Development and validation of machine learning for early mortality in systemic sclerosis.

Clinical predictors of mortality in systemic sclerosis (SSc) are diversely reported due to different healthcare conditions and populations. A simple predictive model for early mortality among patients with SSc is needed as a precise referral tool for general practitioners. We aimed to develop and validate a simple predictive model for predicting mortality among patients with SSc. Prognostic research with a historical cohort study design was conducted between January 1, 2013, and December 31, 2020, in adult SSc patients attending the Scleroderma Clinic at a university hospital in Thailand. The data were extracted from the Scleroderma Registry Database. Early mortality was defined as dying within 5 years after the onset of SSc. Deep learning algorithms with Adam optimizer and different machine learning algorithms (including Logistic Regression, Decision tree, AdaBoost, Random Forest, Gradient Boosting, XGBoost, and Autoencoder neural network) were used to classify SSc mortality. In addition, the model's performance was evaluated using the area under the receiver operating characteristic curve (auROC) and its 95% confidence interval (CI) and values in the confusion matrix. The predictive model development included 528 SSc patients, 343 (65.0%) were females and 374 (70.8%) had dcSSc. Ninety-five died within 5 years after disease onset. The final 2 models with the highest predictive performance comprise the modified Rodnan skin score (mRSS) and the WHO-FC ≥ II for Model 1 and mRSS and WHO-FC ≥ III for Model 2. Model 1 provided the highest predictive performance, followed by Model 2. After internal validation, the accuracy and auROC were good. The specificity was high in Models 1 and 2 (84.8%, 89.8%, and 98.8% in model 1 vs. 84.8%, 85.6%, and 98.8% in model 2). This simplified machine learning model for predicting early mortality among patients with SSc could guide early referrals to specialists and help rheumatologists with close monitoring and management planning. External validation across multi-SSc clinics should be considered for further study.

A comparison of the proportion of early stage cholangiocarcinoma found in an ultrasound-screening program compared to walk-in patients.

BACKGROUND: Patients with cholangiocarcinoma (CCA) usually have no specific symptoms until an advance stage of the disease and curative treatment is not possible. Patients with early stage, operable disease can be found using ultrasonography (US). A US-screening program was implemented in Thailand where CCA incidence is the highest worldwide. Here we evaluate the effectiveness of the program by comparing the proportion of individuals with early stage CCA in the screening group with that of the walk-in group presenting at hospitals with clinical symptoms. METHODS: All patients had a pathological diagnosis of CCA. The difference in the proportions and the 95% confidence interval (CI) were obtained using binomial regression. RESULTS: Of the 762 histologically proven CCA cases, 161 were from the screening group and 601 from the walk-in group. The proportion of early stage CCA (stages 0 to II) diagnosed was 84.5% in the screening and 21.6% in the walk-in groups. After adjustment age, gender, and liver fluke infection, there was a significantly higher proportion (P < 0.001) and higher chance (P < 0.001) of having early stage CCA in the screening group than in the walk-in group. CONCLUSIONS: US-screening is an effective tool for detecting early stage, operable CCA in high incidence areas.

Evaluating a preoperative protocol that includes magnetic resonance imaging for lymph node metastasis in the Cholangiocarcinoma Screening and Care Program (CASCAP) in Thailand.

BACKGROUND: Treatment planning especially liver resection in cholangiocarcinoma (CCA) depends on the extension of tumor and lymph node metastasis which is included as a key criterion for operability. Magnetic resonance imaging (MRI) offers a rapid and powerful tool for the detection of lymph node metastasis (LNM) and in the current manuscript is assessed as a critical tool in the preoperative protocol for liver resection for treatment of CCA. However, the accuracy of MRI to detect LNM from CCA had yet to be comprehensively evaluated. METHODS: The accuracy of MRI to detect LNM was assessed in a cohort of individuals with CCA from the Cholangiocarcinoma Screening and Care Program (CASCAP), a screening program designed to reduce CCA in Northeastern Thailand by community-based ultrasound (US) for CCA. CCA-positive individuals are referred to one of the nine tertiary centers in the study to undergo a preoperative protocol that included enhanced imaging by MRI. Additionally, these individuals also underwent lymph node biopsies for histological confirmation of LNM (the "gold standard") to determine the accuracy of the MRI results. RESULTS: MRI accurately detected the presence or absence of LNM in only 29 out of the 51 CCA cases (56.9%, 95% CI 42.2-70.7), resulting in a sensitivity of 57.1% (95% CI 34.0-78.2) and specificity of 56.7% (95% CI 37.4-74.5), with positive and negative predictive values of 48.0% (95% CI 27.8-68.7) and 65.4% (95% CI 44.3-82.8), respectively. The positive likelihood ratio was 1.32 (95% CI 0.76-2.29), and the negative likelihood ratio was 0.76 (95% CI 0.42-1.36). CONCLUSIONS: MRI showed limited sensitivity and a poor positive predictive value for the diagnosis of LNM for CCA, which is of particular concern in this resource-limited setting, where simpler detection methods could be utilized that are more cost-effective in this region of Thailand. Therefore, the inclusion of MRI, a costly imaging method, should be reconsidered as part of protocol for treatment planning of CCA, given the number of false positives, especially as it is critical in determining the operability for CCA subjects.

Thai nurse cohort study: cohort profiles and key findings.

BACKGROUND: Globally, the nursing profession faces shortages, high turnover, and inequitable distribution. These problems are particularly acute in South East Asia. The present paper describes the design and initial findings of the Thai Nurse Cohort Study (TNCS). METHODS: The TNCS is a longitudinal prospective cohort study comprising multiple age cohorts, initiated in 2009 and expected to run until 2027. Cohorts comprise registered nurses (RN) holding professional licenses granted by the Thailand Nursing and Midwifery Council. Follow-up is at 3-year intervals, with new (younger) TNCS cohorts introduced and older, no-longer eligible members checked out. This maintains the cohort size as representative of the Thai RN population. The first survey round (2009) used a self-administered mailed questionnaire. The second round (2012) provided follow-up of the initial cohort and formed the baseline survey of new entries. RESULTS: The sampling frame for the first round was 142,699 licensed RN; 50,200 age-stratified participants were randomly selected and mailed the questionnaire, and 18,198 questionnaires were returned owing to incorrect addresses. Of the remaining 32,002 participants, 18,756 (58.6 %) responded (average age 43.7 ± 9.8 years). About 15.4 % (equivalent to 20,000 of the current RN population), reported an intention to leave their nursing career. The second round achieved a follow-up rate of 60.2 %. This round included 3020 participants randomly selected from 6402 new RN (response rate, 38.3 %; mean age 23.1 ± 3.5 years). In this round, 11.2 % reported they intended to leave nursing in the next 2 years. CONCLUSIONS: These two survey rounds have highlighted that Thailand is facing critical nurse shortages. A high rate of nurses expressed an intention to leave the profession; the capacity to replace these potential losses is much lower.