Authors' reply

Our article reports on diagnoses of real patients in the real world and hence variability ranges and the diagnostic process may be affected by factors such as psychiatrist or practice characteristics.

Regarding the question of whether full assessments were performed at each visit, we believe that practitioners tend not to update diagnoses if there is no salient clinical change. We hypothesised that clinicians would be less likely to change diagnoses, biasing the data against our reported finding.

Perhaps the most compelling point is that not all diagnoses were unstable. Thus, it is more likely that our findings reflect inconsistencies in our nosological system rather than clinician or practice characteristics, or setting effects. For example, some disorders may not always begin with the features required for diagnosis (e.g. mania in bipolar disorder) and therefore diagnostic instability may reflect the time required to consolidate the diagnosis (Baca-Garcia et al, 2007).

Our nosological system is in constant evolution, with major revisions each 15 years. Unfortunately, administrative procedures change more slowly than psychiatrists. Recording from one ICD system to another may affect the validity of diagnoses but not stability, since any error in the conversion of diagnostic codes would likely be constant, given the use of computerised algorithms.

Diagnoses in pharmacological and clinical studies have good internal validity (appropriate diagnostic schedules and interviews). In general, follow-up periods are short and selection bias is likely since participants are selected from specific programmes or units, often based on meeting specific entry criteria. Of note, Perala et al (2007) recently reported that the National Hospital Discharge Register was the most reliable means of screening for psychotic and bipolar disorder and was much better than the Composite International Diagnostic Interview (CIDI). They concluded that multiple information sources are key to accurate diagnoses. Studies such as ours, where patients are followed over long periods and across several settings, are closer to this approach than clinical trials based on diagnostic schedules and interviews performed in a research unit over a short period or large cross-sectional epidemiological studies based on a single assessment.