We extracted EHR data from the University of Alabama at Birmingham Health System, a multispecialty group practice. The study population included patients with cancer who (1) received medical services between January 2018 and November 2023, (2) were aged 18 to 89 years, (3) were considered potential intervention participants (ie, returned home), and (4) resided in predefined communities (per the study protocol) within the Birmingham metropolitan area of Alabama that were planned locations for LEAP study adaptation and implementation (Figure 1).

Because the health system’s data extraction is operated by zip code, patients were initially identified using zip codes overlapping with target persistent poverty census tracts; as a result, the extracted population included residents of both persistent and nonpersistent poverty areas. Patient addresses were then geocoded to census tracts using institutional geocoding services. Persistent poverty status was determined by linking census tract codes to US Census Bureau persistent poverty designations. The Social Vulnerability Index (SVI) was linked at the census tract level using Centers for Disease Control and Prevention and Agency for Toxic Substances and Disease Registry SVI data. Consistent with the National Cancer Institute definition, survivors of cancer were defined as individuals with a documented cancer diagnosis from the time of diagnosis onward regardless of current treatment status. This definition was operationalized by identifying patients with cancer-related International Classification of Diseases, Ninth Revision; International Classification of Diseases, Tenth Revision; or Systematized Nomenclature of Medicine–Clinical Terms diagnostic codes in their EHR.

EHR data provided clinical information, including cancer type, body weight status, and comorbid conditions, as well as demographic characteristics (eg, age and race or ethnicity) and SDOH (eg, insurance and marital status). Census tract codes were used to identify persistent poverty areas among those living in low-income zip code geographic areas and were used to link the SVI, a composite neighborhood-level index assessing social vulnerabilities [11]. Additional SDOH data were available through the Protocol for Responding to and Assessing Patients’ Assets, Risks, and Experiences (PRAPARE) [12], a patient-reported social risk screener integrated into the EHR system at our institution in the Social Work unit in 2020 [13]. PRAPARE includes 21 questions across 4 domains: personal characteristics, family and home, money and resources, and social and emotional health [12].

Social history notes collected via social work visits were organized in the EHR using Social Interventions Research and Evaluation Network (SIREN) [14] nomenclature. These data were analyzed using NLP to identify key SDOH themes aligned with the Centers for Medicare and Medicaid Services [15]. We applied a rule-based NLP approach to identify events related to unmet needs from both structured questionnaire items and free-text patient responses recorded in the EHR. Text data were first preprocessed through tokenization and removal of stop words, punctuation, and noninformative characters, followed by normalization (eg, lowercasing and stemming) to standardize text across sources [16,17]. This preprocessing was applied consistently to both questionnaire prompts and patient responses.

We then conducted keyword-based text matching to identify SDOH-related events across 5 predefined domains: food insecurity, housing instability, transportation needs, interpersonal safety, and utility difficulties. The domain-specific keywords and phrases used to extract relevant events and identify unmet needs or negative responses are detailed in Table S1 in Multimedia Appendix 1. For each SDOH domain, patients were classified as experiencing unmet needs or insecurity if their most recent response indicated an adverse condition. Patients without such indicators were classified as having no identified unmet need (ie, secure) for that domain.

We summarized findings in tables for community engagement processes (eg, informing co-design and focus group guides) and intervention adaptation (eg, educational content topics, potential delivery channels, and anticipated challenges). Although hypothesis testing was not our primary focus, 2-tailed t tests, and chi-square or Fisher exact tests were used to estimate differences between survivors living in persistent poverty areas and those not living in these areas to examine trends in the data. Distributions of continuous variables (age and BMI) were approximately symmetric, supporting the use of t tests for group comparisons. Given the descriptive nature of this study and the absence of a priori hypotheses, P values are presented to identify potential trends rather than to establish statistical significance. No adjustment for multiple comparisons was applied, and the results should be interpreted as exploratory. All analyses were conducted using R (version 4.3.2; R Foundation for Statistical Computing).

This study was approved by the Institutional Review Board at UAB (approval number IRB-300010921), with informed consent waived. This study was conducted in accordance with the study protocol. Only patients who met the predefined inclusion criteria were included. To protect participant privacy, we avoided reporting overly granular or potentially identifiable information in the results, including zip codes and census tracts. No compensation was provided to the participants.

This study sought to characterize demographic, clinical, and SDOH factors among survivors of cancer in persistent poverty areas using EHR data to inform community-engaged adaptation of lifestyle modification interventions. We found that survivors of cancer in persistent poverty areas within our health system catchment were predominantly Black individuals and older adults with Medicare coverage and experienced high rates of obesity and chronic conditions, including hypertension and diabetes. Among those with social risk screening data, notable proportions reported transportation barriers, health care needs, food instability, fewer social interactions, and stress, consistent with the findings of a previous study [18].

Our data also highlight the burden of social risks in persistent poverty areas, consistent with previous research [2,8,19]. Health disparities among survivors of cancer are driven by a complex interplay of SDOH [20], which also serve as primary barriers to adequate health care access and physician communication [21-24]. Understanding these needs—as we did in this study—in planning tailored interventions will be imperative for sustainability and improved outcomes.

Survivors of cancer in persistent poverty areas often face multiple challenges of cancer survivorship, poverty, and clusters of problems that collectively affect their health. Adapting interventions to address the unique barriers that these populations face is important. In this study, we aggregated data on SDOH to guide the discussions with the planning, implementation, and evaluation working groups (community stakeholders and survivors and cosurvivors of cancer) convened as part of the ongoing LEAP study community engagement processes. This approach aligns with evidence-based cancer control strategies in persistent poverty areas [8]. Our findings support a comprehensive approach to promoting healthy eating and physical activity that addresses clusters of social risks rather than focusing on a single factor (eg, telehealth options to minimize transportation barriers, meal programs or home gardening to address food insecurity, and peer support groups to reduce social isolation).

The growing integration of SDOH screening into EHR systems represents a paradigm shift in how health care systems can support cancer control efforts in underserved communities [25]. Our approach aligns with emerging recommendations from national organizations emphasizing the importance of systematic SDOH data collection and documentation to inform care delivery and population health management [22]. The scoping review by Li et al [26] demonstrated that EHR-embedded SDOH data can be effectively linked to health outcomes and used to design targeted interventions, although standardization of data collection methods remains an ongoing challenge. Future efforts should focus on expanding SDOH screening beyond social work encounters to capture a more representative picture of social risks among survivors of cancer.

Community-based participatory research approaches have demonstrated effectiveness in optimizing health behaviors among survivors of cancer, particularly when interventions are tailored to local community resources and cultural contexts [3]. A recent systematic review found that most community-based interventions for survivors of cancer focused on physical activity and nutrition, with 88% showing improvement in at least one outcome measure; however, the review also highlighted a concerning lack of diversity in participant populations [27]. Our study provides foundational data to inform such community-based participatory research efforts by characterizing the specific demographic, clinical, and social risk profiles of survivors of cancer in our target communities—information essential for adopting evidence-based interventions to address local needs. The Active Living After Cancer program, for example, demonstrated that community-delivered physical activity interventions can successfully improve outcomes among medically underserved survivors of cancer and their caregivers when implemented through trusted community partnerships [28].

Notably, more than half (282/509, 55.4%) of the survivors of cancer in persistent poverty areas in our sample had Medicare coverage, suggesting opportunities to leverage existing Medicare benefits for lifestyle interventions. While Medicare currently covers medical nutrition therapy only for diabetes and chronic kidney disease, recent legislative efforts such as the Medical Nutrition Therapy Act for dietitian-delivered nutrition counseling will be important for this population [29]. In the meantime, health care systems serving persistent poverty areas can explore partnerships with community-based programs and leverage existing Medicare preventive services to address modifiable health behaviors among survivors of cancer.

The strengths of this study include using real-world health system data and NLP to capture both medical and nonmedical factors influencing health outcomes, enabling a more informed approach to addressing disparities prior to intervention development. Furthermore, the aggregated SDOH data were “translated” into a format for nonscience community-based partners to inform the ongoing community-engaged “co-design” of the planned LEAP intervention.

Limitations included the bias in the study sample of survivors of cancer, which was limited to the project target areas defined by low socioeconomic status indicators, and the PRAPARE- and NLP-derived data were also limited to individuals with the most acute social needs (eg, referral to social work). For this reason, we did not conduct formal hypothesis testing and did not focus on those results for interpretation in this report, and the findings may not be generalizable to other medical practice catchments. Our sample reflects survivors of cancer actively engaged in care at our health system, which may underrepresent patients with rapidly fatal cancers and those who have disengaged from care. This “survivor bias” should be considered when interpreting cancer type distributions; however, for the purpose of informing intervention adaptation, characterizing the population currently available for enrollment is appropriate.

Missing data were present for several variables, particularly PRAPARE (2094/2672, 78.4% missing) and NLP-derived SDOH (1076/2672, 40.3% missing) data, which reflects the clinical reality that these data are collected only for patients referred to social work services. This represents a form of data missing not at random, as patients with more acute social needs are more likely to have social work encounters. This limitation should be considered when interpreting prevalence estimates, which likely represent lower bounds for the broader population.

Despite these limitations, this is the first study to use an EHR-embedded social risk screener to provide insightful information for formative work among survivors of cancer in persistent poverty areas, offering valuable insights into their unique challenges and underscoring the importance of adapting interventions to address their specific social and health needs.

This study demonstrates that leveraging SDOH data embedded in health system EHRs provides a feasible approach for characterizing populations of survivors of cancer in persistent poverty areas to inform community-based intervention adaptation. The social risk profiles identified highlight the importance of tailoring lifestyle interventions to address transportation, food access, and safety concerns alongside chronic disease management.