Identifying Sequential Complication and Mortality Patterns in Diabetes Mellitus: Comparisons of Machine Learning Methodologies (preprint)

Background: Diabetes mellitus-related complications adversely affect the quality of life. Better risk-stratified care through mining of sequential complication patterns is needed to enable early detection and prevention. Methods: Univariable and multivariate logistic regression was used to identify significant variables that can predict mortality. A sequence analysis method termed Prefixspan was applied to identify the most common couple, triple, quadruple, quintuple and sextuple sequential complication patterns in the directed comorbidity pathology network. A knowledge enhanced CPT+ (KCPT+) sequence prediction model is developed to predict the next possible outcome along the progression trajectories of diabetes-related complications. Findings: A total of 14,144 diabetic patients (51% males) were included. Acute myocardial infarction (AMI) without known ischaemic heart disease (IHD) (odds ratio [OR]: 2.8, 95% CI: [2.3, 3.4]), peripheral vascular disease (OR: 2.3, 95% CI: [1.9, 2.8]), dementia (OR: 2.1, 95% CI: [1.8, 2.4]), and IHD with AMI (OR: 2.4, 95% CI: [2.1, 2.6]) are the most important multivariate predictors of mortality. KCPT+ shows high accuracy in predicting mortality (F1 score 0.90, ACU 0.88), osteoporosis (F1 score 0.86, AUC 0.82), ophthalmological complications (F1 score 0.82, AUC 0.82), IHD with AMI (F1 score 0.81, AUC 0.85) and neurological complications (F1 score 0.81, AUC 0.83) with a particular prior complication sequence. Interpretation: Sequence analysis identifies the most common pattern characteristics of disease-related complications efficiently. The proposed sequence prediction model is accurate and enables clinicians to diagnose the next complication earlier, provide better risk-stratified care, and devise efficient treatment strategies for diabetes mellitus patients.

Development of a predictive risk model for severe COVID-19 disease using population-based administrative data (preprint)

Background: Recent studies have reported numerous significant predictors for adverse outcomes in COVID-19 disease. However, there have been few simple clinical risk score for prompt risk stratification. The objective is to develop a simple risk score for severe COVID-19 disease using territory-wide healthcare data based on simple clinical and laboratory variables. Methods: Consecutive patients admitted to Hong Kong public hospitals between 1st January and 22nd August 2020 diagnosed with COVID-19, as confirmed by RT-PCR, were included. The primary outcome was composite intensive care unit admission, need for intubation or death with follow-up until 8th September 2020. Results: COVID-19 testing was performed in 237493 patients and 4445 patients (median age 44.8 years old, 95% CI: [28.9, 60.8]); 50% male) were tested positive. Of these, 212 patients (4.8%) met the primary outcome. A risk score including the following components was derived from Cox regression: gender, age, hypertension, stroke, diabetes mellitus, ischemic heart disease/heart failure, respiratory disease, renal disease, increases in neutrophil count, monocyte count, sodium, potassium, urea, alanine transaminase, alkaline phosphatase, high sensitive troponin-I, prothrombin time, activated partial thromboplastin time, D-dimer and C-reactive protein, as well as decreases in lymphocyte count, base excess and bicarbonate levels. The model based on test results taken on the day of admission demonstrated an excellent predictive value. Incorporation of test results on successive time points did not further improve risk prediction. Conclusions: A simple clinical score accurately predicted severe COVID-19 disease, even without including symptoms, blood pressure or oxygen status on presentation, or chest radiograph results.

Identifying main and interaction effects of risk factors to predict intensive care admission in patients hospitalized with COVID-19: a retrospective cohort study in Hong Kong (preprint)

Background: The coronavirus disease 2019 (COVID-19) has become a pandemic, placing significant burdens on the healthcare systems. In this study, we tested the hypothesis that a machine learning approach incorporating hidden nonlinear interactions can improve prediction for Intensive care unit (ICU) admission. Methods: Consecutive patients admitted to public hospitals between 1st January and 24th May 2020 in Hong Kong with COVID-19 diagnosed by RT-PCR were included. The primary endpoint was ICU admission. Results: This study included 1043 patients (median age 35 (IQR: 32-37; 54% male). Nineteen patients were admitted to ICU (median hospital length of stay (LOS): 30 days, median ICU LOS: 16 days). ICU patients were more likely to be prescribed angiotensin converting enzyme inhibitors/angiotensin receptor blockers, anti-retroviral drugs lopinavir/ritonavir and remdesivir, ribavirin, steroids, interferon-beta and hydroxychloroquine. Significant predictors of ICU admission were older age, male sex, prior coronary artery disease, respiratory diseases, diabetes, hypertension and chronic kidney disease, and activated partial thromboplastin time, red cell count, white cell count, albumin and serum sodium. A tree-based machine learning model identified most informative characteristics and hidden interactions that can predict ICU admission. These were: low red cells with 1) male, 2) older age, 3) low albumin, 4) low sodium or 5) prolonged APTT. A five-fold cross validation confirms superior performance of this model over baseline models including XGBoost, LightGBM, random forests, and multivariate logistic regression. Conclusions: A machine learning model including baseline risk factors and their hidden interactions can accurately predict ICU admission in COVID-19.

Are COVID-19 infected children with gastrointestinal symptoms different from those without symptoms? A comparative study of the clinical characteristics and epidemiological trend of 244 pediatric cases from Wuhan (preprint)

Objective: COVID-19 patients presenting with gastrointestinal (GI) symptoms occur in both adults and children. To date, however, no large sample size study focusing on gastrointestinal symptoms in pediatric cases has been published. We analyzed COVID-19 infected children in Wuhan who presented with initial GI symptoms to determine the GI characteristics and epidemiological trend of the disease. Design: We retrospectively analyzed 244 children patients confirmed with COVID-19 at Wuhan Children's Hospital from 21 Jan to 20 Mar 2020. Symptomatic cases were divided into two groups according to whether the patients presented with or without GI symptoms on admission. Demographic, epidemiological, symptoms, and laboratory data were compared. We also analyzed the respective trends of case number changes of GI cases and asymptomatic cases. Results: 34 out of 193 symptomatic children had GI symptoms. They had lower median age and weight, a higher rate of fever, a longer length of stay and more hematological and biochemical abnormalities than patients without GI symptoms. There was no significant difference in chest CT findings or stool SARS-CoV-2 test positive percentages between the two groups. The number of patients admitted with GI symptoms showed an overall downward trend with time. At the time of writing, 242 patients were discharged, one died, and one critically ill patient was still in the intensive care unit. Conclusion: COVID-19 infected children with GI symptoms are prone to presenting with more clinical and laboratory abnormalities than patients without GI symptoms. More attention and timely hospital admission are needed for these patients.