Longitudinal patient-reported outcomes and survival among early-stage non-small cell lung cancer patients receiving stereotactic body radiotherapy

Abstract

Background and Purpose:

The study objective was to determine whether longitudinal changes in patient-reported outcomes (PROs) were associated with survival among early-stage, non-small cell lung cancer (NSCLC) patients undergoing stereotactic body radiation therapy (SBRT).

Materials and Methods:

Data were obtained from January 2015 through March 2020. We ran a joint probability model to assess the relationship between time-to-death, and longitudinal PRO measurements. PROs were measured through the Edmonton Symptom Assessment Scale (ESAS). We controlled for other covariates likely to affect symptom burden and survival including stage, tumor diameter, comorbidities, gender, race/ethnicity, relationship status, age, and smoking status.

Results:

The sample included 510 early-stage NSCLC patients undergoing SBRT. The median age was 73.8 (range: 46.3–94.6). The survival component of the joint model demonstrates that longitudinal changes in ESAS scores are significantly associated with worse survival (HR: 1.04; 95% CI: 1.02–1.05). This finding suggests a one-unit increase in ESAS score increased probability of death by 4%. Other factors significantly associated with worse survival included older age (HR: 1.04; 95% CI: 1.03–1.05), larger tumor diameter (HR: 1.21; 95% CI: 1.01–1.46), male gender (HR: 1.87; 95% CI: 1.36–2.57), and current smoking status (HR: 2.39; 95% CI: 1.25–4.56).

Conclusion:

PROs are increasingly being collected as a part of routine care delivery to improve symptom management. Healthcare systems can integrate these data with other real-world data to predict patient outcomes, such as survival. Capturing longitudinal PROs—in addition to PROs at diagnosis—may add prognostic value for estimating survival among early-stage NSCLC patients undergoing SBRT.

INTRODUCTION

Predicting survival among cancer patients is a challenging task. Studies show that cancer patients and providers tend to overestimate survival likelihood, which negatively affects patient outcomes[1–3]. Prognostic uncertainty is associated with greater utilization of intensive end-of-life care [4], and reduced quality of life [5]. Further, the inability to accurately predict survival hinders patient-provider communication about goals of care and end-of-life preferences[6–10]. Improving our ability to accurately predict survival could inform clinical decision-making (e.g., individualized follow-up and surveillance), enhance patient-provider communication, and ultimately improve end-of-life care delivery. While many survival prediction models have been developed, their clinical utility is often limited due to implementation barriers including manual data entry (e.g., risk score calculators) [9, 11] or reliance on data that is not available at the point of care delivery (e.g., insurance claims data) [12–14]. Leveraging real-world data, such as patient-reported outcomes (PROs) that are captured in routine care delivery, might aid in the development of survival prediction models that could be used in clinical practice.

PROs have demonstrated prognostic value for predicting survival among cancer patients. Several systematic reviews have reported that PROs predict survival across several cancers (e.g., breast, lung, colorectal, and melanoma) [15–18]. However, much of the available research uses clinical trial data, which may not be representative of all cancer patients (e.g., patients with comorbidities) [15–17]. Further, many prior studies examining the association between PROs and survival have been cross-sectional [18, 19]. Cross-sectional measurement of PROs does not account for how PROs change throughout cancer treatment, which may be important for predicting survival. Descriptive studies examining longitudinal changes in PROs suggest that PRO scores change in the months prior to death [20]. For example, one study found that patient reports of moderate-to-severe drowsiness increased by 41% during the last month of life [20]. There has been limited study of whether longitudinal changes in PROs can aid in survival prediction [21]. Available studies that have explored longitudinal changes in PROs and survival have focused on patients with advanced cancer [20–21].

Examining the relationship between longitudinal changes in PROs and survival may be particularly relevant for early-stage, non-small cell lung cancer (NSCLC) patients undergoing stereotactic body radiation therapy (SBRT). Early-stage NSCLC patients undergoing SBRT have a high comorbidity burden compared to patients eligible for surgery [22] and may be at-risk for early death due to competing causes [23]. Therefore, estimation of survival among early-stage NSCLC patients initiating SBRT may inform patient-clinician communication earlier on in care delivery (e.g., supporting goals-of-care conversations before end-of-life). Study results will provide important insights about the role that changing symptom burden may play in these patients’ survival. Such information could be used to inform survival prognostic tools in the future. Therefore, the objective of this study is to examine whether longitudinal changes in PROs is associated with time-to-death among early-stage NSCLC patients undergoing SBRT.

MATERIALS AND METHODS

Study design

We conducted a retrospective, longitudinal cohort study from 2015–2020 using PRO data collected as a part of routine care delivery.

Study population

The sample included patients undergoing SBRT as their primary treatment for early-stage, NSCLC (T1–T3, N0) at Moffitt Cancer Center, a National Cancer Institute (NCI)-designated comprehensive cancer center. SBRT is standard of care for patients who may be ineligible for surgery (e.g., due to frailty or comorbidity burden) or patients who prefer not to receive surgery. While patients with metastatic disease (M1) may also receive SBRT to the thorax (e.g., colorectal cancer with lung oligo-metastasis); we excluded these patients given that symptom burden and survival process are likely distinct from early-stage lung cancer patients. For SBRT, patients were immobilized using a custom BodyFix cradle with vacuum immobilization. A cone-beam CT scan was used for daily image guidance. Patients were treated with planar arcs using Volumetric modulated arc therapy. Post-treatment, SBRT patients are evaluated every 3 months for one year, every 6 months for four years, and then annually thereafter.

Data sources

Moffitt’s Department of Radiation Oncology has curated a database of >5000 patients who received radiation to the thorax since 2005 [24]. The database combines electronic health record, billing, dosimetric, cancer registry, and PRO data. Since early 2015, Moffitt has routinely collected a modified version of the Edmonton Symptom Assessment Scale (ESAS) for all patients receiving care in the Departments of Radiation Oncology and Supportive Care Medicine. The ESAS is administered at all in-person visits, allowing for longitudinal tracking of PROs (described below). Data were obtained from January 2015 through March 2020. The patient cohort was limited to those who received SBRT during this timeframe and who completed at least one ESAS.

Measures

Time-to-death.

We extracted the time in years from the diagnosis date to the date of last follow up or death, using the date that occurred first. Diagnosis date was defined as 1) the date of pathologic confirmation or 2) the date of the imaging study showing a growing lesion, which invoked SBRT treatment if pathologic biopsy was not possible or refused by the patient.

ESAS.

The ESAS assesses nine common symptoms (pain, tiredness, drowsiness, nausea, lack of appetite, shortness of breath, depression, anxiety, overall well-being) [25]. The symptoms are rated on an eleven-point scale from 0 (no symptom burden) to 10 (worst symptom burden). Moffitt uses a modified version of the ESAS that includes two additional symptoms: constipation and sleep disturbance [26]. The ESAS is administered at all in-person visits (e.g., treatment and surveillance visits) at Moffitt's Radiation Oncology clinic (over 99% completion rate). Patients are given a tablet to complete the ESAS questionnaire while they are waiting for their clinic visit and receive assistance from a medical assistant to ensure the questionnaire is complete. The data is automatically scored through a Cerner PowerForm^™ (Cerner Corporation, Kansas City, MO) and integrated into the electronic health record (EHR). All ESAS items were summed and combined into one overall score (potential range: 0 to 121). We chose this approach as opposed to analyzing separate symptom scores given the high collinearity among ESAS symptoms [25–27]. We conducted a sensitivity analysis to determine whether study results differed based on use of the original ESAS versus the modified ESAS. The study results did not differ; therefore, we used the modified version of the ESAS.

Additional covariates.

We included covariates that may affect lung cancer survival and symptom burden including patient demographics (e.g., gender, age, race, ethnicity, marital status), clinical characteristics (e.g., comorbidity burden, maximum tumor diameter, stage), and risk factors (e.g., smoking status). We did not include treatment characteristics (e.g., radiation dose) because there was little variation across the sample (Table 1). Co-morbidity was measured using a weighted Elixhauser score [28], which was computed using the R Comorbidity package using ICD-10-CM (International Classification of Diseases, Tenth Revision, Clinical Modification) codes.[29] The Elixhauser was chosen because it accounts for a greater number of comorbidities and has demonstrated higher classification accuracy for predicting mortality compared to other comorbidity indices [30]. Smoking status included three categories: 1) current smoker, 2) former smoker, and 3) non-smoker.

Table 1.

Sample characteristics

Characteristics	N = 510
Age, median (range)	73.8 (46.3,94.6)
Gender, N (%)
Female	258 (50.6%)
Male	252 (49.4%)
Race and ethnicity, N (%)
Non-Hispanic White	479 (93.9%)
Other	31 (6.08%)
Smoking status, N (%)
Never	43 (8.43%)
Current	108 (21.2%)
Former	359 (70.4%)
Marital Status, N (%)
Married	306 (60.0%)
Divorced or separated	56 (11.0%)
Single	52 (10.2%)
Widowed	96 (18.8%)
Elixhauser Comorbidity Index, median (range)	13.0 (2.0,42.0)
Maximum tumor diameter (cm), median (range)	1.8 (0.4,6.2)
Stage
I	478 (93.7)
II	32 (6.27)
Planned dose
4800 cGy	23 (4.51%)
5000 cGy	474 (92.9%
6000 cGy	13 (2.55%)
Dose per fraction
750 cGy	10 (1.96%)
1000 cGy	474 (92.9%)
1200 cGy	26 (5.10%)
Total planned fractions
4	23 (4.51%)
5	477 (93.5%)
6	10 (1.96%)
ESAS summary score, median (range)	16 (0, 115)

Analytic approach

All analyses were completed in RStudio (R version 3.6.3) from January 2021 – July 2021. There were 572 eligible patients. We excluded patients with data that were likely entered incorrectly into the electronic health record (e.g., ESAS date after date of death) and patients with missing values on variables of interest (e.g., missing date of death) (Figure 1).

Figure 1.

Exclusion of non-sensical or missing data

Descriptive statistics were produced for all variables. For longitudinal predictors, we produced spaghetti plots of scores over time and frequencies on the number of observations per patient (Supplemental File). Given that longitudinal measures of ESAS are likely correlated with time-to-death (i.e., presence of endogeneity), we selected a joint probability model for the primary outcome, time-to-death, and the longitudinal ESAS measurements [31]. The joint distribution of the event times and the longitudinal measurements is modelled via a set of random effects that are assumed to account for the associations between these two outcomes. The joint probability model also deals with data sparseness (e.g., uneven number of ESAS measurements across the cohort). The model was set a priori to include all static covariates in both the longitudinal and survival components to assess the magnitude of any potential effects. Modeling was done using the JM R package [32]. The package computes separate mixed models for the longitudinal predictors, a standard Cox proportional hazards model for time-to-death, and the joint model. Coefficients (β), hazard ratios (HR), 95% confidence intervals (CI), and P values were reported for each model. P values were 2-sided, and statistical significance was set at P < .05. Below we report the results of three models: 1) the marginal linear mixed effects model; 2) the marginal Cox proportional hazards model; and 3) the joint probability model. This study was approved by the Advarra Institutional Review Board (IRB # 18883). The study was reported in concordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [33].

RESULTS

The final sample included 510 early-stage NSCLC patients. The median age was 73.8 (range: 46.3–94.6) (Table 1). Most patients were Non-Hispanic White (93.9%), female (50.6%), married (60.0%), former (70.4%) or current smokers (21.2%), and had stage 1 NSCLC (93.7%). The median maximum tumor diameter was 1.80cm (range: 0.40–6.20cm). The median Elixhauser Comorbidity index score was 13.0 (range: 2.00–42.0). The median ESAS score was 16 (range: 0–115). Patients had a total of 1748 ESAS observations. The median time from last ESAS score and death was 7.6 months for the 190 patients who died in the cohort.

The marginal linear mixed effects model suggests that males had significantly lower ESAS scores (lower symptom burden) compared to females (β = −3.08; P = 0.03) (Table 2). Patients who currently smoke had significantly higher ESAS scores compared to non-smokers (β = 6.85; P = 0.02). Patients with a higher Elixhauser score (higher comorbidity burden) had significantly higher ESAS scores (β = 0.33; P = 0.003). Other covariates (e.g., age, race/ethnicity, marital status, tumor diameter, and stage) were not significantly associated with ESAS scores.

Table 2.

Linear mixed effects model examining factors associated with ESAS score, N=510

Characteristics	β	SE	P-value
Age	−0.17	0.09	0.06
Male gender	−3.08	1.41	0.03
Other race/ethnicity	2.53	2.88	0.39
Smoking status
Never
Current	6.85	2.85	0.02
Former	2.15	2.49	0.39
Marital status
Married
Divorced or separated	2.86	2.23	0.20
Single	−2.64	2.40	0.27
Widowed	3.27	1.94	0.09
Elixhauser comorbidity Index	0.33	0.11	0.003
Maximum tumor diameter	1.61	0.82	0.05
Stage
I
II	−4.01	3.24	0.22
Intercept	22.53	7.20	0.001

There were 190 deaths. The median overall survival was 3.9 years (range: 3.6–4.6) and the median follow-up time was 3.1 years (95% CI: 2.84–3.39). The marginal cox proportional hazards model (with cross-sectional predictors only) suggested that males had significantly worse survival than females (HR: 1.57; 95% CI: 1.16–2.13) (Table 3). Patients who currently smoked had significantly worse survival than patients who had never smoked (HR: 2.93; 95% CI: 1.44–5.97). Older age was associated with significantly worse survival (HR: 1.03; 95% CI: 0.97–1.01). Larger maximum tumor diameter was associated with significantly worse survival (HR: 1.25; 95% CI: 1.05–1.50). Other covariates (e.g., race/ethnicity, marital status, comorbidity score, stage) were not significantly associated with survival.

Table 3.

Cox proportional hazards model examining factors associated with survival, N=510

Characteristics	HR	95% CI	P-value
Age	1.03	(0.97–1.01)	0.001
Male gender	1.57	(1.16–2.13)	0.003
Other race/ethnicity	1.33	(0.77–2.31)	0.30
Smoking status
Never
Current	2.93	(1.44–5.97)	0.003
Former	1.55	(0.80–3.00)	0.19
Marital status
Married
Divorced or separated	1.10	(0.68–1.79)	0.69
Single	1.37	(0.84–2.25)	0.21
Widowed	1.00	(0.66–1.51)	0.99
Elixhauser comorbidity Index	1.02	(1.00–1.04)	0.09
Maximum tumor diameter	1.25	(1.05–1.50)	0.01
Clinical stage
I
II	0.54	(0.26–1.09)	0.09

The estimates from the joint probability model of the longitudinal PROs (mixed model) and the time-to-event (survival component) data are presented in tables 4 and 5 respectively. The mixed model component of the joint model estimating predictors of ESAS scores had similar results to the linear mixed effects model. Men had lower ESAS scores (β = −2.64; P = 0.03) compared to women while patients who reported currently smoking (β = 6.27; P = 0.01), and patients with higher comorbidity burden (β = 0.31; P = 0.001) had significantly higher ESAS scores. A larger maximum tumor diameter (β = 1.75; P = 0.02) was associated with significantly higher ESAS scores. Other covariates (e.g., age, race/ethnicity, marital status, stage) were not significantly associated with ESAS scores.

Table 4.

Longitudinal process estimates of factors associated with ESAS score, N=510

Characteristics	β	SE	P-value
Age	−0.15	0.08	0.06
Male gender	−2.64	1.25	0.03
Other race/ethnicity	2.71	2.89	0.35
Smoking status
Never
Current	6.27	2.45	0.01
Former	1.76	2.09	0.40
Marital status
Married
Divorced or separated	2.27	2.26	0.26
Single	−2.71	2.36	0.25
Widowed	3.30	1.72	0.06
Elixhauser comorbidity Index	0.31	0.09	0.001
Maximum tumor diameter	1.75	0.75	0.02
Stage
I
II	−3.36	2.80	0.23
Intercept	19.27	6.48	0.001

Table 5.

Event process estimates of factors associated with survival, N=510

Characteristics	HR	95% CI	P-value
ESAS score	1.04	(1.02– 1.05)	<0.0001
Age	1.04	(1.03– 1.05)	<0.0001
Male gender	1.87	(1.36– 2.57)	0.0001
Other race/ethnicity	1.16	(0.65– 2.07)	0.61
Smoking status
Never
Current	2.39	(1.25– 4.56)	0.008
Former	1.48	(0.80– 2.75)	0.21
Marital status
Married
Divorced or separated	0.92	(0.55– 1.51)	0.73
Single	1.53	(0.91– 2.56)	0.11
Widowed	0.92	(0.61– 1.40)	0.70
Elixhauser comorbidity Index	1.01	(0.99– 1.03)	0.39
Maximum tumor diameter	1.21	(1.01– 1.46)	0.04
Stage
I
II	0.58	(0.28–1.17)	0.13

The survival component of the joint model demonstrates that longitudinal changes in ESAS scores are significantly associated with worse survival (HR: 1.04; 95% CI: 1.02–1.05). This finding suggests that a one-unit increase in ESAS summary score (indicating worse symptom burden) increased the probability of death by 4%. The joint model found that male gender was significantly associated with worse survival (HR: 1.87; 95% CI: 1.36–2.57). Older age (HR: 1.04; 95% CI: 1.03–1.05) was significantly associated with worse survival. Individuals who reported smoking at diagnosis had significantly worse survival than patients who had never smoked (HR: 2.39; 95% CI: 1.25–4.56). Larger tumor diameter (HR: 1.21; 95% CI: 1.01–1.46) was significantly associated with worse survival. Other covariates (e.g., race/ethnicity, marital status, comorbidity score, stage) were not significantly associated with survival, similar to the marginal Cox proportional Hazards model (which did not account for changes in ESAS scores).

DISCUSSION

The goal of this study was to assess whether longitudinal changes in PROs have prognostic significance for estimating survival in early-stage lung cancer patients receiving SBRT. Overall, our study found that worsening symptom burden over time is associated with survival, even after accounting for other factors. This is the first study to our knowledge to examine the association between longitudinal changes in PROs and survival in early-stage lung cancer patients. PROs are increasingly being collected in routine care delivery and integrated within the EHR to ensure symptoms are adequately managed among cancer patients [34]. These real-world data could be leveraged for secondary purposes, such as predicting survival and other patient outcomes, and guiding the development of clinical decision support tools.

Prior studies have demonstrated that PROs captured in the ESAS (e.g., drowsiness, pain, shortness of breath) as well as overall ESAS score are associated with survival in advanced lung cancer patients [35]. ESAS has also been shown to predict survival in patients with other advanced cancers (e.g., breast, prostate) [36]. The current study builds upon this prior work by demonstrating that the ESAS is a useful prognostic tool for predicting survival among early-stage lung cancer patients. Future studies might test whether the ESAS can predict survival in other patient populations with early-stage cancers or other patient outcomes that may be associated with deterioration in health status (e.g., emergency department [ED] utilization). For example, one cross-sectional study found that certain moderate-to-severe ESAS symptoms (e.g., drowsiness, shortness of breath) were associated with ED utilization among cancer patients receiving palliative care [37]. Future studies could explore whether longitudinal changes in PROs are predictive of ED utilization, given many cancer patients visit the ED multiple times at end-of-life [38]. Like prior studies, our results affirm that patient demographics (e.g., male gender, older age) and risk factors (e.g., current smoking status) negatively affect survival among lung cancer patients [39].

Our study found that certain patient characteristics (e.g., greater comorbidity burden) are associated with greater symptom burden, consistent with prior research. Other studies suggest that comorbidity burden is associated with greater symptom severity in early-stage and advanced lung cancer patients [40]. Research outside of cancer care suggests that certain comorbidities (e.g., chronic obstructive pulmonary disease) are associated with greater symptom severity (e.g., pain, shortness of breath) compared to other comorbidities [41]. Future studies could explore whether certain comorbidity clusters are associated with greater symptom severity and worse survival outcomes among lung cancer patients. Prior work suggests that current comorbidity indices--which capture number and severity of comorbidities--may not accurately capture comorbidity complexity (e.g., interactions between comorbidities) [42]. Additional research could explore the relationships among comorbidity complexity, symptom severity, and survival outcomes. Our results indicate that women report greater symptom severity compared to men, consistent with past research [43]. We also observed that lung cancer patients who currently smoke report worse symptom severity than patients with no smoking history, consistent with prior research [44].

While our study establishes that longitudinal PROs are predictive of survival, more research is needed to understand what events might help explain the relationship between symptom severity and survival. For example, survival among patients with advanced disease can be mediated by acute complications (e.g., pneumonia) [45]. Therefore, it is possible that an acute event may trigger worsening of symptoms and signal clinical deterioration. In addition to acute complications, there may be changes in a patients’ health status over time that are likely to affect symptom severity and survival, such as change in comorbidity burden, frailty, or progressive weight loss [46] [47]. More research is needed to examine factors that may drive progressive worsening of PROs over time among cancer patients.

Limitations

This study has several limitations. First, the study population is predominately non-Hispanic White. Limited racial and ethnic diversity makes it difficult to fully assess the role that race, and ethnicity may play in longitudinal symptom burden and survival. Prior studies suggest that there are racial disparities in symptom burden in cancer patients [48]. Our study is underpowered to detect these differences. Second, our dataset does not have a measure of education or individual-level income, which may play a role in estimating survival [49]. Future studies should account for social determinants of health that are likely to affect lung cancer survival while accounting for smoking status. For example, indices such as the Area Deprivation Index may account for other socioeconomic disparities that drive differences in lung cancer survival (e.g., transportation access) [50]. Third, our study is limited to one healthcare setting and needs to be externally validated in additional healthcare systems. Fourth, we characterized overall smoking status but were unable to account for a more detailed smoking history (e.g., pack years), which is likely to affect lung cancer survival. Fourth, our dataset relies on PROs collected in routine care delivery and has notable limitations (e.g., data sparseness). Fifth, our model does not assess whether baseline PROs are associated with survival—an important area of study for future research. Despite these limitations, this study makes a valuable contribution to the literature by being the first study to assess longitudinal changes in PROs and survival among early-stage lung cancer patients.

CONCLUSION

PROs are increasingly being collected as a part of routine care delivery to improve symptom management. Healthcare systems can integrate these data with other real-world data (e.g., electronic health record) to predict patient outcomes, such as survival. Capturing longitudinal PROs—in addition to PROs at diagnosis—may add prognostic value for estimating survival and help identify deterioration in a patient’s health status. Future studies should build upon this work by integrating PROs into prognostic tools that can be automated in real-time and inform clinical decision making at the point of care.

Study Highlights:

• Longitudinal changes in patient reported outcomes are associated with increased mortality

• Future studies should explore use of patient-reported outcomes data in prognostication tools