Objective: To develop and validate a prognostic model to predict mortality in Parkinson’s disease (PD).

Background: Most previous prognostic studies in PD have used unrepresentative cohorts with selection biases resulting in heterogeneous mortality estimates and limited identification of prognostic factors. Prognosis is best studied by long-term follow-up of community-based incident cohorts. Prognostic models can be used to predict patients’ risk, correct for case mix, and improve clinical trial design (eg, selection of patients or stratification of randomisation by prognosis), but no prognostic model (for any outcome) has been published for use in PD.

Methods: The PINE study and the ParkWest study are both community-based, incidence cohorts of PD with lifelong prospective follow-up in North-East Scotland and Western Norway, respectively. In each study area, all incident cases were identified with multiple, community-based ascertainment strategies (during 2004-6 & 2006-9 in PINE and 2004-7 in ParkWest). Weibull regression was used to create a prognostic model for mortality using data from the PINE cohort with clinical predictors measured at time of diagnosis. It was externally validated by measuring calibration (observed versus predicted survival in quantiles of predicted risk) and discrimination (C-statistic) in the ParkWest cohort.

Results: In the PINE study, 91 of 198 incident PD cases died after follow-up between 6 and 12 years. Independent baseline prognostic factors were age, male sex, higher co-morbidity (with effect in the earlier part, but not later part, of follow-up), more bradykinesia, and more axial relative to limb features (see table

Prognostic model for mortality

Baseline variable	Hazard ratio (95% confidence interval)	P-value
Age at diagnosis (10-year increase)	1.91 (1.43–2.54)	<0.001
Sex (male vs female)	1.84 (1.17–2.91)	0.009
Bradykinesia score (5-point increase)	1.40 (1.14–1.72)	0.002
Axial:limb ratio (1-unit increase)	1.81 (1.22–2.68)	0.003
Charlson score (effect up to 4 years)	1.33 (1.14–1.56)	<0.001
Charlson score (effect after 4 years)	0.83 (0.63–1.09)	0.17
Shape parameter	2.36 (1.96–2.83)	<0.001
Constant	-5.86 (-6.8–-4.89)	<0.001

for hazard ratios). The model had good discrimination (C-statistic=0.77) and had very good calibration in the PINE cohort. In the ParkWest cohort, 37 of 192 PD cases died after follow-up between 6 and 8 years. The model had good discrimination in the ParkWest cohort (C-statistic=0.77) but tended to underestimate survival (see calibration plots in figure).

Conclusions: Overall mortality risk was higher in the PINE study. The prognostic model had good discrimination and can therefore be used for case-mix correction or patient selection/stratification according to prognosis in trials. Because the model tended to underestimate survival in the validation cohort, recalibration of the model may be necessary before it is used for individualised risk prediction in other settings.

« Back to 2016 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/development-and-external-validation-of-a-prognostic-model-of-mortality-in-parkinsons-disease/