Predicting 10-year risk of chronic kidney disease in lithium-treated patients with bipolar disorder: A risk model...

Lithium is a first-line maintenance treatment for bipolar-disorder (BD) but has increased risk for chronic-kidney-disease (CKD). There is a paucity of research on risk-model development predicting CKD during/following lithium treatment, and none was conducted in Asian regions. This study aimed to derive and validate 10-year risk prediction model for CKD-stage 3 in first-diagnosed BD patients receiving ≥ 1 prescription of lithium during 2002-2018 in Hong-Kong, using electronic-medical-record database of public-healthcare services. Literature-informed predictor selection included demographics, physical comorbidities, mean lithium serum-levels and non-lithium psychotropic use. The risk-equation was developed using Least-Absolute-Shrinkage-and-Selection-Operator (LASSO) Cox-proportional hazards regression model with 4-fold internal cross-validation over 1,000 iterations. We identified 2,258 lithium-treated BD patients, with CKD incidence of 12.6 per 1000 person-years (95 %CI=11.1-14.4) over a median follow-up of 7.7 years (interquartile range=3.7-12.3). Our results showed that older age at BD-diagnosis, male sex, physical comorbidities, higher mean lithium serum-level, fewer antipsychotic and mood-stabilizing anticonvulsant use, and greater antidepressant exposure were independent risk factors predicting CKD, with an event-per-variable ratio of 25.2. The 10-year risk prediction model had satisfactory area-under-the-curve (AUC) (0.74 [95 %CI=0.66-0.83]), with good calibration (calibration slope=0.88 [95 %CI=0.61-1.15]; observed/expected risk ratio=1.14 [95 %CI=0.86-1.42]), and discrimination performances (Harrell's C-index=0.75 [95 %CI=0.68-0.82]; Royston and Sauerbrei's D statistic=1.45 [95 %CI=0.99-1.92]). In conclusion, this CKD risk-model for lithium-treated BD patients demonstrated satisfactory prediction performance in a predominantly-Chinese population. Further research including external validation is needed to verify model performance to facilitate implementation of this CKD risk prediction tool for individualized clinical decision-making and outcomes in real-world practice.