Using multi-trait polygenic scores to predict lithium responsiveness in patients with bipolar disorder.

Polygenic scores computed for 60 phenotypes, grouped into five clinical-biological clusters, are combined by machine learning to predict lithium response.

The authors built a multi-trait polygenic score in 2,367 bipolar patients treated with lithium for at least six months within the International Consortium on Lithium Genetics, with response measured on the Alda scale as a continuous total (0-10) and as a categorical outcome (favourable response at 7 or above). Polygenic scores were computed for 60 phenotypes across five clusters: psychiatric and behavioural (23), cardiometabolic (17), autoimmune/inflammatory (5), neurocognitive (8) and renal function (7). The multi-trait score explained between 5.70% (continuous outcome) and 9.11% (categorical outcome) of interindividual variability in response, with an area under the curve of 68.24% (95% CI 64.98-71.66) and a Brier score of 0.257. Psychiatric and behavioural polygenic scores were the most strongly associated, accounting for 3.67-6.52% of variability.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

The gain over single-trait scores is real, but the scale matters: an area under the curve of 68% and under 10% of explained variance cannot decide whether to start lithium in an individual patient. The dominant contribution of psychiatric clusters suggests the score captures the form of the illness as much as drug response. The authors themselves set the horizon at combining it with clinical data, which remains to be done.

Analysis by Dr Thibaut Benquey