Staging Model Research Articles

LGSleepNet: An Automatic Sleep Staging Model Based on Local and Global Representation Learning

LGSleepNet: An Automatic Sleep Staging Model Based on Local and Global Representation Learning

A Developed LSTM-Ladder-Network-Based Model for Sleep Stage Classification.

Sleep staging is crucial for diagnosing sleep-related disorders. The heavy and time-consuming task of manual staging can be released by automatic techniques. However, the automatic staging model would have a relatively poor performance when working on unseen new data due to individual differences. In this research, a developed LSTM-Ladder-Network (LLN) model is proposed for automatic sleep stage classification. Several features are extracted for each epoch and combined with the following epochs to form a cross-epoch vector. The long short-term memory (LSTM) network is added into the basic ladder network (LN) to learn the sequential information of adjacent epochs. The developed model is implemented based on a transductive learning scheme to avoid the issue of accuracy loss caused by individual differences. In this process, the labeled data pre-trains the encoder, and the unlabeled data re- fine the model parameters by minimizing the reconstruction loss. The proposed model is evaluated on the data from public database and hospital. Comparison experiments were conducted where the developed LLN model achieved rather satisfied performance while dealing with the unseen new data. The obtained results demonstrate the effectiveness of the proposed approach in addressing individual differences. This can improve the quality of automatic sleep staging when assessed on different individuals and has strong application potential as a computer aided approach for sleep staging.

Sleep Quality Evaluation Based on Single-Lead Wearable Cardiac Cycle Acquisition Device.

In clinical conditions, polysomnography (PSG) is regarded as the "golden standard" for detecting sleep disease and offering a reference of objective sleep quality. For healthy adults, scores from sleep questionnaires are more reliable than other methods in obtaining knowledge of subjective sleep quality. In practice, the need to simplify PSG to obtain subjective sleep quality by recording a few channels of physiological signals such as single-lead electrocardiogram (ECG) or photoplethysmography (PPG) signal is still very urgent. This study provided a two-step method to differentiate sleep quality into "good sleep" and "poor sleep" based on the single-lead wearable cardiac cycle data, with the comparison of the subjective sleep questionnaire score. First, heart rate variability (HRV) features and ECG-derived respiration features were extracted to construct a sleep staging model (wakefulness (W), rapid eye movement (REM), light sleep (N1&N2) and deep sleep (N3)) using the multi-classifier fusion method. Then, features extracted from the sleep staging results were used to construct a sleep quality evaluation model, i.e., classifying the sleep quality as good and poor. The accuracy of the sleep staging model, tested on the international public database, was 0.661 and 0.659 in Cardiology Challenge 2018 training database and Sleep Heart Health Study Visit 1 database, respectively. The accuracy of the sleep quality evaluation model was 0.786 for our recording subjects, with an average F1-score of 0.771. The proposed sleep staging model and sleep quality evaluation model only requires one channel of wearable cardiac cycle signal. It is very easy to transplant to portable devices, which facilitates daily sleep health monitoring.

Automatic Sleep Staging Based on Single-Channel EEG Signal Using Null Space Pursuit Decomposition Algorithm

Sleep quality is related to people’s physical and mental health, so an accurate assessment of sleep quality is key to recognizing sleep disorders and taking effective interventions. To address the shortcomings of traditional manual and automatic staging methods, such as being time-consuming and having low classification accuracy, an automatic sleep staging method based on the null space pursuit (NSP) decomposition algorithm of single-channel electroencephalographic (EEG) signals is proposed, which provides a new way for EEG signal decomposition and automatic identification of sleep stages. First, the single-channel EEG signal data from the Sleep-EDF database, DREAMS Subject database, and Sleep Heart Health Study database (SHHS), available on PhysioNet, were preprocessed, respectively. Second, the preprocessed single-channel EEG signals were decomposed by the NSP algorithm. Third, we extracted nine features in the time domain of the nonlinear dynamics and statistics from the original EEG signal and the six simple signals that were decomposed. Finally, the extreme gradient boosting (XGBOOST) algorithm was used to construct a classification model to classify and identify the 63 extracted EEG signal features for automatic sleep staging. The experimental results showed that, on the Sleep-EDF database, the accuracy of four and five categories were 93.59% and 92.89%, respectively; on the DREAMS Subject database, the accuracy rates of four and five categories were 91.32% and 90.01%, respectively; on the SHHS database, the accuracy rates of four and five categories were 90.25% and 88.37%, respectively. The experimental results show that the automatic sleep staging model proposed in this work has high classification accuracy and efficiency, as well as strong applicability and robustness.

The Utility of Pre-Treatment Inflammation Markers as Associative Factors to the Adverse Outcomes of Vulvar Cancer: A Study on Staging, Nodal Involvement, and Metastasis Models

Given the role of inflammation in carcinogenesis, this study investigated the utility of pre-treatment inflammatory markers as associative indicators for advanced-stage disease, lymph node metastasis (LNM), and distant metastasis (DM) in vulvar cancer (VC). A cross-sectional study was conducted on 86 women with VC in a single centre in Jakarta, Indonesia. The laboratory data was based on C-reactive protein (CRP), procalcitonin, the erythrocyte sedimentation rate (ESR) and fourteen derived, recorded and calculated ratios: leukocyte-to-platelet (LPR), neutrophil-to-lymphocyte (NLR), derived neutrophil-to-lymphocyte (dNLR), neutrophil-to-monocyte (NMR), platelet-to-monocyte (PLR), lymphocyte-to-monocyte (LMR), basophil-to-monocyte (BLR), systemic immune-inflammation index (SII), body mass index, albumin, and NLR (BAN) score, haemoglobin-to-platelet (HPR), prognostic nutritional index (PNI), modified Glasgow Prognostic Score (mGPS), CRP-to-albumin, and CRP-to-procalcitonin. The optimal cut-off for each marker was determined using receiver operating characteristic (ROC) curve analysis, and their diagnostic indicator performances were assessed. The utility of these ratios as associative factors for three endpoints was further evaluated in multivariate regression models. Investigated inflammatory markers exhibited specific performances for individual adverse outcomes, proving a fair to excellent ability in case finding and screening. After adjustment, the BAN score ≤ 334.89 (OR 9.20, p = 0.001) and ESR ≥ 104 (OR 4.18, p = 0.048) become two advanced-stage associative factors with AUC: 0.769. LNM was solely determined by higher NLR ≥ 2.83 (OR 4.15, p = 0.014) with AUC: 0.615. Meanwhile, BLR ≥ 0.035 (OR 5.67, p = 0.001) and ESR ≥ 84 (OR 6.01, p = 0.003) were contributing factors for DM, with AUC: 0.765. Inflammatory markers are crucial for identifying the deleterious outcomes of VC. Accordingly, yielded models require external validation.

A mechanistic staging model of reward processing alterations in individuals with binge-type eating disorders.

Altered reward processing is thought to characterize binge-type eating disorders, but the exact nature of these alterations is unclear. A more fine-grained understanding of whether specific aspects of reward processing contribute to the development or maintenance of binge eating may point to new therapeutic targets and personalized treatments. The incentive sensitization theory of addiction proposes that repeated use of a substance increases the desire to approach a reward ('wanting') but not pleasure when consuming the reward ('liking'), suggesting that reward processes driving addiction change over time. We hypothesize that the same may be true for binge eating. Further, consistent with the maladaptive scaling hypothesis, reward processing may be heightened for multiple reinforcers in at-risk individuals but become tuned toward food once binge eating is initiated. In this article, we propose a mechanistic staging model of reward processing in binge-type eating disorders that synthesizes existing data and posits that alterations of reward processing depend on illness stage and reward type. We outline translational methods for testing key hypotheses and discuss clinical implications. Considering reward processing alterations in relation to illness stage has the potential to improve treatment outcomes by ensuring that the mechanisms targeted are personalized to the individual patient. PUBLIC SIGNIFICANCE: Individuals with binge-type eating disorders experience alterations in their desire for, and pleasure from, food. We believe that the exact nature of these alterations in reward processing change over the course of illness-from the at-risk state to an established illness. If true, treatments for binge-type eating disorders that target reward processing should be personalized to the illness stage of the patient.

Clinical staging and the differential risks for clinical and functional outcomes in young people presenting for youth mental health care

BackgroundClinical staging proposes that youth-onset mental disorders develop progressively, and that active treatment of earlier stages should prevent progression to more severe disorders. This retrospective cohort study examined the longitudinal relationships between clinical stages and multiple clinical and functional outcomes within the first 12 months of care.MethodsDemographic and clinical information of 2901 young people who accessed mental health care at age 12–25 years was collected at predetermined timepoints (baseline, 3 months, 6 months, 12 months). Initial clinical stage was used to define three fixed groups for analyses (stage 1a: ‘non-specific anxious or depressive symptoms’, 1b: ‘attenuated mood or psychotic syndromes’, 2+: ‘full-threshold mood or psychotic syndromes’). Logistic regression models, which controlled for age and follow-up time, were used to compare clinical and functional outcomes (role and social function, suicidal ideation, alcohol and substance misuse, physical health comorbidity, circadian disturbances) between staging groups within the initial 12 months of care.ResultsOf the entire cohort, 2093 young people aged 12–25 years were followed up at least once over the first 12 months of care, with 60.4% female and a baseline mean age of 18.16 years. Longitudinally, young people at stage 2+ were more likely to develop circadian disturbances (odds ratio [OR]=2.58; CI 1.60–4.17), compared with individuals at stage 1b. Additionally, stage 1b individuals were more likely to become disengaged from education/employment (OR=2.11, CI 1.36–3.28), develop suicidal ideations (OR=1.92; CI 1.30–2.84) and circadian disturbances (OR=1.94, CI 1.31–2.86), compared to stage 1a. By contrast, we found no relationship between clinical stage and the emergence of alcohol or substance misuse and physical comorbidity.ConclusionsThe differential rates of emergence of poor clinical and functional outcomes between early versus late clinical stages support the clinical staging model's assumptions about illness trajectories for mood and psychotic syndromes. The greater risk of progression to poor outcomes in those who present with more severe syndromes may be used to guide specific intervention packages.

Staging the bipolar disorders: Are early stages too early a stage for intervention?

Staging the bipolar disorders: Are early stages too early a stage for intervention?

Newly proposed survival staging system for poorly differentiated thyroid cancer: a SEER-based study.

Despite the recent release of the 8th edition TNM staging system, the risk stratification for poorly differentiated thyroid cancer (PDTC) remains controversial. Retrospective study. SEER database and the First Hospital of China Medical University (FHCMU) database. Between 2004 and 2015, 1201 PDTC patients from the SEER database were enrolled to propose a new staging system. 38 PDTC patients were included from the FHCMU. A retrospective analysis of 1201 PDTC cases was performed, and a new staging classification was developed as follows: stage I: age < 55 and T1/any N/M0 (n = 127, 10.57%); stage II: age < 55 and T2-4/any N/M0 or age ≥ 55 and T1-2/any N/M0 (n = 523, 43.55%); stage III: age < 55 and any T/N0/M1 or age ≥ 55 and any T3/any N/M0 (n = 239, 19.90%); stage IV: age < 55 and any T/N1/M1 or age ≥ 55 and T4/any N/M0, and T/any N/M1 (n = 312, 25.98%). The 10-year disease-specific survival rates of patients in the new stages I, II, III, and IV were 97.9%, 77.9%, 35.3%, and 12.1%, respectively. The proportion of variation explained (PVE) for disease-specific survival of the proposed system was higher than that of the 8th AJCC TNM staging (30.61% vs. 27.15%). The accuracy of the staging system was verified in 38 PDTC patients from the FHCMU. The proposed staging system provided a more accurate risk stratification for PDTC patients. The new staging model may facilitate the design of personalized treatment strategies for PDTC patients.

A machine learning method for improving liver cancer staging

Liver cancer is a common malignant tumor, and its clinical stage is closely related to the clinical treatment and prognosis of patients. Currently, the BCLC staging system revised by the BCLC group of University of Barcelona is the globally recognized staging system for liver cancer. However, with the deepening of related research, the current staging system can no longer fully meet the clinical needs. In this work, we propose a novel machine learning method for constructing an automatic hepatocellular carcinoma staging model that incorporates far more clinical variables than any existing staging system. Our model is based on random survival forests, which generates a unique hazard function for each patient. B-splines are used to embed hazard functions into vectors in low-dimensional space and hierarchical clustering method groups similar patients to form staging cohorts. The resulting staging system significantly outperforms the BCLC system in terms of distinctiveness between patients in different stages.

Validity of the Brief Assessment of Impaired Cognition case-finding instrument for identification of dementia subgroups and staging of dementia.

The aims of this study were to examine the psychometric properties of the Brief Assessment of Impaired Cognition (BASIC) case-finding instrument in clinical settings focusing on (i) test-retest reliability, (ii) the discriminative validity of BASIC and its components for identification of Alzheimer disease (AD) dementia and non-AD dementia, and (iii) the association of expert clinical rating of cognitive status with BASIC performance. The test-retest reliability analysis was based on a sample of general practice patients (n=59) retested with a mean interval of 19 days. Discriminative validity analyses and analysis of the association of cognitive status with BASIC performance were based on data from the primary validation study of BASIC in memory clinics. The test-retest reliability of BASIC was high (r=0.861). No significant difference in discriminative validity was found for identification of AD dementia (sensitivity=0.99, specificity=0.98) and non-AD dementia (sensitivity=0.90, specificity=0.98). All components of BASIC contributed to the high discriminative validity of both AD and non-AD dementia. BASIC performance was significantly correlated with expert clinical rating of the cognitive status of patients. A crude staging model for cognitive status using BASIC score intervals had superior classification accuracy (70%) compared to a Mini-Mental State Examination (MMSE) score range-based model (58% accuracy). BASIC is a reliable and valid case-finding instrument for AD dementia and non-AD dementia in clinical settings. BASIC performance is significantly associated with the degree of cognitive impairment, and BASIC seems to be superior to MMSE for staging of impairment.

Voxel‐wise Staging of Tau Pathology using [18F]RO‐948 PET in the Early AD Continuum

AbstractBackgroundTau positron emission tomography (PET) is a useful biomarker for detecting tau aggregation, one of the hallmark pathological findings in Alzheimer's disease (AD). The goal of this study is to characterize and stage the pattern of spreading of tau pathology across the brain in the early AD continuum.MethodForty‐seven cognitively unimpaired individuals from ALFA+ cohort had [18F]RO‐948 and [18F]flutemetamol PET, T1‐weighted magnetic resonance imaging(MRI), two cognitive assessments, cerebrospinal fluid(CSF), and blood biomarkers. Standardized uptake value ratio(SUVr) maps of [18F]RO‐948 PET images in MNI space created using inferior cerebellum as reference region. Participants were categorized according to the A/T system using pre‐established thresholds for CSF amyloid(A) and tau(T) positivity(Table1). A voxel‐wise hierarchical staging model was constructed based on data from the control group(A‐T‐). Voxel positivity was defined as SUVr higher than mean plus two standard deviations of control group(Figure1). Then, each voxel was assigned to one of 3 stages based on the frequency of its abnormality across the control group, dividing the voxel frequency histogram into 3 equally‐distributed groups. A given stage was assigned to each scan if, at least, 50% of the voxels on the stage mask were positive, hierarchically. That is, to determine stage 2, stage 1 needs to be positive as well. Otherwise, the scan is labeled as non‐stageable. Staging model performance was evaluated by recording the number of non‐stageable scans as well as by performing non‐parametric correlations between tau stages and CSF, blood biomarkers, cortical thickness, Centiloid(CL) as well as changes in cognition composites. Post‐hoc between‐stage comparisons were performed using the Kruskal‐Wallis test followed by a Bonferroni correction.ResultsAll scans were stageable. Twenty‐three scans were assigned to stage 0, 11 to stage 1, and 3 to stage 2(Figure2). Significant correlations between tau stages and CSF, blood biomarkers, Centiloid, cortical thickness as well as memory and attention changes were observed(Table2). Between‐group comparisons of tau stages vs biomarkers and changes in memory and attention were statistically significant(Figure3).ConclusionOur results suggest that the proposed voxel‐wise staging model is sensitive to capturing the spread of early tau pathology and useful for studying early AD pathophysiological mechanisms.

The impact of extra-valvular cardiac damage on mid-term clinical outcome following transcatheter aortic valve replacement in patients with severe aortic stenosis.

To quantify extra-valvular cardiac damage associated with severe aortic valve stenosis (AS), a novel staging model was proposed. This study aimed to validate this model in patients undergoing transcatheter aortic valve replacement (TAVR) as well as to assess its prognostic impact. Based on echocardiographic findings, the following stages were applied: isolated AS (stage 0), left ventricular (LV) damage (stage 1), left atrial or mitral valve damage (stage 2), pulmonary hypertension or tricuspid regurgitation (stage 3), or right ventricular dysfunction (stage 4). The primary endpoint was 2-year all-cause mortality. The distribution across stages was 0.8% at stage 0, 7.5% at stage 1, 63.3% at stage 2, 18.3% at stage 3, and 10.1% at stage 4. All-cause mortality increased at all stages 1-4 (12.1%, 18.2%, 26.6%, and 28.2%; p = 0.023). In the multivariate model, the stage of cardiac damage, age, New York Heart Association (NYHA) class III/IV, peripheral artery disease, and previous pacemaker were independent predictors of the primary endpoint. Patients treated for severe AS show a high prevalence of extra-valvular cardiac damage. An increase in stage is associated with higher 2-year all-cause mortality. The application of this staging model may add value to current treatment algorithms.

Staging models applied in a sample of patients with bipolar disorder: Results from a retrospective cohort study

BackgroundBipolar Disorder (BD) is a life-long illness with compelling evidence of progression. Although different staging models have been proposed to evaluate its course, clinical data remain limited. The aim of the present study was to retrospectively assess applicability of available staging approaches and their pattern of progression in a sample of bipolar patients. MethodsIn a naturalistic sample of 100 BD patients, retrospective assessment of clinical stages was performed at four time points over 10 years, according to four staging models. Staging progression with potential associations between stages and unfavourable illness characteristics were analyzed. ResultsA pattern of stage worsening emerged for each model, with a significant increase at every time point. Greater stage increases emerged in patients with lower educational level, age at first elevated episode ≤35 years, duration of illness ≤25 years, and duration of untreated illness ≤5 years. Lower stage values were associated with BD II, no psychiatric hospitalization, depressive onset and predominant polarity, ≤three lifetime episodes, age at first mood stabilizer >40 years, duration of illness ≤25 years, and engaged/employed status. Higher stage values were associated with lower age at first elevated episode and mood stabilizing treatment instead. LimitationsNaturalistic and retrospective design, recruitment at a 2nd level specialistic clinic. ConclusionsReported findings support the progressive nature of BD and the application of staging models for early intervention, suggesting a conceptualization of a standardized approach to better characterize patients, predict their clinical course, and deliver tailored treatment options.

Preoperative prediction of lymph node status in patients with colorectal cancer. Developing a predictive model using machine learning.

Develop a prediction model to determine the probability of no lymph node metastasis (pN0) in patients with colorectal cancer. We used data from four Danish health databases on patients with colorectal cancer diagnosed between 2001 and 2019. The registries were harmonized into one common data model (CDM). Patients with clinical T4 tumors, undergoing palliative or acute surgery, and patients undergoing neoadjuvant therapy were excluded. Preoperative data was used to train the model. A postoperative model including tumor-specific variables potentially available after local tumor resection was also developed. Additionally, both models were compared with a model based on age, sex, and clinical N stage to resemble current standards. A Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression analysis for prediction was used. In total, 35,812 patients with 16,802 variables were identified in the CDM, and 194 variables affected the probability of pN0 preoperative. The area under the receiver operating characteristic curve (AUROC) was 0.64 (95% CI 0.63-0.66), and the area under the precision-recall curve (AUPRC) was 0.75 (95% CI 0.74-0.76). The mean predicted risk was 0.649, observed risk was 0.650, and calibration-in-large was 0.998. Adding histopathological data from the tumor improved the model slightly by increasing AUROC to 0.69. In comparison, the AUROC of the current standard clinical staging model was 0.57. Using Danish National Patient Registry data in a machine learning-based predictive model showed acceptable results and outperforms current tools for clinical staging in predicting pN0 status in patients scheduled for CRC surgery.

Toward staging differentiation for posttraumatic stress disorder treatment.

Several medical and psychiatric disorders have stage-based treatment decision-making methods. However, international treatment guidelines for posttraumatic stress disorder (PTSD) fail to give specific treatment recommendations based on chronicity or stage of the disorder. There is convincing evidence of a finite range of PTSD symptom trajectories, implying that different phenotypes of the disorder can be distinguished, which are highly relevant for a staging typology of PTSD. State-of-the-art review building on prior work on staging models in other disorders as a mapping tool to identify and synthesize toward PTSD. We propose a four-stage model of PTSD ranging from stage 0: trauma-exposed asymptomatic but at risk to stage 4: severe unremitting illness of increasing chronicity. We favor a symptom description in various chronological characteristics based on neurobiological markers, information processing systems, stress reactivity, and consciousness dimensions. We also advocate for a separate phenomenology of treatment resistance since this can yield treatment recommendations. A staging perspective in the field of PTSD is highly needed. This can facilitate the selection of interventions that are proportionate to patients' current needs and risk of illness progression and can also contribute to an efficient framework to organize biomarker data and guide service delivery. Therefore, we propose that a neurobiologically driven trajectory-based typology of PTSD can help deduct several treatment recommendations leading to a more personalized and refined grid to strategize, plan and evaluate treatment interventions.

Clinical staging of clinicians.

Clinical staging of clinicians.

Bone metastasis risk and prognosis assessment models for kidney cancer based on machine learning.

Bone metastasis is a common adverse event in kidney cancer, often resulting in poor survival. However, tools for predicting KCBM and assessing survival after KCBM have not performed well. The study uses machine learning to build models for assessing kidney cancer bone metastasis risk, prognosis, and performance evaluation. We selected 71,414 kidney cancer patients from SEER database between 2010 and 2016. Additionally, 963 patients with kidney cancer from an independent medical center were chosen to validate the performance. In the next step, eight different machine learning methods were applied to develop KCBM diagnosis and prognosis models while the risk factors were identified from univariate and multivariate logistic regression and the prognosis factors were analyzed through Kaplan-Meier survival curve and Cox proportional hazards regression. The performance of the models was compared with current models, including the logistic regression model and the AJCC TNM staging model, applying receiver operating characteristics, decision curve analysis, and the calculation of accuracy and sensitivity in both internal and independent external cohorts. Our prognosis model achieved an AUC of 0.8269 (95%CI: 0.8083-0.8425) in the internal validation cohort and 0.9123 (95%CI: 0.8979-0.9261) in the external validation cohort. In addition, we tested the performance of the extreme gradient boosting model through decision curve analysis curve, Precision-Recall curve, and Brier score and two models exhibited excellent performance. Our developed models can accurately predict the risk and prognosis of KCBM and contribute to helping improve decision-making.

SHNN: A single-channel EEG sleep staging model based on semi-supervised learning

Sleep staging is an essential step in the diagnosis and treatment of sleep-related diseases. Currently, most supervised learning models face the problem of insufficient labeled data. In addition, most sleep staging models are based on multi-channel EEG, and the models are too complex to be suitable for home sleep monitoring scenarios. To tackle these problems, this study proposes a sleep staging method based on pseudo-label optimization and a single-channel sleep hybrid neural network called SHNN. In the SHNN model, we design a multi-scale convolutional neural network (CNN) to extract the features from the single-channel EEG and use a Bi-directional recurrent gating unit (Bi-GRU) to obtain temporal context information of sleep data sequences. Extensive experiments based on the single-channel EEG (FPz-Cz, Pz-Oz, and Cz-A1) of the Sleep-EDFx and the DREAMS-SUB datasets validate the effectiveness of the SHNN model and the pseudo-label optimization algorithm therein outperforming current single-channel methods regarding the accuracy, kappa, and MF1 Score. Moreover, the pseudo-label optimization algorithm can achieve good results on other sleep staging methods. The SHNN code is available at https://github.com/Caowenpeng/SHNN.

Heterogeneous early illness courses of Korean patients with bipolar disorders: replication of the staging model

BackgroundClinical staging of bipolar disorder (BD) requires application of real-world data, as the next step in hypothesis. This study used the staging model to analyze the long-term course of BD in Korean patients based on clinical features and treatment responses to map the progression of bipolar illness from its early phase after the onset of illness.MethodsA total of 136 patients diagnosed with BD-I (n = 62) or BD-II (n = 74) were recruited. Their progressive stages were retrospectively evaluated. A multi-state model was used to calculate the probability of progression to each stage. Hazard ratios of covariates expected to influence different courses of BD were calculated. Using the Alda score, long-term responses to mood stabilizers depending on the current stage were compared.ResultsSeveral sub-populations showed varied courses during the first five years after the onset of illness, with 41.5% remaining in stage 2 and 53% progressing to higher stages with shortened time for transition. Profiles of patients with BD-I and BD-II were different, suggesting biologically distinct groups. Comorbid psychiatric disorders, such as obsessive-compulsive disorder (OCD) and bulimia nervosa (BN) were associated with a recurrent course (stage 3a or 3b) or a malignant course (stage 3c or 4). Early age of onset, shorter duration of illness, older age at the start of medication, and poor response to lithium affected the illness progression.ConclusionWe were able to apply the stage model based on episode recurrence patterns in early illness courses of Korean patients with BD. The stage progression pattern differed from the early phase in BD-I and BD-II patients. Psychotic comorbidity, age at onset, age at starting psychiatric treatment showed associations with the illness progression.