Prediction of cognition in Parkinson's disease with a clinical-genetic score: a longitudinal analysis of nine cohorts

Abstract

Background: Cognitive decline is a debilitating manifestation of disease progression in Parkinson's disease. We aimed to develop a clinical-genetic score to predict global cognitive impairment in patients with the disease.

Methods: In this longitudinal analysis, we built a prediction algorithm for global cognitive impairment (defined as Mini Mental State Examination [MMSE] ≤25) using data from nine cohorts of patients with Parkinson's disease from North America and Europe assessed between 1986 and 2016. Candidate predictors of cognitive decline were selected through a backward eliminated Cox's proportional hazards analysis using the Akaike's information criterion. These were used to compute the multivariable predictor on the basis of data from six cohorts included in a discovery population. Independent replication was attained in patients from a further three independent longitudinal cohorts. The predictive score was rebuilt and retested in 10 000 training and test sets randomly generated from the entire study population.

Findings: 3200 patients with Parkinson's disease who were longitudinally assessed with 27 022 study visits between 1986 and 2016 in nine cohorts from North America and Europe were assessed for eligibility. 235 patients with MMSE ≤25 at baseline and 135 whose first study visit occurred more than 12 years from disease onset were excluded. The discovery population comprised 1350 patients (after further exclusion of 334 with missing covariates) from six longitudinal cohorts with 5165 longitudinal visits over 12·8 years (median 2·8, IQR 1·6-4·6). Age at onset, baseline MMSE, years of education, motor exam score, sex, depression, and β-glucocerebrosidase (GBA) mutation status were included in the prediction model. The replication population comprised 1132 patients (further excluding 14 patients with missing covariates) from three longitudinal cohorts with 19 127 follow-up visits over 8·6 years (median 6·5, IQR 4·1-7·2). The cognitive risk score predicted cognitive impairment within 10 years of disease onset with an area under the curve (AUC) of more than 0·85 in both the discovery (95% CI 0·82-0·90) and replication (95% CI 0·78-0·91) populations. Patients scoring in the highest quartile for cognitive risk score had an increased hazard for global cognitive impairment compared with those in the lowest quartile (hazard ratio 18·4 [95% CI 9·4-36·1]). Dementia or disabling cognitive impairment was predicted with an AUC of 0·88 (95% CI 0·79-0·94) and a negative predictive value of 0·92 (95% 0·88-0·95) at the predefined cutoff of 0·196. Performance was stable in 10 000 randomly resampled subsets.

Interpretation: Our predictive algorithm provides a potential test for future cognitive health or impairment in patients with Parkinson's disease. This model could improve trials of cognitive interventions and inform on prognosis.

Funding: National Institutes of Health, US Department of Defense.

Figure 1. Flow chart of study design

HBS=Harvard Biomarkers Study, DATATOP=Deprenyl and Tocopherol Antioxidative Therapy of Parkinsonism; PreCEPT=Parkinson Research Examination of CEP-1347 Trial/A Longitudinal Followup of the PRECEPT Study Cohort; CamPaIGN=Cambridgeshire Parkinson’s Incidence from GP to Neurologist; PICNICS=Parkinsonism Incidence, Cognition and Non-motor heterogeneity in Cambridgeshire; DIGPD=Drug Interaction with Genes in PD; PROPARK=PROfiling PARKinson’s disease study; PDBP=Parkinson’s Disease Biomarkers Program; PPMI=Parkinson’s Progression Marker Initiative (PPMI). AUC=area under the curve.

Figure 2. Prediction of global cognitive impairment

(A,B) The cognitive risk score showed high accuracy (quantified by AUC estimates) for predicting, whether a patient will develop global cognitive impairment within ten years from disease onset in the discovery and the validation populations. (A) In the discovery population, 1,350 patients with PD and MMSE > 25 at baseline were followed with 5,165 visits for up to 12.8 (median, 2.8) years. (B) In the independent validation population, 1,132 patients with PD and MMSE > 25 at baseline were followed with 19,127 visits for up to 8.6 (median, 6.5) years. Sensitivity and specificity at the cutoff score of 0.196 (the optimal cutoff identified in the discovery population) are shown for both populations. (C,D) Covariate-adjusted Kaplan-Meier curves for survival free of global cognitive impairment. (C) In the discovery population, 95.8% (95% CI 92.7%–99.1%) of patients with PD in the lowest (first) quartile of scores survived for ten years without global cognitive impairment compared to 34.9% (95% CI 26.5%–46.2%) of those scoring in the highest (fourth) quartile. (D) In the validation population, 96.3% (95% CI 94.1% – 98.6%) of patients in the lowest quartile of scores survived for ten years without global cognitive impairment compared to 27.4% (95% CI 12.6%–59.8%) of patients scoring in the highest quartile scores. To ensure consistency across studies, an MMSE score with the cutoff of ≤ 25 was taken as an indicator of significant global cognitive impairment as recommended by the International Parkinson and Movement Disorders Society (MDS) Task Force.

Figure 3. Prediction of dementia

Beyond cognitive decline, the clinical-genetic score predicts risk of dementia in individuals with PD in validation population. 1,132 patients (without global cognitive impairment at baseline) with 19,127 longitudinal study visits were available for this analysis. The accuracy of the clinical-genetic score for predicting dementia was high with an AUC of 0.877 (95% CI 0.788–0.943). Sensitivity and specificity for predicting dementia at the cutoff (0.196; as predefined in the discovery population).

Figure 4. Stability of the score

To test the stability of the predictive score, we rebuilt and retested the score model in 10,000 randomly generated training and test subsets. In each iteration the entire population of patients was randomly split into a training and a test set pair. In each iteration, we rebuilt the predictive score ab initio in the training set, eliminated predictor variables based on the Akaike information criterion, and used it to predict global cognitive decline in the corresponding test set. (A) In 10,000 iterations, age at onset, enrollment MMSE score, and years of education remained in the score model after stepwise pruning in 100% of iterations, enrollment MDS-UPDRS III in 98.30%, GBA carrier status in 91.79%, depression in 90.61%, and gender in 78.52%. HY stage (which did not make it into our clinical-genetic score) was included in 34.86% of iterations. (B) Across the 10,000 re-sampled test sets, the mean AUC was 0.833 (95% CI, 0.785–0.876) for predicting global cognitive impairment and, even higher, 0.872 (95% CI, 0.817–0.929) for dementia. These data indicate stable variable selection and score performance.

Figure 5. Prediction of longitudinal trajectories of Mini Mental State Exam (MMSE) and Montreal Cognitive Assessment (MoCA) scores

(A) To evaluate longitudinal trajectories of serial MMSE scores in the patients with high (> 0.196) vs. low enrollment cognitive predictive scores (≤ 0.196) at enrollment, we performed a generalized mixed random and fixed effects longitudinal meta-analysis adjusting for disease duration at enrollment. These analyses were conservatively restricted to the validation population. PD patients with high enrollment predictive scores had a significantly more rapid longitudinal decline in MMSE scores over time with p < 0.0001 compared to the patients with low enrollment predictive scores. Illustrative mean MMSE scores across time predicted from the estimated fixed effect parameters in the mixed random and fixed effects model analysis are shown for Parkinson’s patients with low clinical-genetic predictive scores (blue) and those with high scores at enrollment (red). Patients with high scores (measured at enrollment) had a more rapid decline in cognitive function (as measured by serial MMSE) compared to those with low scores with p < 0.0001 adjusting for duration of PD at enrollment. Illustrative MMSE values for a mean disease duration at enrollment are shown. CRS=global cognitive impairment predictive score. (B) Illustrative mean MoCA scores across time predicted from the estimated fixed effect parameters in the mixed random and fixed effects model analysis in Parkinson’s patients with low clinical-genetic predictive scores (≤ 0.196; blue) and those with high scores at enrollment (> 0.196; red). Patients with high clinical-genetic scores at enrollment had a more rapid decline in MoCA scores compared to those with low scores with p < 0.0001 adjusting for disease duration at enrollment. The analysis of MoCA scores was restricted to PPMI and PreCEPT, the two validation cohorts, which had included this scale into their assessment battery. MoCA scores were not collected in DATATOP. Illustrative MoCA values for a mean disease duration at enrollment are shown. (C) Improved power for clinical trials in populations with elevated clinical-genetic scores. Enriching populations based on the clinical-genetic score >0.196 for trials of therapeutics designed to address cognitive impairment in PD will reduce the required sample size by 6-fold compared to an equally powered trial without enrichment. In this hypothetical power estimate, required sample sizes were 137 for the placebo and 137 for the experimental treatment group in order to achieve 80% power. A traditional clinical trial of any PD patients (not enriched based on the clinical-genetic score) would require 801 patients per group to achieve the same power (over the same three-year time period, assuming same α, standard deviation, and test-retest correlations). α= 0.05 for detecting the difference in trajectories for MoCA across time for the placebo vs. the treatment group (group × time interaction).