Dataset Information

Machine learning-derived prediction of in-hospital mortality in patients with severe acute respiratory infection: analysis of claims data from the German-wide Helios hospital network.

ABSTRACT:

Background

Severe acute respiratory infections (SARI) are the most common infectious causes of death. Previous work regarding mortality prediction models for SARI using machine learning (ML) algorithms that can be useful for both individual risk stratification and quality of care assessment is scarce. We aimed to develop reliable models for mortality prediction in SARI patients utilizing ML algorithms and compare its performances with a classic regression analysis approach.

Methods

Administrative data (dataset randomly split 75%/25% for model training/testing) from years 2016-2019 of 86 German Helios hospitals was retrospectively analyzed. Inpatient SARI cases were defined by ICD-codes J09-J22. Three ML algorithms were evaluated and its performance compared to generalized linear models (GLM) by computing receiver operating characteristic area under the curve (AUC) and area under the precision-recall curve (AUPRC).

Results

The dataset contained 241,988 inpatient SARI cases (75 years or older: 49%; male 56.2%). In-hospital mortality was 11.6%. AUC and AUPRC in the testing dataset were 0.83 and 0.372 for GLM, 0.831 and 0.384 for random forest (RF), 0.834 and 0.382 for single layer neural network (NNET) and 0.834 and 0.389 for extreme gradient boosting (XGBoost). Statistical comparison of ROC AUCs revealed a better performance of NNET and XGBoost as compared to GLM.

Conclusion

ML algorithms for predicting in-hospital mortality were trained and tested on a large real-world administrative dataset of SARI patients and showed good discriminatory performances. Broad application of our models in clinical routine practice can contribute to patients' risk assessment and quality management.

SUBMITTER: Leiner J

PROVIDER: S-EPMC9502925 | biostudies-literature | 2022 Sep

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Machine learning-derived prediction of in-hospital mortality in patients with severe acute respiratory infection: analysis of claims data from the German-wide Helios hospital network.

Leiner Johannes J Pellissier Vincent V König Sebastian S Hohenstein Sven S Ueberham Laura L Nachtigall Irit I Meier-Hellmann Andreas A Kuhlen Ralf R Hindricks Gerhard G Bollmann Andreas A

Respiratory research 20220923 1

<h4>Background</h4>Severe acute respiratory infections (SARI) are the most common infectious causes of death. Previous work regarding mortality prediction models for SARI using machine learning (ML) algorithms that can be useful for both individual risk stratification and quality of care assessment is scarce. We aimed to develop reliable models for mortality prediction in SARI patients utilizing ML algorithms and compare its performances with a classic regression analysis approach.<h4>Methods</h ...[more]

PMID: 36151525

Similar Datasets

Project description:BackgroundDuring the early phase of the Covid-19 pandemic, reductions of hospital admissions with a focus on emergencies have been observed for several medical and surgical conditions, while trend data during later stages of the pandemic are scarce. Consequently, this study aims to provide up-to-date hospitalization trends for several conditions including cardiovascular, psychiatry, oncology and surgery cases in both the in- and outpatient setting.Methods and findingsUsing claims data of 86 Helios hospitals in Germany, consecutive cases with an in- or outpatient hospital admission between March 13, 2020 (the begin of the "protection" stage of the German pandemic plan) and December 10, 2020 (end of study period) were analyzed and compared to a corresponding period covering the same weeks in 2019. Cause-specific hospitalizations were defined based on the primary discharge diagnosis according to International Statistical Classification of Diseases and Related Health Problems (ICD-10) or German procedure classification codes for cardiovascular, oncology, psychiatry and surgery cases. Cumulative hospitalization deficit was computed as the difference between the expected and observed cumulative admission number for every week in the study period, expressed as a percentage of the cumulative expected number. The expected admission number was defined as the weekly average during the control period. A total of 1,493,915 hospital admissions (723,364 during the study and 770,551 during the control period) were included. At the end of the study period, total cumulative hospitalization deficit was -10% [95% confidence interval -10; -10] for cardiovascular and -9% [-10; -9] for surgical cases, higher than -4% [-4; -3] in psychiatry and 4% [4; 4] in oncology cases. The utilization of inpatient care and subsequent hospitalization deficit was similar in trend with some variation in magnitude between cardiovascular (-12% [-13; -12]), psychiatry (-18% [-19; -17]), oncology (-7% [-8; -7]) and surgery cases (-11% [-11; -11]). Similarly, cardiovascular and surgical outpatient cases had a deficit of -5% [-6; -5] and -3% [-4; -3], respectively. This was in contrast to psychiatry (2% [1; 2]) and oncology cases (21% [20; 21]) that had a surplus in the outpatient sector. While in-hospital mortality, was higher during the Covid-19 pandemic in cardiovascular (3.9 vs. 3.5%, OR 1.10 [95% CI 1.06-1.15], P<0.01) and in oncology cases (4.5 vs. 4.3%, OR 1.06 [95% CI 1.01-1.11], P<0.01), it was similar in surgical (0.9 vs. 0.8%, OR 1.06 [95% CI 1.00-1.13], P = 0.07) and in psychiatry cases (0.4 vs. 0.5%, OR 1.01 [95% CI 0.78-1.31], P<0.95).ConclusionsThere have been varying changes in care pathways and in-hospital mortality in different disciplines during the Covid-19 pandemic in Germany. Despite all the inherent and well-known limitations of claims data use, this data may be used for health care surveillance as the pandemic continues worldwide. While this study provides an up-to-date analysis of utilization of hospital care in the largest German hospital network, short- and long-term consequences are unknown and deserve further studies.

Project description:Background: After the first COVID-19 infection wave, a constant increase of pulmonary embolism (PE) hospitalizations not linked with active PCR-confirmed COVID-19 was observed, but potential contributors to this observation are unclear. Therefore, we analyzed associations between changes in PE hospitalizations and (1) the incidence of non-COVID-19 pneumonia, (2) the use of computed tomography pulmonary angiography (CTPA), (3) volume depletion, and (4) preceding COVID-19 infection numbers in Germany. Methods: Claims data of Helios hospitals in Germany were used, and consecutive cases with a hospital admission between May 6 and December 15, 2020 (PE surplus period), were analyzed and compared to corresponding periods covering the same weeks in 2016-2019 (control period). We analyzed the number of PE cases in the target period with multivariable Poisson general linear mixed models (GLMM) including (a) cohorts of 2020 versus 2016-2019, (b) the number of cases with pneumonia, (c) CTPA, and (d) volume depletion and adjusted for age and sex. In order to associate the daily number of PE cases in 2020 with the number of preceding SARS-CoV-2 infections in Germany, we calculated the average number of daily infections (divided by 10,000) occurring between 14 up to 90 days with increasing window sizes before PE cases and modeled the data with Poisson regression. Results: There were 2,404 PE hospitalizations between May 6 and December 15, 2020, as opposed to 2,112-2,236 (total 8,717) in the corresponding 2016-2019 control periods (crude rate ratio [CRR] 1.10, 95% CI 1.05-1.15, P < 0.01). With the use of multivariable Poisson GLMM adjusted for age, sex, and volume depletion, PE cases were significantly associated with the number of cases with pneumonia (CRR 1.09, 95% CI 1.07-1.10, P < 0.01) and with CTPA (CRR 1.10, 95% CI 1.09-1.10, P < 0.01). The increase of PE cases in 2020 compared with the control period remained significant (CRR 1.07, 95% CI 1.02-1.12, P < 0.01) when controlling for those factors. In the 2020 cohort, the number of preceding average daily COVID-19 infections was associated with increased PE case incidence in all investigated windows, i.e., including preceding infections from 14 to 90 days. The best model (log likelihood -576) was with a window size of 4 days, i.e., average COVID-19 infections 14-17 days before PE hospitalization had a risk of 1.20 (95% CI 1.12-1.29, P < 0.01). Conclusions: There is an increase in PE cases since early May 2020 compared to corresponding periods in 2016-2019. This surplus was significant even when controlling for changes in potential modulators such as demographics, volume depletion, non-COVID-19 pneumonia, CTPA use, and preceding COVID-19 infections. Future studies are needed (1) to investigate a potential causal link for increased risk of delayed PE with preceding SARS-CoV-2 infection and (2) to define optimal screening for SARS-CoV-2 in patients presenting with pneumonia and PE.

Project description:Mechanical ventilation (MV) is vital for critically ill ICU patients but carries significant mortality risks. This study aims to develop a predictive model to estimate hospital mortality among MV patients, utilizing comprehensive health data to assist ICU physicians with early-stage alerts. We developed a Machine Learning (ML) framework to predict hospital mortality in ICU patients receiving MV. Using the MIMIC-III database, we identified 25,202 eligible patients through ICD-9 codes. We employed backward elimination and the Lasso method, selecting 32 features based on clinical insights and literature. Data preprocessing included eliminating columns with over 90% missing data and using mean imputation for the remaining missing values. To address class imbalance, we used the Synthetic Minority Over-sampling Technique (SMOTE). We evaluated several ML models, including CatBoost, XGBoost, Decision Tree, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Logistic Regression, using a 70/30 train-test split. The CatBoost model was chosen for its superior performance in terms of accuracy, precision, recall, F1-score, AUROC metrics, and calibration plots. The study involved a cohort of 25,202 patients on MV. The CatBoost model attained an AUROC of 0.862, an increase from an initial AUROC of 0.821, which was the best reported in the literature. It also demonstrated an accuracy of 0.789, an F1-score of 0.747, and better calibration, outperforming other models. These improvements are due to systematic feature selection and the robust gradient boosting architecture of CatBoost. The preprocessing methodology significantly reduced the number of relevant features, simplifying computational processes, and identified critical features previously overlooked. Integrating these features and tuning the parameters, our model demonstrated strong generalization to unseen data. This highlights the potential of ML as a crucial tool in ICUs, enhancing resource allocation and providing more personalized interventions for MV patients.

Dataset Information

Machine learning-derived prediction of in-hospital mortality in patients with severe acute respiratory infection: analysis of claims data from the German-wide Helios hospital network.

Background

Methods

Results

Conclusion

Publications

Machine learning-derived prediction of in-hospital mortality in patients with severe acute respiratory infection: analysis of claims data from the German-wide Helios hospital network.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets