Artikel
Towards predictive modelling in an individual patient data meta-analysis (IPD-MA) of older patients with chronic prescriptions in general practice (PROPERmed)
Suche in Medline nach
Autoren
Veröffentlicht: | 20. März 2019 |
---|
Gliederung
Text
Background/research question: Multimorbidity and inappropriate polypharmacy are common in older patients and may lead to medication-related harm resulting in hospitalization and worsening of health-related quality of life (HRQoL). Considering the complex interplay between conditions and medications in this heterogeneous population, we aim to develop and validate an instrument to identify older patients at risk for such harms using an individual participant data meta-analysis (IPD-MA). Meticulous data harmonization is of key importance in advance of prediction modelling and will be reported here.
Methods: This IPD-MA includes data of five cluster-RCTs on medication optimization in general practice, conducted in Germany and The Netherlands from 2009 onwards. Eligible patients were ≥60 years old, had ≥1 chronic prescription and ≥1 health-related problem. Candidate predictors include socio-demographics, morbidity, medication-related parameters (e.g., European list of potentially inappropriate medications, EU-PIM), pain, depression, and HRQoL at baseline.
Results: We included 4,608 patients out of 6,160; main reason for exclusion was not having a chronic disease or prescription at baseline. Median age was 78 years and 58% were female. Most lived at home (95%), 89% had a low/medium educational level, 43% were current smokers, 68% reported pain and 15% had depressive symptoms. Median numbers of chronic diseases, prescriptions and inappropriate prescriptions (EU-PIM) per patient were three, seven and one resp. We either clinically harmonized original study variables measured by different instruments by transforming the original scale (e.g. pain or depressive symptoms) or statistically standardizing them (e.g. functional status). Information on smoking status was missing in two studies, overall there were ≤3% sporadic missing data and the fraction of incomplete cases ranged from 0.005 (age) to 0.631 (Charlson Comorbidity Index).
Conclusions: This will be the first IPD-MA-based prognostic model for older patients with chronic medication use in general practice. Differences in data collection and outcome measures across studies challenged data harmonization. Future work will consist of using multivariable logistic regression to predict worsening of HRQoL and hospital admissions at 6-month follow-up. We will handle systematically and sporadically missing values using multilevel multiple imputation. We will assess risk of bias within and across trials using the PROBAST tool.
Competing interests: The authors declare no competing interest.