Using text mining to extract depressive symptoms and to validate the diagnosis of major depressive disorder from electronic health records

Population-attributable fraction of psychiatric and physical disorders for suicide among older adults in Taiwan

BACKGROUND

This study aimed to estimate the population-attributable fraction (PAF) of psychiatric and physical disorders for suicide among older adults, focusing on sex- and age-specific factors.

METHODS

Data from Taiwan's National Health Insurance Research Data and National Death Registry included 9136 cases of suicide in individuals aged 65+, with 89,439 matched controls. Physical and psychiatric disorders were identified through diagnostic records. Conditional logistic regression assessed risk factors, and PAF was calculated using disorder prevalence and adjusted odds ratios.

RESULTS

Major suicide risk factors among older adults were depressive disorders, anxiety disorders, and sleep disorders. Physical disorders like hypertension, peptic ulcers, and cancer also showed significant PAF values. The combined PAF of physical disorders equaled that of psychiatric disorders. Psychiatric disorders had a greater impact on women and the youngest-old adults, while physical disorders had a higher contribution among men, middle-old adults, and oldest-old adults.

LIMITATIONS

Relying solely on claim data to identify psychiatric and physical disorders may underestimate their prevalence and associations with suicide due to unrecorded cases of individuals not seeking help and the absence of key risk factors like social isolation and family support.

CONCLUSIONS

This study identifies preventable or treatable risk factors for older adult suicide, emphasizing the need to target specific psychiatric and physical disorders in suicide prevention efforts while taking into account sex- and age-specific considerations. It also underscores the importance of establishing social welfare support systems to address the unique challenges older adults face.

An examination of the mediating role of maladaptive emotion regulation strategies in the complex relationship between interpersonal needs and suicidal behavior

Background

Studies have shown that psychological factors, notably interpersonal needs and emotion regulation, play a significant role in suicidal behavior. Interpersonal needs are significant contextual components that affect emotion regulation and contribute to a wide range of dysfunctional behaviors, such as suicidal behavior. It has been postulated that emotion regulation mediates the associations between proximal and distal risk factors of suicidal behavior.

Method

The sample consisted of 340 community-dwelling individuals (62.5% women; SD = 0.48) with an age range of 18 through 55 (M = 30.23; SD = 8.54) who completed the interpersonal needs questionnaire, the suicide behaviors questionnaire-revised, and the cognitive emotion regulation questionnaire. The Structural Equation Modeling (SEM) approach was utilized to evaluate a mediation model.

Results

The findings indicate that interpersonal needs (i.e., perceived burdensomeness r = .55, p <.01 and thwarted belongingness r = .25, p <.01) and putatively maladaptive cognitive emotion regulation strategies (i.e., self-blame; r = .38, p <.01, catastrophizing; r = .55, p <.01, rumination; r = .40, p <.01, and other blame; r = .44, p <.01) have strong associations with suicidal behavior, and these strategies have a mediating effect on the association between interpersonal needs and suicidal behavior.

Conclusions

Our findings show that contextual-interpersonal needs, which underpin suicidal behavior, are significantly influenced by maladaptive emotional processes. Thus, therapeutic outcomes might be enhanced by focusing on the content of the associated cognitions and trying to reduce maladaptive regulatory processes like rumination and catastrophization.

Risks of mental disorders among inpatients with burn injury: A nationwide cohort study

OBJECTIVE

This investigation identified the association between burn injuries and the risk of mental disorders in patients with no documented pre-existing psychiatric comorbidities. We also examined the relationship of injury severity and the types of injury with the likelihood of receiving new diagnoses of mental disorders.

METHODS

This population-based retrospective cohort study used administrative data extracted from the Taiwanese National Health Insurance Research Database (NHIRD) between 2000 and 2013. In total, 10,045 burn survivors were matched with a reference cohort of 40,180 patients without burn injuries and were followed to determine if any mental disorder was diagnosed. Patients diagnosed with mental disorders in the five years before study initiation were excluded to ensure incident diagnoses throughout the research duration. Generalized estimating equations in Cox proportional hazard regression models were used for data analysis.

RESULTS

In general, burn injury survivors have a 1.21-fold risk of being diagnosed with new mental disorders relative to patients without burn injuries. Total body surface area (TBSA) of ≧ 30% (aHR: 1.49, 95% CI: 1.36-1.63) and third- or fourth-degree burns (aHR: 1.49, 95% CI: 1.37-1.63) had a significantly greater risk of being diagnosed with mental disorders in comparison to the reference cohort. Patients TBSA 10-29% (aHR: 0.85, 95% CI: 0.77-0.93) and first- or second-degree burn victims (aHR: 0.89, 95% CI: 0.81-0.97) had relatively lower risk of mental disorders than the reference cohort.

CONCLUSION

Burn injuries were associated with an increased risk of mental disorders. Additional research in this field could elucidate this observation, especially if the inherent limitations of the NHIRD can be overcome.

Clinical outcomes, medical costs, and medication usage patterns of different somatic symptom disorders and functional somatic syndromes: a population-based study in Taiwan

BACKGROUND

Somatic symptom disorders (SSD) and functional somatic syndromes (FSS) are often regarded as similar diagnostic constructs; however, whether they exhibit similar clinical outcomes, medical costs, and medication usage patterns has not been examined in nationwide data. Therefore, this study focused on analyzing SSD and four types of FSS (fibromyalgia, chronic fatigue syndrome, irritable bowel syndrome, functional dyspepsia).

METHODS

This population-based matched cohort study utilized Taiwan's National Health Insurance (NHI) claims database to investigate the impact of SSD/FSS. The study included 2 615 477 newly diagnosed patients with SSD/FSS and matched comparisons from the NHI beneficiary registry. Healthcare utilization, mortality, medical expenditure, and medication usage were assessed as outcome measures. Statistical analysis involved Cox regression models for hazard ratios, generalized linear models for comparing differences, and adjustment for covariates.

RESULTS

All SSD/FSS showed significantly higher adjusted hazard ratios for psychiatric hospitalization and all-cause hospitalization compared to the control group. All SSD/FSS exhibited significantly higher adjusted hazard ratios for suicide, and SSD was particularly high. All-cause mortality was significantly higher in all SSD/FSS. Medical costs were significantly higher for all SSD/FSS compared to controls. The usage duration of all psychiatric medications and analgesics was significantly higher in SSD/FSS compared to the control group.

CONCLUSION

All SSD/FSS shared similar clinical outcomes and medical costs. The high hazard ratio for suicide in SSD deserves clinical attention.

Predicting which patients with cancer will see a psychiatrist or counsellor from their initial oncology consultation document using natural language processing

BACKGROUND

Patients with cancer often have unmet psychosocial needs. Early detection of who requires referral to a counsellor or psychiatrist may improve their care. This work used natural language processing to predict which patients will see a counsellor or psychiatrist from a patient's initial oncology consultation document. We believe this is the first use of artificial intelligence to predict psychiatric outcomes from non-psychiatric medical documents.

METHODS

This retrospective prognostic study used data from 47,625 patients at BC Cancer. We analyzed initial oncology consultation documents using traditional and neural language models to predict whether patients would see a counsellor or psychiatrist in the 12 months following their initial oncology consultation.

RESULTS

Here, we show our best models achieved a balanced accuracy (receiver-operating-characteristic area-under-curve) of 73.1% (0.824) for predicting seeing a psychiatrist, and 71.0% (0.784) for seeing a counsellor. Different words and phrases are important for predicting each outcome.

CONCLUSION

These results suggest natural language processing can be used to predict psychosocial needs of patients with cancer from their initial oncology consultation document. Future research could extend this work to predict the psychosocial needs of medical patients in other settings.

Widowhood and mortality risk in Taiwan: a population-based matched cohort study

BACKGROUND

Studying the causes of death among deceased spouses and surviving partners may provide insights into the underlying mechanisms of the association between widowhood and mortality. This study investigated the mortality risk of widowhood in Taiwan, examined the association of the cause of death between widowed individuals and their deceased spouses and explored potential modifying effects by age, gender and duration after widowhood.

METHODS

This matched cohort study utilized Taiwan's National Health Insurance claims database and National Death Registry. In total, 204 010 widowed men and 596 136 widowed women were identified with a mean follow-up period of 6.9 and 7.9 years, respectively, and 816 040 comparison men and 2 384 544 comparison women were selected.

RESULTS

Widowhood was associated with an increased mortality risk, with widowed men having a 1.32 increased risk and widowed women having a 1.27 increased risk. Age at spousal death and duration modified the associations after widowhood. The widowed individuals are more likely to die by the same cause as the deceased spouse if they died by suicide, accident, endocrine, gastrointestinal disorders or infection.

CONCLUSIONS

The study suggests that healthcare policies and interventions should be developed to improve widowed individuals' health and overall welfare.

Increased risk of acute stress disorder and post-traumatic stress disorder in children and adolescents with autism spectrum disorder: a nation-wide cohort study in Taiwan

Introduction

Children and adolescents with autism spectrum disorder (ASD) may be particularly vulnerable to the impact of traumatic events, yet the association between ASD and the risk of developing acute stress disorder and post-traumatic stress disorder (PTSD) remains uncertain. This study aims to investigate this association, addressing the gap in large-scale evidence on the subject.

Methods

Conducted as a retrospective and matched cohort study, data was sourced from the National Health Insurance Research Database (NHIRD) in Taiwan, spanning from January 1, 2000, to December 31, 2015. The study included patients aged 18 years or under newly diagnosed with ASD (n=15,200) and compared them with a matched control group (n=45,600). The Cox proportional regression model was employed to assess the risk of acute stress disorder and PTSD.

Results

Over the 15-year follow-up period, a total of 132 participants developed either acute stress disorder or PTSD. Among them, 105 cases (0.691% or 64.90 per 100,000 person-years) were in the ASD group, while 27 cases (0.059% or 5.38 per 100,000 person-years) were in the control group. The adjusted hazard ratio for the ASD group was significantly higher compared to the control group (25.661 with 95% CI = 15.913-41.232; P < .001).

Discussion

This study provides compelling evidence that individuals with ASD face an elevated risk of developing acute stress disorder and PTSD. The findings underscore the importance of clinicians recognizing and addressing this vulnerability in ASD individuals exposed to traumatic events. This emphasizes the need for heightened attention to the risk of PTSD and acute stress disorder in the ASD population.

Unlocking the Secrets Behind Advanced Artificial Intelligence Language Models in Deidentifying Chinese-English Mixed Clinical Text: Development and Validation Study

BACKGROUND

The widespread use of electronic health records in the clinical and biomedical fields makes the removal of protected health information (PHI) essential to maintain privacy. However, a significant portion of information is recorded in unstructured textual forms, posing a challenge for deidentification. In multilingual countries, medical records could be written in a mixture of more than one language, referred to as code mixing. Most current clinical natural language processing techniques are designed for monolingual text, and there is a need to address the deidentification of code-mixed text.

OBJECTIVE

The aim of this study was to investigate the effectiveness and underlying mechanism of fine-tuned pretrained language models (PLMs) in identifying PHI in the code-mixed context. Additionally, we aimed to evaluate the potential of prompting large language models (LLMs) for recognizing PHI in a zero-shot manner.

METHODS

We compiled the first clinical code-mixed deidentification data set consisting of text written in Chinese and English. We explored the effectiveness of fine-tuned PLMs for recognizing PHI in code-mixed content, with a focus on whether PLMs exploit naming regularity and mention coverage to achieve superior performance, by probing the developed models' outputs to examine their decision-making process. Furthermore, we investigated the potential of prompt-based in-context learning of LLMs for recognizing PHI in code-mixed text.

RESULTS

The developed methods were evaluated on a code-mixed deidentification corpus of 1700 discharge summaries. We observed that different PHI types had preferences in their occurrences within the different types of language-mixed sentences, and PLMs could effectively recognize PHI by exploiting the learned name regularity. However, the models may exhibit suboptimal results when regularity is weak or mentions contain unknown words that the representations cannot generate well. We also found that the availability of code-mixed training instances is essential for the model's performance. Furthermore, the LLM-based deidentification method was a feasible and appealing approach that can be controlled and enhanced through natural language prompts.

CONCLUSIONS

The study contributes to understanding the underlying mechanism of PLMs in addressing the deidentification process in the code-mixed context and highlights the significance of incorporating code-mixed training instances into the model training phase. To support the advancement of research, we created a manipulated subset of the resynthesized data set available for research purposes. Based on the compiled data set, we found that the LLM-based deidentification method is a feasible approach, but carefully crafted prompts are essential to avoid unwanted output. However, the use of such methods in the hospital setting requires careful consideration of data security and privacy concerns. Further research could explore the augmentation of PLMs and LLMs with external knowledge to improve their strength in recognizing rare PHI.

Natural language processing with machine learning methods to analyze unstructured patient-reported outcomes derived from electronic health records: A systematic review

OBJECTIVE

Natural language processing (NLP) combined with machine learning (ML) techniques are increasingly used to process unstructured/free-text patient-reported outcome (PRO) data available in electronic health records (EHRs). This systematic review summarizes the literature reporting NLP/ML systems/toolkits for analyzing PROs in clinical narratives of EHRs and discusses the future directions for the application of this modality in clinical care.

METHODS

We searched PubMed, Scopus, and Web of Science for studies written in English between 1/1/2000 and 12/31/2020. Seventy-nine studies meeting the eligibility criteria were included. We abstracted and summarized information related to the study purpose, patient population, type/source/amount of unstructured PRO data, linguistic features, and NLP systems/toolkits for processing unstructured PROs in EHRs.

RESULTS

Most of the studies used NLP/ML techniques to extract PROs from clinical narratives (n = 74) and mapped the extracted PROs into specific PRO domains for phenotyping or clustering purposes (n = 26). Some studies used NLP/ML to process PROs for predicting disease progression or onset of adverse events (n = 22) or developing/validating NLP/ML pipelines for analyzing unstructured PROs (n = 19). Studies used different linguistic features, including lexical, syntactic, semantic, and contextual features, to process unstructured PROs. Among the 25 NLP systems/toolkits we identified, 15 used rule-based NLP, 6 used hybrid NLP, and 4 used non-neural ML algorithms embedded in NLP.

CONCLUSIONS

This study supports the potential utility of different NLP/ML techniques in processing unstructured PROs available in EHRs for clinical care. Though using annotation rules for NLP/ML to analyze unstructured PROs is dominant, deploying novel neural ML-based methods is warranted.

Associations between depression and cancer risk among patients with diabetes mellitus: A population-based cohort study

BACKGROUND

The co-occurrence of depression and diabetes mellitus has been linked to an increased risk of developing cancer. This study aimed to investigate whether depression further amplifies the risk of cancer among individuals with diabetes.

METHODS

This population-based matched cohort study utilized Taiwan's National Health Insurance claims database. A total of 85,489 newly diagnosed diabetic patients with depressive disorders were selected, along with 427,445 comparison subjects. The matching process involved age, sex, and the calendar year of diabetes onset. The average follow-up duration for the two cohorts was 6.4 and 6.5 years, respectively. The primary outcome of interest was the occurrence of overall cancer or cancer at specific anatomical sites.

RESULTS

The adjusted hazard ratios for overall cancer incidence were 1.08 (95% CI, 1.05-1.11). For site-specific cancers, depression exhibited significant associations with oropharyngeal, esophageal, liver, gynecological, prostate, kidney, and hematologic malignancies among patients with diabetes. Notably, a severity-response relationship was observed, indicating that patients with recurrent episodes of major depressive disorders exhibited a higher incidence of cancer compared to those diagnosed with dysthymia or depressive disorder not otherwise specified. Furthermore, the strength of the association between depression and cancer risk was more pronounced among younger patients with diabetes as opposed to older adults. However, no significant relationship was observed between adherence to antidepressant treatment and cancer risk.

CONCLUSIONS

The findings of this study indicate a significant association between depression and an elevated risk of cancer among individuals diagnosed with diabetes. Future investigations should replicate our findings, explore the effects of pharmacological and non-pharmacological treatments on cancer risk, and identify the underlying mechanisms.

Population-attributable risk of psychiatric disorders for suicide among adolescents and young adults in Taiwan

BACKGROUND

Youth suicide rates have increased markedly in some countries. This study aimed to estimate the population-attributable risk of psychiatric disorders associated with suicide among Taiwanese youth aged 10-24 years.

METHODS

Data were obtained from the National Death Registry and National Health Insurance (NHI) claims database between 2007 and 2019. Youth who died by suicide were included, and comparisons, 1:10 matched by age and sex, were randomly selected from the Registry for NHI beneficiaries. We used multivariable logistic regression to estimate suicide odds ratios for psychiatric disorders. The population-attributable fractions (PAF) were calculated for each psychiatric disorder.

RESULTS

A total of 2345 youth suicide and 23 450 comparisons were included. Overall, 44.8% of suicides had a psychiatric disorder, while only 7.9% of the comparisons had a psychiatric disorder. The combined PAF for all psychiatric disorders was 55.9%. The top three psychiatric conditions of the largest PAFs were major depressive disorder, dysthymia, and sleep disorder. In the analysis stratified by sex, the combined PAF was 45.5% for males and 69.2% for females. The PAF among young adults aged 20-24 years (57.0%) was higher than among adolescents aged 10-19 years (48.0%).

CONCLUSIONS

Our findings of high PAF from major depressive disorder, dysthymia, and sleep disorder to youth suicides suggest that youth suicide prevention that focuses on detecting and treating mental illness may usefully target these disorders.

Validation of diagnosis codes in healthcare databases in Taiwan, a literature review

PURPOSE

To compile validation findings of diagnosis codes and related algorithms for health outcomes of interest from National Health Insurance (NHI) or electronic medical records in Taiwan.

METHODS

We carried out a literature review of English articles in PubMed® and Embase from 2000 through July 2022 with appropriate search terms. Potentially relevant articles were identified through review of article titles and abstracts, full text search of methodology terms "validation", "positive predictive value", and "algorithm" in Subjects & Methods (or Methods) and Results sections of articles, followed by full text review of potentially eligible articles.

RESULTS

We identified 50 published reports with validation findings of diagnosis codes and related algorithms for a wide range of health outcomes of interest in Taiwan, including cardiovascular diseases, stroke, renal impairment, malignancy, diabetes, mental health diseases, respiratory diseases, viral (B and C) hepatitis, and tuberculosis. Most of the reported PPVs were in the 80% ~ 99% range. Assessment of algorithms based on ICD-10 systems were reported in 8 articles, all published in 2020 or later.

CONCLUSIONS

Investigators have published validation reports that may serve as empirical evidence to evaluate the utility of secondary health data environment in Taiwan for research and regulatory purpose.

Cause-specific mortality after discharge from inpatient psychiatric care in Taiwan: A national matched cohort study

AIMS

We aimed to investigate the trajectories of absolute and relative risks of cause-specific mortality among patients discharged from inpatient psychiatric services.

METHODS

We conducted a national matched cohort study (2002-2013) using data from the Taiwan National Health Insurance database linked to national cause-of-death data files. Patients discharged from inpatient psychiatric care without prior psychiatric hospitalizations were individually matched to 20 comparison individuals based on sex and age. The rates, rate differences, and relative risks (hazard ratios, HRs) of cause-specific mortality were calculated at six follow-up periods post-discharge. Cumulative mortality incidence was assessed at 5 years of follow-up.

RESULTS

The mortality risks of all causes were increased among patients (n = 158 065) relative to comparison individuals (n = 3 161 300). Mortality rate differences were greater for natural causes, while relative risks (HRs) were higher for unnatural causes. Suicide was the leading cause of death within the first year of discharge, while circulatory and respiratory diseases were the leading causes of death from the second year. The mortality rates and HRs for all causes of death (except homicide) were highest during the first 3 months. The elevated risk of unnatural-cause mortality declined rapidly after discharge but remained high in the long term; in contrast, risk elevation for natural-cause mortality was more stable over time. Approximately one-eighth of patients (12.9%, 95% confidence interval 12.7-13.7%) died within 5 years of follow-up.

CONCLUSIONS

Integrated physical and mental health care is needed to reduce excess mortality, particularly during the first 3 months post-discharge, among psychiatric patients.

Neonatal and pregnancy complications following maternal depression or antidepressant exposure: A population-based, retrospective birth cohort study

OBJECTIVES

Depression is common during pregnancy, and antidepressants are often prescribed for treatment. However, depression and antidepressant use both increase the risk of neonatal and pregnancy complications. To separately evaluate the effects of antidepressant use and the underlying depression on pregnancy and neonatal complications by using a robust statistical method to control for confounding by indication.

METHODS

All study data were obtained from Taiwan's National Health Insurance Research Database. Pregnant women were divided into three groups: those with no depression and no antidepressant exposure(n = 1619,198), depression and no antidepressant exposure(n = 2006), and depression and antidepressant exposure(n = 7857). Antidepressant exposure was further divided into that before pregnancy and during each trimester.

RESULTS

Mothers with depression but no antidepressant exposure exhibited increased risks of intrauterine growth restriction and preterm delivery, compared with mothers without depression. In mothers with depression, antidepressant exposure before pregnancy or during the first trimester conferred increased risks of gestational diabetes mellitus, malpresentation, preterm delivery and cardiovascular anomalies, compared with no antidepressant exposure. Moreover, antidepressant exposure during the second or third trimester conferred increased risks of anemia, a low Apgar score, preterm delivery and genitourinary defects. However, antidepressants administered before pregnancy and during all trimesters did not increase the risk of stillbirth.

CONCLUSION

Depression and antidepressant treatment for depression during pregnancy may individually increase the risks of some neonatal and pregnancy complications. Physicians should thoroughly consider the risks and benefits for both the mother and fetus when treating depression during pregnancy by using antidepressants.

Mortality in anorexia nervosa and bulimia nervosa: A population-based cohort study in Taiwan, 2002-2017

OBJECTIVE

To investigate natural- and unnatural-cause mortality at different follow-up time points in Taiwanese patients with anorexia nervosa (AN) and bulimia nervosa (BN).

METHOD

In this longitudinal cohort study, 330,393 patients, including 2143 patients with AN, 13,590 with BN, and 20 times as many respective non-AN and non-BN patients, were followed up for 16 years. We performed conditional Cox regression survival analysis to examine the risk of mortality in the AN and BN groups relative to the comparison group.

RESULTS

A total of 1242 patients died, including 101 and 343 patients with AN and BN, respectively. Mortality rates for AN and BN were 5.42 and 2.90 deaths per 1000 person-years, respectively. Compared with the non-AN group, the AN group had a significantly higher risk of both natural- and unnatural-cause mortality, and the BN group had a significantly higher risk of unnatural-cause mortality. Suicide was the most common cause of death, and suicide risk was significantly higher in both the AN and BN groups. All-cause mortality risk was the highest at the beginning of follow-up and markedly declined in the AN group. In the BN group, all-cause mortality risk was lower but stable at follow-up. The risk of unnatural-cause mortality remained high throughout the follow-up period for both the groups.

CONCLUSIONS

Early detection and treatment for associated physical problems in patients with AN are crucial. Regular monitoring for unnatural-cause mortality events (mainly suicide) in AN and BN over time is also crucial.

PUBLIC SIGNIFICANCE

AN had a significantly higher risk of both natural- and unnatural-cause mortality and BN had a significantly higher risk of death from unnatural causes. All-cause mortality risk was highest at the beginning of follow-up in AN, but unnatural-cause mortality risk remained high throughout the follow-up period for both groups. Our findings imply that early detection and treatment in AN and regular monitoring for unnatural-cause mortality events in AN and BN are crucial.

Comparison of relative areas of interest between major depression disorder and postpartum depression

INTRODUCTION

Postpartum depression (PPD) is defined as a major depressive disorder (MDD) beginning after childbirth. Wide debates aim to better understand PPD's specificities compared with MDD. One of the keys in differentiating PPD from MDD is to systematically study scientific "Areas Of Interest" (AOIs) of these disorders.

METHODS

In November 2021, we performed an extraction and textual computational analysis of associated terms for PPD and MDD, using the biomedical database PubMed. We performed an undirected lexical network analysis to map the 150 first terms in space. Then, we used an unsupervised machine learning technique to detect word patterns and automatically cluster AOIs with a topic-modeling analysis.

RESULTS

We identified 30,000 articles of the 554,724 articles for MDD and 15,642 articles for PPD. Four AOIs were detected in the MDD network: mood disorders and their treatments, risk factors, consequences and quality of life, and mental health and comorbidities. Five AOIs were detected in the PPD network: mood disorders and treatments, risk factors, consequences and child health, patient's background, and the challenges of screening.

DISCUSSION AND CONCLUSION

Limitations are both methodological, in particular due to the qualitative interpretation of AOIs, and are also related to the difficult transferability of these research results to the clinical practice. The partial overlap between AOIs for MDD and for PPD suggest that the latter is a particular form of the former.

Natural Language Processing in Electronic Health Records in relation to healthcare decision-making: A systematic review

BACKGROUND

Natural Language Processing (NLP) is widely used to extract clinical insights from Electronic Health Records (EHRs). However, the lack of annotated data, automated tools, and other challenges hinder the full utilisation of NLP for EHRs. Various Machine Learning (ML), Deep Learning (DL) and NLP techniques are studied and compared to understand the limitations and opportunities in this space comprehensively.

METHODOLOGY

After screening 261 articles from 11 databases, we included 127 papers for full-text review covering seven categories of articles: (1) medical note classification, (2) clinical entity recognition, (3) text summarisation, (4) deep learning (DL) and transfer learning architecture, (5) information extraction, (6) Medical language translation and (7) other NLP applications. This study follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.

RESULT AND DISCUSSION

EHR was the most commonly used data type among the selected articles, and the datasets were primarily unstructured. Various ML and DL methods were used, with prediction or classification being the most common application of ML or DL. The most common use cases were: the International Classification of Diseases, Ninth Revision (ICD-9) classification, clinical note analysis, and named entity recognition (NER) for clinical descriptions and research on psychiatric disorders.

CONCLUSION

We find that the adopted ML models were not adequately assessed. In addition, the data imbalance problem is quite important, yet we must find techniques to address this underlining problem. Future studies should address key limitations in studies, primarily identifying Lupus Nephritis, Suicide Attempts, perinatal self-harmed and ICD-9 classification.

Interactions of Insomnia and Sedative-Hypnotic Drug Use Associated with Frailty Over Time Among Older Adults

BACKGROUND

Insomnia and frailty are prevalent in older adults. This study aimed to elucidate the impact of insomnia and sedative-hypnotic use on the frailty rate over time.

METHODS

We used data from community-dwelling older adults (mean ± SD age = 69.4 ± 8.2 years) from the Healthy Aging Longitudinal Study in Taiwan (HALST). A total of 4,744 participants were included in the study and were followed up for an average of 3.2 years. Frailty was assessed using the Fried criteria. Self-reported sleep problems, sedative-hypnotic use, and claims records from the National Health Insurance database were used. The generalized equation estimation (GEE) approach was applied to account for correlations between repeated measures. The average impact of insomnia and drug use on frailty over time was estimated by adjusting for potential confounding factors using the logic link in the GEE approach.

RESULTS

The adjusted odds ratio (OR) of frailty was 1.41 (95% CI: [1.16, 1.72], Z-test statistics Z = 3.39, p <0.001) for insomnia and 1.52 ([1.16, 2.00], Z = 3.00, p = 0.0027) for sedative-hypnotic use. Interactions between insomnia and sedative-hypnotic use with frailty were not statistically significant. Long sleep duration > 8 hours, daytime sleepiness, and sleep apnea was also associated with an increased likelihood of developing frailty. Notably, a dose-response relationship between sedative-hypnotic drug use and frailty was observed.

CONCLUSIONS

Insomnia and sedative-hypnotic use were independently associated with increased frailty. The implementation of nonpharmacological treatments to attenuate insomnia may reduce frailty rates.

Augmenting DSM-5 diagnostic criteria with self-attention-based BiLSTM models for psychiatric diagnosis

BACKGROUND

Most previous studies make psychiatric diagnoses based on diagnostic terms. In this study we sought to augment Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5) diagnostic criteria with deep neural network models to make psychiatric diagnoses based on psychiatric notes.

METHODS

We augmented DSM-5 diagnostic criteria with self-attention-based bidirectional long short-term memory (BiLSTM) models to identify schizophrenia, bipolar, and unipolar depressive disorders. Given that the diagnostic criteria for psychiatric diagnosis include a certain symptom profile and functional impairment, we first extracted psychiatric symptoms and functional features with two approaches, including a lexicon-based approach and a dependency parsing approach. Then, we incorporated free-text discharge notes and extracted features for psychiatric diagnoses with the proposed models.

RESULTS

The micro-averaged F1 scores of the two automatic annotation approaches were greater than 0.8. BiLSTM models with self-attention outperformed the rule-based models with DSM-5 criteria in the prediction of schizophrenia and bipolar disorder, while the latter outperformed the former in predicting unipolar depressive disorder. Approaches for augmenting DSM-5 criteria with a self-attention-based BiLSTM outperformed both pure rule-based and pure deep neural network models. In terms of classification of psychiatric diagnoses, we observed that the performance for schizophrenia and bipolar disorder was acceptable.

CONCLUSION

This DSM-5-augmented deep neural network models showed good performance in identifying psychiatric diagnoses from psychiatric notes. We conclude that it is possible to establish a model that consults clinical notes to make psychiatric diagnoses comparably to physicians. Further research will be extended to outpatient notes and other psychiatric disorders.

Informatics Research on Mental Health Functioning: Decision Support for the Social Security Administration Disability Program

The disability determination process of the Social Security Administration's (SSA's) disability program requires assessing work-related functioning for individual claimants alleging disability due to mental impairment. This task is particularly challenging because the determination process involves the review of a large file of information, including objective medical evidence and self-reports from claimants, families, and former employers. To improve this decision-making process, SSA entered an interagency agreement with the Rehabilitation Medicine Department, Epidemiology and Biostatistics Section, in the Clinical Center of the National Institutes of Health, intending to use data science and informatics to develop decision support tools. This collaborative effort over the past decade has led to the development of the Work Disability-Functional Assessment Battery and has initiated an approach to applying natural language processing to the review of claimants' files for information on mental health functioning. This informatics research collaboration holds promise for improving the process of disability determination for individuals with mental impairments who make claims at the SSA.

Somatic Changes Perceived by Patients With Heart Failure During Acute Exacerbation: A Qualitative Study Using Text Mining

BACKGROUND

Patients with heart failure (HF) often inadequately perceive their symptoms. This may be because the medical terms do not match the somatic changes experienced by patients. To improve symptom perception, healthcare professionals must understand the somatic changes as perceived by patients.

OBJECTIVE

This study aims to analyze patients' narratives about somatic changes in patients with HF by text mining and to clarify the overall description of somatic changes using patients' expressions.

METHODS

Semistructured interviews were conducted on 21 patients hospitalized for acute exacerbation of HF. Qualitative data obtained from the interviews were analyzed by content analysis through text mining.

RESULTS

Among the 21 patients, 76.2% were men. The mean (SD) age was 71.3 (13.7) years. The most frequently used terms were "breath," "distressed," "feet," and " ha-ha (gasping sound)" (46, 40, 29, and 28 times, respectively). The somatic changes noticed by patients could be categorized into medical jargon such as "dyspnea on exertion," "exercise intolerance," "fatigue," "paroxysmal nocturnal dyspnea," "frequent urination," "increased sputum," "weight gain," "feet and face edema," "abdominal edema," and "ankle edema." However, the expressions of somatic changes used by the patients were diverse.

CONCLUSIONS

The findings of patient-specific expressions of symptoms suggest that there is a need to assess symptoms not only using medical jargon but also by focusing on patient-specific expressions.

Associations between parental psychiatric disorders and autism spectrum disorder in the offspring

Whether parental psychiatric disorders are associated with autism spectrum disorder (ASD) in offspring has remained inconclusive. We examined the associations of parental psychiatric disorders with ASD in offspring. This population-based case-control study used Taiwan's National Health Insurance Research Database to identify a cohort of children born from 2004 to 2017 and their parents. A total of 24,279 children with ASD (diagnostic ICD-9-CM code: 299.x or ICD-10 code F84.x) and 97,715 matched controls were included. Parental psychiatric disorders, including depressive disorders, bipolar spectrum disorders, anxiety disorders, obsessive-compulsive disorder, schizophrenia, substance use disorders, autism spectrum disorder, attention-deficit hyperactivity disorder (ADHD), and adjustment disorders were identified. Conditional logistic regressions with covariate adjustment were performed. The results suggest that parental diagnosis with any of the psychiatric disorders is associated with ASD in offspring (adjusted odds ratio [AOR] = 1.45, 95%CI: 1.40-1.51 for mothers; and AOR = 1.12, 95%CI: 1.08-1.17 for fathers). ASD in offspring was associated with schizophrenia, depressive disorders, obsessive-compulsive disorder, adjustment disorders, ADHD and ASD in both parents. The relationship between parental psychiatric disorders and the timing of the child's birth and ASD diagnosis varied across the different psychiatric disorders. The present study provides supportive evidence that parental psychiatric disorders are associated with autistic children. Furthermore, because the associations between parental psychiatric disorders and the timing of child's birth and ASD diagnosis varied across psychiatric disorders, the observed relationships may be affected by both genetic and environmental factors. Future studies are needed to disentangle the potential influence of genetic and environmental factors on the observed associations.

Extraction and Quantification of Words Representing Degrees of Diseases: Combining the Fuzzy C-Means Method and Gaussian Membership

Background

Due to the development of medical data, a large amount of clinical data has been generated. These unstructured data contain substantial information. Extracting useful knowledge from this data and making scientific decisions for diagnosing and treating diseases have become increasingly necessary. Unstructured data, such as in the Marketplace for Medical Information in Intensive Care III (MIMIC-III) data set, contain several ambiguous words that demonstrate the subjectivity of doctors, such as descriptions of patient symptoms. These data could be used to further improve the accuracy of medical diagnostic system assessments. To the best of our knowledge, there is currently no method for extracting subjective words that express the extent of these symptoms (hereinafter, “degree words”).

Objective

Therefore, we propose using the fuzzy c-means (FCM) method and Gaussian membership to quantify the degree words in the clinical medical data set MIMIC-III.

Methods

First, we preprocessed the 381,091 radiology reports collected in MIMIC-III, and then we used the FCM method to extract degree words from unstructured text. Thereafter, we used the Gaussian membership method to quantify the extracted degree words, which transform the fuzzy words extracted from the medical text into computer-recognizable numbers.

Results

The results showed that the digitization of ambiguous words in medical texts is feasible. The words representing each degree of each disease had a range of corresponding values. Examples of membership medians were 2.971 (atelectasis), 3.121 (pneumonia), 2.899 (pneumothorax), 3.051 (pulmonary edema), and 2.435 (pulmonary embolus). Additionally, all extracted words contained the same subjective words (low, high, etc), which allows for an objective evaluation method. Furthermore, we will verify the specific impact of the quantification results of ambiguous words such as symptom words and degree words on the use of medical texts in subsequent studies. These same ambiguous words may be used as a new set of feature values to represent the disorders.

Conclusions

This study proposes an innovative method for handling subjective words. We used the FCM method to extract the subjective degree words in the English-interpreted report of the MIMIC-III and then used the Gaussian functions to quantify the subjective degree words. In this method, words containing subjectivity in unstructured texts can be automatically processed and transformed into numerical ranges by digital processing. It was concluded that the digitization of ambiguous words in medical texts is feasible.

Comparisons of deep learning and machine learning while using text mining methods to identify suicide attempts of patients with mood disorders

BACKGROUND

Suicide attempt is one of the most severe consequences for patients with mood disorders. This study aimed to perform deep learning and machine learning while using text mining to identify patients with suicide attempts and to compare their effectiveness.

METHODS

A total of 13,100 patients with mood disorders were selected. Two traditional text mining methods, logistic regression and Support vector machine (SVM), and one deep learning model (Convolutional neural network, CNN) were adopted to perform overall analysis and gender-specific subgroup analysis of patients to identify suicide attempts. The classification effectiveness of these models was evaluated by accuracy, F1-value, precision, recall, and the area under Receiver operator characteristic curve (ROC).

RESULTS

CNN's results were greater than the other two for all indicators except recall which was slightly smaller than SVM in male subgroup analysis. The accuracy values of the CNN were 98.4 %, 98.2 %, and 98.5 % in the overall analysis and the subgroup analysis for males and females, respectively. The results of McNemar's test showed that CNN and SVM models' predictions were statistically different from the logistic regression model's predictions in the overall analysis and the subgroup analysis for females (P < 0.050).

LIMITATIONS

A fixed number of features were selected based on document frequency to train models; this was a single-site study.

CONCLUSIONS

CNN model was a better way to detect suicide attempts in patients with mood disorders prior to hospital admission, saving time and resources in recognizing high-risk patients and preventing suicide.

Automatic symptoms identification from a massive volume of unstructured medical consultations using deep neural and BERT models

Automatic symptom identification plays a crucial role in assisting doctors during the diagnosis process in Telemedicine. In general, physicians spend considerable time on clinical documentation and symptom identification, which is unfeasible due to their full schedule. With text-based consultation services in telemedicine, the identification of symptoms from a user's consultation is a sophisticated process and time-consuming. Moreover, at Altibbi, which is an Arabic telemedicine platform and the context of this work, users consult doctors and describe their conditions in different Arabic dialects which makes the problem more complex and challenging. Therefore, in this work, an advanced deep learning approach is developed consultations with multi-dialects. The approach is formulated as a multi-label multi-class classification using features extracted based on AraBERT and fine-tuned on the bidirectional long short-term memory (BiLSTM) network. The Fine-tuning of BiLSTM relies on features engineered based on different variants of the bidirectional encoder representations from transformers (BERT). Evaluating the models based on precision, recall, and a customized hit rate showed a successful identification of symptoms from Arabic texts with promising accuracy. Hence, this paves the way toward deploying an automated symptom identification model in production at Altibbi which can help general practitioners in telemedicine in providing more efficient and accurate consultations.

Perceived barriers to psychiatric help-seeking in South Korea by age groups: text mining analyses of social media big data

BACKGROUND

The psychiatric treatment gap is substantial in Korea, implying barriers in seeking help.

OBJECTIVES

This study aims to explore barriers of seeing psychiatrists, expressed on the internet by age groups.

METHODS

A corpus of data was garnered extensively from internet communities, blogs and social network services from 1 January 2016 to 31 July 2019. Among the texts collected, texts containing words linked to psychiatry were selected. Then the corpus was dismantled into words by using natural language processing. Words linked to barriers to seeking help were identified and classified. Then the words from web communities that we were able to identify the age groups were additionally organized by age groups.

RESULTS

97,730,360 articles were identified and 6,097,369 were included in the analysis. Words implying the barriers were selected and classified into four groups of structural discrimination, public prejudice, low accessibility, and adverse drug effects. Structural discrimination was the greatest barrier occupying 34%, followed by public prejudice (27.8%), adverse drug effects (18.6%), and cost/low accessibility (16.1%). In the analysis by age groups, structural discrimination caused teenagers (51%), job seekers (64%) and mothers with children (43%) the most concern. In contrast, the public prejudice (49%) was the greatest barriers in the senior group.

CONCLUSIONS

Although structural discrimination may most contribute to barriers to visiting psychiatrists in Korea, variation by generations may exist. Along with the general attempt to tackle the discrimination, customized approach might be needed.

Associations between Allergic and Autoimmune Diseases with Autism Spectrum Disorder and Attention-Deficit/Hyperactivity Disorder within Families: A Population-Based Cohort Study

Autism spectrum disorder (ASD) and attention-deficit/hyperactivity disorder (ADHD) are commonly comorbid with allergic and autoimmune diseases in children. The aim of the current study was to investigate the association between children’s and first-degree relatives’ (i.e., mother, father, and full sibling) allergic and autoimmune diseases and children’s ASD and ADHD. We enrolled participants from Taiwan’s Maternal and Child Health Database. We used the Cox regression model to examine the associations of familial, siblings’ and children’s allergic and autoimmune diseases with children’s ASD and/or ADHD. In total, we included 1,386,260 children in the current study. We found the significant association between familial allergic or autoimmune disease and development of ASD or ADHD among children. We also identified the predominant impact of familial aggregation on the above associations. The associations between some parental diagnoses of autoimmune or allergic diseases in children’s ASD and/or ADHD were stronger in mothers than those in fathers. Early assessment of the possibility of ASD and ADHD is required for children who have a parent with an allergic or autoimmune disease.

Real-World Evidence for the Association Between Pneumonia-Related Intensive Care Unit Stay and Dementia

OBJECTIVE

There is limited clarity concerning the risk of dementia after pneumonia with intensive care unit (ICU) stay. We conducted a nationwide cohort study, which aimed to investigate the impact of dementia after pneumonia with and without intensive care unit admission.

METHODS

Data was obtained from Taiwan's National Health Insurance Research Database between 2000 and 2015. A total of 7,473 patients were identified as having pneumonia required ICU stay, along with 22,419 controls matched by sex and age. After adjusting for confounding factors, multivariate Cox regression model analysis was used to compare the risk of developing dementia during the 15-years follow-up period.

RESULTS

The enrolled pneumonia patients with ICU admission had a dementia rate of 9.89%. Pneumonia patients without ICU admission had a dementia rate of 9.21%. The multivariate Cox regression model analysis revealed that the patients with ICU stay had the higher risk of dementia, with a crude hazard ratio of 3.371 (95% confidence interval, 3.093-3.675; p<0.001).

CONCLUSION

This study indicated that pneumonia with ICU stay is associated with an increased risk of dementia. A 3-fold risk of dementia was observed in patients admitted to the ICU compared to the control group.

Association Between Antibiotic Treatment of Leptospirosis Infections and Reduced Risk of Dementia: A Nationwide, Cohort Study in Taiwan

Background

To explore the association between leptospirosis, the risk of dementia, and the potential protective role of antibiotic treatment.

Methods

We conducted a retrospective cohort nationwide, population-based study, from Taiwan’s National Health Insurance Research Database (NHIRD). We enrolled 1,428 subjects aged 50 years or above, in the index year of 2000, which included those retrieved from the NHIRD record. Dementia diagnosis and incidence over 16 years follow-up was retrieved from the NHIRD records. The Fine and Gray survival analysis was used to determine the risk of dementia, and the results were presented as a sub-distribution hazard ratio (SHR) with a 95% confidence interval.

Results

In the study period, 43 of the 357 leptospirosis patients developed dementia, as compared to 103 of the control group (930.90 vs. 732.49 per 10⁵ person-years). By the Fine and Gray survival analysis, the leptospirosis was associated with the risk of dementia, and the adjusted SHR was 1.357 (95% confidence interval [CI]: 1.213–1.519, P < 0.001), across 16-year of the follow-up period. To exclude the protopathic bias, the sensitivity analysis was conducted. This analysis revealed that the leptospirosis was associated with the increased risk of dementia, even after excluding the dementia diagnosis within the first year (adjusted SHR = 1.246, 95%CI: 1.114–1.395, P < 0.001) or within the first 5 years (adjusted SHR = 1.079, 95%CI: 1.023–1.152, P = 0.028), antibiotic treatment for leptospirosis was associated with the reduced risk of dementia (P = 0.001).

Conclusion

Leptospirosis was associated with an increased risk for dementia, and antibiotic treatment was associated with a reduced risk. Further research will be necessary to explore the underlying mechanisms of this association.

LASSO Regression Modeling on Prediction of Medical Terms among Seafarers’ Health Documents Using Tidy Text Mining

Generally, seafarers face a higher risk of illnesses and accidents than land workers. In most cases, there are no medical professionals on board seagoing vessels, which makes disease diagnosis even more difficult. When this occurs, onshore doctors may be able to provide medical advice through telemedicine by receiving better symptomatic and clinical details in the health abstracts of seafarers. The adoption of text mining techniques can assist in extracting diagnostic information from clinical texts. We applied lexicon sentimental analysis to explore the automatic labeling of positive and negative healthcare terms to seafarers’ text healthcare documents. This was due to the lack of experimental evaluations using computational techniques. In order to classify diseases and their associated symptoms, the LASSO regression algorithm is applied to analyze these text documents. A visualization of symptomatic data frequency for each disease can be achieved by analyzing TF-IDF values. The proposed approach allows for the classification of text documents with 93.8% accuracy by using a machine learning model called LASSO regression. It is possible to classify text documents effectively with tidy text mining libraries. In addition to delivering health assistance, this method can be used to classify diseases and establish health observatories. Knowledge developed in the present work will be applied to establish an Epidemiological Observatory of Seafarers’ Pathologies and Injuries. This Observatory will be a collaborative initiative of the Italian Ministry of Health, University of Camerino, and International Radio Medical Centre (C.I.R.M.), the Italian TMAS.

Risk of anxiety and depression in patients with lichen planus: A nationwide population-based study

BACKGROUND

This study aims to determine the risk of developing anxiety and/or depression among patients with lichen planus.

METHODS

Based on the Longitudinal Health Insurance Database of Taiwan National Health Insurance Research Database, a total of 4012 patients with lichen planus and 16,048 matched controls (1:4) were enrolled between January 1, 2000, and December 31, 2015. After controlling for the risk variables, multivariate Cox proportional hazard regression and the log-rank test with Kaplan-Meier method were performed to assess the influence of anxiety/depression among individuals with lichen planus under a maximum follow-up period of 16 years.

RESULTS

The subsequent anxiety or depression incidence of the lichen planus group and the comparisons was 19.67% (1962.70 per 10⁵ person-years) and 10.11% (982.23 per 10⁵person-years), respectively. Additionally, after adjustment of the risk variables, the hazard ratios for anxiety, depression, anxiety without depression, depression without anxiety, anxiety or depression, and both anxiety and depression combined were 1.779 (95%CI: 1.289-2.477, p < 0.001), 2.010 (95%CI: 1.454-2.790, p < 0.001), 2.015 (95%CI: 1.463-2.799, p < 0.001), 2.356 (95%CI: 1.705-3.286, p < 0.001), 2.011 (95%CI: 1.457-2.793, p < 0.001), and 1.515 (95%CI: 1.100-2.134, p < 0.001), respectively.

LIMITATIONS

Individuals with lichen planus were unable to be classified into oral subtype and cutaneous subtype based on the ICD-9-CM. Moreover, the results of our study could not demonstrate the mechanism between lichen planus and anxiety and/or depression.

CONCLUSION

Patients with lichen planus was positively associated with developing anxiety or depression. Physicians should to be aware of the signs of anxiety and/or depression while facing the patients with lichen planus during the clinical practices.

Trajectories of Vital Signs and Risk of In-Hospital Cardiac Arrest

Background: Little is known about the trajectories of vital signs prior to in-hospital cardiac arrest (IHCA), which could explain the heterogeneous processes preceding this event. We aimed to identify clinically relevant subphenotypes at high risk of IHCA in the emergency department (ED).

Methods: This retrospective cohort study used electronic clinical warehouse data from a tertiary medical center. We retrieved data from 733,398 ED visits over a 7-year period. We selected one ED visit per person and retrieved patient demographics, triage data, vital signs (systolic blood pressure [SBP], heart rate [HR], body temperature, respiratory rate, oxygen saturation), selected laboratory markers, and IHCA status. Group-based trajectory modeling was performed.

Results: There were 37,697 adult ED patients with a total of 1,507,121 data points across all vital-sign categories. Three to four trajectory groups per vital-sign category were identified, and the following five trajectory groups were associated with a higher rate of IHCA: low and fluctuating SBP, high and fluctuating HR, persistent hypothermia, recurring tachypnea, and low and fluctuating oxygen saturation. The IHCA-prone trajectory group was associated with a higher triage level and a higher mortality rate, compared to other trajectory groups. Except for the persistent hypothermia group, the other four trajectory groups were more likely to have higher levels of C-reactive protein, lactic acid, cardiac troponin I, and D-dimer. Multivariable analysis revealed that hypothermia (adjusted odds ratio [aOR], 2.20; 95% confidence interval [95%CI], 1.35–3.57) and recurring tachypnea (aOR 2.44; 95%CI, 1.24–4.79) were independently associated with IHCA.

Conclusions: We identified five novel vital-sign sub-phenotypes associated with a higher likelihood of IHCA, with distinct patterns in clinical course and laboratory markers. A better understanding of the pre-IHCA vital-sign trajectories may help with the early identification of deteriorating patients.

Serum Lactate for Predicting Cardiac Arrest in the Emergency Department

Objectives: Early recognition and prevention of in-hospital cardiac arrest (IHCA) play an increasingly important role in the Chain of Survival. However, clinical tools for predicting IHCA in the emergency department (ED) are scanty. We sought to evaluate the role of serum lactate in predicting ED-based IHCA. Methods: Data were retrieved from 733,398 ED visits over a 7-year period in a tertiary medical centre. We selected one ED visit per person and excluded out-of-hospital cardiac arrest, children, or those without lactate measurements. Patient demographics, computerised triage information, and serum lactate levels were extracted. The initial serum lactate levels were grouped into normal (≤2 mmol/L), moderately elevated (2 < lactate ≤ 4), and highly elevated (>4 mmol/L) categories. The primary outcome was ED-based IHCA. Results: A total of 17,392 adult patients were included. Of them, 342 (2%) developed IHCA. About 50% of the lactate levels were normal, 30% were moderately elevated, and 20% were highly elevated. In multivariable analysis, the group with highly elevated lactate had an 18-fold increased risk of IHCA (adjusted odds ratio [OR], 18.0; 95% confidence interval [CI], 11.5-28.2), compared with the normal lactate group. In subgroup analysis, the poor lactate-clearance group (<2.5%/h) was associated with a 7.5-fold higher risk of IHCA (adjusted OR, 7.5; 95%CI, 3.7-15.1) compared with the normal clearance group. Conclusions: Elevated lactate levels and poor lactate clearance were strongly associated with a higher risk of ED-based IHCA. Clinicians may consider a more liberal sampling of lactate in patients at higher risk of IHCA with follow-up of abnormal levels.

Association between depressive symptoms and diagnosis of diabetes and its complications: A network analysis in electronic health records

OBJECTIVES

Diabetes and its complications are commonly associated with depressive symptoms, and few studies have investigated the diagnosis effect of depressive symptoms in patients with diabetes. The present study used a network-based approach to explore the association between depressive symptoms, which are annotated from electronic health record (EHR) notes by a deep learning model, and the diagnosis of type 2 diabetes mellitus (T2DM) and its complications.

METHODS

In this study, we used anonymous admission notes of 52,139 inpatients diagnosed with T2DM at the first affiliated hospital of Nanjing Medical University from 2008 to 2016 as input for a symptom annotation model named T5-depression based on transformer architecture which helps to annotate depressive symptoms from present illness. We measured the performance of the model by using the F1 score and the area under the receiver operating characteristic curve (AUROC). We constructed networks of depressive symptoms to examine the connectivity of these networks in patients diagnosed with T2DM, including those with certain complications.

RESULTS

The T5-depression model achieved the best performance with an F1-score of 91.71 and an AUROC of 96.25 compared with the benchmark models. The connectivity of depressive symptoms in patients diagnosed with T2DM (p = 0.025) and hypertension (p = 0.013) showed a statistically significant increase 2 years after the diagnosis, which is consistent with the number of patients diagnosed with depression.

CONCLUSION

The T5-depression model proposed in this study can effectively annotate depressive symptoms in EHR notes. The connectivity of annotated depressive symptoms is associated with the diagnosis of T2DM and hypertension. The changes in the network of depressive symptoms generated by the T5-depression model could be used as an indicator for screening depression.

Validation of Machine Learning-Based Individualized Treatment for Depressive Disorder Using Target Trial Emulation

This study aims to develop and validate the use of machine learning-based prediction models to select individualized pharmacological treatment for patients with depressive disorder. This study used data from Taiwan’s National Health Insurance Research Database. Patients with incident depressive disorders were included in this study. The study outcome was treatment failure, which was defined as psychiatric hospitalization, self-harm hospitalization, emergency visits, or treatment change. Prediction models based on the Super Learner ensemble were trained separately for the initial and the next-step treatments if the previous treatments failed. An individualized treatment strategy was developed for selecting the drug with the lowest probability of treatment failure for each patient as the model-selected regimen. We emulated clinical trials to estimate the effectiveness of individualized treatments. The area under the curve of the prediction model using Super Learner was 0.627 and 0.751 for the initial treatment and the next-step treatment, respectively. Model-selected regimens were associated with reduced treatment failure rates, with a 0.84-fold (95% confidence interval (CI) 0.82–0.86) decrease for the initial treatment and a 0.82-fold (95% CI 0.80–0.83) decrease for the next-step. In emulation of clinical trials, the model-selected regimen was associated with a reduced treatment failure rate.

Risk of depression in patients with oral cancer: a nationwide cohort study in Taiwan

This study investigates an association between oral cancers and the risk of developing depression. We conducted a total of 3031 patients with newly diagnosed oral cancers and 9093 age-, sex-, and index year-matched controls (1:3) from 2000 to 2013 were selected from the National Health Insurance Research Database (NHIRD) of Taiwan. After adjusting for confounding factors, multivariate Cox proportional hazards analysis was used to compare the risk of depression over a 13-year follow-up. Of the patients with oral cancer, 69 (2.28%, or 288.57 per 105 person-years) developed depression compared to 150 (1.65%, 135.64 per 105 person-years) in the control group. The Cox proportional hazards regression analysis showed that the adjustment hazard ratio (HR) for subsequent depression in patients with oral cancer diagnosed was 2.224 (95% Confidence Interval [CI] 1.641–3.013, p < 0.001). It is noteworthy that in the sensitivity analysis is the adjusted HR in the group with depression diagnosis was 3.392 and in the oral cancer subgroup of “Tongue” was 2.539. This study shows oral cancer was associated with a significantly increased risk for developing subsequent depression and early identification and treatment of depression in oral cancer patients is crucial.

Associations Between Antidepressant Use and Advanced Diabetes Outcomes in Patients with Depression and Diabetes Mellitus

CONTEXT

Comorbid depression in patients with diabetes deteriorates the prognosis. Antidepressants might attenuate the adverse effects of depression; however, they are associated with cardiometabolic adverse effects.

OBJECTIVE

This study aimed to explore the association between antidepressant treatment and advanced diabetic complications and mortality among patients with depression and diabetes mellitus.

METHODS

We conducted a nationwide retrospective cohort study of 36 276 patients with depression and newly treated diabetes mellitus using Taiwan's universal health insurance database. Antidepressant treatment patterns within a 6-month window were classified into none, poor, partial, and regular use, and we accounted for time-dependent variables in the Cox proportional hazards regression analysis with adjustment for time-dependent comorbidity and concomitant use of medications. Different classes of antidepressants were compared. Macro- and microvascular complications, as well as all-cause mortality, were the main outcomes. Benzodiazepines were chosen as negative control exposure.

RESULTS

Compared with poor use of antidepressants, regular use was associated with a 0.92-fold decreased risk of macrovascular complications and a 0.86-fold decreased risk of all-cause mortality but not associated with microvascular complications. Regular use of selective serotonin reuptake inhibitors was associated with a 0.83- and 0.75-fold decreased risk of macrovascular complications and all-cause mortality, respectively. Regular use of tricyclic or tetracyclic antidepressants was associated with a 0.78-fold decreased risk of all-cause mortality. Regular use of benzodiazepine showed no association with diabetic outcomes.

CONCLUSION

Regular antidepressant use was associated with lower risk of advanced diabetic complications compared with poor adherence. Clinicians should emphasize antidepressant treatment adherence among patients with depression and diabetes mellitus.

Factors associated with a high-risk return visit to the emergency department: a case-crossover study

BACKGROUND AND IMPORTANCE

Although factors related to a return emergency department (ED) visit have been reported, few studies have examined 'high-risk' return ED visits with serious adverse outcomes. Understanding factors associated with high-risk return ED visits may help with early recognition and prevention of these catastrophic events.

OBJECTIVES

We aimed to (1) estimate the incidence of high-risk return ED visits, and (2) to investigate time-varying factors associated with these revisits.

DESIGN

Case-crossover study.

SETTINGS AND PARTICIPANTS

We used electronic clinical warehouse data from a tertiary medical center. We retrieved data from 651 815 ED visits over a 6-year period. Patient demographics and computerized triage information were extracted.

OUTCOME MEASURE AND ANALYSIS

A high-risk return ED visit was defined as a revisit within 72 h of the index visit with ICU admission, receiving emergency surgery, or with in-hospital cardiac arrest during the return ED visit. Time-varying factors associated with a return visit were identified.

MAIN RESULTS

There were 440 281 adult index visits, of which 19 675 (4.5%) return visits occurred within 72 h. Of them, 417 (0.1%) were high-risk revisits. Multivariable analysis showed that time-varying factors associated with an increased risk of high-risk revisits included the following: arrival by ambulance, dyspnea, or chest pain on ED presentation, triage level 1 or 2, acute change in levels of consciousness, tachycardia (>90/min), and high fever (>39°C).

CONCLUSIONS

We found a relatively small fraction of discharges (0.1%) developed serious adverse events during the return ED visits. We identified symptom-based and vital sign-based warning signs that may be used for patient self-monitoring at home, as well as new-onset signs during the return visit to alert healthcare providers for timely management of these high-risk revisits.

Development and validation of the dialysis dementia risk score: A retrospective, population-based, nested case-control study

Background

Dementia is prevalent and underdiagnosed in the dialysis population. We aimed to develop and validate a simple dialysis dementia scoring system to facilitate identification of individuals who are at high risk for dementia.

Methods

We applied a retrospective, nested case‐control study design using a national dialysis cohort derived from the National Health Insurance Research Database in Taiwan. Patients aged between 40 and 80 years were included and 2940 patients with incident dementia were matched to 29,248 non‐dementia controls. All subjects were randomly divided into the derivation and validation sets with a ratio of 4:1. Conditional logistic regression models were used to identify factors contributing to the risk score. The cutoff value of the risk score was determined by Youden's J statistic and the graphic method.

Results

The dialysis dementia risk score (DDRS) finally included age and 10 comorbidities as risk predictors. The C‐statistic of the model was 0.71 (95% confidence interval [CI] 0.70–0.72). Calibration revealed a strong linear relationship between predicted and observed dementia risk (R² = 0.99). At a cutoff value of 50 points, the high‐risk patients had an approximately three‐fold increased risk of having dementia compared to those with low risk (odds ratio [OR] 3.03, 95% CI 2.78–3.31). The DDRS performance, including discrimination (C‐statistic 0.71, 95% CI 0.69–0.73) and calibration (p value of Hosmer−Lemeshow test for goodness of fit = 0.18), was acceptable during validation. The OR value (2.82, 95% CI 2.37–3.35) was similar to those in the derivation set.

Conclusion

The DDRS system has the potential to serve as an easily accessible screening tool to determine the high‐risk groups who deserve subsequent neurological evaluation in daily clinical practice.

Use of antipsychotic drugs and cholinesterase inhibitors and risk of falls and fractures: self-controlled case series

OBJECTIVE

To evaluate the association between the use of antipsychotic drugs and cholinesterase inhibitors and the risk of falls and fractures in elderly patients with major neurocognitive disorders.

DESIGN

Self-controlled case series.

SETTING

Taiwan's National Health Insurance Database.

PARTICIPANTS

15 278 adults, aged ≥65, with newly prescribed antipsychotic drugs and cholinesterase inhibitors, who had an incident fall or fracture between 2006 and 2017. Prescription records of cholinesterase inhibitors confirmed the diagnosis of major neurocognitive disorders; all use of cholinesterase inhibitors was reviewed by experts.

MAIN OUTCOME MEASURES

Conditional Poisson regression was used to derive incidence rate ratios and 95% confidence intervals for evaluating the risk of falls and fractures for different treatment periods: use of cholinesterase inhibitors alone, antipsychotic drugs alone, and a combination of cholinesterase inhibitors and antipsychotic drugs, compared with the non-treatment period in the same individual. A 14 day pretreatment period was defined before starting the study drugs because of concerns about confounding by indication.

RESULTS

The incidence of falls and fractures per 100 person years was 8.30 (95% confidence interval 8.14 to 8.46) for the non-treatment period, 52.35 (48.46 to 56.47) for the pretreatment period, and 10.55 (9.98 to 11.14), 10.34 (9.80 to 10.89), and 9.41 (8.98 to 9.86) for use of a combination of cholinesterase inhibitors and antipsychotic drugs, antipsychotic drugs alone, and cholinesterase inhibitors alone, respectively. Compared with the non-treatment period, the highest risk of falls and fractures was during the pretreatment period (adjusted incidence rate ratio 6.17, 95% confidence interval 5.69 to 6.69), followed by treatment with the combination of cholinesterase inhibitors and antipsychotic drugs (1.35, 1.26 to 1.45), antipsychotic drugs alone (1.33, 1.24 to 1.43), and cholinesterase inhibitors alone (1.17, 1.10 to 1.24).

CONCLUSIONS

The incidence of falls and fractures was high in the pretreatment period, suggesting that factors other than the study drugs, such as underlying diseases, should be taken into consideration when evaluating the association between the risk of falls and fractures and use of cholinesterase inhibitors and antipsychotic drugs. The treatment periods were also associated with a higher risk of falls and fractures compared with the non-treatment period, although the magnitude was much lower than during the pretreatment period. Strategies for prevention and close monitoring of the risk of falls are still necessary until patients regain a more stable physical and mental state.

Risk of dementia in patients with toxoplasmosis: a nationwide, population-based cohort study in Taiwan

Background

Approximately 25–30% of individuals worldwide are infected with Toxoplasma gondii (T. gondii), which is difficult to detect in its latent state. We aimed to evaluate the association between toxoplasmosis, the risk of dementia, and the effects of antibiotics in Taiwan.

Methods

This nationwide, population-based, retrospective cohort study was conducted using the Longitudinal Health Insurance Database containing the records of 2 million individuals retrieved from Taiwan’s National Health Insurance Research Database. Fine–Gray competing risk analysis was used to determine the risk for the development of dementia in the toxoplasmosis cohort relative to the non-toxoplasmosis cohort. A sensitivity analysis was also conducted. The effects of antibiotics (sulfadiazine or clindamycin) on the risk of dementia were also analyzed.

Results

We enrolled a total of 800 subjects, and identified 200 patients with toxoplasmosis and 600 sex- and age-matched controls without toxoplasmosis infection in a ratio of 1:3, selected between 2000 and 2015. The crude hazard ratio (HR) of the risk of developing dementia was 2.570 [95% confidence interval (CI) = 1.511–4.347, P < 0.001]. After adjusting for sex, age, monthly insurance premiums, urbanization level, geographical region, and comorbidities, the adjusted HR was 2.878 (95% CI = 1.709–4.968, P < 0.001). Sensitivity analysis revealed that toxoplasmosis was associated with the risk of dementia even after excluding diagnosis in the first year and the first 5 years. The usage of sulfadiazine or clindamycin in the treatment of toxoplasmosis was associated with a decreased risk of dementia.

Conclusions

This finding supports the evidence that toxoplasmosis is associated with dementia and that antibiotic treatment against toxoplasmosis is associated with a reduced risk of dementia. Further studies are necessary to explore the underlying mechanisms of these associations.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s13071-021-04928-7.

Effect of medication on risk of traumatic brain injury in patients with bipolar disorder: A nationwide population-based cohort study

BACKGROUND

Increased traumatic brain injury (TBI) risk was found in patients with bipolar disorder (BPD). Whether the medications for BPD and dosage moderate the risk of TBI is not clear.

AIM

This study aimed to determine whether an association exists between BPD and TBI and whether the prescription of psychotropics moderates TBI risk.

METHODS

A total of 5606 individuals who had received diagnoses of BPD between January 1, 1997 and December 31, 2013 and 56,060 matched controls without BPD were identified from Taiwan's National Health Insurance Research Database. Cases and controls were followed until the date of TBI diagnosis.

RESULTS

BPD was associated with a high risk of TBI (adjusted hazard ratio (aHR): 1.85; 95% CI: 1.62-2.11). Patients with BPD, with or without a history of psychiatric hospitalization, had increased risks of TBI (aHR: 1.94, 95% CI: 1.57-2.4 and aHR: 1.82, 95% CI: 1.55-2.1, respectively). The prescription of typical antipsychotics (0 < defined daily dose (DDD) < 28: hazard ratio (HR) = 1.52, 95% CI: 1.19-1.94; ⩾28 DDD: HR = 1.54, 95% CI: 1.15-2.06) and tricyclic antidepressants (TCAs) (0 < DDD < 28: HR = 1.73, 95% CI: 1.26-2.39; ⩾28 DDD: HR = 1.52, 95% CI: 1.02-2.25) was associated with higher TBI risk. Patients receiving higher doses of benzodiazepines (BZDs) (cumulative dose ⩾28 DDD) had a higher TBI risk (HR = 1.53, 95% CI: 1.13-2.06).

CONCLUSION

Patients with BPD have a higher risk of TBI. The use of typical antipsychotics, TCAs, or high-dose BZDs increases the risk of TBI in BPD.

Additive interaction of mid- to late-life depression and cerebrovascular disease on the risk of dementia: a nationwide population-based cohort study

Background

Dementia is a progressive neurocognitive disease with a substantial social burden. No apparent breakthroughs in treatment options have emerged so far; thus, disease prevention is essential for at-risk populations. Depression and cerebrovascular disease (CVD) are independent risk factors for dementia, but no studies have examined their interaction effect on dementia risk. This study aimed to identify the association of depression and CVD with the risk of dementia and evaluate whether dementia risk among patients with comorbid depression and CVD is higher than the sum of the individual risk due to each condition.

Methods

A population-based cohort study was conducted to analyze the Korean National Health Insurance Service-National Sample Cohort data of all individuals over 50 years of age. Individuals who had not been diagnosed with dementia at baseline were included and followed up from January 1, 2005, to December 31, 2013. A time-varying Cox proportional hazard regression model adjusted for potential confounding factors was used for the analysis. The interaction between depression and CVD was estimated based on the attributable proportion (AP), relative excess risk due to interaction (RERI), synergy index (SI), and multiplicative-scale interaction.

Results

A total of 242,237 participants were included in the analytical sample, of which 12,735 (5.3%) developed dementia. Compared to that for participants without depression or CVD, the adjusted hazard ratio for the incidence of dementia for those with depression alone was 2.35 (95% confidence interval [CI] 2.21–2.49), CVD alone was 3.25 (95% CI 3.11–3.39), and comorbid depression and CVD was 5.02 (95% CI 4.66–5.42). The additive interaction between depression and CVD was statistically significant (AP—0.08, 95% CI 0.01–0.16; RERI—0.42, 95% CI 0.03–0.82; SI—1.12, 95% CI 1.01–1.24). The multiplicative interaction was significant too, but the effect was negative (0.66, 95% CI 0.60–0.73).

Conclusions

In this population-based nationwide cohort with long-term follow-up, depression and CVD were associated with an increased risk of dementia, and their coexistence additively increased dementia risk more than the sum of the individual risks.

Clinical notes as prognostic markers of mortality associated with diabetes mellitus following critical care: A retrospective cohort analysis using machine learning and unstructured big data

BACKGROUND

Clinical notes are ubiquitous resources offering potential value in optimizing critical care via data mining technologies.

OBJECTIVE

To determine the predictive value of clinical notes as prognostic markers of 1-year all-cause mortality among people with diabetes following critical care.

MATERIALS AND METHODS

Mortality of diabetes patients were predicted using three cohorts of clinical text in a critical care database, written by physicians (n = 45253), nurses (159027), and both (n = 204280). Natural language processing was used to pre-process text documents and LASSO-regularized logistic regression models were trained and tested. Confusion matrix metrics of each model were calculated and AUROC estimates between models were compared. All predictive words and corresponding coefficients were extracted. Outcome probability associated with each text document was estimated.

RESULTS

Models built on clinical text of physicians, nurses, and the combined cohort predicted mortality with AUROC of 0.996, 0.893, and 0.922, respectively. Predictive performance of the models significantly differed from one another whereas inter-rater reliability ranged from substantial to almost perfect across them. Number of predictive words with non-zero coefficients were 3994, 8159, and 10579, respectively, in the models of physicians, nurses, and the combined cohort. Physicians' and nursing notes, both individually and when combined, strongly predicted 1-year all-cause mortality among people with diabetes following critical care.

CONCLUSION

Clinical notes of physicians and nurses are strong and novel prognostic markers of diabetes-associated mortality in critical care, offering potentially generalizable and scalable applications. Clinical text-derived personalized risk estimates of prognostic outcomes such as mortality could be used to optimize patient care.

Electronic Medical Record-Based Case Phenotyping for the Charlson Conditions: Scoping Review

Background

Electronic medical records (EMRs) contain large amounts of rich clinical information. Developing EMR-based case definitions, also known as EMR phenotyping, is an active area of research that has implications for epidemiology, clinical care, and health services research.

Objective

This review aims to describe and assess the present landscape of EMR-based case phenotyping for the Charlson conditions.

Methods

A scoping review of EMR-based algorithms for defining the Charlson comorbidity index conditions was completed. This study covered articles published between January 2000 and April 2020, both inclusive. Embase (Excerpta Medica database) and MEDLINE (Medical Literature Analysis and Retrieval System Online) were searched using keywords developed in the following 3 domains: terms related to EMR, terms related to case finding, and disease-specific terms. The manuscript follows the Preferred Reporting Items for Systematic reviews and Meta-analyses extension for Scoping Reviews (PRISMA) guidelines.

Results

A total of 274 articles representing 299 algorithms were assessed and summarized. Most studies were undertaken in the United States (181/299, 60.5%), followed by the United Kingdom (42/299, 14.0%) and Canada (15/299, 5.0%). These algorithms were mostly developed either in primary care (103/299, 34.4%) or inpatient (168/299, 56.2%) settings. Diabetes, congestive heart failure, myocardial infarction, and rheumatology had the highest number of developed algorithms. Data-driven and clinical rule–based approaches have been identified. EMR-based phenotype and algorithm development reflect the data access allowed by respective health systems, and algorithms vary in their performance.

Conclusions

Recognizing similarities and differences in health systems, data collection strategies, extraction, data release protocols, and existing clinical pathways is critical to algorithm development strategies. Several strategies to assist with phenotype-based case definitions have been proposed.

Paths to the first-time diagnoses of anorexia nervosa and bulimia nervosa in Taiwan

OBJECTIVE

This study aimed to examine the characteristics of psychiatrists and the hospital settings in relation to the first-time diagnoses of anorexia nervosa (AN) and bulimia nervosa (BN) and depict medical utilization and the detection rate before diagnosis of patients with AN and BN.

METHOD

We extracted data of individuals with AN or BN, as defined by the International Classification of Diseases, Ninth Revision, Clinical Modification, from a national health insurance database. Individuals with AN (n = 1,893) or BN (n = 10,542) who were first-time diagnosed by psychiatrists from 2002 to 2013 were included. Individuals with schizophrenia were selected as control groups that were matched with the incident AN and BN cases for sex, age stratum, and year of diagnosis.

RESULTS

AN was more likely to be diagnosed by female psychiatrists. Patients with AN were more frequently diagnosed in medical centers while patients with BN were mostly diagnosed in primary care clinics. Nearly all patients with AN and BN had sought treatment for physical problems but less than half had sought help for mental health problems in the year preceding the diagnosis. Individuals with AN, BN, and schizophrenia were all under-detected by nonpsychiatric medical professionals. Notably, BN was least likely to be recognized by both psychiatrists and other medical professionals.

DISCUSSION

Our findings underscore the importance of educational programs that are designed to improve the detection and management of eating disorders by medical professionals in Taiwan. Advanced educational programs that target differential diagnosis and the tailored management of different eating disorders should be highlighted among psychiatrists.