Algorithm Accurately Predicts Cognitive Decline in Parkinson's Disease
The investigators hope that the risk score will be a good tool for clinical trial design, and eventually clinical practice.
Prediction of cognitive decline within 10 years of onset of Parkinson's disease (PD) is now possible through an analytic model developed by a multinational collaborative of investigators. The algorithm provides a set of clinical-genetic scores that accurately predict future dementia or disabling cognitive impairment, according to results of a large-scale longitudinal analysis published in the Lancet Neurology.
Cognitive decline is 1 of the long-term outcomes of PD that has been largely unmanageable, and clinical trials are hampered by the highly varied rates of progression to cognitive clinical endpoints. The predictive algorithm designed by the International Genetics of Parkinson's Disease Progression Consortium can facilitate effective risk stratification of patient populations to improve comparative results of clinical trials for disease-modifying therapies.
A total of 3200 patients from 9 international study cohorts (based on visits from 1986 to 2016) were longitudinally assessed for eligibility, including a discovery population of 1350 patients (5165 first visits over the course of 12.8 years) from 6 cohorts and a replication population of 1132 patients (19,127 follow-up visits over the course of 8.6 years) from the remaining 3 cohorts.
The investigators combined data from 7 known contributing factors to cognitive impairment, including age at diagnosis of PD, sex, educational level, and GBA mutation status, as well as depression, Mini-Mental State Examination, and Movement Disorders Society Unified Parkinson's Disease Rating Scale Part III scores at the time of enrollment, in the algorithm to arrive at 4 quartiles of risk: 1 (0 to <0.0954), 2 (0.0954 to <0.1958), 3 (0.1958 to <0.3789), and 4 (0.3789 to ≤1).
Risk for global cognitive impairment within 10 years of PD onset was highest for patients scoring in the highest quartile (4) compared with the lowest (hazard ratio [HR] 21.9; 95% CI, 6.5-73.1). Conversely, survival for 10 years without global cognitive impairment was dramatically higher in the lowest quartile compared with the highest (95.8% [95% CI, 92.7%-99.1%] vs 34.9% [95% CI, 26.5%-46.2%). Similarly, those in the highest quartile had an increased hazard ratio for dementia (HR, 21.9; 95% CI, 6.5-73.1) compared with those in the lowest quartile. Notably, 98.9% of patients in the lowest quartile survived for 10 years without dementia compared with 48.3% of patients in the highest quartile (P <.0001).
The accuracy of prediction of global cognitive impairment and dementia in the model was high, with an area under the curve of 0.86 (95% CI, 0.82-0.90) for cognitive impairment in the discovery cohort, showing high sensitivity (0.87 area under the curve; 95% CI, 0.80-0.92) and good specificity (0.72 area under the curve; 95% CI, 0.65-0.78). The area under the curve for prediction of dementia was even greater, with a sensitivity of 0.86 (95% CI, 0.72-0.94) and specificity of 0.72 (95% CI, 0.59-0.84). Similar rates were reported in the replication cohort. Performance of the algorithm showed excellent stability when retested in 10,000 random training sets.
This predictive algorithm demonstrates good potential for use to address ongoing problems in clinical trial design for PD, as well as ultimate utility in general clinical settings. The consortium plans to develop the cognitive risk score calculator further to capture new predictive factors as they are discovered.
Liu G, Locascio JJ, Corvol JC, et al. Prediction of cognition in Parkinson's disease with a clinical-genetic score: a longitudinal analysis of nine cohorts. Lancet Neurol. 2017;16:620-629.