Data-Focused Strategies to Prevent COVID-19 Related Hospitalizations

With the risk of COVID-19 overwhelming hospital systems, many healthcare organizations are considering how best to use their existing resources (primary care providers, advanced practitioners, nurses, care managers, etc.) to help prevent COVID-19 related hospital admissions. This is an optimization challenge that benefits from using data and technology to steer the right resources to the right people.

Having worked with several organizations on this topic, we thought it might be useful to share our considerations and recommendations.

This document focuses first on how to use data to prioritize patient outreach. In the course of working with our customers, several related strategies were introduced that have the potential to reduce unnecessary admissions. They are presented in the Additional Strategies section.

Patient Outreach Prioritization Strategies

The strategies below are listed in order of suggested priority, moving from the most urgent and easiest to identify down to larger populations that require more creative use of data to organize.

1. Prioritize outreach to patients with suspected, presumed, or confirmed COVID-19 infections.

Follow up with presumed or positive COVID-19 cases to assess which can recover safely at home versus which may require hospitalization. Organizations should track this list closely to ensure trained clinicians follow up routinely to mitigate community spread.

2. If possible, create a list of patients in nursing homes or long-term care facilities.

This information will be available to payers, and may be available to healthcare providers that have financial risk-sharing arrangements with payers (e.g., Medicare Advantage plans, NextGen ACOs) or organizations that represent both the payer and provider (e.g., duals plans, PACE programs). Organizations without access to such data can proactively reach out to local facilities to assess the needs of patients and deploy resources accordingly.

3. Prioritize patients whose needs were previously met with scheduled clinical interactions.

Among patients with chronic, complex diseases are a subpopulation whose health is dependent on routine interaction with the healthcare system. These individuals can be identified in data using queries on usage of durable medical equipment, care management interactions, specific medications, assignment to home care agencies, meal and / or medication delivery, transportation, routine in-person medical services, etc.

Not coincidentally, they are likely to have several conditions that put them at risk for hospitalization if infected. Most critically, their existing medical needs, if unmet, increase their odds of exposure. As with the previous tiers, this population is likely to be small enough and with enough acute medical need to warrant proactive outreach by clinicians (nurses, PCPs, etc.)

4. For the remaining majority, do not over-invest in precision.

The majority of patients over the age of 65 will have one or more comorbidity and therefore fall into this high-risk category. While there are known risk factors that one can sort by, and we provide recommendations accordingly, the number at risk is so high that we recommend teams complement prioritization strategies with ways to get in touch with as many people in this group as possible. In other words, while prediction is helpful for finding needles in the haystack, in this population there may be more needles than hay. Expectations and approaches should be planned accordingly.

5. Focus on simple queries and common sense, rather than machine learning algorithms and risk scores.

We encourage teams to recognize that the risk of hospitalization due to COVID-19 is not the same as the risk of hospitalization, as traditionally treated by risk scores. Exposure is caused by a variety of patient needs and contexts. Therefore, consider more than just clinical need in any data strategy.

Research is emerging on the topic of using machine learning to predict hospitalization and mortality from COVID-19 infection. While we will see significant progress in this area in the future, there are not yet adequate training sets available for building and evaluating models. As a result, most of the emerging machine learning/COVID-19 research is either biased or does not present reliable results.

Traditional risk scores will prioritize patients that have some of the conditions recognized by the CDC as being correlated with severe disease due to COVID-19, but users should be cautious of their shortcomings. Most rely heavily on prior cost and utilization. Research has shown that this introduces racial biases for outreach toward white, often affluent populations on whom more healthcare dollars are spent. Critically, this may overlook minority populations and those living in more dense, socioeconomically challenged areas, who may be more likely to spread and suffer from severe disease.

For these reasons and the high risk posed to everyone 65 or older, we suggest writing simple queries that take into consideration age and the following factors:

CDC-published drivers of severe COVID-19 disease (for those age 65+). The conditions identified thus far by the CDC with a higher risk of severe illness from COVID-19 that can be identified via ICD-10 or problem list query include chronic lung disease, heart disease, liver disease, chronic kidney disease, diabetes, HIV/AIDS, asthma, and cancer.

Obesity (BMI over 40) and history of smoking are correlates of severe disease that may be identified within the electronic medical record. Though again, with so many at risk, the benefit of prioritizing by EMR-based factors may be outweighed by the cost depending on how each institution’s data is structured and how accessible it is.

Procedures. Examples of procedures that can be identified via common procedural terminology (CPT) codes that are correlated with severe disease from COVID-19 include dialysis, bone marrow or organ transplantation.

Zip code. Population density and lower socioeconomic status are correlated with disease spread and severity. Certain zip codes can be prioritized accordingly.

Again, there are more sophisticated algorithms and additional details that can be extracted from other sources, such as electronic medical records. However, organizations should keep the cost versus benefit of each source and approach in mind in light of the high risk posed to everyone over the age of 65.

Additional Strategies

In the course of working with customers to help define strategies for prioritization, several complementary approaches were discussed that have the potential to prevent unnecessary hospitalizations. They are provided below.

Inbound inquiries

One way to reduce unnecessary exposure is to ensure that inbound inquiries, whether via a visit to the organization’s web page, a call to a PCP or call center, or a message in an organization’s online patient portal, are properly handled. The ability to identify and address patient needs via inbound traffic will be instrumental and teams should plan triage strategies accordingly. This means going beyond questions related to screening for COVID-19 infection to include screening for patient needs that can lead to exposure.

Prepare to discover and address more than clinical needs

Health system representatives interacting with patients should be trained to discover and address needs other than clinical ones and route patients to appropriate services whenever possible. One way to discover and triage needs is for all health system representatives, whether PCPs or call center reps, to ask a standard set of questions.

For example:

Do you have any concerns about your medication(s)?

Do you have a way to receive fresh food without leaving your home?

Do you live with anyone who is unable to follow social distancing recommendations?

Do you find yourself experiencing stretches of sadness, nervousness, or fear that concern you?

Do you need help with daily activities (bathing, dressing, eating)?

Would you like to be contacted by a clinician?

To reach as many people as possible and triage to the appropriate levels of service, organizations should consider high volume outbound channels, such as text and email.

Answers to these questions can be routed to the appropriate level of support

Support may be most efficiently provided by a volunteer, social worker, specialist, care manager, or other appropriate clinician. These various providers can offer specific services, designed to eliminate patient need and reduce risk of exposure including:

• Testing information
• Remote care
• Rx refills
• Food delivery services
• Caregiver support
• Transportation needs
• Disinfection processes/education
• Loneliness support

Closing Thoughts

We are incredibly grateful to all who put themselves in harm’s way to care for those most in need and appreciate the opportunity to contribute in any way possible. The strategies above attempt to deliver maximum benefit at minimum cost with the recognition that speed is critical. The above strategies are likely to remain relevant at least until a vaccine is available and should be easily repurposed for other population health goals once the threat of COVID-19 has been mitigated.

Please do not hesitate to contact any of us to ask questions or discuss ideas.

Caitlin Brennan, PhD, NP

VP of Clinical Improvement, Cyft Inc.

Visiting Scholar, Boston College

[email protected]

Adin Shniffer, MBA, MSc

Engagement Manager, Cyft Inc.

[email protected]

Leonard D’Avolio, PhD

CEO, Cyft Inc.

Asst. Professor, Harvard Medical School and Brigham and Women’s Hospital

[email protected]