufen Liu, Email: liujufen@bjmu.edu.cn. Junliang Yuan, Email: junliangyuan@bjmu.edu.cn. Xiaoying Zheng, Email: zhengxiaoying@sph.pumc.edu.cn. Supplementary information The online version contains supplementary material available at 10.1038/s41531-026-01419-3. References Associated Data Supplementary Materials Supplementary Materials STROBE-checklist Data Availability Statement Data for the study came from the SHARE and ELSA project and are available to all researchers for purely scientific purposes upon request on the website: https://share-eric.eu/data/ and https://www.elsa-project.ac.uk/. The underlying code for this study is available on GitHub and can be accessed via https://github.com/RenzySleeping/Physical-psychological-and-cognitive-decline-before-and-after-Parkinson-s-disease. Detailed codes are available from the corresponding author upon reasonable request.