Lessons for ICD-11 coming after DSM-5

ICD Insights

The Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) has arrived, and we await the International Classification of Diseases, 11th Revision (ICD-11). Without trying to be exhaustive, here are suggestions for improving matters.

1. Avoid the nominalist fallacy (I name, therefore I know). Classifications are always provisional, to be improved over time by convergent validation and new science.

2. Expand the frame of reference for convergent validation. In DSM-III through DSM-5, four key sources of information were mostly excluded. These are: course of illness, response to treatment, family history, and biomarkers (Robins and Guze, 1970).

3. Avoid the appearance of closure. Like classifications, diagnoses also are always provisional. Diagnoses are casewise probability statements, open to revision as new information arrives (Carroll, 2013).

4. Maintain a clear distinction between making a diagnosis and defining a disorder. This speaks to a fundamental category error that appeared soon after DSM-III and that still persists in the form of checklist menu diagnoses – ironically mostly in research settings. DSM-5 did not fix that (Carroll, 2012a).

5. Allow for uncertainty. DSM-5 gives the impression that initial, cross-sectional diagnoses are the norm. Neurologists, in contrast, do not allow the diagnosis of probable Parkinson’s disease (PD) without at least 3 years of observation (Gelb et al., 1999). In keeping with this provisional approach, the likelihood of PD is specified initially as possible, probable, or definite. During the 3 years of observation in patients with possible PD, neurologists document significant clinical events that would modify or support the possible diagnosis. A similar approach is needed for psychiatric presentations with psychosis, mood disturbance, anxiety, cognitive decline, and more.

6. Get numerate! The perversion of checklist diagnoses is fostered by the innumerate style of DSM-5, which provides no sensitivity and specificity data for definitional signs and symptoms. In making clinical diagnoses, not all signs and symptoms carry equal weight, but learners would never know that from reading the diagnostic criteria in DSM-5. Neurologists have operationalized this understanding by specifying cardinal symptoms and accessory symptoms (Gelb et al., 1999).

7. Do not allow reliability to trump validity. As DSM-III taught us, that course is a fool’s trade-off, evidenced most notably by 35 wasted years of research into a fictive entity labeled major depressive disorder. DSM-5 did not fix that.

8. Get serious about diagnostic unreliability! It is not enough just to declare mediocre kappa values – or worse – and quickly move on, as happened in DSM-5. One can demonstrate conclusively by mathematical modeling that such poor diagnostic agreement is an insurmountable obstacle in research settings (Carroll, 1989).

9. Get serious about diagnostic instability! We have known for decades that stability data are poor, even with contemporary operational criteria for diagnosis, but we averted our gaze. An essential part of every description of a disorder needs to be an account of how the diagnosis is known to change over time. Systematic data on this issue for most diagnoses are hard to find.

10. Balance the idiographic and nomothetic aspects of nosology. It would be a plus for ICD-11 to keep the case vignette approach of ICD-10, while avoiding the DSM-5 emphasis on fixed numbers of decontextualized qualifying symptoms.

11. Get Bayesian! A key part of Bayesian thinking is recognition of casewise prior probabilities. That is crucial in the process of sorting through differential diagnoses (Carroll, 2013).

12. Get with biomarkers! Biomarkers are absent from DSM-5, yet current biomarkers perform at least as well as many definitional signs and symptoms. Adding biomarkers to signs and symptoms in defining disorders will bring psychiatry into alignment with the rest of medicine. On the other hand, demanding perfection of biomarkers while tolerating demonstrably insensitive and non-specific definitional signs and symptoms will be self-defeating (Carroll, 2013). That is especially true if we remain blind to the problems of diagnostic unreliability and longitudinal instability.

13. Keep commerce out of the process. Nobody owns diagnostic criteria. The American Psychiatric Association (APA) threw its weight around with threats of SLAPP (strategic lawsuits against public participation) lawsuits and the like during the DSM-5 process, aiming to deter critics in advance of publication of the manual. This uncollegial behavior was driven by the corporate commercial interest of the APA (Carroll, 2012b).

14. Keep the process truly open. Science progresses bottom-up rather than top-down. So, the organizers of ICD-11 should view themselves as facilitating editors rather than as bosses who call the shots. In this respect, the ICD process seems preferable to the DSM process. There should be no commercial subtexts or undeclared professional guild issues in the ICD process.