Predictive models can help stratify patient risk during and after COVID-19
A pair of studies published in the Journal of the American Medical Association show how predictive models can assist with risk stratification when it comes to treating patients with COVID-19 and scheduling elective procedures after the pandemic.
The first study, which focused on 2,511 hospitalized COVID-19 patients in eastern Massachusetts, examined how laboratory studies, in conjunction with sociodemographic features and prior diagnosis, helped identify individuals at particularly high risk.
The second study, a Duke University study that developed predictive models from the electronic health records of 42,199 elective surgery patients, found that modeling, along with other factors, can be used to inform how to recommence elective inpatient procedures.
"The novel coronavirus disease 2019 has changed the provision of hospital- and clinic-based surgical care," noted the authors of the Duke University study.
"As hospitals prepared for possible surges of infected patients requiring admission and possible intensive care stay, entire institutions and health systems took stock of their resources to meet an uncertain demand," they continued.
"This included estimating an ever-fluctuating number of available beds, securing sufficient personal protective equipment and ventilators, minimizing staff shortages, and establishing protocols to mitigate against nosocomial infection."
WHY IT MATTERS
During the COVID-19 crisis, hospitals have faced crammed ICUs and shortages of beds, ventilators and personal protective equipment – making it more important than ever for systems to manage supply chains effectively.
"Given the constrained resources for treatment of COVID-19, particularly with regard to mechanical ventilation, simple approaches to stratifying morbidity and mortality risk at time of hospitalization are needed," wrote the authors of the Massachusetts study.
"Electronic health records may facilitate a rapid and efficient investigation of clinical cohorts and may form the basis of efforts by consortia to address COVID-19 at scale," they continued.
For that study, researchers applied data from three community hospitals in eastern Massachusetts to generate models estimating the risk of a severe hospital course – characterized by the need for mechanical ventilation, ICU care or death risk.
In general, the team found that abnormal hematologic measures and diminished renal function were associated with a greater risk of severe hospital course. "Predictions may be most useful during the initial week of hospitalization; a useful next-step study could examine whether rerunning models with additional laboratory studies, or incorporating other biomarkers, can improve longer-term prediction," the authors noted.
In the Duke study, researchers used predictive modeling on EHR data regarding case type, patient demographics, service utilization history, comorbidities and medications.
The team used the modeling to produce a clinical decision support tool to assess the risk of high resource utilization for scheduled cases, identifying those at highest risk to avoid exceeding hospital capacity.
The tool produces predictions of four outcomes: length of stay, ICU length of stay, mechanical ventilator requirement and discharge to a skilled nursing facility.
"To prevent the underestimation of resource needs and the misclassification of high-risk patients as low-risk patients, we set the low-risk threshold to include less than 5% of those with resource needs," the Duke researchers wrote. "Our low-risk categories all had high negative predictive values (approximately 99%), allowing us to safely consider that those designated as low risk are in fact low risk."
The most predictive values were demographic, service utilization and procedural factors. Although clinical information was part of the overall models, it was not among the top predictors.
The researchers implemented the CDS tool on June 17. Since its launch, they've made modifications to speed up the data flow process and amend aspects of the decision rule to better meet user needs.
In both studies, researchers noted the limitations that come from relying on somewhat limited patient data. The eastern Massachusetts study excluded hospital transfers because of potentially biased comorbidity documentation, but such missing data would likely diminish the predictive power of any diagnosis.
The Duke model was constructed and validated using a single hospital system, limiting its external validity. Furthermore, it noted that other contextual factors (such as local COVID-19 positivity rates) must be considered as well.
"This type of tool supports a system for determining whether hospital resources are at risk of being overwhelmed on any given day or week; it is to be used together with the background COVID-19 rate in the community, a current inpatient and ICU census, an understanding of deferred cases, and the medical necessity of any one patient," they wrote.
THE LARGER TREND
As more data has become available about COVID-19 patient needs, researchers have developed several predictive models to help with resource management. Relatively early in the pandemic, the team at analytics vendor Health Catalyst created a capacity planning tool forecasting when hospitals would reach their capacity.
"Essentially, what we're trying to do for you here is to create some time," said John Hansmann, senior vice president for professional services at Health Catalyst, during a HIMSS20 Digital demonstration of the product this spring.
Other systems have turned to vendors such as TeleTracking in response to their COVID-19 patient flood.
ON THE RECORD
"To the extent hospital resources are constrained, the ability to target resources to highest-risk individuals is likely to be valuable, and expansion and refinement of risk models may represent a useful approach to optimizing care," wrote the Massachusetts researchers.