To the Editor: We recently discovered an artifact bearing on the effects of time from illness onset to the start of lithium maintenance treatment and measures of treatment responsiveness in patients with bipolar disorder (1). Such findings are pertinent to early intervention in bipolar disorder, a matter of considerable clinical and public health importance. In 1998 we noted an association between shorter latency to treatment and apparently superior treatment response, as measured by the percentage of time ill during lithium maintenance therapy subtracted from the percentage of time ill before treatment. In a later study (2) we analyzed the relationship of treatment latency or pretreatment episode number to morbidity during maintenance treatment in the same clinical population. Response was defined by a survival analysis using length of the first interepisode wellness interval and the percentage of time ill during the treatment. Neither outcome was associated with treatment latency or number of pretreatment episodes.

These inconsistencies led us to reanalyze treatment outcomes in an expanded study group from the same clinical group. We found a striking inverse association between treatment latency and percentage of time ill before treatment ( r_s=–0.67, N=376, p<0.0001) but no relationship to illness during treatment (r_s=–0.03, N=376, p=0.51). That is, a shorter time to treatment was strongly associated with greater pretreatment morbidity. In turn, outcomes evaluated as change in percentage of time ill were inflated at shorter treatment latencies (r_s=–0.55, N=376, p<0.0001). This effect no doubt contributed to an impression that earlier intervention yielded superior outcomes (1). Instead, this finding appears to derive from an association of a greater treatment-associated change with a greater level of pretreatment morbidity.

Interpretation of the association of greater morbidity with shorter treatment latency is not entirely clear. It may reflect a clinical urgency to start treatment early with very ill patients, or it may represent a mathematically higher proportion of time ill with shorter exposure times. We apologize for any confusion occasioned by our seemingly inconsistent findings and urge caution in use of change in morbidity to evaluate treatment response. In general, there is a need for wider consensus on measures of treatment effectiveness in studies of bipolar disorder (3). Finally, we strongly support efforts at early recognition and clinical intervention in this potentially disabling or lethal illness, without prejudice about potential treatment response based on delay of treatment.