The CAD-MDD: A Computerized Adaptive Diagnostic Screening Tool for Depression

By Robert D. Gibbons, Giles Hooker, M. Finkelman, David J. Weiss, P. Pilkonis, E. Frank, Tara Moore, David J. Kupfer. 2013

73 citations recorded by the source corpus.

Objective To develop a computerized adaptive diagnostic (CAD) screening tool for depression that decreases patient and clinician burden and increases sensitivity and specificity for clinician-based DSM-IV diagnosis of major depressive disorder (MDD). Method 656 individuals with and without minor and major depression were recruited from a psychiatric clinic, community mental health center, and through public announcements (controls without depression). The item bank consists of 88 depression scale items drawn from 73 depression measures. The focus of this study was the development of the CAD-MDD diagnostic screening tool based on a decision-theoretic approach (random forests and decision trees). Sensitivity and specificity for predicting clinician-based SCID DSM-IV diagnoses of MDD were the primary outcomes. Diagnostic screening accuracy was then compared to the PHQ-9. Results An average of 4 items per participant was required (maximum of 6 items). Overall sensitivity and specificity were 0.95 and 0.87 respectively. For the PHQ-9, sensitivity was 0.70 and specificity was 0.91. Conclusion High sensitivity and reasonable specificity for a clinician-based DSM-IV diagnosis of depression can be obtained using an average of 4 adaptively administered self-report items in less than one minute. Relative to the currently used PHQ-9, the CAD-MDD dramatically increased sensitivity while maintaining similar specificity. As such, the CAD-MDD will identify more true positives (lower false negative rate) than the PHQ-9 using half the number of items. Inexpensive (relative to clinical assessment), efficient and accurate screening of depression in primary care settings, psychiatric epidemiology, molecular genetics, and global health are all direct applications of the current system.

Subfields: Applied Psychology, Pharmacology, Social Psychology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Sarah Tan
2. Giles Hooker
3. R. Caruana
4. Yin Lou
5. Cynthia Rudin
6. Yichen Zhou
7. Ala Al-Fuqaha
8. Khansa Rasheed

Citations in the corpus

• Distill-and-Compare: Auditing Black-Box Models Using Transparent Model Distillation (2017, background): “For example, decision trees are transparent, reasonably accurate, and can perform both regression and classi€cation, but are notoriously unstable [12] which in turn leads to instability in inferences of how a feature contributes to the prediction.”
• Interpreting Models via Single Tree Approximation (2016, methodology): “We now demonstrate our method using the MDD data set which motivated Gibbons et al. [2013].”
• Auditing Black-Box Models Using Transparent Model Distillation With Side Information (2017, methodology): “They were used explicitly in [21] and more recently in [11].”
• Explainable, trustworthy, and ethical machine learning for healthcare: A survey (2021, methodology): “Gibbons et al. [86] used a hybrid approach for incorporating both the benefit of individual DT and the efficiency of ensemble DTs.”