Li et al investigated the levels of brain-derived neurotrophic factor (BDNF) in patients with a first major depressive episode to assess the impact of these levels on the development of bipolar disorder during 3-year follow-up. ^{Reference Li, Zhang, Fan, Yuan, Huang and Chen1} They found that a combination of the levels of BDNF messenger RNA (mRNA) and plasma BDNF predicted a switch from depression to bipolar disorder in the following 3 years, with an area under the receiver operating characteristics (ROC) curve of 0.80. Li et al claimed that BDNF levels serve as a differential diagnostic biomarker for bipolar disorder in patients with a first depressive episode. If physicians could predict future development of bipolar disorder during a first depressive episode using biomarkers, it would allow us to select optimum treatment strategies. However, some caveats should be noted in Li et al’s study.

First, in the sample studied by Li et al, the diagnostic conversion rate from depression to bipolar disorder during the 3-year follow-up was as high as 12.8% (i.e. a shift to bipolar disorder occurred in 4.3% of the patients per year). This rate is amazingly high compared with rates in previous studies: for example, 1.5% per year ^{Reference Angst, Sellaro, Stassen and Gamma2} and 2.3% per year. ^{Reference Holma, Melartin, Holma and Isometsa3} If the patients who dropped out were all assumed to remain unipolar depressed, the rate would be still high (10.3% in 3 years). The high conversion rate in this sample might be attributable to an inclusion of patients with previous (hypo)manic episodes that had been overlooked, ^{Reference Bschor, Angst, Azorin, Bowden, Perugi and Vieta4} and/or a liberal threshold for the diagnosis of a hypomanic or manic episode in the study. Caution should be exercised when generalising the findings of this study.

Second, the positive predictive value (PPV) of the combination of the levels of BDNF mRNA and plasma BDNF in detecting the future development of bipolar disorder was low, even if the high diagnostic conversion rate (12.8% in 3 years) was used; according to the ROC curve in Fig. 3, ^{Reference Li, Zhang, Fan, Yuan, Huang and Chen1} the PPV was 48%, with sensitivity of 71% and specificity of 80%. This suggests that, of every two patients predicted to develop bipolar disorder within 3 years, using this biological index, one may be mislabelled as having latent bipolar disorder despite remaining unipolar depressed. Provided the diagnostic switch is assumed to occur in 6% of patients with a first depressive episode over 3 years after the onset of depression, as expected from previous studies, ^{Reference Angst, Sellaro, Stassen and Gamma2,Reference Holma, Melartin, Holma and Isometsa3} the PPV in this case would fall further, to 26%. Feasibility and clinical applicability cannot be undervalued.

We agree with the authors that BDNF can be linked with the pathophysiology of mood disorders, and that the impact may be more evident in bipolar disorder. However, before considering BDNF as a differential diagnostic biomarker in clinical settings, the low conversion rate from depression to bipolar disorder and the resulting low PPV ought to be taken into account.