key: cord-0048022-3s0x8yxi authors: Schnake-Mahl, Alina S.; Carty, Marcy G.; Sierra, Gerardo; Ajayi, Toyin title: Identifying Patients with Increased Risk of Severe Covid-19 Complications: Building an Actionable Rules-Based Model for Care Teams date: 2020-05-04 journal: NEJM Catal Innov Care Deliv DOI: 10.1056/cat.20.0116 sha: a94c43b98a5ec25d5f2b4c1a79165c4f6692bffd doc_id: 48022 cord_uid: 3s0x8yxi The team at Cityblock Health is building, expanding, and regularly updating its rules-based, adaptable model to identify Covid-19 patients at highest risk. Recognizing the importance of a coordinated response and shared learnings, they wanted to produce an open-source tool to help other providers and health care organizations identify their patients at highest risk of hospitalization, ICU use, and death from the coronavirus pandemic. Other studies have since employed these categories and compared patient characteristics by category.3 , 5 , 11 Our rules-based model aggregates evidence from across studies and uses simple rules to identify which characteristics are likely to put a member in one of these categoriesif they contract the disease. We identified conditions and factors that put members at high risk (those most likely to experiencecritical disease as defined by WHO), moderate risk (those most likely to experiencesevere disease) and low risk (those not at elevated risk for critical or severe disease). See Table 1 for a full list of the conditions and factors associated with each risk level. Our approach is intentionally simplistic. We avoid any risk scoring beyond the high, moderate, and low categories because the existing data are very preliminary and based on small sample sizes, as well as observational studies that cannot imply causality. We believe the risk factors we have incorporated into the algorithm flag members with increased risk of complications as a consequence of contracting Covid-19. The model likely overidentifies those at highest risk. Without more detailed evidence, we erred on the side of overidentification of highest risk cases for priority, person-level engagement, and further risk assessment. We are actively tracking and providing care, where appropriate, to our lab-confirmed positive, likely Covid-positive based on symptoms, and asymptomatic but exposed members. Over the coming months we will use a combination of our tracker, and encounter data from claims, to validate the model in our population. We expect that higher percentages of our population will experience critical and severe Covid-19 than demonstrated so far in the general population. We derived the model criteria from available studies and government reports, preprint and published, between January 1, 2020, and April 1, 2020, which we gathered by searching PubMedand medRxiv (preprint repository) and reviewing all titles/abstracts that included epidemiologic profiles and comorbidities of Covid-positive individuals. For non-peer reviewed preprint articles we performed our own peer review before including information from these articles in the model. Using the selected articles, we compared the characteristics of patients who died, went to the ICU, were hospitalized, or had moderate symptoms. A large proportion of our members also face unstable housing, uncertain food access, or lack of social support. They utilize the emergency department, inpatient admissions, and skilled nursing facilities at higher rates than the general population." After finalizing the relevant conditions and risk factors (e.g., age, homelessness), our clinical and data teams partnered to create the relevant logic and apply it to the data we have on our members. We used a combination of relevant Hierarchical Condition Categories (HCCs), ICD-10 codes, CPT codes, National Drug Code (NDC) categories, and place of service (POS) codes to best identify conditions in our data. We include both the criteria (e.g., diagnosis or characteristic) and the logic used to identify individuals, shown in Table 2 . The model requires medical claims or electronic health record data, which capture comorbidities and medical encounter history, but will be more comprehensive if both data sources are included. We developed three models between March 12 and April 1, based on the evidence available at the time. Our original model, developed on March 12, included individuals over age 50 in the high-and moderate-risk categories, as available studies were primarily based on data sets from China and Italy, where the majority of the hospitalized populations were older than 50.3 , 9 , 25 Serious illness requiring hospitalization or intensive care and deaths have since been reported among younger individuals in the United States.1 , 2 , 8 Most of the existing reports and studies including younger populations either are missing data on comorbidities or only assess each risk factor independently, so we cannot determine interaction effects between age and comorbidities. Recognizing these limitations, we interpreted the evidence to suggest that, on average, younger individuals with chronic conditions remain at proportionally lower risk of advanced illness from Covid-19 than older individuals with the same conditions. " Hypertension ICD10: H35031, H35032, H35033, H35039, I10, I110, I119, I120, I129, I130, I1310, I1311, I132, I150, I151, I152, I158, I159 We encourage other social determinants-oriented providers to similarly adopt simplified tools to maximize understanding of the highest-risk variables during the Covid-19 pandemic, and offer our tool as an option for use." In our second model, we removed smoking as an independent risk factor26 and added end-stage liver disease,20 decreasing the number of members in the high-or moderate-risk categories by 2.2 percentage points. For our third model update, we opened the age range to members 18 or older for those with uncontrolled asthma, severe disability indicators, and a small number of immunocompromised and autoimmune conditions. These additions were based on validation from our clinicians' insight, and supported by recommendations from specialty associations and recent news reports suggesting high infection and case fatality rates in group homes for disabled populations.27 Given the low incidence of these conditions in the general population, few studies have assessed these diagnoses as independent risks for severe Covid-19 complications among the small number of Covid-positive individuals with these conditions. 4 We also note limited examination in the literature thus far on social factors, other than homelessness and living in a group home,8 which may elevate risk of adverse outcomes. We will continue to monitor the literature and update the model as new information becomes available. We developed the model to ensure our highest-risk members and their loved ones are comprehensively educated and monitored by our care teams. Across our markets, we identified a large percentage of our population as high or moderate risk, which is not surprising given that we care for our partner health plans' highest-cost and highest-risk members. We anticipate that other organizations with less complex patients will identify a smaller percentage of their population as high risk for Covid-19 complications. We provided our care teams with a list of all high-and moderate-risk members immediately following the first model output, and began telephonic, SMS, and video-based outreach. Additionally, we incorporated the risk score into Commons, our custom-built care facilitation platform, and added the risk level on each member's profile to help direct member care. During outreach, we ask members about any challenges to staying safe and healthy while sheltering in place. We conduct an assessment including a standard Covid-19 symptom and exposure screening, as well as using a social needs tool (Figure 1 ) to identify potential social risk factors caused or exacerbated by the current Covid-19 societal ramifications (e.g., job loss, inability to access food or medication, home care discontinuation). " We've built dashboards to track assessment completion and responses, and we use assessment answers to help target clinical and social services, resources, and Covid-19 education to members and their caregivers, first prioritizing those high-and moderate-risk members and then conducting the full assessment with low-risk members. Positive responses to symptom-related questions trigger clinical evaluation. We are seeing a substantial increase in members unable to pay for or access food and necessities such as medications, medical and cleaning supplies, and diapers." Notably, this Covid tool is a meaningful departure from Cityblock's usual member screening and assessment approach, which provides for substantially deeper understanding of our members' needs through nuanced branching logic and comprehensive assessments. We elected for a simple, Covid-focused tool to ensure consistency of use and to maximize volume of member assessments completed in a short period of time. We encourage other social determinants-oriented providers to similarly adopt simplified tools to maximize understanding of the highest-risk variables during the Covid-19 pandemic, and offer our tool as an option for use. At a population level, changing trends in aggregate member needs (food, medication access, etc.) are reviewed by practice leadership to ensure appropriate focus on specific strategies. Within 2 weeks, we connected with 67% of the high-and moderate-risk members in our active population across all markets. Consistent with our model, we regularly reach out to the remainder of our members and do a full Covid assessment when we make contact with them. For members with a positive Covid test, positive Covid symptom screen, or known exposure, we have built a Covid-19 triage and escalation tool to track and follow up with members. Covidpositive or likely positive members receive clear and tailored instructions about red flag signs and symptoms to watch out for, and daily telemedicine follow-ups from our clinical team until symptom resolution. To support telemedicine visits, some members are provided with remote monitoring tools such as pulse oximeters and thermometers. We are also using Cityblock palliative care clinicians if members want to discuss and document goals of care and plan supportive care in place, as well as urgent in-home visits if symptoms worsen, behavioral health virtual visits if urgent behavioral health needs are identified, and emergency medical services if clinically indicated. We are seeing a substantial increase in members unable to pay for or access food and necessities such as medications, medical and cleaning supplies, and diapers. To address these barriers and help members stay safe and in their homes, particularly the high-and moderate-Covid-19 risk populations, we've developed targeted programs to deliver food, and have connected with courier services and local mutual aid providers to deliver other necessary goods. We anticipate continued social challenges for our members, even as the case rate begins to flatten. To measure impacts on member outcomes, we are also conducting evaluations of our overall Covid-19 response and specific programs such as palliative care. Below we highlight key organizational needs that ensure model outputs are actionable for care teams and members: Clear messaging about the model output. Care teams were initially confused about the purpose of the risk model. We recommend clearly communicating that the model isNOT " intended to identify those at highest risk ofcontracting Covid-19, but instead those at highest risk of complicationsif they contract the disease. We also recommend emphasizing that it is possible for someone who falls into thelow-risk category to experience serious morbidity/mortality if they contract Covid-19, but we believe theirrisk of these outcomes is materially lower. Use the model for prioritization of proactive outreach. As our model is based on constantly changing evidence, it is possible we miss some members or flag others as high-risk incorrectly.We strongly recommend using the model to inform which members are prioritized for outreach, education, and monitoring, butNOT to determine who does and does not receive care. We recommend that all members across risk categories be engaged and assessed, especially those in lower-income communities where social needs are significantly exacerbated by the Covid-19 pandemic. We The risk model can help other provider groups and hospital systems proactively identify their patients at the highest risk of serious morbidity or mortality from Covid-19. The model is simple to understand. It can help providers prioritize proactive outreach and monitoring to keep patients safe and at home, and allow for rapid care escalation, including in-home care if needed. The model may be helpful for economic projections and ensuring equitable resource distribution based on estimated needs. For hospital-based systems and providers, the model might be useful to help forecast hospitals' needs and potential hospital bed shortages; insurers might use the model to " estimate likely hospitalization and ICU use among their membership. Insurers and providers with regional or citywide coverage can employ the model to estimate community-specific incidence of likely hospitalizations based on the underlying comorbidity burden of their populations and infection rates, and they can equitably distribute care and resourcing according to likely need. If combined with demographic data such as race/ethnicity and insurance type, the model can help researchers and policy makers identify socioeconomic and racial inequities in the risk of adverse outcomes for the Covid-positive population. The model will be updated in the coming weeks as the global medical and public health communities learn more about the disease and disease risk factors, and we hope our risk tool can contribute to this global knowledge base. We strongly encourage other organizations, even those who typically commercialize risk-identification and analytic tools, to do the same. We will make efforts to update the code set and NDC codes as our shared learning continues to evolve (Appendix). If users of our model find value and adapt it further, we ask that they share that information with the corresponding author (ASM) so that we may keep running notes of additional use cases and insights. Despite the evolving evidence, the current model has given us crucial information to target our highest-risk members and support their medical and social needs during a period of escalating vulnerability. New York City Department of Health and Mental Hygiene. City of New York Severe outcomes among patients with coronavirus disease 2019 (COVID-19) -United States Clinical characteristics of coronavirus disease 2019 in China COVID-19 Response Team. Preliminary estimates of the prevalence of selected underlying health conditions among patients with coronavirus disease 2019 -United States Prevalence of comorbidities in the novel Wuhan coronavirus (COVID-19) infection: a systematic review and meta-analysis Int J Infect Dis Epidemiological and clinical characteristics of 99 cases of 2019 novel coronavirus pneumonia in Wuhan, China: a descriptive study Clinical characteristics of 138 hospitalized patients with 2019 novel coronavirus-infected pneumonia in Wuhan, China COVID-19): People Who Are at Higher Risk for Severe Illness. Washington: U.S. Department of Health and Human Services Report of the WHO-China Joint Mission on Coronavirus Disease 2019 (COVID-19) China Concealed Extent of Virus Outbreak, U.S. Intelligence Says. Bloomberg Characteristics of and important lessons from the coronavirus disease 2019 (COVID-19) outbreak in China: Summary of a report of 72 314 cases from the Chinese Center for Disease Control and Prevention Guidance on Shielding and Protecting People Defined on Medical Grounds as Extremely Vulnerable from COVID-19 Association of cardiac injury with mortality in hospitalized patients with COVID-19 in Wuhan, China Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study Association of coronavirus disease 2019 (COVID-19) with myocardial injury and mortality Epidemiology of Covid-19 in a long-term care facility in King County, Washington Risk factors associated with acute respiratory distress syndrome and death in patients with coronavirus disease 2019 pneumonia in Wuhan, China Coronavirus disease 2019: implications of emerging infections for transplantation Clinical features and progression of acute respiratory distress syndrome in coronavirus disease 2019 UPDATED: What People With Liver Disease Need to Know About the New Coronavirus Intensive Care National Audit and Research Centre. ICNARC report on COVID-19 in critical care Association of British Neurologists Guidance on COVID-19 for People with Neurological Conditions. Their Doctors and Carers. Association of British Neurologists Rome: Instituto Superiore Di Sanita Kidney disease is associated with in-hospital death of patients with COVID-19 COVID-19 and Italy: what next? Active smoking is not associated with severity of coronavirus disease 2019 (COVID-19) It's Hit Our Front Door': Homes for the Disabled See a Surge of Covid-19 NDC Code Keyword Search.