Exploring the Relationship of Radiation Dose Exposed to the Length of Esophagus and Weight Loss in Patients with Lung Cancer

Abstract

Purpose:

We investigate whether esophageal dose–length parameters (L_dose) can robustly predict significant weight loss—≥5% weight loss during radiation therapy (RT) compared with the weight before RT—in patients with lung cancer treated with definitive intent.

Methods and Materials:

Patients with lung cancer treated with conventionally fractionated RT between 2010 and 2018 were retro-spectively identified. L_Fdose and L_Pdose, the length of full- and partial-circumferential esophagus receiving greater than a threshold dose in Gy, respectively, were created. Multivariate logistic regression examined the associations between individual L_dose and weight loss after adjusting for clinical parameters and correcting for multiple comparisons. Ridge logistic regression examined the relative importance of L_dose compared with dose–volume (V_dose), mean dose (D_mean), and clinical parameters in determining weight loss. Univariate logistic regression examined the unadjusted probability of weight loss for important L_dose parameters.

Results:

Among the 214 patients identified, median age was 66.9 years (range, 31.5–88.9 years), 50.5% (n = 108) were male, 68.2% (n = 146) had stage III lung cancer, median RT dose was 63 Gy (range, 60–66 Gy), and 88.3% (n = 189) received concurrent chemotherapy. Esophagus lengths receiving high full-circumferential (L_F50-L_F60) and high partial-circumferential doses (L_P60) were associated with significant weight loss (P ≤.05). L_F65 and L_P65 reached near significance (P = .06 and .053, respectively). L_F65 > L_F60 > L_P65 were the most important dose parameters in determining weight loss compared with other L_dose, V_dose, and D_mean parameters.

Conclusions:

Esophageal L_dose parameters are an efficient way of interpreting complex dose parameters in relation to weight loss toxicity among patients with lung cancer receiving definitive RT.

Introduction

Treatment-related weight loss is strongly associated with worse survival among patients with locally advanced lung cancer treated with definitive radiation therapy (RT).^1,2 According to the National Comprehensive Cancer Network guidelines, weight loss of 5% or more during treatment compared with weight before RT is considered severe³ and may necessitate supportive clinical interventions during RT.^4,5 Consequently, identifying patients who may be at increased risk for significant weight loss before the start of RT is important for monitoring and timely clinical intervention.

The reported incidence of grade 2+ esophagitis ranges from 27% to 39% during definitive RT, with a higher incidence among patients receiving concurrent chemotherapy and when the location of the esophagus is within or adjacent to the RT field.^1,2,6,7 Several studies have examined the association between RT dosimetric parameters and treatment-related esophagitis in an attempt to identify the most predictive RT parameter for esophageal toxicity. These parameters include dose to the esophageal surface area,⁸ mean esophageal dose,⁶ and percent volume of esophagus irradiated.^8,9 According to the Quantitative Analyses of Normal Tissue Effects in the Clinic, limiting exposure of the whole esophagus, including the mean esophageal dose (D_mean <34 Gy) and percent volumetric dose parameters (V₃₅ <50%, V₅₀ <40%, V₇₀ <20%), will lower rate of grade 2+ or 3+ esophagitis after RT to less than 20% to 30%.¹⁰ Although these observations show strong association between RT parameters and esophagitis, inconsistencies in esophageal contouring practices may affect the accuracy of mean and relative volumetric esophageal constraints in predicting treatment-related esophageal toxicity.¹¹

We hypothesized that a length parameter describing the length of esophagus exposed to a threshold dose may more robustly characterize RT-related esophageal toxicity. The length of esophageal dose exposure is a visually intuitive dosimetric parameter that is independent of the delineation of the whole esophagus and may more effectively reflect the esophageal dose exposure for thoracic malignancies. Moreover, studies are needed to characterize the relative importance of esophageal dosimetric parameters in relation to clinical parameters to best identify patients with lung cancer at highest risk of significant weight loss from treatment.

Herein, we introduce an esophageal dose–length parameter (L_dose), which summarizes the length of esophagus receiving a threshold dose circumferentially. We then assess (1) the association and (2) the robustness of esophageal L_dose compared with esophageal volumetric parameters, esophageal mean dose, and clinical factors in predicting treatment-related weight loss.

Methods and Materials

Patients

Consecutive patients with lung cancer treated with conventionally fractionated RT (1.8–2.0 Gy/d) at a single institution from 2010 to 2018 were identified as a part of an institutional review boardeapproved database. Patients were treated with 3-dimensional conformal RT, intensity modulated RT, and volumetric modulated arc therapy. All patients underwent 4-dimensional planning scans. The treating physician verified organ-at-risk contours on the free-breathing scan before RT planning. The esophagus was delineated from the region of the postcricoid esophagus to the gastroesophageal junction. RT was delivered with daily cone beam image guidance. Patients treated with twice-daily radiation or with unavailable electronic radiation plans were excluded.

Demographic, clinical, and treatment data were retrospectively obtained from patient medical records. Baseline demographic and clinical data included age, sex, race, marital status, body mass index (BMI; kg/m²), pre-RT albumin (g/dL), pretreatment weight loss (yes, no), Eastern Cooperative Oncology Group (ECOG) performance status, and Charlson Comorbidity Index (CCI). Treatment data included use of concurrent systemic therapy (none, sensitizing chemotherapy [eg, weekly carboplatin/paclitaxel], or full-dose chemotherapy [eg, carboplatin/pemetrexed, cisplatin/etoposide, cisplatin/pemetrexed, cisplatin/vinblastine]), targeted agents, and immunotherapy. RT data included 3-dimensional dose, esophagus contours, planning target volumes (PTVs, cm³), esophageal doseevolume parameters (V_dose, the percent volume of the esophagus receiving at least x Gy), and esophageal mean dose (D_mean).

Weight loss and dose length parameter (L_dose)

Significant weight loss was defined as ≥5% weight loss at completion of RT relative to baseline weight. The baseline weight was measured within 7 days before the start of treatment.

A radiomorphologic feature generation pipeline¹² was used to create a feature set detailing dose exposed to the contoured esophageal subvolumes. First, we segmented the esophagus contours extracted from delivered RT plans into slices (subvolumes) in the axial 2-dimensional plane, based on computed tomography (CT) slice thickness. Next, we extracted dose parameters for each subvolume. Full-circumferential dose (F_dose) and partial-circumferential (P_dose) dose were defined as the dose delivered to 90% and 50% of the specified esophageal subvolume, respectively. Lastly, the L_dose parameters were created by summing the number of subvolumes times the slice thickness that received partial or full-circumferential dose exposures (Fig E1, available online at https://doi.org/10.1016/j.prro.2020.03.002). L_F(x) and L_P(x) denoted the length of full- and partial-circumferential esophagus receiving greater than a threshold dose of x Gy, respectively.

Statistical analysis

We created dose–length histograms (DLHs) to visualize the relationship between L_dose parameters and development of significant weight loss. Next, we stratified aggregated L_dose parameters by severity of weight loss experienced (<0%, 0% to <5%, 5% to <10%, and ≥10%) using generalized additive models.

We evaluated associations between individual L_dose parameters and weight loss, after adjusting for clinical variables and multiple comparisons, using multivariate logistic regression with Bonferroni corrections. Multivariate logistic regression models were adjusted for the following clinical parameters: age (years), sex (female vs male), race (white vs nonwhite), marital status (partnered vs nonpartnered), BMI (kg/m²), albumin (<3 g/dL vs ≥3 g/dL), ECOG performance status, CCI, PTV (cm³), and concurrent chemotherapy type (no, sensitizing dose, full dose).^13,14 P-adjusted values were calculated given Bonferroni corrections. Higher odds ratios indicated greater risk of significant weight loss.

Weight loss prediction performance of L_dose, V_dose, and D_mean parameters, also adjusted for clinical variables, was determined using logistic regression with ridge regularization.¹⁵ Ridge logistic regression regularizes coefficients by assigning larger weights to more important parameters, thus making it the preferred method when there is high correlation between L_dose parameters and V_dose parameters.¹⁵ Parameters’ coefficients (β) were learned by regularization hyperparameter tuning using 10-fold cross-validation. β indicates how a single unit increase in a parameter influences a change in the “odds” of significant weight loss on a log scale. β >0 suggests that the increase in the parameter is correlated with increased probability of significant weight loss, and β <0 suggests that the increase in the parameter is correlated with decreased probability of significant weight loss. We ranked the dose parameters and the clinical parameters by the magnitude of each parameter’s coefficients (β) relative to the most important respective parameter found on ridge logistic regression. All the input features were standardized.

To determine how L_dose compares to V_dose in predicting weight loss, 3 comparative ridge logistic regression models were fit to predict weight loss using (1) L_dose, (2) V_dose + D_mean, and (3) L_dose + V_dose + D_mean separately. We used nested cross-validation to assess the model performance of L_dose compared with V_dose in predicting weight loss.¹⁶ Cross-validation was performed with 10 folds in the inner loop and 10 folds in the outer loop. Weight loss prediction performances were measured using area under the curve, where higher area under the curve indicates better model performance.

Lastly, we calculated the probability of significant weight loss using the most important L_dose parameters identified by multivariate logistics regression analysis. Unadjusted probability of weight loss was determined using a univariate logistic regression.

All statistical tests were 2-sided, and a P < .05 was considered to be statistically significant.

Results

Patient characteristics

Two hundred and fourteen patients with lung cancer treated with conventionally fractionated RT between 2010 and 2018 were identified. The median age was 66.9 years (range, 31.5–88.9 years). The majority of patients were male 50.5% (n = 108), and 49.5% (n = 106) were white. At the time of initial presentation, 68.2% (n = 146) of patients had stage III disease, 99.1% (n = 212) had ECOG performance status ≤2, 52.6% (n = 120) had BMI <25 kg/m², 65.9% (n = 141) had concomitant severe comorbid disease (CCI ≥5), and 59.1% (n = 124) were partnered.

The median RT dose was 63 Gy (range, 60–66 Gy), and 99.1% (n = 208) received intensity modulated RT/volumetric modulated arc therapy with daily image guidance. A majority of patients (88.3%, n = 189) received concurrent chemotherapy: 62% full-dose and 37.6% (n = 71) sensitizing chemotherapy. The median length of contoured esophagus was 25.2 cm (range, 15–37.2 cm). Figure 1 highlights the variations in contoured esophageal lengths that exist in clinical practice.

Figure 1.

The variations in esophageal length seen in clinical practice. The esophageal length varied in the study population (median,2.5 cm; range, 15–37.2 cm). For each interval on the x-axis, the parenthesis demarcates an excluded endpoint value, whereas the bracket demarcates an included endpoint value. For example, (17.7, 20.4] is the interval of real numbers between 17.7 and 20.4, excluding 17.7 and including 20.4.

Clinical characteristics and weight loss

Of the cohort, 25.2% of patients (n = 54) developed significant weight loss. Of the 54 patients who developed ≥5% weight loss, 20.4% (n = 11) and 3.7% (n = 2) of patients developed >10% and >15% weight loss relative to baseline status, respectively. Of note, no patient required a feeding tube during RT due to treatment toxicity.

Patients who developed significant weight loss had larger mean PTV (358 cm³ vs 265 cm³; P = .032) and greater percent volumetric and mean RT dose exposure to the esophagus (V₃₅ = 33% vs 23%; V₅₀ = 37% vs 13%; V₆₀ = 16% vs 2%; D_mean = 22.7 Gy vs 16.8 Gy; all P <.001) compared with patients who did not develop significant weight loss. There were no other differences in clinical or treatment characteristics between those who developed significant weight loss and those who did not (P > .05; Table 1).

Table 1.

Patient characteristics by weight-loss group

Variable	No significant weight loss (N = 160)	Significant weight loss (N = 54)	P
Age, y*	66.6 (58.9–73.3)	68.9 (58.8–75.7)	.43
Race†			.53
White	99 (61.9)	36 (66.7)
Sex			.81
Male	80 (50.0)	28 (51.9)
Marital status			.46
Partnered	95 (60.5)	29 (54.7)
BMI, kg/m2	26.5 (5.4)	26.2 (8.0)	.24
Overall stage
I	15 (9.4)	2 (3.7)	.11
II	18 (11.3)	4 (7.4)
III	102 (63.7)	44 (81.5)
IV	25 (15.6)	4 (7.4)
Pretreatment albumin, g/dL			.24
≥3	145 (92.4)	47 (87.0)	.64
Pretreatment weightless
Yes	65 (40.6)	20 (37.0)
ECOG			.15
0–1	144 (90.0)	52 (96.3)
2+	16 (10.0)	2 (3.7)
Charlson Comorbidity Index
0	9 (5.6)	2 (3.7)
1 and 2	4 (2.5)	2 (3.7)
3 and 4	41 (25.6)	15 (27.8)
≥5	106 (66.3)	35 (64.8)
Concurrent chemotherapy
No	22 (13.8)	3 (5.6)
Concurrent	138 (86.3)	51 (94.4)
PTV, cm3	265.3 (169.0, 454.3)	357.7 (248.8, 479.6)	.03
Total RT dose, Gy	63 (60–66)	63 (61.2–66)	.34
Dose per fraction			.26
180 cGy	66 (41.3)	27 (50.0)
200 cGy	94 (58.8)	27 (50.0)
Esophagus V35	23.4 (5.7–32.6)	32.5 (25.7–41.2)	<.001
Esophagus V50	13.2 (0.3–24.2)	26.6 (16.4–33.4)	<.001
Esophagus V60	2.3 (0.0–13.8)	16.4 (1.5–25.5)	<.001
Esophagus V70	0.0 (0.0–0.0)	0.0 (0.0–0.0)	.50
V35 ≥50%	4 (2.5)	2 (3.7)	.64
V50 ≥40%	25 (15.6)	16 (29.6)	.02
Mean esophagus dose, Gy	16.8 (1004–2235)	22.7 (17.3–27.6)	<.001
Mean esophagus dose over 34 Gy	2 (1.3)	1 (1.9)	.75

Abbreviations: BMI = body mass index; ECOG = Eastern Cooperative Oncology Group; IQR = interquartile range; PTV = planning target volume; RT = radiation; SD = standard deviation; Vx = the percent volume of esophagus receiving x dose.

The values in bold indicate P < .05.

^*All continuous variables are displayed as mean (SD) or median (IQR).

^†All categorical variables are displayed as n (%).

Esophageal L_dose parameter and weight loss

The length of esophagus receiving a threshold circumferential dose in relation to degree of weight loss was examined descriptively in a DLH (Fig 2). There was separation of the aggregated dose–length parameters for patients experiencing different degrees of weight loss (<0%, 0% to <5% vs 5% to <10%, >10%). Patients who developed significant weight loss had longer lengths of full- and partial-circumferential esophagus exposed to each dose level compared with patients who did not experience significant weight loss. Correspondingly, patients who developed significant weight loss were exposed to higher RT dose at each full- and partial-circumferential esophageal length compared with patients who did not experience significant weight loss (Fig 2).

Figure 2.

The aggregated esophageal dose–length parameters L_dose, stratified by the severity of weight loss. Gray zones indicate the 95% confidence interval for each group respectively. A point on the dose–length histogram represents the length of esophagus (y-axis) receiving greater than a threshold dose of x Gy (x-axis). (a) L_Fx full-circumferential dose–length parameter, (b) L_Px partial circumferential dose–length parameter.

Importance of L_dose relative to clinical factors in predicting for weight loss

In multivariate analysis, the lengths of full-circumferential esophagus receiving high doses between 50 and 60 Gy were associated with weight loss (L_F50, L_F55, and L_F60; P < .05; Table E1, available online at https://doi.org/10.1016/j.prro.2020.03.002). The lengths of partial-circumferential esophagus receiving 60 Gy was associated with weight loss (L_P60; P < .05; Table E1, available online at https://doi.org/10.1016/j.prro.2020.03.002). The association between weight loss and lengths of full-circumferential esophagus receiving 65 Gy was marginally significant (L_F65; P = .06; Table E1, available online at https://doi.org/10.1016/j.prro.2020.03.002). The association between weight loss and lengths of partial-circumferential esophagus receiving 55 Gy and 65 Gy was marginally significant (L_P55 and L_P65; both P = .053; Table E1, available online at https://doi.org/10.1016/j.prro.2020.03.002). Other clinical variables were not significantly associated with weight loss.

Ranked comparison of L_dose, V_dose, and clinical parameters in predicting weight loss

We fitted a ridge logistic regression model to rank the relative importance of dose–length parameters to clinically used RT parameters (V_dose: V₃₅, V₅₀, V_60, V₇₀; esophagus D_mean) and clinical parameters (Table 2, Table E2, available online at https://doi.org/10.1016/j.prro.2020.03.002). Pretreatment albumin was the most important clinical variable was (β = –0.001, indicating that lower baseline albumin increased the chance of developing significant weight loss; Table E2, available online at https://doi.org/10.1016/j.prro.2020.03.002). L_F65 was the most important RT dose parameter (β = 0.001, Table 2, column 2) and the second most important clinical parameter. L_F60 was the next most important RT dose parameter (β = 0.0005). Dosimetric variables were ranked separately among other dosimetric parameters because these variables, unlike clinical variables, are potentially modifiable at the time of treatment planning. Among RT dose parameters, L_dose parameters notably ranked higher in importance than V_dose (highest relative rank is 12) or D_mean parameters (highest relative rank is 0) in predicting weight loss. The prediction performance of the model using L_dose parameters is similar to the model using V_dose + D_mean and the model using L_dose + V_dose + D_mean (Table E3, available online at https://doi.org/10.1016/j.prro.2020.03.002).

Table 2.

The ranked importance of esophageal-specific L_dose, V_dose, and D_mean on weight loss

Parameters	β coeffcient*	Importance among RT parameters only	Importance among RT and clinical parameters
LF65	9.78E-04	100	84
LF60	5.25E-04	54	45
LP65	4.89E-04	50	42
LF55	4.10E-04	42	35
LP60	3.59E-04	37	31
LF50	2.75E-04	28	24
LP55	2.65E-04	27	23
LF45	2.07E-04	21	18
LP50	2.00E-04	20	17
LF35	1.94E-04	20	17
LF30	1.93E-04	20	17
LF40	1.78E-04	18	15
LP45	1.75E-04	18	15
LF25	1.59E-04	16	14
LP40	1.57E-04	16	13
LP35	1.45E-04	15	12
LF20	1.33E-04	13	11
LP30	1.28E-04	13	11
LF5	1.23E-04	12	10
V60	1.23E-04	12	10
LF10	1.21E-04	12	10
LP20	1.11E-04	11	9
LP25	1.11E-04	11	9
LF15	1.10E-04	11	9
LP15	1.10E-04	11	9
LP5	1.09E-04	11	9
LP10	1.06E-04	11	9
V50	8.30E-05	8	7
V35	7.00E-05	7	6
V70	−1.40E-05	1	1
Mean esophagus dose	1.40E-06	0	0
PTV	2.10E-06	0	0

Abbreviations: L_Fdose and L_Pdose = the length of full- and partial-circumferential esophagus receiving greater than a threshold dose in Gy, respectively; PTV planning treatment volume; RT = radiation therapy.

^*β coefficients from ridge logistic regression models used to rank the importance of dosimetric parameters in relation to RT and clinical parameters (pretreatment albumin, performance status, concurrent chemotherapy, body mass index, comorbidity index, age, sex, marital status

Probability of significant weight loss due to important L_dose parameters alone

The probabilities of weight loss associated with important dose–length parameters identified on multivariate analyses and ridge regression (L_F65, L_F60, L_F55, L_F50, and L_P65, L_P60, L_P55) are shown in Figure 3. The probabilities of significant weight loss for L_F65 at 3 cm, 2 cm, and 1 cm are 54%, 42%, and 31%, respectively (Fig 3a; Table E4, available online at https://doi.org/10.1016/j.prro.2020.03.002). The probabilities of significant weight loss for L_P65 at 3 cm, 2 cm, and 1 cm are 35%, 30%, and 25%, respectively (Fig 3b; Table E4, available online at https://doi.org/10.1016/j.prro.2020.03.002) and for L_F60 at 3 cm, 2 cm, and 1 cm are 33%, 27%, and 22%, respectively (Fig 3a; Table E4, available online at https://doi.org/10.1016/j.prro.2020.03.002).

Figure 3.

Probability of ≥5% weight loss. (a) L_Fdose and (b) L_Pdose, the length of full- and partial-circumferential esophagus receiving greater than a threshold dose in Gy, respectively. Solid lines indicate full-circumferential dose, whereas dashed lines indicate partial-circumferential dose. *P < .05. L_F65 reached near significance (P = .06) and was the most important dosimetric parameter by ridge logistic regression. L_P65 and L_P55 reached near significance, P = .053.

Discussion

In this series of patients with lung cancer who developed significant weight loss ≥5% at completion of definitive conventionally fractionated radiation, we demonstrate that a range of esophageal dose–length parameters (L_F50-L_F60 and L_P60, the length of full- and partial-circumferential esophagus receiving greater than a threshold dose of 50–60 Gy and 60 Gy, respectively) are associated with significant weight loss. Among this range of potentially important dose–length parameters, we show that L_F65 > L_F60 > L_P65 were ranked the most important RT dose parameters in determining weight loss. Notably, these specific L_dose parameters were relatively more important in predicting weight loss than commonly used esophageal dose parameters such as V₆₀ and mean esophageal dose. Lastly, overall model performance of L_dose was comparable to V_dose in predicting weight loss.

Prior studies have evaluated the length of esophagus irradiated in relation to esophagitis,^17,18 but few have evaluated L_dose in relation to weight loss. A retrospective study of 50 patients with lung cancer examined the association between esophageal dose–length parameters and significant weight loss but only focused on the univariate impact of dose–length parameters because of the limited sample size.⁴ Our results build on this study in that we demonstrate the multivariate impact and evaluate the relative importance of L_dose parameters in the context of clinically used V_dose parameters, D_mean, and comorbid clinical parameters for predicting weight loss. Furthermore, to increase clinical applicability of L_dose, we provide threshold probabilities of weight loss for each incremental increase in important L_dose parameters identified in multivariate analysis. For example, patients with L_F65 ≥1 cm, L_P65 ≥2 cm, and L_F60 ≥3 cm were likely to have significant weight loss probability over 30%. Our findings support that an increased risk of toxicity is associated with dose escalation above 60 Gy for patients with non-small cell lung carcinoma (NSCLC) receiving definitive RT.² If high RT dose exposure to the esophagus is unavoidable, contralateral sparing of the esophagus from 60 Gy is desirable.¹⁹

D_mean and V_dose parameters are susceptible to variation in esophageal contouring practice.¹¹ This study highlights that the length of esophagus contours from delivered radiation plans demonstrates real-life variations (range, 15.0–37.2 cm) and reflects differences patient heights and contour consistency. The L_dose calculation method proposed using our radiomorphologic feature generation pipeline is robust because it depends on the absolute length of maximum dose exposure from each segmented “slice” of the esophagus. L_dose can be applied to any partially contoured at-risk esophageal volume and is not dependent on the definition of the full esophagus contour.

Several points deserve further consideration. We compared the ranked importance of L_dose to V_dose and D_mean in predicting an objective outcome of weight loss, a different outcome than esophagitis. Of note, the majority of D_mean and V_dose parameters have been validated for esophagitis but not for weight loss. For example, esophageal D_mean, high-dose volumetric parameters (V₆₀ as a continuous variable), and Quantitative Analyses of Normal Tissue Effects in the Clinic constraints (D_mean <34 Gy, V₃₅ >50%, V₅₀ >40%, V₇₀ >20%) predict for acute esophagitis after RT.^2,8,10,20,21 In a meta-analysis of published retrospective series of patients with NSCLC who received definitive RT, among V_dose and D_mean parameters, V₆₀ was the best predictor of esophagitis on multivariate analysis, and all V_dose parameters had a stronger association with developing esophagitis than D_mean on univariate analysis (odds ratio, 1.12–1.72 vs 1.09).²⁰ Thus, although D_mean <34 Gy is a commonly used dosimetric constraint in evaluating RT plans, D_mean may have limited utility compared with V_dose in predicting esophagitis, as seen in Palma et al.²⁰ In our series, 98% of RT plans met the D_mean <34 Gy constraint. Although the D_mean differed among those patients experiencing ≥5% versus <5% weight loss (Table 1), D_mean ranked relatively the least important among RT dose parameters in determining weight loss (Table 2). Our data support that D_mean may be less robustly associated with RT-related esophageal toxicity compared with dose parameters (L_F65, V₆₀) that specifically reflect high dose delivered to a small region of esophagus and support the categorization of the esophagus as a serial organ at risk.

Esophagitis and its relationship with weight loss is influenced by multiple factors and varies with the toxicity scoring system used. For example, in a series of 151 patients with stage III/IV NSCLC who received definitive-intent RT, 17% presented with ≥5% weight loss before lung cancer treatment start.²² In our series, pretreatment albumin was correspondingly found to be the most important clinical factor associated with ≥5% weight loss at the end of RT compared with baseline. Notably, the definition of esophagitis varies with the toxicity scoring system used. For example, older studies used the Radiation Therapy Oncology Group toxicity grading system to define acute esophagitis, where grade 3 acute toxicity was defined as ≥15% weight loss from baseline.²³ However, more modern studies use the National Cancer Institute’s Common Terminology for Adverse Events grading system, which does not include weight loss in its definition of esophagitis (version 3 and later).²⁴ Thus, although weight loss as an endpoint is a limitation in benchmarking our results with established literature that use esophagitis toxicity scores, weight loss is arguably a more generalizable outcome measure that is not subject to observer bias intrinsic in toxicity scales nor type of toxicity scoring system used. Future research will explore the strength of association between L_dose and esophagitis and compare it to the strength of association between V_dose and esophagitis.

Next, exact L_dose values are not currently summarized in treatment planning systems. However, as a proxy, RT plans could be easily reviewed in sagittal or coronal planes with attention to the length of full- or partial-circumferential esophagus covered by the 60 Gy or 65 Gy isodose lines, respectively. This practical approach in determining L_dose, when used in concert with the calculated probability of weight loss (Fig 3) for important dose–length parameters, may facilitate identification of patients who may benefit from either treatment-planning modifications or early nutritional interventions at the time of RT plan approval. As next steps, we are creating an electronic plan quality evaluation output that will report L_dose statistics and corresponding probabilities of weight loss to further facilitate identification of patients at risk for significant weight loss.

Last, because this is a retrospective study with data limited to a single institution, our multivariate model may lack power to detect the significance of additional clinical variables in predicating weight loss. For example, there were only 16 patients with an esophageal L_F65 >0 cm. As such, L_F65 was found to be the most sensitive relative dosimetric variable in predicting weight loss in ridge logistic regression modeling but was not found to be significant in multivariable logistic regression modeling. Among dose parameters, although the ridge logistic regression model demonstrates that L_dose is more sensitive than V_dose on treatment-related weight loss, further study is needed in larger patient cohorts and across multiple institutions to validate the superiority of specific esophageal L_dose parameters in comparison to specific V_dose and D_mean parameters.

Conclusions

Esophagus lengths receiving high full- and partial-circumferential doses are significantly associated with weight loss. Furthermore, our novel esophageal dose–length parameter is more sensitive for treatment-related weight loss than currently used dosimetric parameters such as esophageal V₆₀ or D_mean Esophageal DLHs are an efficient way of visualizing and interpreting complex and highly associated esophageal dose parameters in relation to weight loss toxicity outcomes among patients with lung cancer receiving definitive RT