Background
Social support is a complex construct defined as interpersonal relationships that protect people from the deleterious effects of stress, such as a threat to health.1 Studies in community populations have shown that patients’ self-reporting of their social support correlates with all-cause and cardiovascular mortality.2–4 In oncology, data suggest that patients’ social support may play a prognostic role in maintaining health-related quality of life.5 However, data are lacking that examine the relationship between patients’ social support and outcomes such as survival and healthcare utilization.6 Social support may be especially critical for outcomes in patients with aggressive hematologic malignancies, because these patients often require intensive and complex treatment that results in substantial toxicities. In addition, these patients face significant illness burden and high healthcare utilization.7–13 Despite the need for intensive treatment, a significant illness burden, and the suspected importance of social support, the relationships among social support, survival, and healthcare utilization in patients with aggressive hematologic malignancies are poorly understood.
The literature focused on social support in oncology has been hindered by important limitations, including difficulties measuring the extent of social support, the prevalence of ceiling effects in measurements of perceived social support,14 missing data in patient-reported outcomes studies,15,16 and limited sample sizes to assess the impact of social supports on clinical outcomes.6,17 Notably, prior investigations have primarily relied on marital status as a proxy for social support.18 However, social support can emerge from other sources, including family, friends, peers, and community. Thus, novel efforts are needed to better capture information about patients’ social support, such as natural language processing (NLP). NLP includes information extraction methods that rapidly process and analyze written text, and has been used to assess end-of-life quality indicators and functional status documentation.19–21 NLP can be used to rapidly scan vast quantities of electronic health records (EHRs) to detect prespecified indicators,21 and therefore represents a novel method to potentially identify and assess the extent of social support utilizing EHRs.
In the present study, we sought to use NLP to identify and describe the extent of social support for hospitalized patients with aggressive hematologic malignancies. We also examined associations between the extent of social support with overall survival (OS) and healthcare utilization in this population.
Methods
Study Procedures
This study was approved by the Dana-Farber/Harvard Cancer Center Institutional Review Board. The study is a cross-sectional secondary data analysis of a parent study that enrolled 1,580 adult patients with cancer who were hospitalized for an unplanned hospital admission at Massachusetts General Hospital (MGH) in a longitudinal cohort study from September 2014 to April 2017.22–24 We identified and recruited consecutive patients with an unplanned hospital admission (index hospitalization) during the study period by screening the daily inpatient oncology census. Each participant contributed one unique hospitalization. A research assistant obtained written informed consent from eligible patients within 1 to 5 days of the hospitalization. In this cross-sectional analysis, we focused on 251 patients with aggressive hematologic malignancies.
Participants
Participants in this secondary analysis were eligible if they were adults (aged ≥18 years) diagnosed with an aggressive hematologic malignancy and admitted to MGH. Aggressive hematologic malignancy was defined as an aggressive non-Hodgkin lymphoma, Hodgkin lymphoma, acute leukemia, chronic myeloid leukemia, myelodysplastic syndrome, or other myeloproliferative neoplasm. Participants also had to be able to read and speak English well enough to independently complete study questionnaires. Our exclusion criteria were similar to those of the parent study. Specifically, we excluded patients admitted for elective or planned hospitalizations, defined as hospital admissions for chemotherapy, planned surgeries or other elective procedures, chemotherapy desensitization, or bone marrow transplantation.
Study Measures
Sociodemographic and Clinical Factors
We conducted an EHR review to collect demographic information (age, sex, race, education, insurance) and clinical factors (Charlson comorbidity index score,25 time from diagnosis of cancer to index hospitalization, cancer diagnosis). For the Charlson comorbidity index score, the patients’ hematologic malignancy was not included in the score.
Social Support
We used ClinicalRegex NLP software (Lindvall Lab) to search all clinical documentation data from the EHR for social support documentation, similar to prior studies using NLP.19–21 We focused on social support documentation that occurred within 6 months of the date of index hospitalization. Our ontology for identifying social support documentation included a keyword library prioritizing sensitivity over specificity and involved 3 keyword categories: social support, living situation, and caregivers (supplemental eTable 1, available with this article at JNCCN.org.). To identify the keywords and categories, 2 researchers (P.C. Johnson, N.H. Markovitz) manually reviewed 30 medical records to identify documentation of social support. The researchers identified any words or key terms that could be related to social support and organized these keywords into categories as described earlier. The keyword library was then used to enumerate clinical documentation about social support for each patient. We reviewed the clinical documentation enumerated by NLP and assessed social support as being limited or adequate for each patient (supplemental eTable 2). Two independent coders (P.C. Johnson, N.H. Markovitz) validated the interrater reliability of the NLP coding method on a subset of 50 patients, achieving a kappa of 0.94 (95% CI, 0.92–1.00). For a subset of patients (12/50), we could not obtain the clinical documentation file necessary to run the NLP algorithm. For these patients, 2 coders (P.C. Johnson, N.H. Markovitz) used manual review of the electronic documentation to assess the extent of social support as being limited or adequate. These 2 coders discussed disagreement in the manual review process with a consensus panel (P.C. Johnson, N.H. Markovitz, A. El-Jawahri) until reaching consensus, similar to prior studies.26
Outcome Measures
We conducted an EHR review to determine the date of death and date of last follow-up for all patients. We also conducted an EHR review to determine readmission within 90 days of index hospitalization discharge (yes or no) and date of hospital readmission. For healthcare utilization outcomes, we excluded patients who died during the index hospitalization. We created a composite outcome of death and/or readmission within 90 days versus patients who were alive without a readmission within 90 days to account for early mortality, as done in prior studies.23 To further account for mortality given that early death impacts the time at risk for readmission, we used time to first readmission within 90 days of hospital discharge as an outcome measure. This outcome was defined as the number of days from hospital discharge to the first unplanned readmission within 90 days, as done in prior studies.23 Length of stay (LoS) for the index hospitalization was determined for each patient.
Statistical Analysis
We used descriptive statistics to summarize patients’ sociodemographic and clinical characteristics along with healthcare utilization and mortality. To investigate the relationship between social support and OS, we conducted Cox proportional hazards regression analyses, adjusting for the following covariates that were a priori defined based on our review of the literature and prior studies: age, sex, race, education, insurance, comorbidities (Charlson comorbidity index score), time since cancer diagnosis, and cancer diagnosis.22,23,27–33 We did not include marital status as a covariate given that marital status was included in the keyword library to evaluate social support. We used logistic regression models, adjusting for the same covariates described earlier to assess the relationship between social support and the odds of death or readmission within 90 days. In addition, we used Cox proportional hazards regression models, adjusting for the same covariates described earlier to assess the relationship between social support and time to readmission within 90 days. We conducted linear regression models, adjusting for the same covariates to assess the relationship between social support and index hospitalization LoS. All reported P values were 2-sided, with a P value <.05 considered statistically significant. We performed statistical analyses using STATA, version 14.2 (StataCorp LLP).
Results
Study Participants
Table 1 describes the clinical characteristics of the patients (n=251) in this analysis. The median age of the study population was 64 years (range, 19–93 years) and most patients were White (89.6%; n=225), male (68.9%; n=173), and married (65.3%; n=164). The median Charlson comorbidity index score of patients was 0 (range, 0–10). A plurality of patients had leukemia (42.2%), followed by lymphoma (37.9%) and myelodysplastic syndrome/myeloproliferative neoplasm (19.9%). Overall, 8.8% (n=22) of patients had limited social support.
Patient Characteristics
Healthcare Utilization, Symptom Burden, and Survival
Among all patients, 45.0% (n=113) had a hospital readmission within 90 days of discharge from index hospitalization. The median LoS for index hospitalization was 6 days (range, 1–111 days). With a median follow-up of 187 days (range, 0–597 days), 33.9% (n=85) of patients died. Overall, 17.9% (n=45) of patients died either during index admission or within 90 days of discharge from index hospitalization.
Association Between Social Support and OS
In an unadjusted Cox regression, we found that limited social support was not significantly associated with OS (hazard ratio [HR], 1.66; 95% CI, 0.88–3.13; P=.117). In a multivariable Cox regression model, we found that patients with limited social support had lower OS (HR, 2.00; 95% CI, 1.03–3.90; P=.042) (Table 2).
Association of Social Support With Overall Survival
Association Between Social Support and Healthcare Utilization
In unadjusted analyses, limited social support was not associated with death or readmission within 90 days of discharge from index hospitalization (odds ratio, 2.78; 95% CI, 0.98–7.87; P=.054), time to readmission within 90 days (HR, 1.50; 95% CI, 0.82–2.74; P=.183), or index hospitalization LoS (β=1.33; 95% CI, –3.83 to 6.49; P=.612). In multivariable models, limited social support was associated with a higher likelihood of death or readmission within 90 days of discharge (odds ratio, 3.11; 95% CI, 1.04–9.31; P=.043) but not time to readmission within 90 days or with index hospitalization LoS (Tables 3 and 4).
Association of Social Support With Death or Readmission Within 90 Days of Discharge From Index Hospitalization
Association of Social Support With Time to Readmission Within 90 Days
Discussion
In this study, we found that limited social support was associated with worse OS and increased likelihood of death or readmission within 90 days of index hospitalization discharge in hospitalized patients with aggressive hematologic malignancies. We did not find associations of limited social support with time to hospital readmission or hospital LoS. These findings highlight the importance of assessing social support in this population to identify patients at high risk of worse survival.
To our knowledge, this is the first study to show an association of limited social support with worse OS in patients with aggressive hematologic malignancies. Prior research has reported an association between being socially isolated and OS in patients with solid tumors.17 In addition, a meta-analysis found a relationship between perceived social support, social network size, and marital status with OS in patients with cancer.30 However, this meta-analysis included few patients with hematologic malignancies, and social support was primarily measured using marital status, further limiting the scope of the work.30,34–37 Thus, our findings are consistent with prior studies of social support in oncology, yet we highlight novel findings of the impact of social support on survival in patients with aggressive hematologic malignancies. Patients with aggressive hematologic malignancies are at high risk of complications, morbidity, and intensity of healthcare use; therefore, social support may play a particularly critical role in this population.7,9,12,38
The mechanisms underlying the association of limited social support with reduced OS remain unclear. Social support could impact health behaviors, access to the healthcare system, ability to receive aggressive therapies, overall patient psychological well-being, or stress-related biological changes affecting tumor proliferation.17,30,39–41 Future research in larger patient cohorts should explore the mechanism of the association between social support and OS in patients with aggressive hematologic malignancies. Nonetheless, our findings underscore the importance of evaluating social support in patients with aggressive hematologic malignancies and can help clinicians identify a vulnerable group of patients at high risk of poor outcomes. Future studies should examine the role of supportive care interventions focused on improving social support as an innovative strategy to improve cancer outcomes. Although some aspects of social support, such as familial relationships, are not modifiable factors, interventions utilizing peer support, trained healthcare professionals (eg, psychologists, social workers), or care navigators may provide patients with an extra layer of needed support to improve their quality of life and care. For example, online social support interventions have shown feasibility in caregivers of patients with dementia and are currently being evaluated to help with feelings of caregiver competence, perceived social support, quality of life, and psychological symptoms in those populations.42,43 Ultimately, social support interventions tailored to the needs of patients with aggressive hematologic malignancies have great potential to help improve outcomes in this vulnerable population.
Contrary to our hypothesis, we did not detect an association between social support and time to hospital readmission or LoS in this population. In studies of older adults in the general medicine population, social support has been shown to be associated with early hospital readmissions and longer hospital LoS.44 Patients with hematologic malignancies have high healthcare utilization,7,11 and thus we hypothesized that social support would correlate with significantly increased time to hospital readmission and decreased LoS in this population. Plausibly, our study lacked adequate statistical power to fully explore the relationship between social support and time to hospital readmission or LoS among patients with aggressive hematologic malignancies. In addition, patients with aggressive hematologic malignancies represent a heterogeneous population, and future studies examining the impact of social support on healthcare utilization in these groups are warranted.
We show in this study that NLP can be successfully applied to EHR data to identify the extent of social support in patients with cancer. The NLP method in our analysis resulted in high agreement between 2 coders, suggesting excellent reliability. Our findings underscore the utility of NLP to assess complex health constructs and highlight its usefulness as a novel method to examine the extent of social support more comprehensively in patients with cancer. NLP provides the capacity to measure social support encompassing multiple dimensions, beyond factors such as marital status or family support, and is not dependent upon patient reporting of social support, which is vulnerable to ceiling effects.14 The use of NLP to analyze data in the EHR provides a unique opportunity to harness EHR documentation as an innovative tool to better inform clinicians and identify high-risk populations that may benefit from additional support to improve their outcomes. Our findings underscore the need for larger-scale studies evaluating the use of NLP to analyze social support. Healthcare systems could integrate NLP into the EHR as a future application of our work to further examine social support in patients with cancer.
Our study has several limitations worth considering. First, we conducted a secondary analysis of patients at a single large academic center that lacked racial diversity, and thus our findings may not generalize to other populations. Second, rule-based NLP models only detect phrases in notes if they match specified keywords, which means that we could have missed some social support documentation. However, our NLP algorithm prioritized sensitivity to ensure that we captured any documentation regarding the social support construct. Third, we were limited to information about patients’ healthcare utilization that was available in our EHR, and therefore we lacked data about hospital admissions and healthcare utilization at other institutions. Fourth, we could only analyze associations, and thus we cannot assume causation between the relationships observed. Fifth, the extent of EHR documentation regarding social support could potentially reflect the degree of support received from healthcare providers. In addition, we did not have data regarding the particular providers (ie, physicians, nurses, social workers) documenting social support, which can also be an important variable in these analyses. Finally, our sample size limited the number of predictors that we could analyze in our multivariate analyses, and the cell size for limited social support was small; thus, our model may not fully account for all possible confounders and is hypothesis-generating but should be evaluated further in larger studies.
Conclusions
We have shown that limited social support assessed by NLP is associated with important clinical outcomes, including OS and likelihood of death or readmission within 90 days of discharge in hospitalized patients with aggressive hematologic malignancies. These findings support the use of NLP to successfully assess social support in patients with cancer. Identifying patients with limited social support can provide clinicians with important prognostic information regarding patients at higher risk of death, and larger-scale studies should further examine the impact of social support on outcomes in patients with aggressive hematologic malignancies.
References
- 1.↑
Wortman CB. Social support and the cancer patient. Conceptual and methodologic issues. Cancer 1984;53(10 Suppl):2339–2362.
- 2.↑
Uzuki T, Konta T, Saito R, et al. Relationship between social support status and mortality in a community-based population: a prospective observational study (Yamagata study). BMC Public Health 2020;20:1630.
- 3.↑
Freeborne N, Simmens SJ, Manson JE, et al. Perceived social support and the risk of cardiovascular disease and all-cause mortality in the Women’s Health Initiative Observational Study. Menopause 2019;26:698–707.
- 4.↑
Holt-Lunstad J, Smith TB, Layton JB. Social relationships and mortality risk: a meta-analytic review. PLoS Med 2010;7:e1000316.
- 5.↑
Leung J, Pachana NA, McLaughlin D. Social support and health-related quality of life in women with breast cancer: a longitudinal study. Psychooncology 2014;23:1014–1020.
- 6.↑
Nausheen B, Gidron Y, Peveler R, et al. Social support and cancer progression: a systematic review. J Psychosom Res 2009;67:403–415.
- 7.↑
El-Jawahri A, Nelson AM, Gray TF, et al. Palliative and end-of-life care for patients with hematologic malignancies. J Clin Oncol 2020;38:944–953.
- 8.↑
Peyrade F, Jardin F, Thieblemont C, et al. Attenuated immunochemotherapy regimen (R-miniCHOP) in elderly patients older than 80 years with diffuse large B-cell lymphoma: a multicentre, single-arm, phase 2 trial. Lancet Oncol 2011;12:460–468.
- 9.↑
Crump M, Neelapu SS, Farooq U, et al. Outcomes in refractory diffuse large B-cell lymphoma: results from the international SCHOLAR-1 study. Blood 2017;130:1800–1808.
- 10.↑
Neelapu SS, Locke FL, Bartlett NL, et al. Axicabtagene ciloleucel CAR T-cell therapy in refractory large B-cell lymphoma. N Engl J Med 2017;377:2531–2544.
- 11.↑
El-Jawahri AR, Abel GA, Steensma DP, et al. Health care utilization and end-of-life care for older patients with acute myeloid leukemia. Cancer 2015;121:2840–2848.
- 12.↑
El-Jawahri A, LeBlanc T, VanDusen H, et al. Effect of inpatient palliative care on quality of life 2 weeks after hematopoietic stem cell transplantation: a randomized clinical trial. JAMA 2016;316:2094–2103.
- 13.↑
El-Jawahri AR, Traeger LN, Kuzmuk K, et al. Quality of life and mood of patients and family caregivers during hospitalization for hematopoietic stem cell transplantation. Cancer 2015;121:951–959.
- 14.↑
Birkeland MS, Holt T, Ormhaug SM, et al. Perceived social support and posttraumatic stress symptoms in children and youth in therapy: a parallel process latent growth curve model. Behav Res Ther 2020;132:103655.
- 15.↑
Reisinger MW, Moss M, Clark BJ. Is lack of social support associated with a delay in seeking medical care? A cross-sectional study of Minnesota and Tennessee residents using data from the Behavioral Risk Factor Surveillance System. BMJ Open 2018;8:e018139.
- 16.↑
Molassiotis A, Chan CW, Yam BM, et al. Quality of life in Chinese women with gynaecological cancers. Support Care Cancer 2000;8:414–422.
- 17.↑
Kroenke CH, Kubzansky LD, Schernhammer ES, et al. Social networks, social support, and survival after breast cancer diagnosis. J Clin Oncol 2006;24:1105–1111.
- 18.↑
Jatoi A, Novotny P, Cassivi S, et al. Does marital status impact survival and quality of life in patients with non-small cell lung cancer? Observations from the Mayo Clinic lung cancer cohort. Oncologist 2007;12:1456–1463.
- 19.↑
Agaronnik N, Lindvall C, El-Jawahri A, et al. Use of natural language processing to assess frequency of functional status documentation for patients newly diagnosed with colorectal cancer. JAMA Oncol 2020;6:1628–1630.
- 20.↑
Agaronnik ND, Lindvall C, El-Jawahri A, et al. Challenges of developing a natural language processing method with electronic health records to identify persons with chronic mobility disability. Arch Phys Med Rehabil 2020;101:1739–1746.
- 21.↑
Lindvall C, Lilley EJ, Zupanc SN, et al. Natural language processing to assess end-of-life quality indicators in cancer patients receiving palliative surgery. J Palliat Med 2019;22:183–187.
- 22.↑
Nipp RD, El-Jawahri A, Moran SM, et al. The relationship between physical and psychological symptoms and health care utilization in hospitalized patients with advanced cancer. Cancer 2017;123:4720–4727.
- 23.↑
Newcomb RA, Nipp RD, Waldman LP, et al. Symptom burden in patients with cancer who are experiencing unplanned hospitalization. Cancer 2020;126:2924–2933.
- 24.↑
Nipp RD, El-Jawahri A, D’Arpino SM, et al. Symptoms of posttraumatic stress disorder among hospitalized patients with cancer. Cancer 2018;124:3445–3453.
- 25.↑
Charlson ME, Pompei P, Ales KL, et al. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40:373–383.
- 26.↑
Johnson PC, Xiao Y, Wong RL, et al. Potentially avoidable hospital readmissions in patients with advanced cancer. J Oncol Pract 2019;15:e420–427.
- 27.↑
Thompson T, Rodebaugh TL, Pérez M, et al. Perceived social support change in patients with early stage breast cancer and controls. Health Psychol 2013;32:886–895.
- 28.↑
Haviland J, Sodergren S, Calman L, et al. Social support following diagnosis and treatment for colorectal cancer and associations with health-related quality of life: results from the UK ColoREctal Wellbeing (CREW) cohort study. Psychooncology 2017;26:2276–2284.
- 29.↑
Reynolds P, Boyd PT, Blacklow RS, et al. The relationship between social ties and survival among black and white breast cancer patients. Cancer Epidemiol Biomarkers Prev 1994;3:253–259.
- 30.↑
Pinquart M, Duberstein PR. Associations of social networks with cancer mortality: a meta-analysis. Crit Rev Oncol Hematol 2010;75:122–137.
- 31.↑
Melissant HC, van Uden-Kraan CF, Lissenberg-Witte BI, et al. Body changes after cancer: female cancer patients’ perceived social support and their perspective on care. Support Care Cancer 2019;27:4299–4306.
- 32.↑
Kumar A, Rivera-Hernandez M, Karmarkar AM, et al. Social and health-related factors associated with enrollment in Medicare Advantage plans in older adults. J Am Geriatr Soc 2020;68:313–320.
- 33.↑
Ellis L, Canchola AJ, Spiegel D, et al. Trends in cancer survival by health insurance status in California from 1997 to 2014. JAMA Oncol 2018;4:317–323.
- 34.↑
Foster LW, McLellan L, Rybicki L, et al. Validating the positive impact of in-hospital lay care-partner support on patient survival in allogeneic BMT: a prospective study. Bone Marrow Transplant 2013;48:671–677.
- 35.↑
Frick E, Motzke C, Fischer N, et al. Is perceived social support a predictor of survival for patients undergoing autologous peripheral blood stem cell transplantation? Psychooncology 2005;14:759–770.
- 36.↑
Colón EA, Callies AL, Popkin MK, et al. Depressed mood and other variables related to bone marrow transplantation survival in acute leukemia. Psychosomatics 1991;32:420–425.
- 37.↑
Cronin-Fenton DP, Sharp L, Deady S, et al. Treatment and survival for non-Hodgkin’s lymphoma: influence of histological subtype, age, and other factors in a population-based study (1999–2001). Eur J Cancer 2006;42:2786–2793.
- 38.↑
El-Jawahri A, Keenan T, Abel GA, et al. Potentially avoidable hospital admissions in older patients with acute myeloid leukaemia in the USA: a retrospective analysis. Lancet Haematol 2016;3:e276–283.
- 39.↑
Duric V, Stockler M. Patients’ preferences for adjuvant chemotherapy in early breast cancer: a review of what makes it worthwhile. Lancet Oncol 2001;2:691–697.
- 40.↑
Gonzalez-Saenz de Tejada M, Bilbao A, Baré M, et al. Association between social support, functional status, and change in health-related quality of life and changes in anxiety and depression in colorectal cancer patients. Psychooncology 2017;26:1263–1269.
- 41.↑
Dong X, Liu A. Variations between sources of social support and cancer screen behaviors in U.S. Chinese older adults. J Gerontol A Biol Sci Med Sci 2017;72(Suppl 1):S26–31.
- 42.↑
Dam AEH, de Vugt ME, van Boxtel MPJ, et al. Effectiveness of an online social support intervention for caregivers of people with dementia: the study protocol of a randomised controlled trial. Trials 2017;18:395.
- 43.↑
Dam AEH, van Boxtel MPJ, Rozendaal N, et al. Development and feasibility of Inlife: a pilot study of an online social support intervention for informal caregivers of people with dementia. PLoS One 2017;12:e0183386.
- 44.↑
Valtorta NK, Moore DC, Barron L, et al. Older adults’ social relationships and health care utilization: a systematic teview. Am J Public Health 2018;108:e1–10.
