Insights
1 February 2023
The role of cognitive functions in the diagnosis of bipolar disorder: A machine learning model
We caught up with Ercan Altınöz, Associate Professor of Psychiatry at Eskişehir Osmangazi University School of Medicine in Eskişehir, Turkey, who told us how combining CANTAB® tasks with machine learning may help to improve diagnosis of bipolar disorder.
Bipolar disorder, when considered as a spectrum, affects approximately 6% of the population1. Data from scientific studies show that approximately 1/3rd of individuals who have this disorder have significant cognitive impairments in several domains such as executive functions, attention, processing speed and memory, which has negative effects on both the daily life and vocational functionality of the patient2. The contrasting nature of this disorder and its comorbidities with other mental disorders are some of the barriers to making an accurate diagnosis. Many people with bipolar disorder experience misdiagnosis, with some waiting up to 10 years before they receive the correct diagnosis3 . This can prevent those affected from receiving proper and early treatment. Making an accurate diagnosis and initiating early treatment are substantial in helping these patients.
Data from scientific studies show that identifying specific cognitive deficits can help diagnose bipolar disorder in its early stages. On the other hand, studies using machine learning for early diagnosis of bipolar disorder seem promising. To the best of our knowledge, there has only been one study combining these two approaches4. In this study the authors were able to distinguish bipolar disorder from healthy controls with 71% accuracy by assessing cognitive findings through machine learning.
What was your study investigating?
In this study, Sonkurt HO, Altınöz AE and Köşger F from Eskişehir Osmangazi University Psychiatry Department, together with researchers Çimen E and Öztürk G from Eskişehir Technical University, combined machine learning with a broader cognitive assessment for the purpose of making a correct, early and objective diagnosis of bipolar disorder. We used a modified version of one-class Polyhedral Conic Function algorithm as the classifier in order to detect bipolar patients. For pre-processing, we have used 1R feature selection algorithm in order to select the best features. The study evaluated bipolar 1 patients, those who have experienced both mania and depression in their lifetime, and excluded bipolar 2 patients, those who have experienced both hypomania and depression in their lifetime.
Why did you choose CANTAB® for your study?
We opted for Cambridge Neuropsychological Test Automated Battery (CANTAB®), an innovative, standardised measure of cognitive functioning because it helps to eliminate rater bias, it is accurate, and it provides a vast amount of data that can work with machine learning algorithms.
Which CANTAB® tasks did you use?
After reviewing the literature on the cognitive profile of bipolar disorder, we aimed to assess cognition with a broad cognitive battery. Thus, we used Motor Screening Task (MOT), Spatial Working Memory (SWM), Delayed Matching to Sample (DMS), One Touch Stockings of Cambridge (OTS), Emotion Recognition Task (ERT) and Rapid Visual Information Processing (RVP) from the CANTAB® software. In this way, we evaluated the participants’ motor processing speed, working memory, executive functions, visual memory, emotion recognition and sustained attention abilities. With these data, we used the Polyhedral Conic Functions algorithm to classify the participants.
What did you discover?
1R method sorts all outcome measures (features) from CANTAB® tests according to features’ contribution to the classifier. Using the list of features ordered by their variable importance for classification, the features were added one by one to the classification model. The increase in accuracy was monitored until adding more features no longer increased the classifier accuracy. Seven features included outcome measures from SWM, DTS and OTS tests. Our results were able to distinguish individuals with bipolar disorder from healthy individuals with 78% accuracy in specific areas of executive functions, strategy setting, and spatial working memory. Our approach in machine learning and choosing specific tasks that measure executive functions and strategy setting made it possible to improve the findings in the literature, which is discussed in detail in our article.
What are the next steps?
In our study, since only bipolar 1 patients were evaluated, it is difficult to generalize the findings to all bipolar patients. Including patients with bipolar 2 disorder may provide more reliable data on cognition across the spectrum of the disorder. In addition, if we account for areas that show overlap between schizophrenia and bipolar disorder such as disorganization, psychoticism, and dysfunctionality, cognitive function could be used to make a correct, accurate and early diagnosis of bipolar disorder in future studies.
References
1. Merikangas, K. R., Akiskal, H. S., Angst, J., Greenberg, P. E., Hirschfeld, R. M., Petukhova, M., & Kessler, R. C. (2007). Lifetime and 12-month prevalence of bipolar spectrum disorder in the National Comorbidity Survey replication. Archives of general psychiatry, 64(5), 543-552.
2. Bourne, C., Aydemir, Ö., Balanzá‐Martínez, V., Bora, E., Brissos, S., Cavanagh, J. T. O., … & Goodwin, G. M. (2013). Neuropsychological testing of cognitive impairment in euthymic bipolar disorder: an individual patient data meta‐analysis. Acta Psychiatrica Scandinavica, 128(3), 149-162.
3. Hirschfeld, L. Lewis, L.A. Vornik, Perceptions and impact of bipolar disorder:how far have we really come? Results of the National Depressive and Manic-Depressive Association 2000 survey of individuals with bipolar disorder, J. Clin.Psychiatry (2003)
4. M.J. Wu, I.C. Passos, I.E. Bauer, L. Lavagnino, B. Cao, G.B. Zunta-Soares, F. Kapczinski, B. Mwangi, J.C. Soares, Individualized identification of euthymic bipolar disorder using the Cambridge Neuropsychological Test Automated Battery (CANTAB) and machine learning, Journal of Affective Disorders, 192 (2016) 219-225.
Our latest insights:
Author:
