Forty et al's study ^{Reference Forty, Smith, Jones, Jones, Caesar, Cooper, Fraser, Gordon-Smith, Hyde, Farmer, McGuffin and Craddock1} revisits a familiar question and reports some statistically significant differences in the frequency of clinical features between unipolar and bipolar depression. The report then moves beyond description to emphasise the clinical importance of these differences. The three symptoms most predictive of bipolar depression were presence of psychosis, diurnal mood variation and hypersomnia. To be considered important, these symptoms when present would need to influence clinical decisions.

The sensitivity of these three symptoms for bipolar depression ranged from 0.3 to 0.59. The specificity ranged from 0.5 to 0.9. The positive predictive value (PPV) ranged from 0.55 to 0.69. As the prevalence of bipolar disorder in the sample was 0.43, these PPV results do not greatly increase the probability of bipolar disorder above the base rate. Likewise, the negative predictive value ranged from 0.63 to 0.69, while the base rate of unipolar depression was 0.57. Again, there is little gain of information.

Because this differential diagnosis relies on pattern recognition rather than on discrete, pathognomonic symptoms, it may be more helpful to examine cases in which all three ‘important clinical differences’ were present. When all three features are required, beginning with the highest PPV (psychotic features), then the middle PPV (hypersomnia), then the lowest PPV (diurnal variation), and assuming the three symptoms are independent, then 34 bipolar cases and 7 unipolar cases would stand out. From this result, the sensitivity of the triune pattern would be 0.08 and the specificity 0.99. The PPV is 0.83 and the negative predictive value is 0.59. How ‘important’ is a clinical symptom pattern that detects only 8% of bona fide bipolar cases and that does not positively rule in unipolar cases? Moreover, there is no guarantee that latent bipolar depression has the same symptom profile as fully expressed bipolar depression.

All in all, these results underscore the limitations of parsing clinical symptoms for the purpose of classification. The ‘important clinical differences’ give little added information to clinicians for treatment planning. That is why efforts continue to discover biomarkers or endophenotypes or genetic markers.