Association of Social-Ecological Factors With Delay in Time to Initiation of Postoperative Radiation Therapy

Key Points

Question

How are individual and community-level factors associated with risk for postoperative radiation (PORT) delays in patients with head and neck squamous cell carcinoma (HNSCC)?

Findings

In this prospective cohort study of 171 participants who received primary surgery and PORT for HNSCC, lower levels of written health literacy were significantly associated with PORT delays when controlling for demographic and clinical factors. The addition of health literacy and the community-level area deprivation index improved the model’s prediction of PORT delay risk.

Meaning

Precise measures of baseline risk for PORT delay at multiple social-ecologic levels are necessary to develop and implement precise resource-conscious interventions for at-risk patients.

This cohort study examines individual and community-level factors associated with delay in initiation of postoperative radiation therapy among patients with head and neck squamous cell carcinoma.

Abstract

Importance

Timely initiation of postoperative radiation therapy (PORT) is associated with reduced recurrence rates and improved overall survival in patients with head and neck squamous cell carcinoma (HNSCC). Measurement of the association of social-ecological variables with PORT delays is lacking.

Objective

To assess individual and community-level factors associated with PORT delay among patients with HNSCC.

Design, Setting, and Participants

This prospective cohort study carried out between September 2018 and June 2022 included adults with untreated HNSCC who were enrolled in a prospective registry at a single academic tertiary medical center. Demographic information and validated self-reported measures of health literacy were obtained at baseline visits. Clinical data were recorded, and participant addresses were used to calculate the area deprivation index (ADI), a measure of community-level social vulnerability. Participants receiving primary surgery and PORT were analyzed. Univariable and multivariable regression analysis was performed to identify risk factors for PORT delays.

Exposures

Surgical treatment and PORT.

Main Outcomes and Measures

The primary outcome was PORT initiation delay (>42 days from surgery). Risk of PORT initiation delay was evaluated using individual-level (demographic, health literacy, and clinical data) and community-level information (ADI and rural-urban continuum codes).

Results

Of 171 patients, 104 patients (60.8%) had PORT delays. Mean (SD) age of participants was 61.0 (11.2) years, 161 were White (94.2%), and 105 were men (61.4%). Insurance was employer-based or public among 65 (38.5%) and 75 (44.4%) participants, respectively. Mean (SD) ADI (national percentile) was 60.2 (24.4), and 71 (41.8%) resided in rural communities. Tumor sites were most commonly oral cavity (123 [71.9%]), with 108 (63.5%) classified as stage 4 at presentation. On multivariable analysis, a model incorporating individual-level factors with health literacy in addition to community-level factors was most predictive of PORT delay (AOC= 0.78; R², 0.18).

Conclusions and Relevance

This cohort study provides a more comprehensive assessment of predictors of PORT delays that include health literacy and community-level measures. Predictive models that incorporate multilevel measures outperform models with individual-level factors alone and may guide precise interventions to decrease PORT delay for at-risk patients with HNSCC.

Introduction

Timely initiation of radiation therapy following primary surgical ablation of head and neck squamous cell carcinoma (HNSCC) is critical to achieving locoregional control and improved survival in patients with advanced-stage and high-risk disease.^1,2 Although initiation of postoperative radiation therapy (PORT) within 6 weeks of surgery is the standard to achieve optimal oncologic outcomes, most patients do not begin adjuvant therapy within this window.^3,4,5 In early 2022, the Commission on Cancer made initiation of radiation within 6 weeks of surgery its first and only head and neck cancer quality measure.⁶ Factors associated with an increased risk of PORT delay in prior studies include oral cavity and advanced-stage tumors, PORT outside the surgical facility, Medicaid insurance or uninsured status, minority race and ethnicity, rural geographic location, lower education level, lack of interdisciplinary care coordination, and lower social support.^7,8,9,10,11

Prior studies^12,13,14 have demonstrated that delays in initiating PORT are associated with worse overall survival and increased recurrence. The interval between primary surgery and initiation of PORT is the primary driver of total treatment package time, affecting survival independent of diagnosis-to-treatment interval or duration of radiation therapy.^12,15,16 Delays in PORT initiation disproportionately affect racial and ethnic minorities, those with Medicaid insurance, and those in medically underserved areas.^13,17,18,19

To guide targeted interventions to improve the delivery of timely, equitable, guideline-adherent PORT, better predictive tools are needed to identify patients at risk for delay. The social-ecological model, originally developed to guide violence prevention, recognizes the influence of individual characteristics, interpersonal or relationship influences, community factors, and society-level influences on health behavior and outcomes.^20,21 The influence of individual contexts and community environments on health behaviors has been studied extensively over the past several decades.²² Likewise, the social-ecological model has been applied across studies in public health and otolaryngology, including understanding disparities in human papilloma virus (HPV) vaccination uptake.^23,24 Prior qualitative work has even referenced the social-ecological model to evaluate and conceptualize barriers to timely PORT in head and neck cancer.²⁵ By identifying disparities in PORT delay at the individual, interpersonal, community, and society level, the social-ecological model may guide how to address key drivers of prevention, timely diagnosis, timely completion of treatment, and support for survivorship at each level of influence. Although a number of studies have identified individual-level factors associated with PORT delay using data from national databases,⁵ they have been limited in their consideration of community-level factors affecting delays. Moreover, to our knowledge, no studies have applied these measures in a prospective cohort to quantitatively assess factors associated with PORT delay in patients with head and neck cancer. The aim of this study was to evaluate factors from multiple levels of influence among patients with HNSCC treated at a single Midwestern institution to develop a more comprehensive baseline risk assessment for PORT delay. We hypothesize that based on the social-ecological framework, a model that incorporates baseline measures at the individual and community levels will outperform a model with individual-level factors alone when predicting delays in PORT initiation.

Methods

This study received approval by the institutional review board at the University of Kansas Medical Center (KUMC) (#00001732), and all participants provided written informed consent. Adult patients eligible for enrollment into the KUMC Head and Neck Registry were those receiving primary evaluation and treatment for HNSCC at KUMC, a tertiary care referral center. This registry has been described in detail in prior work.²⁶ Patients who completed radiation therapy between September 2018 and June 2022 were included in the analysis. These dates corresponded to the earliest registry enrollment with available radiation records. Patients included in the current study were those requiring primary surgical ablation followed by PORT with or without chemotherapy. Patients were enrolled at their initial consultation with a head and neck oncologic surgeon. The decision to pursue PORT was determined by consensus recommendations of a multidisciplinary tumor board, which reflected the National Comprehensive Cancer Center (NCCN) guidelines for treatment of head and neck mucosal squamous cell carcinomas. Surgery was performed at the primary facility, and patients received adjuvant radiation in or outside KUMC, depending on patient preference. Data regarding PORT initiation were captured prospectively from the electronic medical record or end of treatment summaries. Eight patients were excluded because radiation records were unavailable.

Individual-Level Measures

In the social-ecological framework, variables measured at the individual level included clinical data such as tumor site and stage, health literacy and education levels, race and ethnicity, age, sex, insurance, Charlson Comorbidity Index (CCI) score, and whether the patient required free flap reconstruction. The CCI was coded as a categorical variable, excluding the index tumor. A CCI score of 0 to 1 was considered low, a CCI score of 2 to 4 as intermediate, and a CCI score of 5 or greater as high.

The Brief Health Literacy Screening Tool (BRIEF) provides a short assessment of confidence in understanding written and verbal health information.²⁷ In this 4-item self-reported survey, scores range from 4 to 20, with 4 to 12 being considered limited health literacy, 13 to 16 being marginal health literacy, and 17 to 20 being adequate health literacy. Mean BRIEF scores in the delayed and nondelayed groups were calculated and compared as continuous measures.

Community-Level Measures

In the social-ecological model, the Rural-Urban Continuum Code (RUCC) and Area Deprivation Index (ADI) were considered community-level variables. The RUCC uses measures of population density and urbanization to classify US counties.²⁸ The latest classifications from 2013 were used to calculate the RUCC. The ADI is a measure of socioeconomic disadvantage of a neighborhood in relation to either the corresponding state or US as a whole, and incorporates 17 measures of housing quality, education, income/employment, and poverty. The ADI uses the census block group as the geographic unit of construction, which is a subdivision in a census tract and more closely approximates the “neighborhood” of a given address than the county or zip code reported in national databases.²⁹ Patient addresses were obtained at baseline and used to calculate the ADI and RUCC. The ADI was calculated using the American Community Survey from 2018. Current addresses were extracted from the electronic medical record and geocoded to latitude and longitude coordinates using Geoapify,³⁰ an online geocoding tool, and then merged with the corresponding polygon associated with their census block using ArcGIS Pro (Esri, version 2.9). Only participants residing in Kansas or Missouri were included. Eight (4.5%) participants provided only PO box addresses, and for these ADI was manually mapped to the PO box location. The ADI national percentile was scored on a continuous scale (1-99, with 1 the least disadvantaged, and 99 the most disadvantaged).

Data Collection and Quantitative Analysis

Demographic information, insurance status, and BRIEF survey were collected at baseline. Clinical data including pathologic tumor stage, site, and radiation start date were obtained from the electronic medical record. All data were stored in a secure online database.^31,32 Patients were divided into 2 groups based on whether PORT was initiated within or after 42 days.

Univariable logistic regression was performed to generate crude (unadjusted) odds ratios (ORs) based on patient demographic information, clinical information, and survey scores. To better outline specific aspects of the BRIEF screen in relation to delays, individual question scores were compared between cohorts. Multivariable logistic regression generated adjusted odds ratios (aORs) after controlling for other variables in each model. With planned analysis of 8 variables spanning the levels of the social-ecological model, an enrollment target of 135 patients was set, provided an expected incidence of delays of 60% based on historical experience at our institution. This would provide the “rule of thumb” of at least 10 events per variable prior to analysis.³³

Hypothesis Testing

To test the central hypothesis, we generated several multivariable models that built toward including measures from multiple social-ecological levels of influence. In the first multivariable model, individual-level measures from the social-ecological model were assessed, which included age, insurance, overall stage, tumor site, and CCI score category (high, intermediate, or low CCI scores). A second individual-level model was generated, incorporating the “learning” score from the BRIEF to evaluate its effect on the model’s strength, as the effect of health literacy on PORT delay risk has not been directly evaluated in prior studies. A third model was generated incorporating the community-level measure ADI. Finally, a fourth model incorporated both community-level ADI and RUCC.

Multiple imputation was used to account for missing data (0.96% of the total data set); most were missing BRIEF scores (n = 11). A description of values imputed is provided in eTable 1 in Supplement 1. Multivariable imputation by chained equations was performed using the mice package in R. Twenty-five data sets were generated, and sequential regression imputation calculated each missing value based on all other variables in the regression equation. Residual variance was added to the parameter estimates in the regression model using the Bayesian method. A 2-sided α threshold of α = .05 was applied for all tests of statistical significance. All statistical analysis was performed using R Studio (version 4.2.1; R Foundation).

Results

A total of 171 patients who received primary surgery and adjuvant radiation therapy were included in the overall cohort. Of these, 104 patients (60.8%) experienced a delay in initiating PORT. The mean (SD) age for the overall cohort was 61.0 (11.2) years. Overall, 161 patients were White (94.2%), and 105 were men (61.4%) (Table 1), which is representative of the population of patients with head and neck cancer seen at KUMC. This study categorized 14.97% of the US population as rural, in agreement with published federal data.³⁴ In 154 patients who completed the BRIEF, overall health literacy score did not predict PORT delay on univariable analysis (Table 2). When examining individual subscores of the BRIEF, a lower self-reported ability to learn written health information predicted PORT delay (OR, 1.07; 95% CI, 1.00-1.16). In addition, a high CCI score category predicted delay (OR, 1.29; 95% CI, 1.06-1.58).

Table 1. Demographics and Associations With Postoperative Radiation Therapy Delays.

Characteristic	No. (%)			Crude OR (95% CI)
	All (N = 171)	PORT delay
		Yes (N = 104)	No (N = 67)
Sex
Female	66 (38.6)	43 (41.3)	23 (34.3)	1.07 (0.92-1.25)
Male	105 (61.4)	61 (58.7)	44 (65.7)	1 [Reference]
White	161 (94.2)	63 (94.0)	98 (94.2)	1.01 (0.74-1.38)
RUCC, rurala	71 (41.8)	24 (36.4)	47 (45.2)	1.09 (0.94-1.27)
Age, mean (SD), y	61.0 (11.2)	62.3 (11.1)	59.0 (11.1)	1.01 (1.00-1.01)
Insurance
Employer-based	65 (38.5)	33 (32.4)	32 (47.8)	1 [Reference]
Public	75 (44.4)	50 (49.0)	25 (37.3)	1.17 (1.00-1.38)
Other	29 (17.2)	19 (18.6)	10 (14.9)	1.16 (0.94-1.43)
Employment
Full time	59 (35.3)	31 (30.7)	28 (42.4)	1 [Reference]
Part time	19 (11.4)	11 (10.9)	8 (12.1)	1.05 (0.82-1.36)
Unemployed	20 (12.0)	17 (16.8)	3 (4.5)	1.38 (1.08-1.77)
Retired	69 (41.3)	42 (41.6)	27 (40.9)	1.09 (0.92-1.29)
Education
High school	53 (31.9)	31 (31.3)	22 (32.8)	1 [Reference]
Some college	55 (33.1)	35 (35.4)	20 (29.9)	1.05 (0.87-1.27)
Completed college	35 (21.1)	21 (21.2)	14 (20.9)	1.02 (0.82-1.25)
Graduate education	23 (13.9)	12 (12.1)	11 (16.4)	0.94 (0.74-1.20)
Tumor site
Oral cavity	123 (71.9)	82 (78.8)	41 (61.2)	1 [Reference]
Oropharynx, HPV+	23 (13.5)	9 (8.7)	14 (20.9)	0.76 (0.61-0.94)
Oropharynx, HPV−	4 (2.3)	1 (1.0)	3 (4.5)	0.66 (0.41-1.07)
Larynx	13 (7.6)	8 (7.7)	5 (7.5)	0.95 (0.72-1.25)
Other	8 (4.7)	4 (3.8)	4 (6.0)	0.85 (0.60-1.20)
AJCC clinical stage
1	19 (11.2)	7 (6.7)	12 (18.2)	1 [Reference]
2	14 (8.2)	9 (8.7)	5 (7.6)	1.32 (0.94-1.84)
3	29 (17.1)	18 (17.3)	11 (16.7)	1.29 (0.97-1.70)
4	108 (63.5)	70 (67.3)	38 (57.6)	1.32 (1.04-1.68)
Charlson Comorbidity Index (CCI) scoreb
Low (0-1)	34 (19.9)	17 (16.3)	17 (25.4)	1 [Reference]
Intermediate (2-4)	75 (43.9)	40 (38.5)	35 (52.2)	1.03 (0.85-1.26)
High (≥5)	62 (36.3)	47 (45.2)	15 (22.4)	1.29 (1.06-1.58)
Free flap reconstruction, yes	115 (67.3)	75 (72.1)	40 (59.7)	1.14 (0.98-1.34)

Abbreviations: AJCC, American Joint Commission on Cancer; CCI, Charlson Comorbidity Index; HPV, human papilloma virus; PORT, postoperative radiation therapy; RUCC, Rural Urban Continuum Codes.

^a Urban corresponds to RUCC 1 to 2, and rural corresponds to RUCC 3 to 9.

^b Charlson Comorbidity Index was coded excluding the index cancer.

Table 2. Association of Area Deprivation Index and Health Literacy With Postoperative Radiation Therapy Delay.

Variable	Mean (SD)			Crude OR (95%CI)
	All	PORT delay
		Yes	No
No.	171	104	67	NA
ADI, national percentile, mean (SD)a	60.7 (24.2)	63.2 (22.6)	56.8 (26.2)	1.00 (1.00-1.01)
Health literacy	Alle	NA	NA	NA
No.	154	96	58	NA
BRIEF score, mean (SD)b	15.8 (3.8)	15.6 (3.7)	16.0 (4.0)	1.01 (0.99-1.03)
How often do you have…
Someone to help you read hospital materials?c	3.7 (1.4)	3.8 (1.4)	3.6 (1.5)	0.98 (0.93-1.03)
Problems learning about your medical condition because of difficulty understanding written information?c	4.1 (1.1)	4.0 (1.1)	4.3 (1.0)	1.07 (1.00-1.16)
A problem understanding what is told to you about your medical condition?c	4.1 (1.0)	4.1 (0.9)	4.2 (1.1)	1.01 (0.93-1.09)
How confident are you filling out medical forms by yourself?d	3.8 (1.2)	3.7 (1.2)	3.9 (1.2)	1.03 (0.96-1.10)

Abbreviations: ADI, area deprivation index; BRIEF, BRIEF Health Literacy Screening Tool; NA, not applicable; OR, odds ratio; PORT, postoperative radiation therapy.

^a National percentile ranges from 1 (least disadvantaged) to 99 (most disadvantaged).

^b Scores range from 4 to 12 (limited health literacy), 13 to 16 (marginal health literacy), 17 to 20 (adequate health literacy).

^c Scores range from 1 (always [lower level of health literacy]) to 5 (never [higher level of health literacy]).

^d Scores range from 1 (not at all [lower level of health literacy]) to 5 (extremely [higher level of health literacy]).

^e Imputation was not performed for univariable analysis.

A series of multivariable logistic regression equations were generated, sequentially adding measures from different levels of influence in the social-ecological model. The first predictive model incorporating individual-level measures without the BRIEF yielded a model with an area under the curve (AUC) of 0.74 and R² of 0.15 (eTable 2 in Supplement 1). A second predictive model incorporating the BRIEF with other individual-level measures yielded an improved model (AUC, 0.77; R² = 0.17), and in which a low BRIEF learning score significantly predicted delay (aOR, 1.45; 95% CI, 1.02-2.13) (eTable 3 in Supplement 1). Including the community-level ADI yielded a better-fitting model (AUC, 0.78; R² = 0.18; eTable 4 in Supplement 1). The strength of the model was largely unchanged with the addition of RUCC (Table 3). Although mean ADI was higher among delayed patients, this result was not significant on multivariable analysis (Table 2, Table 3). Controlling for all other factors, a high CCI score category significantly predicted delay (aOR, 4.21; 95% CI, 1.11–17.60) (Table 3).

Table 3. Multivariable Logistic Regression Model of Postoperative Radiation Therapy Delay Risk Based on Individual and Community Factors^a.

Patient characteristic	aOR (95%CI)
Age, y	1.03 (0.98-1.08)
Insurance
Employer-based	1 [Reference]
Public	1.23 (0.47-3.22)
Other	2.33 (0.84-6.79)
CCI
Low	1 [Reference]
Intermediate	0.63 (0.19-2.03)
High	4.21 (1.11-17.6)
AJCC clinical stage
1	1 [Reference]
2	4.40 (0.77-21.80)
3	1.67 (0.29-9.70)
4	1.35 (0.25-7.12)
Tumor site
Oral cavity	1 [Reference]
Oropharynx, HPV+	0.10 (0.02-0.56)
Oropharynx, HPV−	0.21 (0.01-2.37)
Larynx	0.66 (0.17-2.65)
Other	0.26 (0.04-1.60)
BRIEF, learning	1.43 (1.00-2.08)
ADI, national percentile	1.01 (0.99-1.03)
RUCC, rural	1.21 (0.53-2.75)

Abbreviations: ADI, area deprivation index; AJCC, American Joint Commission on Cancer; aOR, adjusted odds ratios; AUC, area under the curve; BRIEF, BRIEF Health Literacy Screening Tool; CCI, Charlson Comorbidity Index; HPV, human papilloma virus; PORT, postoperative radiation therapy; RUCC, Rural Urban Continuum Codes.

^a Multiple imputation was used to fill missing data (0.96% of data set). Using multivariable logistic regression with individual-level factors and community-level factors, a significant model was found with an area under the curve (AUC), 0.78; R² of 0.18.

In addition, HPV-positive oropharyngeal tumors were protective of delays on multivariable analysis (aOR, 0.10; 95% CI, 0.02-0.56) (Table 3). Compared with all other tumor sites including HPV-negative oropharyngeal cancers, participants with HPV-positive oropharyngeal tumors were less often women (2 [9%] vs 64 [43%]; difference 34%; 95% CI, 18%-51%), less often insured through a public payer (4 [17%] vs 71 [49%]; difference 32%; 95% CI, 11%-51%), less likely to present with stage 3 to 4 disease (3 [9%] vs 135 [92%]; difference 83%; 95% CI, 68%-98%), and had significantly lower ADI (50th vs 62nd percentile; difference 12 percentiles, 95% CI, 2nd percentile to 22nd percentile). Age, race and ethnicity, education level, rural status, and BRIEF scores were not significantly different in participants with HPV-positive oropharyngeal tumors.

Further analysis was performed to assess the absolute differences in median days to PORT initiation between groups for each variable measured. Overall, median (IQR) days to PORT initiation in the delayed group was 53.5 (16.5) days. Patients with public insurance were significantly more delayed in median (IQR) days to PORT at 48.0 (18.5) days compared with 43.0 (16.0) days among patients with private insurance (P = .04). Subsequent analysis of patients at extremes of delay revealed that 40 (30.5%) patients were delayed past 56 days. Notably, a significant association was found between low levels on the BRIEF learning score and extreme PORT delay (OR, 1.09; 95% CI, 1.03-1.16) (eTable 5 in Supplement 1).

Discussion

The prevalence of PORT delay in the present study (104 [60.8%]) supports the finding from national database studies that most patients with HNSCC do not receive timely initiation of adjuvant radiation therapy.¹⁰ The University of Wisconsin Population Health Institute has estimated social determinants of health and physical environment traced to an individual’s zip code account for 50% of health outcomes.³⁵ When considering health behaviors such as tobacco and alcohol use, interactions with the health care environment are only estimated to be driving 20% of outcomes. Likewise, PORT delays are multifactorial, contributing to the complexity of predicting and mitigating their risk. Although not explicitly applying the social-ecological framework, prior studies have considered individual-level factors and added a level to the framework, with organization-level factors increasing risk of PORT delay. Specifically, authors include payers, separate surgical and PORT facilities, postoperative length of stay, advanced stage at presentation, and tumor site.^11,17,36 Contrary to these prior investigations, the current study assessed the individual- and community-level influences on delays and found that a predictive model incorporating multiple levels of influence from the social-ecological framework outperformed a model based on individual-level factors alone.

In addition, incorporating health literacy strengthened the predictive model in this study. Disparities in cancer care have been associated with lower levels of health literacy.³⁷ Though objective measures of health literacy have not been used to predict PORT delay, the BRIEF screen has previously been used to predict self-directed behaviors in head and neck cancer, such as active engagement, positive attitude and self-monitoring, and skill acquisition.³⁸ Lower scores have been associated with decreased functional well-being, socio-emotional quality of life, and disease-specific health-related quality of life.^38,39 Among individual-level factors in the current study, a lower score specifically on the learning component of the BRIEF assessment was significantly associated with PORT delay and even extreme delays (past 56 days). Given the overwhelming information presented to patients with a new cancer diagnosis, the tools in place to assist with information retention may not benefit individuals with lower written health literacy. For example, these individuals may be less likely to reference or understand information in after-visit summaries and online medical record portals. Interventions targeting treatment delays could benefit from improved communication methods via illustrations, videos, and teach-back methodologies employed by care coordinators reiterating priorities and tracking progress through treatment. This is the first study to incorporate the BRIEF in a discussion of PORT delay risk, and the strength of the predictive model emphasizes its application in prospective risk stratification.

Rural residence, as measured by the RUCC classification, was not associated with increased odds of PORT delay in the current study. These findings are contrary to what Shew and Levy et al found using data from the National Cancer Database (NCDB).^8,11 There are a number of potential explanations for these discrepant findings. First, county-level classification may not have been sensitive enough to detect the true influence of rurality in this cohort, where significant differences in population density and urbanization exist in 1 county and where rurality may simply be a proxy for critical access communities with limited health resources. Second, 71 (42%) participants in the cohort resided in rural communities, which is much higher than in prior studies.^8,11 Unfortunately, no optimal measure has been developed to predict rural-urban health disparities in all contexts.⁴⁰

Although ADI alone did not significantly predict delay in initiating PORT, its interaction with individual-level factors strengthened the multivariable model. Factors potentially affecting its significance include the homogenous cohort, influence from historical exposures to disadvantage, or societal-level influences such as Medicaid expansion, which was not assessable in the nonexpansion state examined in this study. The ADI was used to estimate social vulnerability, as it is a more precise measure than income quartiles from zip code information available in the NCDB. Agreement exists in identifying more significant delays and worse survival among individuals residing in lower-income areas.^17,36,41 Furthermore, prior work across other disciplines suggests an individual’s address alone may predict risk without the need for individually reported household income information.^42,43,44 Future multicenter prospective studies may provide more robust assessments of community-level influence, but the current study findings highlight the importance of environment on health behavior.

Structure and process factors relating to care coordination and availability of consultation services have been shown to be independent factors affecting delays, but are also affected by social determinants of health emphasized in this work. For example, the decision to pursue PORT at an outside facility is likely influenced by distance from facility, social support, and access to transportation.¹⁷ Lower baseline social support, health literacy, and access to a main tertiary hospital are later compounded by fragmented coordination between surgical and radiation facilities.⁴⁵ Similarly, clinical factors previously associated with PORT delay in head and neck cancer have included stage 4 disease and oral cavity site.¹¹ Timely adjuvant therapy in patients with higher oncologic risk is critical, and appropriate risk stratification may guide earlier dental and radiation oncology referrals. In addition, patients with higher baseline comorbidity status were at significantly higher risk of delay in this study, which may reflect an effect on downstream postsurgical factors such as length of stay or readmission not assessed in this study. By contrast, the favorable individual social circumstances plus insurance and income status in patients with HPV-mediated oropharyngeal cancers in this study were likely the primary influencers contributing to fewer delays, consistent with prior literature.⁴⁶ Although NCDB-based studies have been effective in outlining the broad effect these factors have on delay risk, the present study offers a more precise understanding of the interplay between clinical and social factors to guide precise interventions in clinical practice.

Equity-Driven Interventions

Preliminary trials aimed at targeting risk factors for PORT delay have focused on improving care coordination through multidisciplinary referrals and tracking, clarifying patient navigator roles, community-based travel support, and patient education to emphasize goals.^47,48 These interventions have also targeted patients known to experience disparities in access to care through intentional recruitment. Pilot data from these interventions shows promising improvements in timely PORT initiation, and larger studies are under way to validate these findings.⁴⁸ Though not specifically addressed in this study, the Medical Outcomes Study—Social Support Inventory (MOS-SSS) is a measure of social support evaluated in our prior work.⁴⁵ Considered an interpersonal-level factor in the social-ecological model, social support has been shown to contribute to resilience through cancer treatment, and has been associated with increased follow-up, decreased travel burden, and improved depressive symptoms and psychological distress associated with physical symptoms.^{7,17,49,50,51} In our prior study consisting of a smaller cohort from the same patient population, a higher score on the MOS-SSS was indeed found to be a significant predictor of PORT delay, even when controlling for other baseline individual factors. In strengthening existing nomograms of risk factors for PORT delay, studies may consider including individual, interpersonal, and community measures employed in this study and our prior work. Future trials could adopt these measures to better stratify patient risk factors for delays, and offer more precise interventions that address both system care processes and negative social determinant factors critical to timely therapy.

Limitations

This is a single-institution study at a tertiary academic center with a relatively homogenous patient cohort of White men, thus limiting generalizability across patient populations. Although the BRIEF has been validated for use in more representative samples, the present results are subject to selection bias and should be replicated in other patient populations. Inherent social acceptability and voluntary enrollment bias also exists in self-reported information obtained via in-person surveys. We considered the potential for selection bias relating to participant consent for enrollment and patient factors influencing the likelihood for consent. However, we offer enrollment to every new patient in our clinic, with a participation rate of no less than 90%, limiting this source of bias. Estimates of rurality were made based on data from the 2013 edition of RUCC, which may not account for recent changes in populations. Among the 8 participants with only PO box locations provided, we acknowledge the potential discrepancy between ADI closer to their physical residence. However, because this was such a small proportion of participants, it did not significantly influence our findings. In addition, residence, insurance, and employment status were captured at a single time point and do not account for changes between the initial consultation and initiation of PORT. As PORT facility relative to surgical facility has been shown to affect delays, we evaluated this variable and found it to be significant on multivariable analysis. However, because PORT facility is often not known at baseline, we chose to include the RUCC and ADI scores as community-level measures, which likely influence the choice of PORT facility. Moreover, this study was not powered to assess additional variables that may affect baseline PORT delay risk, such as BMI and smoking history. Future studies may evaluate the influence of these individual-level variables on a predictive model of risk.

Conclusions

Ample evidence exists to demonstrate the systemic inequities that predispose certain individuals to PORT delay. As shown in iterative predictive models presented in this prospective cohort study, precise community-level measures of vulnerability provide additional predictive value, and may be considered in multicenter studies examining the social determinants of delays and outcomes. The current study presents patient-centered precise measures of risk that aid in identifying those at highest risk for delays. Future studies may also apply these measures to develop resource-conscious interventions, with the goal of reducing the effect of nonmodifiable risk factors on PORT delay.

Supplement 1.

eTable 1. Values imputed for missing data

eTable 2. Multivariable analysis of PORT delay risk with individual-level measures only

eTable 3. Multivariable analysis of PORT delay risk with individual-level measures only, including the “learning” component of the BRIEF health literacy screen

eTable 4. Multivariable analysis of PORT delay risk with individual-level measures and community-level ADI

eTable 5. Associations with Extreme Post-operative Radiation Therapy (PORT) Delays Past 56 days

Supplement 2.

Data Sharing Statement