Natural speech algorithm applied to baseline interview data can predict which patients will respond to psilocybin for treatment-resistant depression

doi:10.1016/j.jad.2018.01.006

Journal of Affective Disorders

Volume 230, 1 April 2018, Pages 84-86

https://doi.org/10.1016/j.jad.2018.01.006 Get rights and content

Highlights

•
Interview data from 17 patients with treatment-resistant depression was recorded.
•
Automated Emotional Analysis, a natural language processing method, was used to quantify emotional content of baseline interviews.
•
A machine learning algorithm was used to identify patterns in emotional analysis results.
•
Detected patterns predict therapeutic effectiveness of psilocybin for treatment-resistant depression.

Abstract

Background

Natural speech analytics has seen some improvements over recent years, and this has opened a window for objective and quantitative diagnosis in psychiatry. Here, we used a machine learning algorithm applied to natural speech to ask whether language properties measured before psilocybin for treatment-resistant can predict for which patients it will be effective and for which it will not.

Methods

A baseline autobiographical memory interview was conducted and transcribed. Patients with treatment-resistant depression received 2 doses of psilocybin, 10 mg and 25 mg, 7 days apart. Psychological support was provided before, during and after all dosing sessions. Quantitative speech measures were applied to the interview data from 17 patients and 18 untreated age-matched healthy control subjects. A machine learning algorithm was used to classify between controls and patients and predict treatment response.

Results

Speech analytics and machine learning successfully differentiated depressed patients from healthy controls and identified treatment responders from non-responders with a significant level of 85% of accuracy (75% precision).

Conclusions

Automatic natural language analysis was used to predict effective response to treatment with psilocybin, suggesting that these tools offer a highly cost-effective facility for screening individuals for treatment suitability and sensitivity.

Limitations

The sample size was small and replication is required to strengthen inferences on these results.

Introduction

Quantitative analyses of natural speech have undergone significant advances in recent years (Mikolov et al., 2013, Michel et al., 2011, Carrillo et al., 2015, Cambria and White, 2014)and are beginning to be applied in psychiatry (Wang and Krystal, 2014, Wiecki et al., 2015, Mota et al., 2016, Carrillo et al., 2014, Huys et al., 2011, Mundt et al., 2012). For example, automatic analysis of speech incoherence has been used as a biomarker of schizophrenia; in a proof-of-concept experiment with a small sample size, a machine learning algorithm predicted conversion to psychosis in ‘at-risk’ individuals with a 100% accuracy (Bedi et al., 2015).

In mood disorders, a recently developed measure of emotion in speech was found to accurately sort between bipolar patients and control subjects (Carrillo et al., 2016) and related tools have proved effective in identifying depression in interview-based speech (Pestian et al., 2008) and social media-based text (De Choudhury et al., 2013, De Choudhury et al., 2013). These studies highlight the power of natural language analytics to diagnose and prognose mental illness and its response to treatment.

In the present study, we sought to build on this work by testing whether natural speech analytics combined with machine learning could predict clinical responses to psilocybin in patients with treatment-resistant depression (TRD), defined here as failure of at least two different antidepressants of differing pharmacology, within the same depressive episode. Psilocybin is a serotonin 2A receptor agonist and classic psychedelic drug that is currently showing promise for the treatment of a range of psychiatric conditions, including depression (Carhart-Harris and Goodwin, 2017).

Section snippets

Methods

This trial received a favorable opinion from NRES London-West London, was sponsored by Imperial College London, and was carried out in accordance with Good Clinical Practice Guidelines. It was an open-label design in which patients with TRD received two doses of psilocybin (10 mg and 25 mg) one week apart. The autobiographical memory test (AMT) (Williams and Scott, 1988) was performed by patients (n = 17) and age and sex matched matched controls (n = 18), who were recruited separately. For more

Results

Before we addressed our main question, we asked whether our method can distinguish between controls and patients. A significant between-group difference was found in the rate of positive words used in participants’ AMT interview responses, with patients using significant fewer positive words: controls AVG P = 0.0532 $\pm$ 0.013 and patients AVG P = 0.0384 $\pm$ 0.011 (t-test p = 0.0011). The AVG N did not differ significantly between both groups (p = 0.4). Using a machine learning classifier, with a 7

Discussion

In the present study, natural speech analytics combined with machine learning was able to differentiate depressed patients from healthy controls and predict responders versus non-responders in a clinical trial of psilocybin for treatment-resistant depression. The AMT interviews that produced the data on which these analyses were performed took little longer than 10 min to perform, yet were able to identify depression from health and predict treatment response with a significant level of

Acknowledgments

Robin L. Carhart-Harris is supported by the Alex Mosley Charitable Trust and that the UK MRC (MR/J00460X/1) funded the trial. Diego Fernández Slezak is sponsored by the Microsoft Faculty Fellowship 2014. MS is sponsored by James McDonnell Foundation 21st Century Science Initiative in Understanding Human Cognition.

References (20)

R.L. Carhart-Harris
Psilocybin with psychological support for treatment-resistant depression: an open-label feasibility study
Lancet Psychiatry
(2016)
Q.J. Huys et al.
Are computational models of any use to psychiatry?
Neural Netw.
(2011)
J.C. Mundt et al.
Vocal acoustic biomarkers of depression severity and treatment response
Biol. Psychiatry
(2012)
X.-J. Wang et al.
Computational psychiatry
Neuron
(2014)
G. Bedi
Automated analysis of free speech predicts psychosis onset in high-risk youths
NPJ Schizophr.
(2015)
E. Cambria et al.
Jumping nlp curves: a review of natural language processing research
IEEE Comput. Intell. Mag.
(2014)
R.L. Carhart-Harris et al.
The therapeutic potential of psychedelic drugs: past, present and future
Neuropsychopharmacology
(2017)
Carrillo, F., et al. 2014. Automated speech analysis for psychosis evaluation. In: International Workshop on Machine...
F. Carrillo et al.
Fast distributed dynamics of semantic networks via social media
Comput. Intell. Neurosci.
(2015)
Carrillo, F. et al. 2016. Emotional intensity analysis in bipolar subjects. arXiv preprint...

There are more references available in the full text version of this article.

Cited by (29)

Predicting the outcome of psilocybin treatment for depression from baseline fMRI functional connectivity
2024, Journal of Affective Disorders
Psilocybin is a serotonergic psychedelic drug under assessment as a potential therapy for treatment-resistant and major depression. Heterogeneous treatment responses raise interest in predicting the outcome from baseline data.
A machine learning pipeline was implemented to investigate baseline resting-state functional connectivity measured with functional magnetic resonance imaging (fMRI) as a predictor of symptom severity in psilocybin monotherapy for treatment-resistant depression (16 patients administered two 5 mg capsules followed by 25 mg, separated by one week). Generalizability was tested in a sample of 22 patients who participated in a psilocybin vs. escitalopram trial for moderate-to-severe major depression (two separate doses of 25 mg of psilocybin 3 weeks apart plus 6 weeks of daily placebo vs. two separate doses of 1 mg of psilocybin 3 weeks apart plus 6 weeks of daily oral escitalopram). The analysis was repeated using both samples combined.
Functional connectivity of visual, default mode and executive networks predicted early symptom improvement, while the salience network predicted responders up to 24 weeks after treatment (accuracy≈0.9). Generalization performance was borderline significant. Consistent results were obtained from the combined sample analysis. Fronto-occipital and fronto-temporal coupling predicted early and late symptom reduction, respectively.
The number of participants and differences between the two datasets limit the generalizability of the findings, while the lack of a placebo arm limits their specificity.
Baseline neurophysiological measurements can predict the outcome of psilocybin treatment for depression. Future research based on larger datasets should strive to assess the generalizability of these predictions.
Clinical and biological predictors of psychedelic response in the treatment of psychiatric and addictive disorders: A systematic review
2021, Journal of Psychiatric Research
Citation Excerpt :
The same procedure was followed in order to obtain the average negativity. Through this speech analysis, the authors showed that they could identify responders from non-responders with a significant level of 85% accuracy (Carrillo et al., 2018). One study has investigated the pre-existing state or trait markers in patients with TRD (Erritzoe et al., 2018).
The use of psychedelic treatments has shown very promising results in some psychiatric and addictive disorders, but not all patients achieved a response.
The aim of this review is to explore the clinical and biological factors which could predict the response to psychedelics in psychiatric and addictive disorders.
A systematic research was performed on MEDLINE, PsycInfo, Web of science, and Scopus databases from January 1990 to May 2020. All studies investigating the predictive factors of response to psychedelics regardless of psychiatric or addictive disorders, were included.
Twenty studies investigating addictive disorder, treatment-resistant depression, obsessive-compulsive disorder and depressive and anxiety symptoms in patients with life-threatening cancer were included in this review. We found that, in all indications, the main predictive factor of response to psychedelics is the intensity of the acute psychedelic experience. Indeed, we found this factor for alcohol and tobacco use disorders, treatment-resistant depression, and anxiety and depressive symptoms in patients with life-threatening cancer, but not for obsessive-compulsive disorder.
The intensity of the acute psychedelic experience was the main predicting factor of response. The action mechanism of this experience was not clear, but some hypotheses could be made, such as a modulation of serotoninergic system by 5-HT2A receptors agonism, a modulation of the default mode network (DMN) with an acute modular disintegration of the DMN followed by a re-integration of this network with a normal functioning, or an anti-inflammatory effect of this treatment.
The entropic tongue: Disorganization of natural language under LSD
2021, Consciousness and Cognition
Citation Excerpt :
Indeed, in terms of screening and prediction, the most valuable data would be collected ahead of dosing sessions. The value of natural language processing for predicting treatment response is supported by a recent proof of concept study that used a machine learning classifier applied to interview data collected at baseline to differentiate and potentially predict responders from non-responders in patients suffering from major depression treated with psilocybin therapy (Carrillo et al., 2018). In conclusion, here we characterized natural language under the acute effects of LSD.
Serotonergic psychedelics have been suggested to mirror certain aspects of psychosis, and, more generally, elicit a state of consciousness underpinned by increased entropy of on-going neural activity. We investigated the hypothesis that language produced under the effects of lysergic acid diethylamide (LSD) should exhibit increased entropy and reduced semantic coherence. Computational analysis of interviews conducted at two different time points after 75 μg of intravenous LSD verified this prediction. Non-semantic analysis of speech organization revealed increased verbosity and a reduced lexicon, changes that are more similar to those observed during manic psychoses than in schizophrenia, which was confirmed by direct comparison with reference samples. Importantly, features related to language organization allowed machine learning classifiers to identify speech under LSD with accuracy comparable to that obtained by examining semantic content. These results constitute a quantitative and objective characterization of disorganized natural speech as a landmark feature of the psychedelic state.
Using text mining to extract depressive symptoms and to validate the diagnosis of major depressive disorder from electronic health records
2020, Journal of Affective Disorders
Many studies have used Taiwan's National Health Insurance Research database (NHIRD) to conduct psychiatric research. However, the accuracy of the diagnostic codes for psychiatric disorders in NHIRD is not validated, and the symptom profiles are not available either. This study aimed to evaluate the accuracy of diagnostic codes and use text mining to extract symptom profile and functional impairment from electronic health records (EHRs) to overcome the above research limitations.
A total of 500 discharge notes were randomly selected from a medical center's database. Three annotators reviewed the notes to establish gold standards. The accuracy of diagnostic codes for major psychiatric illness was evaluated. Text mining approaches were applied to extract depressive symptoms and function profiles and to identify patients with major depressive disorder.
The accuracy of the diagnostic code for major depressive disorder, schizophrenia, and dementia was acceptable but that of bipolar disorder and minor depression was less satisfactory. The performance of text mining approach to recognize depressive symptoms is satisfactory; however, the recall for functional impairment is lower resulting in lower F-scores of 0.774–0.753. Using the text mining approach to identify major depressive disorder, the recall was 0.85 but precision was only 0.69.
The accuracy of the diagnostic code for major depressive disorder in discharge notes was generally acceptable. This finding supports the utilization of psychiatric diagnoses in claims databases. The application of text mining to EHRs might help in overcoming current limitations in research using claims databases.
Can Machine Learning help us in dealing with treatment resistant depression? A review
2019, Journal of Affective Disorders
Citation Excerpt :
Among the studies aiming at predicting the response to a specific treatment in a TRD population, two studies focused on response to repetitive transcranial magnetic stimulation (rTMS) (Bailey et al., 2018, 2019), one study on the response to SSRI (Khodayari-Rostamabad et al., 2013) and one to psilocybin (Carrillo et al., 2018). Interestingly, three of these studies employed EEG features as predictors (Khodayari-Rostamabad et al., 2013; Bailey et al., 2018, 2019), while the remaining one used features extracted by an automated algorithm for quantifying the emotional content of spoken or written text (Carrillo et al., 2018). Specifically, in two subsequent studies, Bailey et al. (2018, 2019) predicted response to rTMS in a small sample of TRD by means of a combination of clinical and cognitive variables (MADRS and Working Memory accuracy) and EEG measures (alpha, theta and gamma powers; alpha theta and gamma connectivity; theta gamma coupling) assessed at baseline and after one week.
About one third of patients treated with antidepressant do not show sufficient symptoms relief and up to 15% of patients remain symptomatic even after multiple trials are applied, configuring a state called treatment resistant depression (TRD). A clear definition of this state and the understanding of underlying mechanisms contributing to chronic disability caused by major depressive disorder is still unknown. Therefore, Machine Learning (ML) techniques emerged in the last years as interesting approaches to deal with such complex problems.
We performed a bibliographic search on Pubmed, Google Scholar and Medline of clinical, imaging, genetic and EEG ML classification studies on treatment-responding depression and TRD as well as studies trying to predict response to a specific treatment in already established TRD. The inclusion criteria were met by eleven studies. Seven focused on the definition of predictors of TRD onset while four attempted to predict the response to specific treatments in TRD.
The results showed that it seems possible to classify between responders MDD and TRD with good accuracies based on clinical variables. Moreover, some studies reported the possibility of using EEG measures to predict response to different pharmacological and non-pharmacological treatments in established TRD.
The definition of TRD, the selection of variables together with ML algorithms and pipelines varies across the studies, ultimately determining the unfeasibility to implement these models in clinical practice.
The findings suggest that ML could be a valid approach to increase our understanding of TRD and to better classify and stratify this disorder, which may ultimately help clinicians in the assessment of major depressive disorders.
Rapid-acting antidepressants
2019, Advances in Pharmacology
Conventional antidepressants (biogenic amine mechanisms) are not fully efficacious (e.g., symptoms remain after treatment, not all patients respond), produce effects only after weeks of daily dosing, and do not impact all disease symptoms. In contrast, a new class of antidepressants has been emerging since 2006 that has demonstrated rapid onset, large effect size, activity after only a single or few dose applications, and positive impact in treatment refractory patients and against some treatment-resistant symptoms (e.g., anhedonia). Rapid-acting antidepressant drug action has been demonstrated in controlled clinical studies for ketamine, a few other NMDA receptor antagonists, and scopolamine. Less clinical data are currently available for psychedelic drugs such as psilocybin, lysergic acid diethylamide, and ayahuasca. The mechanisms of action of rapid-acting antidepressants are not fully understood. However, a general triggering mechanism appears to involve the potentiation of AMPA receptor function. Although the durability of antidepressant effects of ketamine and scopolamine is limited, psychedelic drugs have been reported to produce effects for many months. The primary impediment to generating a medicine of this type for depressed patients is side effects and the lack of methods to ensure enduring antidepressant effects. Thus, further exploration of drug possibilities continues. Esketamine ((S)-ketamine) was recently FDA approved. Compounds currently in clinical development include the NMDA receptor antagonist (R)-ketamine, the NMDA receptor modulator, GLYX-13 (Rapastinel), and the AMPA receptor potentiator TAK-653. Additional pharmacological classes have produced effects in the preclinical laboratory to suggest their potential as rapid-acting agents. These include mGlu2/3 receptor antagonists, AMPA receptor potentiators, and negative allosteric modulators of GABA_A(α5) receptors. In all cases, molecules exist that could be used to provide clinical proof of concept testing.

View all citing articles on Scopus

View full text

Natural speech algorithm applied to baseline interview data can predict which patients will respond to psilocybin for treatment-resistant depression

Highlights

Abstract

Background

Methods

Results

Conclusions

Limitations

Introduction

Section snippets

Methods

Results

Discussion

Acknowledgments

Lancet Psychiatry

Neural Netw.

Biol. Psychiatry

Neuron

Automated analysis of free speech predicts psychosis onset in high-risk youths

NPJ Schizophr.

Jumping nlp curves: a review of natural language processing research

IEEE Comput. Intell. Mag.

The therapeutic potential of psychedelic drugs: past, present and future

Neuropsychopharmacology

Fast distributed dynamics of semantic networks via social media

Comput. Intell. Neurosci.