A process mining- deep learning approach to predict survival in a cohort of hospitalized COVID-19 patients

Background

Various machine learning and artificial intelligence methods have been used to predict outcomes of hospitalized COVID-19 patients. However, process mining has not yet been used for COVID-19 prediction. We developed a process mining/deep learning approach to predict mortality among COVID-19 patients and updated the prediction in 6-h intervals during the first 72 h after hospital admission.

Methods

The process mining/deep learning model produced temporal information related to the variables and incorporated demographic and clinical data to predict mortality. The mortality prediction was updated in 6-h intervals during the first 72 h after hospital admission. Moreover, the performance of the model was compared with published and self-developed traditional machine learning models that did not use time as a variable. The performance was compared using the Area Under the Receiver Operator Curve (AUROC), accuracy, sensitivity, and specificity.

Results

The proposed process mining/deep learning model outperformed the comparison models in almost all time intervals with a robust AUROC above 80% on a dataset that was imbalanced.

Conclusions

Our proposed process mining/deep learning model performed significantly better than commonly used machine learning approaches that ignore time information. Thus, time information should be incorporated in models to predict outcomes more accurately.

Prediction of unplanned 30-day readmission for ICU patients with heart failure

BACKGROUND

Intensive Care Unit (ICU) readmissions in patients with heart failure (HF) result in a significant risk of death and financial burden for patients and healthcare systems. Prediction of at-risk patients for readmission allows for targeted interventions that reduce morbidity and mortality.

METHODS AND RESULTS

We presented a process mining/deep learning approach for the prediction of unplanned 30-day readmission of ICU patients with HF. A patient's health records can be understood as a sequence of observations called event logs; used to discover a process model. Time information was extracted using the DREAM (Decay Replay Mining) algorithm. Demographic information and severity scores upon admission were then combined with the time information and fed to a neural network (NN) model to further enhance the prediction efficiency. Additionally, several machine learning (ML) algorithms were developed to be used as the baseline models for the comparison of the results.

RESULTS

By using the Medical Information Mart for Intensive Care III (MIMIC-III) dataset of 3411 ICU patients with HF, our proposed model yielded an area under the receiver operating characteristics (AUROC) of 0.930, 95% confidence interval of [0.898-0.960], the precision of 0.886, sensitivity of 0.805, accuracy of 0.841, and F-score of 0.800 which were far better than the results of the best baseline model and the existing literature.

CONCLUSIONS

The proposed approach was capable of modeling the time-related variables and incorporating the medical history of patients from prior hospital visits for prediction. Thus, our approach significantly improved the outcome prediction compared to that of other ML-based models and health calculators.

Inverse association between cigarette and water pipe smoking and hypertension in an elderly population in Iran: Bushehr elderly health programme

The collected data in Bushehr Elderly Health (BEH) Program which had detailed the data on participants' smoking status and habits, was analysed to investigate the association between smoking of both water pipes and cigarettes and hypertension in an elderly population. Three thousand elderly men and women who participated in the baseline assessment of the BEH Program-a prospective population-based study being conducted in Bushehr, Iran-were selected randomly through a multistage, stratified cluster sampling method. Systolic and diastolic blood pressures were measured twice using a mercury sphygmomanometer, and researchers asked participants about medical history of hypertension as well as history of cigarette and water pipe smoking. Researchers used binary logistic regression models to assess the association of hypertension and smoking, and found an inverse, statistically significant association between current smoking and hypertension (odds ratio (OR)=0.50 (95% confidence interval (CI)=0.41, 0.60)). The association remained statistically significant after controlling for age, education and body mass index (OR=0.54 (95% CI=0.45, 0.66)). Findings were consistent for cigarette and water pipe smoking by sex (all ORs were inverse and statistically significant). Both cigarette and water pipe smoking were associated with reduced hypertension among older people, but the strength of association was different between men and women and also between cigarette and water pipe smoking. The reasons behind the association as well as the differences observed need to be investigated through more comprehensive, longitudinal studies.

Abstract P2-06-06: A clinical model for assessment of the individual breast cancer risk

Background. Most mammography screening programs are not individualized. To efficiently screen for breast cancer the individual risk of the disease should be determined. We describe a model that estimate the 2-year risk of breast cancer and could be used at most mammography screening units without adding substantial cost.

Methods. The study was based on the population based prospective Karma cohort including 70,877 participants. Mammograms were collected up to three years following baseline mammogram. A prediction protocol was developed using mammographic features (density, microcalcifications and masses), use of hormone replacement therapy (HRT), family history of breast cancer, menopausal status, age, and body mass index. Relative risks were calculated using conditional logistic regression. Absolute risks were calculated using the iCARE protocol.

Results. Comparing women at highest and lowest mammographic density yielded a 5-fold higher risk of breast cancer for women at highest density. When adding microcalcifications and masses to the model, high-risk women had a nearly 9-fold higher risk of breast cancer compared to those at lowest risk. The difference in microcalcifications and masses between left and right breast was a better predictor of breast cancer than number of microcalcifications and masses in the breasts.

When calculating the absolute 2-year risk of breast cancer we stratified women using the NICE guidelines for 10-year risk divided by 5 (Table 1). The mean absolute 2-year risk of breast cancer in the different risk categories was 0.12%, 0.33%, 0.83% and 1.95% for women at low, moderate, general, and high risk. In most countries with established mammography screening programs approximately 5 women in a 1000 are diagnosed with breast cancer at regular screening. We managed to identify a low risk group of approximately 10% of all women where 1 woman in a 1000 will be diagnosed with breast cancer, contrasting the 2% of all women at highest risk where 20 women out of a 1000 will be diagnosed with breast cancer within 2 years of last negative screen (Table 1).

In the full model, taking HRT use, family history of breast cancer and menopausal status into consideration, area under the curve (AUC) reached 0.71.

Conclusions. Our model includes three mammographic features that could easily be derived from clinically available software. By adding information on some few established risk factors it is possible to improve clinical care by identifying women in need of additional examination procedures. At the same time there is a substantial proportion of women that will have very little benefit from mammography screening due to their low risk of breast cancer.

Table 1. Absolute 2-year risk of breast cancer in women stratified in to risk categories based on the NICE guidelinesAbsolute 2-year risk (risk group)Percent women at riskMean absolute 2-year risk2Stratified 2-year risk0-0.15 (low)10.30.121.0 (reference)0.15-<0.6 (general)64.80.332.750.6-<1.6 (moderate)22.90.826.83≥1.6 (high)2.01.9516.2

Citation Format: Eriksson M, Czene K, Pawitan Y, Leifland K, Darabi H, Hall P. A clinical model for assessment of the individual breast cancer risk [abstract]. In: Proceedings of the 2016 San Antonio Breast Cancer Symposium; 2016 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2017;77(4 Suppl):Abstract nr P2-06-06.

FGF receptor genes and breast cancer susceptibility: results from the Breast Cancer Association Consortium

BACKGROUND

Breast cancer is one of the most common malignancies in women. Genome-wide association studies have identified FGFR2 as a breast cancer susceptibility gene. Common variation in other fibroblast growth factor (FGF) receptors might also modify risk. We tested this hypothesis by studying genotyped single-nucleotide polymorphisms (SNPs) and imputed SNPs in FGFR1, FGFR3, FGFR4 and FGFRL1 in the Breast Cancer Association Consortium.

METHODS

Data were combined from 49 studies, including 53 835 cases and 50 156 controls, of which 89 050 (46 450 cases and 42 600 controls) were of European ancestry, 12 893 (6269 cases and 6624 controls) of Asian and 2048 (1116 cases and 932 controls) of African ancestry. Associations with risk of breast cancer, overall and by disease sub-type, were assessed using unconditional logistic regression.

RESULTS

Little evidence of association with breast cancer risk was observed for SNPs in the FGF receptor genes. The strongest evidence in European women was for rs743682 in FGFR3; the estimated per-allele odds ratio was 1.05 (95% confidence interval=1.02-1.09, P=0.0020), which is substantially lower than that observed for SNPs in FGFR2.

CONCLUSION

Our results suggest that common variants in the other FGF receptors are not associated with risk of breast cancer to the degree observed for FGFR2.

Effectiveness of methoprene, an insect growth regulator, against malaria vectors in Fars, Iran: a field study

Methoprene, an insect growth regulator, was evaluated under field conditions against the main malaria vectors in the Islamic Republic of Iran. The effect of 5, 10 and 20 kg ha(-1) concentration ofmethoprene granule formulation and 100 and 200 mL ha(-1) concentration of EC formulation was measured to determine any changes in Anophelini larval abundance and IE ratio in both rice fields and artificial ponds. In artificial ponds, granular methoprene at a dose of 20 kg ha(-1) inhibited adult emergence by 77.1% after 1 day and 65.9% after 3 days. The emulsifiable concentrate formulation of methoprene at 200 mL ha(-1) inhibited adult emergence by 83.7% after 1 day and 32.2% after 3 days. In rice fields, inhibition of emergence was 44.3% at 20 kg ha(-1) granule and 35.8% for emulsifiable concentrate at 200 mL ha(-1) after 3 days. The results vary depending on the mosquito species, treatment methods, breeding places and type of formulation.

Mixed infections of Vibrio cholerae O1 Ogawa EL Tor with Shigella dysenteriae

Mixed infections caused by enteric pathogens such as bacteria, virus, protozoa and helminthes were reported in different literatures. This report also describes the co-infections caused by Vibrio cholerae O1 Ogawa EL Tor with Shigella dysenteriae in a patient. A 36-year-old man was admitted in Fatemeh Zahra Hospital of Bushehr Iran with fever, vomiting and dysentery. His stool sample was cultured, for identification purposes TCBS, XLD and other media were used. V. cholerae and S. dysenteriae were identified. Both species were resistant to ampicillin and sensitive to nalidixic acid and trimethoprim-sulfamethoxazol. Shigella was resistant to tetracycline. The results of the study showed that places where diarrheal diseases especially cholera are endemic, it is better to examine for those patients with dysentery for the presence of the V. cholerae O1. That will prevent the spread of pathogenic organism in the community.

Compositional changes of PBL population in patients with chronic hepatitis B virus infection

In this report we have analysed the peripheral blood lymphocyte of several patients with chronic hepatitis B virus infection with flow cytometry. Based on the presence and absence of the HBeAb, patients were divided into two groups. In both, all the patients were HBsAg positive with normal range of serum alanine aminotranferase (23.9 +/- 17.8). We have found that the immunophenotypic profiles of patients were different from healthy donors with significant decrease in CD(3)(+) T cells, specially CD(8)(+) T cells and a significant increase in the CD(19)(+) B cells. The differences were seen in other subset of T cells (CD(4)(+)) or NK cells (CD(56)(+)/CD(16)(+)) and HLA-DR markers were not significant. When the phenotypic profiles of both groups were compared with each other, such changes were more dominant in group II, with HBeAb positive than in group I, with HBeAb negative. Also, we have seen a correlation between the increase of CD(19)(+) B cells and the decrease of CD CD(3)(+) T cells. No such correlation was observed with other cells.