Cluster analysis of COVID-19 recovery center patients at a clinic in Boston, MA 2021–2022: impact on strategies for access and personalized care

COVID-19 disproportionately impacts historically disadvantaged communities of color and patients from socioeconomically under-resourced backgrounds [1]. Individuals with non-White race/ethnicity suffer from more COVID-19 infections and higher COVID-related mortality rates [2‐4]. Factors associated with increased risk of COVID infections among racial/ethnic minorities are multifactorial social determinants of health and include underlying comorbidities, occupations that require frequent interactions with the public, housing conditions with limited opportunity for social distancing [2], disparities in access to quality healthcare and historical medical racism resulting in lack of trust in healthcare institutions [5]. These inequities are in part a function of differential resource investment and structural racism resulting in various vulnerabilities [6]. Prior to COVID-19, non-White patients had less reliable access to specialty care [7] and to long-term acute care facilities (mediated by differences in insurance coverage) [8]. These disparities have extended to COVID-19 care. For example, recent evidence suggests that there may be disparities in referral to physical/occupational therapy during an inpatient stay [9], and that living in a neighborhood with greater social vulnerability is associated with organ dysfunction/failure and need for mechanical ventilation [10].

Many post-COVID clinics and multidisciplinary recovery centers have been opened to meet the demand of patients recovering from COVID-19. However, without targeted interventions, post-COVID clinics may perpetuate existing healthcare inequities [11]. Some COVID Recovery clinics have recognized the impact of social determinants of health on recovery and made investments in ensuring equity [12]. Other groups have described recovery processes focused on resilience and rebuilding for communities and health systems, not only for individuals [13, 14]. Indeed, there has been growing awareness of the possibility of disparities in post-COVID care. Despite this raised awareness, little disparities-related quantitative information exists regarding post-COVID symptoms, referrals to recovery centers, and utilization of resources.

We created a COVID Recovery Center (CRC) that not only provides streamlined multi-disciplinary care to post-COVID patients, but also incorporates strategies to address inequalities by facilitating access to socioeconomic status (SES)-targeted interventions. We also gather quantitative data on the effectiveness of these measures. Machine learning algorithms offer an opportunity to infer subpopulations from complex SES data and help to identify high SES-risk subgroups. We hypothesized that clustering using SES risk factors would identify patient subpopulations within the CRC who are more likely to have more symptoms and worse clinical outcomes, and who are less likely to utilize CRC resources. This quantitative approach supports targeted interventions for patients most in-need of post-COVID care resources.

In this study of over 1,200 post-COVID patients recruited from a newly designed ambulatory COVID Recovery Center (CRC), we used four easily obtainable clinical variables (sex, self-reported race, language, insurance type) to identify three patient clusters with differences in the need for ICU admission, symptom reporting, and resource utilization. We found two clusters of predominantly non-White patients who were more likely to experience an ICU admission and report persistent symptoms, but one cluster who were less likely to utilize resources even after a CRC referral. These results highlight the need to improve not only access to multi-disciplinary specialty care, but also to address barriers to access within post-COVID care clinics.

We observed that the majority of CRC patients are English-speaking White women under 50 years of age with managed care, which highlights the need to improve CRC referral to expand inclusion of patients who demographically had higher rates of infection. Further, we identified one subset of patients (Cluster 1) who were primarily non-English-speaking, had government insurance, and a higher odds of ICU admission and symptoms, but who were less likely to access resources after CRC referral. We also identified a subset of primarily English-speaking and Black/African American patients (Cluster 3) with commercial insurance who had an increased odds of ICU admission but had similar resource utilization compared to our reference cluster. Taken together, these results suggest that 1) we need to improve referral volume of groups with a disproportionate burden of COVID infection, and 2) certain minority patients are not able to fully utilize the offered resources.

Our results demonstrating increased ICU admission in minority-predominant clusters are consistent with prior literature showing that minority populations share the disproportionate burden of severe disease [21]. A prior study of multihospital hospitalized patients in Michigan, combining clinical data with social vulnerability indices (SVI), found that patients from high SVI areas (who are more likely to be Black/African American or Hispanic) were more likely to have an ICU admission [10] and have acute organ dysfunction and failure. Both Cluster 1 and 3 patients experience greater ICU admissions and persistent symptoms. These associations may be explained by the fact that pulmonary sequala symptoms appear to be closely associated with acute infection severity [22], although the relationship may be nonlinear [23]. Understanding pulmonary impairments among racial/ethnic groups is also an area of ongoing research, with some data suggesting that Black adult patients hospitalized with COVID-19, at 6 months follow-up had lower percent predicted DLCO compared to Hispanic and White patients, after controlling for various predictors such as smoking status, ICU admission and history of chronic lung disease [24]. These data further support the need to purposefully engage and include minority patients in post-COVID evaluation and COVID recovery care.

We observed that patients referred to the CRC who were primarily non-English speaking individuals were less likely to utilize the offered resources, which suggests that post-COVID centers must be cognizant of within-center barriers to care. Recent data suggests that patients recovering from COVID-19 have increased risk of new onset mental health disorders, including anxiety [25]. It will be imperative to ensure that patients are comfortable reporting new mental health changes and to encourage resource utilization. The decreased utilization of psychoeducation and support groups is also notable because patients may experience mental health symptoms (Cluster 1 for example had increased odds of reporting anxiety) but be unlikely to seek specific support for a variety of reasons. The reason for this observation is unclear but may be attributable to cost of work-up and management/coverage concerns with reporting symptoms, language barriers, a paucity of Spanish-speaking healthcare providers, mistrust of social support systems, and stigma against psychoeducation. Other data during the pandemic also highlight the impact of broader societal events [26] on mental health. Improving access to interpreter services for healthcare providers, care coordination managers, social workers, and all other employees at the CRC and other similar centers will be critical to providing equitable care.

Despite an increase in ICU admissions and pulmonary symptoms, Cluster 3 patients reported similar non-pulmonary symptoms and appear effective at accessing support services, which may allude to multiple SES-related care barriers. There was a trend toward increased utilization of services for navigating government benefits; attempts to access support services suggests that these predominantly Black/African American and Latino/a patients may be living in areas with high social vulnerability index [10] and thus have a need for these specific resources. Whether Cluster 3 patients under-reported or did not experience non-pulmonary post-COVID-19 symptoms (e.g. anxiety) is unclear. Symptom underreporting amongst minority communities is well-described in oncological literature and is attributed to a variety of individual and societal factors [27]. Additionally, patients may not report symptoms during initial intake which is frequently remote (via electronic health record or telephone), therefore this divide may modify both report and capture of this information. This barrier to care may be present across many post-COVID care centers. The optimal strategies to mitigate underreporting are unclear, but include standardized approaches to symptom monitoring, actively offering resources and providing extra assistance in accessing resources, recruitment and retention of healthcare providers from minority communities, and building long-term trustworthy partnerships within the community. Lastly, more granular documentation of request and provision of services would be helpful to better monitor needs. Ensuring minority patients have community mental health resources will be important to support their recovery. Reducing housing insecurity and supporting affordable housing in the context of COVID-19 and beyond also presents an opportunity for focused discussion and policies. Further investigation into the socioeconomic factors affecting this patient subgroup is needed.

Strengths of this study include a relatively large, well-characterized sample of post-COVID patients in whom data were collected prospectively for research purposes and the subsequent application of a clustering algorithm to identify patient subgroups. To our knowledge, this is the first study of outpatient COVID recovery care to apply quantitative machine-learning methods to understand differential resource utilization and symptom reporting within a recovery center. Such a clustering method can be used to categorize individuals into high-risk groups in other populations. As more data analytics emerge to identify patients who may have post-COVID symptoms [28] for a variety of clinical and research purposes, ensuring equity will be of high importance. Limitations of this study include that, by design, it is a single cohort study of an ongoing project. Replication in other centers is needed as well as the development and validation of SES-risk prediction tools. We were able to identify specific care barriers (i.e... access to interpreter services and within center resource utilization), but we have yet to measure the effects of interventions targeting these factors. Future studies can address these issues.

In conclusion, we used a clustering algorithm to define patient subpopulations using sex, race, language, and insurance status in a post-COVID clinic, and demonstrated a high-risk subgroup who had more ICU admissions and more symptoms, but less resource utilization. Overall, these data allow us to consider patient-centered approaches for individual patients and subpopulations of patients with shared needs. Future studies will focus on targeted approaches to promote resource utilization which may ultimately improve individual outcomes and provide a framework to improve the equity of care across the hospital system.