Risk prediction models that use routinely collected electronic health data: generalisable and useful in heterogeneous settings?
2019
Background Healthcare sites (practices) may vary in quality of data recording and populations. Risk predictions models that use routinely collected data do not test generalisability to heterogeneous practices. Aim To assess the extent of variability between practices on individual patients’ risk of cardiovascular disease (CVD) that is not taken into account by the risk prediction model QRISK3. Method A longitudinal cohort study from 1 Jan 1998 to Jan 2015 in 392 general practices (including 3.6 million patients) from the Clinical Practice Research Datalink (CPRD). This was a shared frailty model to incorporate QRISK3 predictors, practice variability, and simulations to measure random variability. Results There was considerable variation in data recording between general practices. Practices on 5th percentile of missingness of body mass index have 18.7% patients with missing values and 60.1% on the 95th percentile. The crude incidence rates also varied considerably between practices (from 0.4 to 1.3 CVD events per 100 patient-years, respectively). The estimates of individual CVD risks with the random effect model were inconsistent with the estimated QRISK3 risk. For patients with a QRISK3 CVD risk of 10%, the 95% range of predicted risks were between 7.2% and 13.7% with the random effects model. Random variability only explained a small part of this. The random effects model was similar to QRISK3 for discrimination and calibration. Conclusion Risk prediction models that use routinely collected electronic health data can have limited generalisability and accuracy in predicting individual patient risks in heterogeneous settings. They need to be based on more robust evidence on causal risk factors.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
0
References
0
Citations
NaN
KQI