Savable but lost lives when ICU is overloaded: a model from 733 patients in epicenter Wuhan, China

Ziwei Zhu, Yang Li, Tingting Dan, Xijie Chen, Wuxiu Quan, Yu Hu, Guihua Tao, Lei Zhu, Jijin Zhu, Yuyan Jin, Longgeng Li, Chaokai Liang, Yangming Ou, Hongmin Cai, Hanchun WenResearch Square 2020 · DOI

Abstract

Background: Coronavirus Disease (COVID-19) causes a sudden clinical deterioration at specific checkpoints, necessitating Intensive Care Unit (ICU) intervention. The surge in demand for ICUs during the initial outbreak posed significant risks to medical systems. Estimating the mortality of critical patients not admitted to the ICU (MI-mortality) is vital for optimizing resource management. Methods: This retrospective study analyzed 733 COVID-19 inpatients at Huangpi Hospital of Traditional Chinese Medicine (Wuhan, China) as of March 18, 2020. We aimed to estimate MI-mortality and build a machine learning model to identify critical patients. Demographic, clinical, and laboratory data were analyzed to find prognostic factors. Results: A prognostic classification model was built to identify patients needing ICU care. Predictive accuracies on the whole, training, and testing sets were 0.8513, 0.8935, and 0.8288, respectively, with AUCs of 0.8844, 0.8941, and 0.9120. Analysis showed MI-mortality was 41% and ICU-mortality was 32%. Conclusions: In our cohort, 25 patients were admitted to the ICU (8 died). Another 25 patients predicted to need ICU care did not receive it due to shortages, resulting in a 41% MI-mortality rate.

Introduction

During the early 2020 outbreak in Wuhan, the medical system faced a catastrophic surge in critical patients, making ICU beds a life-or-death bottleneck. Clinicians were forced to make agonizing triage decisions with limited predictive tools. This study addresses two major gaps: objectively identifying patients likely to deteriorate and quantifying the mortality cost of an overloaded ICU system. The authors define "Missing-ICU mortality" (MI-mortality) as the death rate of patients who met ICU criteria but remained in general wards due to lack of space.

Methods and Patient Cohort

The researchers conducted a retrospective study using a cohort of 733 laboratory-confirmed COVID-19 patients. The dataset included 143 continuous variables and 51 categorical factors.

Demographic and clinical characteristics of the 733-patient cohort, comparing ICU and non-ICU care groups.

As shown in the table above, significant differences ($p < 0.0001$) were observed between ICU and non-ICU groups in age and specific laboratory findings, such as high-sensitivity troponin I and myoglobin levels. The modeling approach involved:

  • Feature Engineering: Extraction of 909 features, including longitudinal laboratory results.
  • Machine Learning: A Support Vector Machine (SVM) classifier was employed. To address class imbalance (only 25 ICU admissions), an ensemble learning strategy with balance-sampling was used.
  • Feature Selection: The top 10 prognostic factors were identified based on Area Under the Curve (AUC) to create a deployable "lean" model.

Results

The model demonstrated robust predictive power, particularly when utilizing the refined set of 10 key features.

Performance metrics of the prognostic classification model across training, testing, and whole datasets.

The quantitative performance metrics indicate high reliability, with the AUC reaching 0.9120 on the testing set using the top 10 features. Notably, the model achieved a perfect sensitivity of 1.0000 on the test set, ensuring no patients requiring ICU care were missed.

The Mortality Gap and Biomarkers

The most striking finding is the survival difference. Patients who received ICU care had a mortality rate of 32%. However, for those identified by the model as needing ICU care but who remained in general wards, mortality jumped to 41%. This 9% gap represents "savable" lives lost to system overload. Key indicators of crisis identified by the model include high-sensitivity troponin I (hs-cTnI), myoglobin (Mb), D-Dimer, and IgM.

Discussion and Conclusion

Statistical analysis highlighted age as a critical threshold; patients over 60 were significantly more likely to require ICU admission. By focusing on the variance and mean of laboratory results over time, the model serves as an early warning system.

Limitations include the small positive sample size (25 ICU patients) and the single-center nature of the study. Future work should validate these biomarkers across different variants and healthcare systems. Integrating this model into hospital management could facilitate rational ICU resource planning during future surges. In conclusion, the 41% MI-mortality rate underscores the critical need for adequate ICU capacity and early identification of at-risk patients.

Summary updated

Abstract

<title>Abstract</title> <bold>Background. </bold>Coronavirus Disease (COVID-19) causes a sudden turn over to bad at some check-point and thus needs intervention of intensive care unit (ICU). This resulted in urgent and large needs of ICUs posed great risks to the medical system. Estimating the mortality of critical in-patients who were not admitted to the ICU (MI-mortality) will be valuable to optimize the management and assignment of ICU.<bold>Methods.</bold> Retrospective, of the 733 in-patients diagnosed with COVD-19 at Huangpi Hospital of Traditional Chinese Medicine (Wuhan, China), as of March 18, 2020. This study aims to estimate the MI-mortality and build a model to identify the critical in-patients. Demographic, clinical and laboratory results were collected and analyzed. The mortality rate for the patients who failed to receive ICU and unfortunately died was analyzed. To this end, the key factors for prognostic of patients who may need ICU care were found. A prognostic classification model using machine learning was built to identify the patient who may need ICU. <bold>Results. </bold>Considering the shortage of ICU beds at the beginning of disease emergence, we defined the mortality for those patients who were predicted to be in needing of ICU treatment yet they did not as MI-mortality. Patients who entered the ICU and died were defined as ICU-mortality. To estimate MI-mortality, a prognostic classification model was built to identify the in-patients who may need ICU care based on the medical factors collected in-hospital. Its predictive accuracies on whole patient set (733 [25 708]), training set (586 [20 566]) and testing set (147 [5 142]) dataset were 0.8513, 0.8935 and 0.8288, with the AUC of 0.8844, 0.8941 and 0.9120, respectively. Our analysis had shown that the MI-mortality is 41% and the ICU-mortality is 32%, implying that enough bed of ICU in treating patients in critical conditions. <bold>Conclusions. </bold>On our cohort of 733 patients, 25 in-patients were admitted to ICU, among them 8 patients died. 25 in-patients who have been predicted by our model that they should need ICU care, yet they did not enter ICU due to lack of shorting ICU wards. The MI-mortality is 41%.
BibTeX
@inproceedings{zhu2020savable,
  author = {Ziwei Zhu and Yang Li and Tingting Dan and Xijie Chen and Wuxiu Quan and Yu Hu and Guihua Tao and Lei Zhu and Jijin Zhu and Yuyan Jin and Longgeng Li and Chaokai Liang and Yangming Ou and Hongmin Cai and Hanchun Wen},
  title = {{Savable but lost lives when ICU is overloaded: a model from 733 patients in epicenter Wuhan, China}},
  booktitle = {Research Square},
  year = {2020},
  doi = {10.21203/rs.3.rs-39310/v1},
}
Figure