Personalised prediction of institutionalisation in Parkinson's: prognostic factor identification and model development and validation using IPD meta-analysis.

BACKGROUND

People with Parkinson's (PwP) who lose independence may need care in nursing homes or similar institutions if home care is insufficient. Institutionalisation has major social and financial implications. Better understanding of which PwP are most likely to be institutionalised would improve information provision, clinical risk stratification, and healthcare planning.

OBJECTIVES

To identify risk factors for institutionalisation in PwP and develop models predicting individual institutionalisation risk.

METHODS

We described institutionalisation in the Parkinson's Incidence Cohorts Collaboration, comprising 6 European incidence cohorts. We identified prognostic factors by two-stage individual-participant-data meta-analysis. Prognostic models predicting risk of institutionalisation within 7 years and 10 years were developed using the Royston-Parmar model. Heterogeneity in model performance was assessed using internal-external cross validation (IECV).

RESULTS

In 1046 PwP, the cumulative incidence of institutionalisation by 10 years was 37.2%. The incidence rate ranged from 1.7 to 6.2 per 100 person-years. Older age, higher MDS-UPDRS part 3 and lower MMSE at baseline independently predicted higher institutionalisation risk. IECV showed good discrimination in the 10-year (C-statistics 0.73-0.81) and 7-year (0.71-0.84) models. However, calibration (agreement between predictions and observed outcomes) showed under- and over-prediction across studies. After updating model intercept and coefficients (recalibration), the calibration improved.

CONCLUSION

37% of PwP entered institutional care within ten years from diagnosis. Older age, higher MDS-UPDRS part 3 and lower MMSE predicted institutionalisation. The prognostic models discriminated well, but calibration varied between cohorts. We recommend further validation before applying the models in other settings.