A Brazilian version of the Montgomery-Asberg Depression Rating Scale, or MADRS, showed good consistency in a study of people with unipolar and bipolar mood disorders, but the results did not support treating the scale as a perfectly one-dimensional measure. The analysis also pointed to two different useful score thresholds: 9 for screening or classification and 18 for a stronger balance between detecting cases and ruling out non-cases.
The two thresholds reflect different scoring criteria in the analysis. The score of 9 produced very high sensitivity but lower specificity. The score of 18 offered a more even trade-off between the two measures. Neither threshold should be treated as established for populations or clinical settings outside the study.
What the researchers tested
The researchers carried out a cross-sectional psychometric analysis, meaning they examined measurements taken at one point rather than following participants over time. Volunteers were recruited through social media and public psychiatric outpatient clinics in Northeast Brazil. Of 448 people assessed, 145 were excluded and 303 completed the full protocol.
The final group had a mean age of 37.03 years. Women made up 84.16% of the participants, while 206 people had unipolar disorder and 97 had bipolar disorder. The study used a convenience sample rather than random recruitment, which can limit how confidently the findings apply beyond the people who took part.
The scale's internal consistency was good. Cronbach's alpha was 0.862 and McDonald's omega was 0.869, with reliability remaining stable across the items. In practical terms, the questions tended to work together as a measure, though that result alone does not show that they capture only one distinct dimension.
A mostly single pattern, with important qualifications
Tests of the data supported looking for an underlying factor, or broad pattern shared by the items. The analysis favored one factor, which explained 53.03% of the variation in the real data, compared with 24.15% in simulated data. That supports a predominant single structure for the scale in this sample.
The more detailed fit checks were mixed. Two measures met their stated thresholds, with RMSR at 0.074 and CFI at 0.975, while RMSEA was 0.103, above the reported acceptable threshold of 0.08. The H-latent and H-observed measures, which address whether a construct can be reproduced, were 0.924 and 0.934, both above the study's 0.8 threshold.
Other measures did not establish strict unidimensionality, the stronger claim that one underlying trait accounts for the scale in a sufficiently clean way. The full-scale UniCo was 0.91, explained common variance was 0.85, and IREAL was 0.23, but the authors concluded that the evidence was not robust enough for that conclusion.
One item drew particular attention: the question about reduced appetite had weaker factor loading and sampling adequacy. Removing it lowered RMSR from 0.074 to 0.069 and nudged CFI from 0.975 to 0.976, but RMSEA worsened from 0.103 to 0.113. The authors say the item may warrant future refinement, but the study does not establish that it should be removed.
What the score tracked, and what remains uncertain
MADRS scores had a moderate positive correlation with GAD-7 scores, with a Spearman correlation of 0.55 and a p-value below 0.001. That pattern supported convergent validity, meaning the two measures moved together in a way the researchers expected. Correlations with nicotine and alcohol subscales were much weaker, at 0.13 and 0.12, but still fell just outside the study's formal discriminant-validity threshold of an absolute correlation no greater than 0.10.
For the cutoff analysis, the Youden criterion selected a MADRS score of 18 as the best sensitivity-specificity balance in this sample. At that threshold, sensitivity was 85.71% and specificity was 83.08%; the positive predictive value was 94.88% and the negative predictive value was 61.36%.
A different scoring goal favored 9. That threshold had sensitivity of 99.58%, with specificity of 47.69%. The positive predictive value was 87.45%, the negative predictive value was 96.88%, and the F1 score was 0.931. The pattern is consistent with the authors' interpretation of 9 as useful for screening and classification.
The researchers caution that the findings are tied to a convenience, cross-sectional sample and may be affected by selection bias. The study did not administer the HAM-D, the MINI 7.02 did not provide severity criteria for the depressive episode, and the analysis was not designed to compare the psychometric properties of unipolar and bipolar disorders separately. The proposed cutoffs therefore remain findings from this sample until they are tested in independent clinical settings.
Paper data and sources
Original title: New Insights About the Psychometric Properties of the Montgomery‐Asberg Depression Rating Scale
Authors: Neto S, Cavalcante E, Leão M et al.
Journal/Repository: International journal of methods in psychiatric research
Status: Peer-reviewed
First online: 2026-10-11
DOI: Not available
Original paper · Full text