Psychiatric practice routinely uses semistructured and/or unstructured free text to record the behavior and mental state of patients. Many of these data are unstructured, lack standardization, and are difficult to use for analysis. Thus, it is difficult to quantitatively analyze a patient's illness trajectory over time and his or her responsiveness to treatment, and it is also difficult to compare different patients quantitatively. In this article, experts in the field of psychiatry, along with machine learning models, have collaboratively transformed patient data available in status assessments generated by physicians into binary vector representations. Data from patients with mental health disorders collected within a real-world clinical setting from one of the largest behavioral electronic health record (EHR) systems in the United States have been used for generating these representations. The binary vector representation of these health records is shown to be useful in various clinical tasks, such as disease phenotyping, characterizing the suicidality of patients, and inferring diagnoses. To summarize, this approach can transform semistructured free-text summaries of patients' status assessments into a structured, quantifiable format, which enriches the data that reside within EHR systems. This allows for effective intra-and interpatient quantifications and comparisons, which are much needed in the field of mental health. With the aid of these binary representations, patients' mental states can be systematically tracked over time, as can their responses to medications at the individual and population levels.
To describe attrition patterns of opioid use disorder (OUD) patients treated with buprenorphine (BUP) and to assess how clinical, sociodemographic, or BUP medication dosing features are associated with attrition. Patients and Methods: Electronic health records of adults (16+ year-olds) with OUD treated with BUP from 23 different substance use or mental health care programs across 11 US states were examined for one year following BUP initiation in inpatient (IP), intensive outpatient (IOP), or outpatient (OP) settings. Treatment attrition was declared at >37 days following the last recorded visit. Survival analyses and predictive modelling were used. Results: Retention was consistently 2-3 times higher following BUP initiation in OP (n = 2409) than in IP/IOP (n = 2749) settings after 2 (50% vs 25%), 6 (27% vs 9%) and 12 months (14% vs 4%). Retention was higher for females, whites (vs blacks), and those with less severe OUD, better global function, or not using non-psychotropic medications. Comorbid substance use, other psychiatric disorders, and the number of psychotropic medications were variously related to retention depending on the setting in which BUP was initiated. Predictive modelling revealed that a higher global assessment of functioning and a smaller OUD severity based on the Clinical Global Impression -Severity led to longer retentions, a higher initial BUP dose led to higher retention in a few cases, an OP setting of BUP initiation led to longer retentions, and a lower total number of psychotropic and nonpsychotropic medications led to longer retentions. These were the most important parameters in the model, which identified 75.2% of patients who left BUP treatment within three months post-initiation, with a precision of 90.5%. Conclusion: Of all the OUD patients who began BUP, 50-75% left treatment within three months, and most could be accurately identified. This could facilitate patient-centered management to better retain OUD patients in BUP treatment.
To train robust deep neural networks (DNNs), we systematically study several target modification approaches, which include output regularisation, self and non-self label correction (LC). Three key issues are discovered: (1) Self LC is the most appealing as it exploits its own knowledge and requires no extra models. However, how to automatically decide the trust degree of a learner as training goes is not well answered in the literature. (2) Some methods penalise while the others reward low-entropy predictions, prompting us to ask which one is better. (3) Using the standard training setting, a trained network is of low confidence when severe noise exists, making it hard to leverage its high-entropy self knowledge. To resolve the issue (1), taking two well-accepted propositions-deep neural networks learn meaningful patterns before fitting noise and minimum entropy regularisation principle-we propose a novel end-to-end method named ProSelfLC, which is designed according to learning time and entropy. Specifically, given a data point, we progressively increase trust in its predicted label distribution versus its annotated one if a model has been trained for enough time and the prediction is of low entropy (high confidence). For the issue (2), according to ProSelfLC, we empirically prove that it is better to redefine a meaningful low-entropy status and optimise the learner toward it. This serves as a defence of entropy minimisation. To address the issue (3), we decrease the entropy of self knowledge using a low temperature before exploiting it to correct labels, so that the revised labels redefine a low-entropy target state. We demonstrate the effectiveness of ProSelfLC through extensive experiments in both clean and noisy settings, and on both image and protein datasets. Furthermore, our source code is available at https://github.com/XinshaoAmosWang/ProSelfLC-AT.
Purpose Life engagement encompasses concepts such as life fulfillment, well-being, and participation in meaningful activities, encompassing cognitive, physical, social, and emotional dimensions. Patients with MDD experience impaired functioning across multiple domains of life engagement and have ranked concepts related to life engagement and fulfillment as important predictors of treatment success. Post-hoc analyses of three clinical trials of patients with MDD treated adjunctively with brexpiprazole have reported a significantly greater improvement in life engagement. This study investigated improvements in life engagement among patients with MDD following initiation of brexpiprazole treatment using a real-world dataset. Patients and Methods Information was extracted from semi-structured clinical notes of the Mental Status Examination (MSE) of patients in a real-world setting to develop an outcome measure for quantifying life engagement of psychiatric patients. Measures of life engagement and its four sub-domains (emotional, physical, social, and cognitive) were calculated at each clinical visit for 624 adult patients with MDD during the 6 months following brexpiprazole initiation. Paired t-tests assessed differences between the index event and time periods within 6 months of the index event. Kaplan–Meier survival analyses were used to quantify the improvement in life engagement scores following brexpiprazole initiation. Results The study identified 54 clinical features associated with life engagement. Statistically significant improvements were observed from as early as 1 month following brexpiprazole initiation, with 20.6%, 37.9%, and 53.9% of the patients demonstrating improved life engagement scores within 1, 3, and 6 months, respectively. The improvements were particularly apparent for the emotional and social sub-domains. Conclusion The results of this study provide evidence of improved life engagement following brexpiprazole initiation in a real-world dataset.
To train robust deep neural networks (DNNs), we systematically study several target modification approaches, which include output regularisation, self and non-self label correction (LC). Three key issues are discovered: (1) Self LC is the most appealing as it exploits its own knowledge and requires no extra models. However, how to automatically decide the trust degree of a learner as training goes is not well answered in the literature. (2) Some methods penalise while the others reward low-entropy predictions, prompting us to ask which one is better. (3) Using the standard training setting, a trained network is of low confidence when severe noise exists, making it hard to leverage its high-entropy self knowledge. To resolve the issue (1), taking two well-accepted propositions--deep neural networks learn meaningful patterns before fitting noise and minimum entropy regularisation principle--we propose a novel end-to-end method named ProSelfLC, which is designed according to learning time and entropy. Specifically, given a data point, we progressively increase trust in its predicted label distribution versus its annotated one if a model has been trained for enough time and the prediction is of low entropy (high confidence). For the issue (2), according to ProSelfLC, we empirically prove that it is better to redefine a meaningful low-entropy status and optimise the learner toward it. This serves as a defence of entropy minimisation. To address the issue (3), we decrease the entropy of self knowledge using a low temperature before exploiting it to correct labels, so that the revised labels redefine a low-entropy target state. We demonstrate the effectiveness of ProSelfLC through extensive experiments in both clean and noisy settings, and on both image and protein datasets. Furthermore, our source code is available at https://github.com/XinshaoAmosWang/ProSelfLC-AT.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.