Elemental psychopathology: distilling constituent symptoms and patterns of repetition in the diagnostic criteria of the DSM-5

BACKGROUND

The DSM-5 features hundreds of diagnoses comprising a multitude of symptoms, and there is considerable repetition in the symptoms among diagnoses. This repetition undermines what we can learn from studying individual diagnostic constructs because it can obscure both disorder- and symptom-specific signals. However, these lost opportunities are currently veiled because symptom repetition in the DSM-5 has not been quantified.

METHOD

This descriptive study mapped the repetition among the 1419 symptoms described in 202 diagnoses of adult psychopathology in section II of the DSM-5. Over a million possible symptom comparisons needed to be conducted, for which we used both qualitative content coding and natural language processing.

RESULTS

In total, we identified 628 distinct symptoms: 397 symptoms (63.2%) were unique to a single diagnosis, whereas 231 symptoms (36.8%) repeated across multiple diagnoses a total of 1022 times (median 3 times per symptom; range 2-22). Some chapters had more repetition than others: For example, every symptom of every diagnosis in the bipolar and related disorders chapter was repeated in other chapters, but there was no repetition for any symptoms of any diagnoses in the elimination disorders, gender dysphoria or paraphilic disorders. The most frequently repeated symptoms included insomnia, difficulty concentrating, and irritability - listed in 22, 17 and 16 diagnoses, respectively. Notably, the top 15 most frequently repeating diagnostic criteria were dominated by symptoms of major depressive disorder.

CONCLUSION

Overall, our findings lay the foundation for a better understanding of the extent and potential consequences of symptom overlap.

Clinical named entity recognition and relation extraction using natural language processing of medical free text: A systematic review

BACKGROUND

Natural Language Processing (NLP) applications have developed over the past years in various fields including its application to clinical free text for named entity recognition and relation extraction. However, there has been rapid developments the last few years that there's currently no overview of it. Moreover, it is unclear how these models and tools have been translated into clinical practice. We aim to synthesize and review these developments.

METHODS

We reviewed literature from 2010 to date, searching PubMed, Scopus, the Association of Computational Linguistics (ACL), and Association of Computer Machinery (ACM) libraries for studies of NLP systems performing general-purpose (i.e., not disease- or treatment-specific) information extraction and relation extraction tasks in unstructured clinical text (e.g., discharge summaries).

RESULTS

We included in the review 94 studies with 30 studies published in the last three years. Machine learning methods were used in 68 studies, rule-based in 5 studies, and both in 22 studies. 63 studies focused on Named Entity Recognition, 13 on Relation Extraction and 18 performed both. The most frequently extracted entities were "problem", "test" and "treatment". 72 studies used public datasets and 22 studies used proprietary datasets alone. Only 14 studies defined clearly a clinical or information task to be addressed by the system and just three studies reported its use outside the experimental setting. Only 7 studies shared a pre-trained model and only 8 an available software tool.

DISCUSSION

Machine learning-based methods have dominated the NLP field on information extraction tasks. More recently, Transformer-based language models are taking the lead and showing the strongest performance. However, these developments are mostly based on a few datasets and generic annotations, with very few real-world use cases. This may raise questions about the generalizability of findings, translation into practice and highlights the need for robust clinical evaluation.

Trends in Opioid Medication Adherence During the COVID-19 Pandemic: Retrospective Cohort Study

BACKGROUND

The recent pandemic had the potential to worsen the opioid crisis through multiple effects on patients' lives, such as the disruption of care. In particular, good levels of adherence with respect to medication for opioid use disorder (MOUD), recognized as being important for positive outcomes, may be disrupted.

OBJECTIVE

This study aimed to investigate whether patients on MOUD experienced a drop in medication adherence during the recent COVID-19 pandemic.

METHODS

This retrospective cohort study used Medicaid claims data from 6 US states from 2018 until the start of 2021. We compared medication adherence for people on MOUD before and after the beginning of the COVID-19 pandemic in March 2020. Our main measure was the proportion of days covered (PDC), a score that measures patients' adherence to their MOUD. We carried out a breakpoint analysis on PDC, followed by a patient-level beta regression analysis with PDC as the dependent variable while controlling for a set of covariates.

RESULTS

A total of 79,991 PDC scores were calculated for 37,604 patients (age: mean 37.6, SD 9.8 years; sex: n=17,825, 47.4% female) between 2018 and 2021. The coefficient for the effect of COVID-19 on PDC score was -0.076 and was statistically significant (odds ratio 0.925, 95% CI 0.90-0.94).

CONCLUSIONS

The COVID-19 pandemic was negatively associated with patients' adherence to their medication, which had declined since the beginning of the pandemic.

Using machine learning to study the effect of medication adherence in Opioid Use Disorder

BACKGROUND

Opioid Use Disorder (OUD) and opioid overdose (OD) impose huge social and economic burdens on society and health care systems. Research suggests that Medication for Opioid Use Disorder (MOUD) is effective in the treatment of OUD. We use machine learning to investigate the association between patient's adherence to prescribed MOUD along with other risk factors in patients diagnosed with OUD and potential OD following the treatment.

METHODS

We used longitudinal Medicaid claims for two selected US states to subset a total of 26,685 patients with OUD diagnosis and appropriate Medicaid coverage between 2015 and 2018. We considered patient age, sex, region level socio-economic data, past comorbidities, MOUD prescription type and other selected prescribed medications along with the Proportion of Days Covered (PDC) as a proxy for adherence to MOUD as predictive variables for our model, and overdose events as the dependent variable. We applied four different machine learning classifiers and compared their performance, focusing on the importance and effect of PDC as a variable. We also calculated results based on risk stratification, where our models separate high risk individuals from low risk, to assess usefulness in clinical decision-making.

RESULTS

Among the selected classifiers, the XGBoost classifier has the highest AUC (0.77) closely followed by the Logistic Regression (LR). The LR has the best stratification result: patients in the top 10% of risk scores account for 35.37% of overdose events over the next 12 month observation period. PDC score calculated over the treatment window is one of the most important features, with better PDC lowering risk of OD, as expected. In terms of risk stratification results, of the 35.37% of overdose events that the predictive model could detect within the top 10% of risk scores, 72.3% of these cases were non-adherent in terms of their medication (PDC <0.8). Targeting the top 10% outcome of the predictive model could decrease the total number of OD events by 10.4%.

CONCLUSIONS

The best performing models allow identification of, and focus on, those at high risk of opioid overdose. With MOUD being included for the first time as a factor of interest, and being identified as a significant factor, outreach activities related to MOUD can be targeted at those at highest risk.

Conversation dynamics in a multiplayer video game with knowledge asymmetry

Despite the challenges associated with virtually mediated communication, remote collaboration is a defining characteristic of online multiplayer gaming communities. Inspired by the teamwork exhibited by players in first-person shooter games, this study investigated the verbal and behavioral coordination of four-player teams playing a cooperative online video game. The game, Desert Herding, involved teams consisting of three ground players and one drone operator tasked to locate, corral, and contain evasive robot agents scattered across a large desert environment. Ground players could move throughout the environment, while the drone operator’s role was akin to that of a “spectator” with a bird’s-eye view, with access to veridical information of the locations of teammates and the to-be-corralled agents. Categorical recurrence quantification analysis (catRQA) was used to measure the communication dynamics of teams as they completed the task. Demands on coordination were manipulated by varying the ground players’ ability to observe the environment with the use of game “fog.” Results show that catRQA was sensitive to changes to task visibility, with reductions in task visibility reorganizing how participants conversed during the game to maintain team situation awareness. The results are discussed in the context of future work that can address how team coordination can be augmented with the inclusion of artificial agents, as synthetic teammates.

Assessing Team Effectiveness by How Players Structure Their Search in a First-Person Multiplayer Video Game

People working as a team can achieve more than when working alone due to a team's ability to parallelize the completion of tasks. In collaborative search tasks, this necessitates the formation of effective division of labor strategies to minimize redundancies in search. For such strategies to be developed, team members need to perceive the task's relevant components and how they evolve over time, as well as an understanding of what others will do so that they can structure their own behavior to contribute to the team's goal. This study explored whether the capacity for team members to coordinate effectively can be related to how participants structure their search behaviors in an online multiplayer collaborative search task. Our results demonstrated that the structure of search behavior, quantified using detrended fluctuation analysis, was sensitive to contextual factors that limit a participant's ability to gather information. Further, increases in the persistence of movement fluctuations during search behavior were found as teams developed more effective coordinative strategies and were associated with better task performance.

Augmentation of Electronic Medical Record Data for Deep Learning

Data imbalance is a well-known challenge in the development of machine learning models. This is particularly relevant when the minority class is the class of interest, which is frequently the case in models that predict mortality, specific diagnoses or other important clinical end-points. Typical methods of dealing with this include over- or under-sampling training data, or weighting the loss function in order to boost the signal from the minority class. Data augmentation is another frequently employed method - particularly for models that use images as input data. For discrete time-series data, however, there is no consensus method of data augmentation. We propose a simple data augmentation strategy that can be applied to discrete time-series data from the EMR. This strategy is then demonstrated using a publicly available data-set, in order to provide proof of concept for the work undertaken in [1], where data is unable to be made open.

Predicting Suicidal Behavior Without Asking About Suicidal Ideation: Machine Learning and the Role of Borderline Personality Disorder Criteria

OBJECTIVE

Identifying predictors contributing to suicide risk could help prevent suicides via targeted interventions. However, using only known risk factors may not yield accurate enough results. Furthermore, risk models typically rely on suicidal ideation, even though people often withhold this information.

METHOD

This study examined the contribution of various predictors to the accuracy of six machine learning models for identifying suicidal behavior in a prison population (n = 353), including borderline personality disorder (BPD) and antisocial personality disorder (APD) criteria, and compared how excluding data about suicidal ideation affects accuracy.

RESULTS

Results revealed that gradient tree boosting accurately identified individuals with suicidal behavior, even without relying on questions about suicidal ideation (AUC = 0.875, F1 = 0.846). Furthermore, the model maintained this accuracy with only 29 predictors. Meeting five or more diagnostic criteria of BPD was an important risk factor for suicidal behavior. APD criteria, in the presence of other predictors, did not substantially improve accuracy. Additionally, it may be possible to implement a decision tree model to assess individuals at risk of suicide, without focusing upon suicidal ideation.

CONCLUSIONS

These findings highlight that modern classification algorithms do not necessarily require information about suicidal ideation for modeling suicide and self-harm behavior.

Image Captioning using Facial Expression and Attention

Benefiting from advances in machine vision and natural language processing techniques, current image captioning systems are able to generate detailed visual descriptions. For the most part, these descriptions represent an objective characterisation of the image, although some models do incorporate subjective aspects related to the observer’s view of the image, such as sentiment; current models, however, usually do not consider the emotional content of images during the caption generation process. This paper addresses this issue by proposing novel image captioning models which use facial expression features to generate image captions. The models generate image captions using long short-term memory networks applying facial features in addition to other visual features at different time steps. We compare a comprehensive collection of image captioning models with and without facial features using all standard evaluation metrics. The evaluation metrics indicate that applying facial features with an attention mechanism achieves the best performance, showing more expressive and more correlated image captions, on an image caption dataset extracted from the standard Flickr 30K dataset, consisting of around 11K images containing faces. An analysis of the generated captions finds that, perhaps unexpectedly, the improvement in caption quality appears to come not from the addition of adjectives linked to emotional aspects of the images, but from more variety in the actions described in the captions.