Abstract: TH-PO008
Predicting the Risk of Albuminuria in Patients with Diabetes: A Systematic Review and Development Study
Session Information
- Augmented Intelligence for Prediction and Image Analysis
October 24, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Janse, Roemer J., Leids Universitair Medisch Centrum, Leiden, Zuid-Holland, Netherlands
- Oomen, Marretje W., Leids Universitair Medisch Centrum, Leiden, Zuid-Holland, Netherlands
- Ramspek, Chava L., Leids Universitair Medisch Centrum, Leiden, Zuid-Holland, Netherlands
- Dekker, Friedo W., Leids Universitair Medisch Centrum, Leiden, Zuid-Holland, Netherlands
- Carrero, Juan Jesus, Karolinska Institutet, Stockholm, Stockholm, Sweden
- van Diepen, Merel, Leids Universitair Medisch Centrum, Leiden, Zuid-Holland, Netherlands
Background
Patients with type 2 diabetes mellitus (T2DM) are at high risk of kidney disease, which may be ameliorated by early detection of albuminuria. Clinical prediction models (CPMs) can tailor guideline screening to the individual patient. To allow this, we aimed to 1) identify available CPMs and their shortcomings and 2) predict albuminuria risk, albuminuria-free time, and progression through albuminuria stages.
Methods
We systematically identified albuminuria CPMs and appraised their risk of bias (ROB). New models were developed in the Stockholm Creatinine Measurements cohort. We divided Stockholm residents with an albumin/creatinine test between 2006-2021 with T2DM and no albuminuria into a development and temporal validation cohort. Predictors were selected based on clinical expertise, literature, and previous CPMs. We predicted micro- and macroalbuminuria within 3 years. We predicted 1) the risk using a Fine-Gray CPM accounting for the competing risk of death, 2) the expected albuminuria-free days using an accelerated failure time (AFT) CPM, and 3) the risk of different albuminuria stages using a multi-state CPM. CPM discrimination and calibration were assessed in the development and validation cohorts.
Results
We identified 9 studies reporting on 11 CPMs. Most models were at high ROB. We developed new models on 38649 individuals with 6904 events and validated these in 45009 individuals with 6499 events. The Fine-Gray CPM had adequate discrimination internally (C-statistic, 95%CI; 0.64, 0.64-0.65) and temporally (0.66, 0.66-0.67). Calibration was good. The AFT CPM had adequate discrimination (internally: 0.63, 0.62-0.63; temporally: 0.65, 0.64-0.65), but poor calibration. An individual prediction from the multi-state CPM can be seen in Figure 1.
Conclusion
Predicting albuminuria in T2DM patients allows tailoring albuminuria screening to the individual. We developed multiple CPMs that provide different information on the risk of albuminuria within 3 years. These models can serve to improve albuminuria screening and ameliorate the risk of kidney damage in patients with T2DM.