The Use of Natural Language Processing to Assess Social Support in Patients With Advanced Cancer

Sunil Bhatt; P Connor Johnson; Netana H Markovitz; Tamryn Gray; Ryan D Nipp; Nneka Ufere; Julia Rice; Matthew J Reynolds; Mitchell W Lavoie; Madison A Clay; Charlotta Lindvall; Areej El-Jawahri

doi:10.1093/oncolo/oyac238

The Use of Natural Language Processing to Assess Social Support in Patients With Advanced Cancer

Oncologist. 2023 Feb 8;28(2):165-171. doi: 10.1093/oncolo/oyac238.

Authors

Sunil Bhatt¹, P Connor Johnson^{1

2}, Netana H Markovitz¹, Tamryn Gray^{2

3}, Ryan D Nipp^{1

2}, Nneka Ufere^{2

4}, Julia Rice⁵, Matthew J Reynolds⁵, Mitchell W Lavoie⁵, Madison A Clay⁵, Charlotta Lindvall^{2

3}, Areej El-Jawahri^{1

2}

Affiliations

¹ Department of Medicine, Division of Hematology & Oncology, Massachusetts General Hospital, Boston, MA, USA.
² Harvard Medical School, Boston, MA, USA.
³ Department of Psychosocial Oncology and Palliative Care, Dana-Farber Cancer Institute, Boston, MA, USA.
⁴ Division of Gastroenterology, Department of Medicine, Massachusetts General Hospital, Brigham and Women's Hospital, Boston, MA, USA.
⁵ Department of Psychiatry, Massachusetts General Hospital, Boston, MA, USA.

Abstract

Background: Data examining associations among social support, survival, and healthcare utilization are lacking in patients with advanced cancer.

Methods: We conducted a cross-sectional secondary analysis using data from a prospective longitudinal cohort study of 966 hospitalized patients with advanced cancer at Massachusetts General Hospital from 2014 through 2017. We used NLP to identify extent of patients' social support (limited versus adequate as defined by NLP-aided review of the Electronic Health Record (EHR)). Two independent coders achieved a Kappa of 0.90 (95% CI: 0.84-1.00) using NLP. Using multivariable regression models, we examined associations of social support with: 1) OS; 2) death or readmission within 90 days of hospital discharge; 3) time to readmission within 90 days; and 4) hospital length of stay (LOS).

Results: Patients' median age was 65 (range: 21-92) years, and a plurality had gastrointestinal (GI) cancer (34.3%) followed by lung cancer (19.5%). 6.2% (60/966) of patients had limited social support. In multivariable analyses, limited social support was not significantly associated with OS (HR = 1.13, P = 0.390), death or readmission (OR = 1.18, P = 0.578), time to readmission (HR = 0.92, P = 0.698), or LOS (β = -0.22, P = 0.726). We identified a potential interaction suggesting cancer type (GI cancer versus other) may be an effect modifier of the relationship between social support and OS (interaction term P = 0.053). In separate unadjusted analyses, limited social support was associated with lower OS (HR = 2.10, P = 0.008) in patients with GI cancer but not other cancer types (HR = 1.00, P = 0.991).

Conclusion: We used NLP to assess the extent of social support in patients with advanced cancer. We did not identify significant associations of social support with OS or healthcare utilization but found cancer type may be an effect modifier of the relationship between social support and OS. These findings underscore the potential utility of NLP for evaluating social support in patients with advanced cancer.

Keywords: hospitalization; natural language processing; social support; survival analysis.

Publication types

Research Support, Non-U.S. Gov't

MeSH terms

Aged
Cross-Sectional Studies
Humans
Longitudinal Studies
Natural Language Processing*
Neoplasms* / therapy
Prospective Studies