Abstract
Background:
Personalised radiotherapy can improve treatment outcomes of patients with head and neck cancer (HNC), where currently a ‘one-dose-fits-all’ approach is the standard. The aim was to establish individualised outcome prediction based on multi-institutional international ‘big-data’ to facilitate risk-based stratification of patients with HNC.
Methods:
The data of 4611 HNC radiotherapy patients from three academic cancer centres were split into four cohorts: a training (n = 2241), independent test (n = 786), and external validation cohorts 1 (n = 1087) and 2 (n = 497). Tumour- and patient-related clinical variables were considered in a machine learning pipeline to predict overall survival (primary end-point) and local and regional tumour control (secondary end-points); serially, imaging features were considered for optional model improvement. Finally, patients were stratified into high-, intermediate-, and low-risk groups.
Results:
Performance score, AJCC8th stage, pack-years, and Age were identified as predictors for overall survival, demonstrating good performance in both the training cohort (c-index = 0.72 [95% CI, 0.66–0.77]) and in all three validation cohorts (c-indices: 0.76 [0.69–0.83], 0.73 [0.68–0.77], and 0.75 [0.68–0.80]). Excellent stratification of patients with HNC into high, intermediate, and low mortality risk was achieved; with 5-year overall survival rates of 17–46% for the high-risk group compared to 92–98% for the low-risk group. The addition of morphological image feature further improved the performance (c-index = 0.73 [0.64–0.81]). These models are integrated in a clinic-ready interactive web interface: https://uic-evl.github.io/hnc-predictor/
Conclusions:
Robust model-based prediction was able to stratify patients with HNC in distinct high, intermediate, and low mortality risk groups. This can effectively be capitalised for personalised radiotherapy, e.g., for tumour radiation dose escalation/de-escalation.
Keywords: Head and neck cancer, Overall survival, Machine learning, Image biomarkers, Decision support tool
1. Introduction
Head and neck cancer (HNC) affects almost 650,000 individuals and causes 350,000 deaths worldwide annually [1]. Historically, the main etiological HNC risk factor was smoking; hence, HNC incidence rates were expected to decrease along with the decline in societal smoking [2–5]. Yet, HNC cases increased due to a relatively new epidemiological subtype, human papilloma virus (HPV)-related HNC, which affects relatively younger patients and is associated with much better prognosis compared to HPV-negative HNC [6,7].
Radiotherapy is a cornerstone for curative HNC treatment. To date, a ‘one-dose-fits-all’ approach is deployed, i.e., all patients receive roughly similar tumour radiation dose prescription based mainly on historic pre-HPV clinical trials. Currently, personalising radiation dose to optimise tumour control is relatively unexplored. For instance, only tumour stage (i.e., early stage versus locally advanced) is used to select eligible patients in recent dose-escalation clinical trials, aiming to improve treatment control by increasing the radiation tumour dose [8–11]. The risk of severe radiation-induced sequelae from dose-escalation [10] makes improved selection a vital unmet need. On the other hand, patients with a low risk of treatment failure might benefit from de-intensified treatment, e.g., MR-guided dose de-escalation [12]. To date, attempts at therapeutic de-intensification in large heterogenous cohorts without patient-specific criteria have been uncompelling [13–15]; consequently, granular treatment outcome estimation for directed dose modification remains a substantive opportunity for HNC treatment personalisation.
Robust treatment outcome prediction based on multifactorial clinical variables is thus crucial to improve treatment success and establish effective personalised radiotherapy [16,17]. While clinical models have been developed [18–21], they are largely unused; clinical implementation has been hampered due to the lack of clinically useful prediction tools that are backed by large representative multi-institutional dataset for training and validation. Additionally, radiomics features – tumour-specific characteristics quantified from medical images – have been shown to improve HNC treatment outcome prediction [22–24]. An approach to add imaging features to well-established clinical models is needed for robust radiomics applications.
The main aim was to establish a large-scale multi-institutional standard for a more individualised outcome prediction in patients with HNC of overall survival (OS) and oncologic outcomes (i.e., local [LC] and regional control [RC]) following radiotherapy using large high-quality international datasets (>4500 patients with HNC). Additionally, an interactive web-based risk prediction tool was pursued to make the models direct clinically actionable for clinicians. Finally, we present a serial prediction model approach, where the clinical models can be enriched by an optional imaging component (Fig. 1A).
Fig. 1. Study overview.
(A) Serial prediction model design. The ‘fundamental clinical model’ component is the core component as it is based data of >4500 patients; the ‘predicted risk(%)’ can be refined with the ‘optional imaging component’, using radiomics features to improve the outcome risk prediction (‘refined Predicted Risk (%)’) to stratify patients in low-, intermediate-, and high-risk patients. The imaging component can be dynamically updated with future technical developments. (B) Datasets for clinical model training, validation, and external validation. Partial cases are patient that are missing at least one variable. Only complete cases were used for the validation of the models.
2. Methods
2.1. Patient considerations
The MD Anderson Cancer Center (MDACC) Big Data Radiotherapy HNC collection effort has been initiated for this study. The prospective and retrospective data collection was approved by the MDACC Institutional Review Board [PA14-0947/RCR03-0800]. This dataset was used for training and independent validation. Prospectively collected data from the University Medical Center Groningen (UMCG) were used for external validation (Standardized Follow-up Program: NCT02435576). The publicly available data from Princes Margret Hospital (PMH) on The Cancer Imaging Archive (TCIA) were used for additional external validation [25].
Inclusion criteria for all cohorts included: (1) proven squamous cell carcinoma of the head and neck, (2) treatment with definitive or adjuvant radiotherapy with/without chemotherapy, and (3) no prior head and neck radiation. Patients were treated from 2001 to 2019, 2007 to 2020, and 2005 to 2010 at MDACC, UMCG, and PMH, respectively. Prescribed tumour doses were 60–72 Gy, as detailed previously by each institution [23,26,27].
2.2. Outcome measures
The primary prediction end-point was OS. The secondary end-points were LC and RC, which were defined as recurrent, progressive, or residual disease of the primary tumour or regional lymph nodes after radiotherapy, respectively (with death as a censor). Time-to-event was measured from start of radiotherapy until the event, alternatively data were censored at last follow-up date. Systematic follow-up was part of the standard of care in both treatment centres: every 3 months in year 1, followed by every 6 months thereafter.
2.3. Clinical variables definitions
The clinical variables (and categorisations) considered in this study were demarcated as follows: gender (female, male); age (<55, 55–65, 65–75, >75); performance score (0, 1, ≥2); smoking status (current, former, never); pack-years (<5, 5–25, 26–50, >50); T-stage (T0-1, T2, T3, T4); N-stage (N0-2a/b, N2c, N3); tumour site (oropharynx [OPC], larynx, hypopharynx, nasopharynx, oral cavity); HPV status (positive and negative), and tumour stage AJCC8th (I, II, III, IV) [28]. The AJCC8th staging was generated from the T-stage, N-stage, tumour site and HPV status with in-house developed algorithm (eMethods). If HPV status was unknown/unspecified, it was assumed as HPV-negative for non-OPC cases. Categorisation was determined on the Kaplan–Meier curves in the training data to meet adequate proportionality testing (eFig. 1).
2.4. Statistical analysis
The MDACC dataset was split into a training and independent validation cohort for the clinical model development (Fig. 1B). The data with all variables collected (i.e., complete cases) were split with a 60:40 ratio into training:validation data. Cases with missing variables (i.e., partial cases) were added to the training set. Only complete cases were considered for the independent and external validation cohorts.
Step-wise forward variable selection was employed to select variables for the Cox regression OS, LC, and RC model based on likelihood ratio test with a Bonferroni-corrected significance level of p < 0.005. Repeated selection was performed on 10 imputed datasets using Multivariate Imputation by Chained Equations (R-package ‘mice’ v3.13.0) with predictive mean matching across 25 iterations [29]. Based on the variable selection and intervariable correlation results, potential models were tested in the validation cohorts. The final models were used for patient stratification. The final OS model was compared with a model based on AJCC8th alone with the likelihood ratio test.
2.5. Risk-based patient stratification
Patients were stratified into high-, intermediate-, and low-risk groups based on the predicted 2-year mortality risk derived from the Cox regression clinical models. These 2-year mortality risk thresholds were visually determined in the training cohort by evaluating the Kaplan–Meier curves for the different risk groups.
2.6. Imaging prediction component
For a subset of patients with available pre-treatment contrast-enhanced CT scans, image characteristics of the primary tumour were quantified in geometric and texture radiomics features using previously developed libraries [30,31], according to the Image Biomarker Standardisation Initiative [32]. Features were selected with bootstrapped forward stepwise variable selection (1000 samples). Subsequently, model improvement was tested for the addition of these features to the clinical risk prediction (i.e., linear predictor).
3. Results
3.1. Patients
A total of 4611 HNC patients were used for the analyses: training (MDACC; n = 2241), independent test (MDACC; n = 786), external validation cohort 1 (UMCG; n = 1087), and external validation cohort 2 (PMH; n = 497). Patient characteristics per cohort are shown in Table 1. Noteworthy differences between cohorts were seen in HPV status (ranging from 16 to 71%), OPC incidence (30–100%), and pack-years (μ = 20–31). Imputation of clinical variables was only performed in the training cohort for pack-years (5% missing), performance score (16%), and HPV status (19%). The overall median follow-up time was 3.6 year (interquartile range [IQR]: [1.6–6.0]), and for censored patients (i.e. excluding patients that die) only 4.3 year [IQR: 2.1–6.7] (site specific, MDACC: 4.1 [2.1–6.6], UMCG: 3.2 [1.7–5.1], and PMH: 8.0 [6.1–9.3]).
Table 1.
Demographics for training, independent validation, and two validation cohorts.
| MDACC training | MDACC validation | UMCG validation | TCIA PMH validation | p-value | |||||
|---|---|---|---|---|---|---|---|---|---|
| N | 2241 | 786 | 1087 | 497 | |||||
| Age (mean (SD)) | 59.48 | (10.14) | 59.74 | (9.74) | 64.17 | (10.56) | 60.16 | (9.90) | <0.001 |
| Sex (%) | |||||||||
| Female | 373 | (17) | 130 | (17) | 324 | (30) | 105 | (21) | <0.001 |
| Male | 1868 | (83) | 656 | (83) | 763 | (70) | 392 | (79) | |
| T stage (%) | <0.001 | ||||||||
| T0 | 53 | (2) | 17 | (2) | 2 | (0) | 0 | (0) | |
| T1 | 507 | (23) | 178 | (23) | 196 | (18) | 90 | (18) | |
| T2 | 788 | (35) | 288 | (37) | 262 | (24) | 162 | (33) | |
| T3 | 453 | (20) | 162 | (21) | 251 | (23) | 146 | (29) | |
| T4 | 414 | (18) | 136 | (17) | 376 | (35) | 99 | (20) | |
| Tx | 26 | (1) | 5 | (1) | 0 | (0) | 0 | (0) | |
| N stage (%) | <0.001 | ||||||||
| N0 | 487 | (22) | 152 | (19) | 435 | (40) | 82 | (16) | |
| N1 | 276 | (12) | 116 | (15) | 130 | (12) | 48 | (10) | |
| N2a-b | 1081 | (48) | 356 | (45) | 279 | (26) | 202 | (41) | |
| N2c | 318 | (14) | 141 | (18) | 202 | (19) | 123 | (25) | |
| N3 | 79 | (4) | 21 | (3) | 37 | (3) | 42 | (8) | |
| HPV status (%) | <0.001 | ||||||||
| Negative | 617 | (28) | 288 | (37) | 912 | (84) | 142 | (29) | |
| Positive | 990 | (44) | 498 | (63) | 175 | (16) | 355 | (71) | |
| Unknown | 634 | (28) | 0 | (0) | 0 | (0) | 0 | (0) | |
| Site (%) | <0.001 | ||||||||
| Oropharynx | 1382 | (62) | 462 | (59) | 328 | (30) | 497 | (100) | |
| Larynx | 420 | (19) | 179 | (23) | 446 | (41) | 0 | (0) | |
| Oral Cavity | 314 | (14) | 95 | (12) | 263 | (24) | 0 | (0) | |
| Hypopharynx | 50 | (2) | 32 | (4) | 26 | (2) | 0 | (0) | |
| Nasopharynx | 22 | (1) | 0 | (0) | 23 | (2) | 0 | (0) | |
| Unkown primary | 53 | (2) | 18 | (2) | 1 | (0) | 0 | (0) | |
| AJCC8th stage (%) | <0.001 | ||||||||
| I | 605 | (27) | 271 | (34) | 159 | (15) | 156 | (31) | |
| II | 368 | (16) | 157 | (20) | 163 | (15) | 137 | (28) | |
| III | 349 | (16) | 143 | (18) | 247 | (23) | 106 | (21) | |
| IVa | 472 | (21) | 206 | (26) | 491 | (45) | 87 | (18) | |
| IVb | 29 | (1) | 9 | (1) | 27 | (2) | 11 | (2) | |
| Unknown | 418 | (19) | 0 | (0) | 0 | (0) | 0 | (0) | |
| Performance score (%) | <0.001 | ||||||||
| 0 | 850 | (38) | 397 | (51) | 619 | (57) | 323 | (65) | |
| 1 | 620 | (28) | 319 | (41) | 350 | (32) | 125 | (25) | |
| >2 | 181 | (8) | 70 | (9) | 118 | (11) | 49 | (10) | |
| Unknown | 590 | (26) | 0 | (0) | 0 | (0) | 0 | (0) | |
| Smoking status (%) | <0.001 | ||||||||
| Never | 773 | (34) | 301 | (38) | 175 | (16) | 144 | (29) | |
| Former | 998 | (45) | 360 | (46) | 457 | (42) | 198 | (40) | |
| Current | 453 | (20) | 125 | (16) | 427 | (39) | 155 | (31) | |
| Unknown | 17 | (1) | 0 | (0) | 28 | (3) | 0 | (0) | |
| Pack years (mean (SD)) | 22.03 | (33.69) | 20.01 | (28.19) | 30.77 | (23.90) | 24.35 | (24.67) | <0.001 |
| Chemotherapy (%) | <0.001 | ||||||||
| None | 446 | (20) | 115 | (15) | 696 | (64) | 254 | (51) | |
| Concurrent | 1060 | (47) | 410 | (52) | 389 | (36) | 243 | (49) | |
| Induction | 218 | (10) | 100 | (13) | 1 | (0) | 0 | (0) | |
| Induction + concurrent | 480 | (21) | 161 | (20) | 1 | (0) | 0 | (0) | |
| Unknown | 37 | (2) | 0 | (0) | 0 | (0) | 0 | (0) | |
| Technique (%) | <0.001 | ||||||||
| 3DCRT | 211 | (9) | 9 | (1) | 14 | (1) | 0 | (0) | |
| IMRT | 1496 | (67) | 450 | (57) | 517 | (48) | 497 | (100) | |
| VMAT | 466 | (21) | 292 | (37) | 401 | (37) | 0 | (0) | |
| IMPT | 68 | (3) | 35 | (4) | 111 | (10) | 0 | (0) | |
| Unknown | 0 | (0) | 0 | (0) | 44 | (4) | 0 | (0) | |
| Radiotherapy type (%) | <0.001 | ||||||||
| Primary | 1727 | (77) | 644 | (82) | 852 | (78) | 497 | (100) | |
| Post-operative | 251 | (11) | 40 | (5) | 230 | (21) | 0 | (0) | |
| Unknown | 263 | (12) | 102 | (13) | 5 | (0) | 0 | (0) | |
| Mortality events (%) | 635 | (28) | 148 | (19) | 402 | (37) | 206 | (41) | <0.001 |
| Local failure events (%) | 233 | (10) | 70 | (9) | 149 | (14) | 46 | (9) | <0.001 |
| Regional failure events (%) | 182 | (8) | 48 | (6) | 105 | (10) | 31 | (6) | 0.005 |
Abbreviations: SD: standard deviation; HPV, human papilloma virus; 3DCRT, three-dimensional conformal radiotherapy; IMRT, intensity-modulated radiotherapy; VMAT, volumetric-modulated arc therapy; IMPT, intensity-modulated proton therapy.
3.2. Association of clinical variables and treatment outcome
For OS, univariable analyses showed that all clinical variables were significant (p < 0.0001), except gender (eTable 1). For LC or RC, all variables were significant, except age and gender (p > 0.106), and N-stage for LC (p = 0.189).
For comprehensive multivariable model analyses and iterations, please refer to eResults 1.
For OS, the final model included the following clinicodemographic variables: performance score, AJCC8th stage, pack-years, and age (Table 2); note that AJCC8th stage is based on T- and N-stage, tumour site, and HPV-status. The performance of the OS clinical model was good in both the MDACC training (c-index = 0.72 95% CI [0.66–0.77]) and independent validation cohort (c-index = 0.76 [0.69–0.83]). External validation showed good performance in both the UMCG cohort (c-index = 0.73 [0.68–0.77]) and PMH cohort (c-index = 0.75 [0.68–0.80]). AJCC8th staging alone was significantly inferior (p < 0.0001) to clinical OS model with c-indices: training 0.65 [0.59–0.71]; test 0.72[0.64–0.80]; UMCG 0.67 [0.62–0.72]; PMH 0.69 [0.62–0.76].
Table 2.
Clinical model parameters and c-index model performance.
| Overall survival (OS) | ||||
|---|---|---|---|---|
| Variables | Category | Coefficients | Hazard ratio | p value |
| Performance score | 0 | 0 | 1 | ref |
| 1 | 0.469 | 1.6 (1.28–1.99) | <0.0001 | |
| ≥2 | 0.781 | 2.18 (1.51–3.16) | 0.0001 | |
| AJCC8th stage | I | 0 | 1 | ref |
| II | 0.117 | 1.12 (0.76–1.65) | 0.5545 | |
| III | 0.679 | 1.97 (1.42–2.74) | 0.0001 | |
| IVa | 0.793 | 2.21 (1.66–2.94) | <0.0001 | |
| IVb | 1.509 | 4.52 (2.79–7.33) | <0.0001 | |
| Pack years | ≤5 | 0 | 1 | ref |
| 5–25 | 0.267 | 1.31 (1.01–1.7) | 0.0459 | |
| 26–50 | 0.499 | 1.65 (1.3–2.08) | <0.0001 | |
| >50 | 0.867 | 2.38 (1.78–3.17) | <0.0001 | |
| Age | ≤55 | 0 | 1 | ref |
| 56–65 | 0.085 | 1.09 (0.89–1.33) | 0.4113 | |
| 65–75 | 0.400 | 1.49 (1.2–1.85) | 0.0003 | |
| >75 | 0.753 | 2.12 (1.56–2.89) | <0.0001 | |
| Local control (LC) | ||||
| Variables | Category | Coefficients | Hazard ratio | p value |
|
| ||||
| T stage | T1 | 0 | 1 | ref |
| T2 | 1.432 | 4.19 (2.19–8.03) | <0.0001 | |
| T3 | 1.473 | 4.36 (2.22–8.58) | <0.0001 | |
| T4 | 1.613 | 5.02 (2.56–9.83) | <0.0001 | |
| HPV status | positive = 1 | −0.694 | 0.5 (0.34–0.73) | 0.0003 |
| Performance score | 0 | 0 | 1 | ref |
| 1 | 0.421 | 1.52 (1.05–2.22) | 0.0276 | |
| ≥2 | 0.801 | 2.23 (1.38–3.59) | 0.0010 | |
| Pack years | ≤5 | 0 | 1 | ref |
| 5–25 | −0.039 | 0.96 (0.58–1.6) | 0.8807 | |
| 26–50 | 0.294 | 1.34 (0.87–2.08) | 0.1858 | |
| >50 | 0.496 | 1.64 (1.02–2.64) | 0.0403 | |
| Regional control (RC) | ||||
| Variables | Category | Coefficients | Hazard ratio | p value |
|
| ||||
| AJCC8th stage | I | 0 | 1 | ref |
| II | 0.442 | 1.56 (0.7–3.46) | 0.2774 | |
| III | 0.984 | 2.68 (1.28–5.59) | 0.0089 | |
| IVa | 1.567 | 4.79 (2.34–9.81) | <0.0001 | |
| IVb | 2.565 | 13 (4.76–35.55) | <0.0001 | |
| Performance score | 0 | 0 | 1 | ref |
| 1 | 0.573 | 1.77 (1.15–2.73) | 0.0093 | |
| ≥2 | 0.793 | 2.21 (1.27–3.84) | 0.0049 | |
| Tumour site | Hypopharynx | 0 | 1 | ref |
| Larynx | −0.118 | 0.89 (0.45–1.75) | 0.7343 | |
| Oropharynx | −0.648 | 0.52 (0.25–1.11) | 0.0898 | |
| Oral cavity | −0.853 | 0.43 (0.21–0.88) | 0.0203 | |
| Unknown Prim | −1.140 | 0.32 (0.07–1.51) | 0.1493 | |
| Nasopharynx | −4.995 | 0.01 (0–21498.48) | 0.9932 | |
| Model performance (c-index [95%CI]) | ||||
| MDACC Training | MDACC validation | UMCG external validation 1 | MGH external validation 2 | |
|
| ||||
| Overall Survival (OS) | 0.72 [0.66–0.78] | 0.76 [0.68–0.83] | 0.73 [0.68–0.78] | 0.75 [0.69–0.81] |
| Local control (LC) | 0.74 [0.67–0.82] | 0.71 [0.58–0.84] | 0.70 [0.62–0.77] | 0.75 [0.61–0.90] |
| Regional control (RC) | 0.74 [0.64–0.83] | 0.73 [0.57–0.89] | 0.7 [0.62–0.78] | 0.74 [0.56–0.91] |
Abbreviations: HPV, human papilloma virus; CI, confidence interval.
The final LC model contained T-stage, HPV status, performance score, and pack-years, with resultant c-indices: training: 0.74 [0.70–0.78]; testing: 0.71 [0.58–0.84]; external validation: 0.70 [0.62–0.76] (UMCG); and 0.74 [0.59–0.89) (PMH). T-stage (HR: T2, 4.19 [2.19–8.03]); T3, 4.36 [2.22–8.58]; T4, 5.02 [2.56–9.83]) and HPV status (HR: 0.5 [0.34–0.73]) were the most dominant factors in predicting LC.
The final RC model included AJCC8th stage, tumour site, and performance score as component variables (Table 2). Resultant c-indices showed training: 0.74 [0.69–0.78]; testing: 0.73 [0.57–0.89]; external validation: 0.7 [0.62–0.77] (UMCG) and 0.71 [0.48–0.94] (PMH). While N-stage can be expected to be an important predictor for RC, the combination of tumour characteristics in the AJCC8th outperformed N-stage alone.
Overall, the calibration plots and HosmereLemeshow analyses showed good calibration of the models in the comparator cohort (eFig. 2). Yet, significant calibration deviation was seen for the OS model in the external cohorts.
3.3. Model-based patient stratification
The survival curves of patients stratified based on their model-based predicted 2-year mortality risk (2y-risk) are shown in Fig. 2. Based on the training cohort, the best separation was seen for predicted 2y-risk lower than 5% (low-risk), between 5 and 25% (intermediate-risk), and higher than 25% (high risk). The average observed 5-year OS was 95% (range: 93–98%) for the low-risk group, 65% (58–79%) for the intermediate-risk group, and 29% (17–42%) for the high-risk group. Notably, the proportion of MDACC and PMH patients stratified as low risk (20% and 26%) was substantially larger compared to the UMCG patients (8%). See eFigures R1.2 and R1.3 for LC and RC analyses.
Fig. 2. Patient stratification based on predicted mortality risk.
Survival curves for low risk (in green; 2-year mortality risk<5%), intermediate risk (in orange; risk≥5 and <25%), and high risk (in blue; ≥25%) in training, validation, and two external validation cohort. Note: Follow-up time was truncated at 6 years for UMCG and 10 years for MDACC and PMH data. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)
Prediction based on AJCC8th staging alone gives a single 2y-risk per category (x-axis Fig. 3-left), while a sizeable spread can be seen per category in 2y-risk calculated by the clinical model (y-axis). Fig. 3 shows that only a select portion of the Stage I is low risk (2y-risk<5%), and limited number of Stage III-IV patients are high risk (2y-risk>25%). The ‘by-the-model-identified’ high-risk patients were correctly classified as the majority of these patients died (Fig. 3-right).
Fig. 3. Predicted overall survival risk based on clinical model versus AJCC8th staging.
Predicted 2-year mortality risk (y-axis) depicted per AJCC8th stage group (left); percentages on x-axis are risks predicted based on staging alone. Survival curves show clear split with model-based risk stratification both in patients with low (right, top) and advanced AJCC8th stage (right, bottom) patients. These figures are based on the MDACC data.
3.4. Web interface prediction and stratification tool
The clinically usable prediction tool was implemented in an interactive web interface https://uic-evl.github.io/hnc-predictor/employing the final clinical models. Here, the clinical variables of a new patient (e.g., age) can be interactively submitted, whereafter the patient-specific predicted OS, LC, or RC curves can be calculated. Finally, by submitting the desired 2-y risk threshold, the new patient is stratified into being low, intermediate, high risk of OS, LC, and/or RC.
3.5. Models in tumour site sub cohorts
The clinical models performed well in two largest sub-cohorts: OPC (n = 2930 patients) and larynx (n = 1257) with c-indices of 0.77/0.76/0.71 and 0.70/0.63/0.73 for OS/LC/RC, respectively (eFig. 3). The model performance (c-index: 0.66/0.67/0.64) was lower for the oral cavity patients (n = 805). Overall, the calibration of the models was good, yet the actual mortality risk was higher than predicted for the OPC and oral cavity patients (HosmereLemeshow p-value<0.05), which was comparable to the total cohort. The number of hypopharynx (n = 136), nasopharynx (n = 56), and unknown primary (n = 73) patients was too low to draw reliable conclusions (eFig. 3).
3.6. Imaging component
For the radiomics features, 455 MDACC patients were used for training, and 229 UMCG and 430 PMH patients for external validation. The bootstrapped stepwise forward selection identified the ‘minor axis length’ of the primary tumour as the most frequently selected geometric predictor for OS (eResults 2). This image feature significantly added (likelihood ratio test; p = 0.004) to predicted risk from clinical model (i.e., linear predictor). Compared to the clinical model (c-index = 0.72 [0.63–0.81]), the performance of this combined model increased slightly (c-index 0.73 [0.64–0.81]). While the validation c-index increase was more pronounced in the UMCG cohort (from 0.71 [0.62–0.81] to 0.74 [0.64–0.83]), no performance improvement was seen in the PMH validation cohort (from 0.74 [0.67–0.80] to 0.74 [0.67–0.81]). No robust features could be identified for LC and RC (eResults 2).
4. Discussion
The clear stratification of patients with HNC into high, intermediate, and low risk of mortality (Fig. 2) by the models can be effectively used for personalised radiotherapy, e.g., selecting high-risk patients for tumour radiation dose escalation or low-risk patients for dose de-escalation. The impressive survival differences for patients who are nominally in the same AJCC (including HPV) risk category allows for more directive and granular patient-by-patient risk differentiation. For example, OPC HPV-positive patients are considered for de-escalation trials [13–15], yet our findings show that 4% and 14% of these patients have a 2-y mortality of >25% and >15%, respectively, for which dose de-escalation may not be advisable. By using this international big dataset of more than 4500 patients, this study establishes a benchmark for robust OS, LC, and RC prediction in patients with HNC. Additionally, the clinic-ready web-based tool calculates and visualises the expected survival and tumour outcome for new individual patients (https://uic-evl.github.io/hnc-predictor/). The underlying model code, radiomics, and clinical data are publicly shared in a Figshare repository: https://doi.org/10.6084/m9.figshare.21303000.
All final clinical models included the patient’s performance score; that poor(er) performance scores are associated with poorer survival has been long recognised [33,34], yet that tumour control is associated with performance status is less intuitive. The composite variable AJCC8th staging together with pack-years, age, and performance score were included in the OS model; hence, all clinical variables were directly or indirectly incorporated in this model, except gender. Similar OS risk factors have been observed in previous studies, age, tumour location, smoking status, T and N-stage [20,35], and later HPV status [18,19]. Beesley et al. developed a US-trained/EU-validated multistate Bayesian clinical prediction model for radiotherapy OPC patients to predict event likelihood parameters [36]. While the modelling procedure was quite different, similar input predictors were identified: T, N-stage, HPV status, age, smoking status; notably, tobacco pack-years and performance score were not included. Overall, these findings suggest that despite distinct modelling approaches and datasets, convergent phenomena have been observed.
For the LC prediction, T-stage, HPV status, performance score, and pack-years were selected. Since HPV status was highly correlated to tumour site (Rho = 0.89; p < 0.0001), it is difficult to determine the impact of tumour location on LC. In contrast, for RC, tumour site showed added predictive value to AJCC8th staging, which is interesting as it based on the tumour site. This is likely due to the difference of the lymphatic tumour spread per tumour location [37].
Outcome prediction was robust across multi-institutional cohorts, even though they had distinct patient demographic profiles (Table 1); particularly, the HPV-positive HNC incidence was substantially lower in European compared to the North American cohorts. Additionally, OPC, larynx, and oral cavity cancer sub-analyses (eFig. 3) showed clinical applicable levels of model performance and calibration. For the hypopharynx, nasopharynx, and unknown primary cancer sub-cohorts, caution is advised when applying these models due to the sparse patient numbers.
The serial approach of building the prediction model presented in this study (Fig. 1A) allows for flexible addition of imaging features. Higher OS risk was associated with larger minor axis length of the tumour [32], which represents an intuitive metric for tumour size. Previous studies showed the relation between OS and features indicating larger or more irregular tumours [22,23]. Texture features, in contrast to prior works [22–24], failed to improve our model discrimination (eResults 2.2); similar to a previous study [38]. This may be due to the sensitivity of intensity/texture features to image acquisition discrepancies [39], arguing for improved image harmonisation, standardisation, and image quality.
Limitations of the study cohort are that the majority of tumour locations were OPC, larynx, and oral cavity, underrepresenting hypopharynx, nasopharynx, and unknown primary cases. While this is a representative of the HNC clinical incidence, this may mean that the presented models are not sufficiently tested for under-represented tumour sites. Another challenge is the definition of local and RC, for which an event was broadly defined as recurrent, progressive, or residual disease. The detection residual/returning disease can be challenging [40] and is further complicated when no salvage treatments are available or when patients are lost from follow-up, and thus, no pathologic conformation, clinical progression, or imaging can be obtained. This may therefore potentially result in an underdetection bias of disease control in the cohorts, which can influence accuracy of the LC and RC models.
As with multi-site data aggregation and risk modelling efforts at large scale, there are intrinsic limitations as function of data availability, e.g., anaemia identified by Beesley et al. was not recorded in these datasets [36]. Consequently, the utility of this (or any) predictive model is necessarily predicated on input variables and could be modified or altered with updated or augmented data. Moreover, stage migration considerations between AJCC 7th and 8th edition should be noted; for example, extranodal extension was not always specified/recorded as a formalised component of AJCC 7th ed. and may have been obscured. Improved incorporation could improve the models, or alternatively, it could be added as a separate variable [36]. While we focus on OS, LC, and RC, future work will focus on predicting distant metastases and disease-free survival.
Nonetheless, this study is to our knowledge based on the largest head and neck extant multi-site dataset, which allowed for the development of statistically robust and clinic-ready HNC risk models. This provides a benchmark platform for extended future developments of image-incorporating prediction methods, such as deep learning. Moreover, the end-user-enabled web interface (GUI) provides an accessible decision support tool for patient-individual risk stratification for therapeutic selection.
5. Conclusion
Developed and assessed in this international ‘big-data’-set, our prediction models presented excellent capacity to stratify patients with HNC at high, intermediate, and low mortality risk – outperforming AJCC8th staging. This work sets a benchmark for robust OS, LC, and RC risk prediction in radiotherapy HNC patients, which can effectively be capitalised for personalised radiotherapy with the clinic-ready web-based tool prediction tool for new patients that does not require under-the-hood knowledge of model mechanics (https://uic-evl.github.io/hnc-predictor/).
Supplementary Material
Acknowledgements
The authors thank the MD Anderson Stiefel Center for Oropharyngeal Cancer Research and Education (MDA-SCORE) for the additional prospectively collected data. The authors also thank the MDACC summer students of 2019 for the aid in collecting part of the clinical variable data.
Funding statement
Dr. van Dijk received funding and salary support from the Dutch organization NWO ZonMw for the execution of this study via the Rubicon Individual career development grant.
Dr. Mohamed and Dr. Fuller received funding and salary support from directly related to this project from: NIH National Institute of Dental and Craniofacial Research (NIDCR) Academic Industrial Partnership Grant (R01DE028290); NIDCR Establishing Outcome Measures for Clinical Studies of Oral and Craniofacial Diseases and Conditions award (R01DE025248); NIH/NSF NCI Smart Connected Health Program (R01CA257814).
Dr. Fuller received funding and salary support from directly unrelated to this project from: NCI Parent Research Project Grant (R01CA258827); NCI Ruth L. Kirschstein RSNA, United states Institutional Research Training Grant (T32CA261856); NIH NIDCR Exploratory/Developmental Research Grant Program (R21DE031082); National Institutes of Health (NIH) National Cancer Institute (NCI) Early Stage Development of Technologies in Biomedical Computing, Informatics, and Big Data Science Program (R01CA214825); NSF/NIH Joint Initiative on Quantitative Approaches to Biomedical Big Data program (R01CA225190); NIH National Institute of Biomedical Imaging and Bioengineering (NIBIB) Research Education Programs for Residents and Clinical Fellows Grant (R25EB025787); NCI Early Phase Clinical Trials in Imaging and Image-Guided Interventions Program (1R01CA218148); NIH/NCI Cancer Center Support Grant (CCSG) Pilot Research Program Award from the UT MD Anderson CCSG Radiation Oncology and Cancer Imaging Program (P30CA016672); Small Business Innovation Research Grant Program a sub-award from Oncospace, Inc. (R43CA254559); The Human BioMolecular Atlas Program (HuBMAP) Integration, Visualization and Engagement (HIVE) Initiative (OT2OD026675) sub-award; Patient-Centered Outcomes Research Institute (PCS-1609-36,195) sub-award from Princess Margaret Hospital; National Science Foundation (NSF) Division of Civil, Mechanical, and Manufacturing Innovation (CMMI) grant (NSF 1933369). Dr. Fuller receives grant and infrastructure support from MD Anderson Cancer Center via: the Charles and Daneen Stiefel Center for Head and Neck Cancer Oropharyngeal Cancer Research Program; the Program in Image-guided Cancer Therapy; and the NIH/NCI Cancer Center Support Grant (CCSG) Radiation Oncology and Cancer Imaging Program (P30CA016672).
Mr. Nipu and Dr. Marai received funding unrelated to this project during the period of the study execution from: the U.S. National Institutes of Health (NIH) National Cancer Institute (NCI) (R01CA258827, R01CA225190), from the NIH National Library of Medicine (NLM) (R01LM012527), and from the U.S. National Science Foundation (CDS&E-1854815, CNS-1828265).
Mr. Wahid is supported by the Dr. John J. Kopchick Fellowship through The University of Texas MD Anderson UTHealth Graduate School of Biomedical Sciences, the American Legion Auxiliary Fellowship in Cancer Research, and an NIH/National Institute for Dental and Craniofacial Research (NIDCR) F31 fellowship (1 F31DE031502-01).
Prof. Langendijk received unrelated funding from the Dutch Cancer Society and the European Union during this study.
Footnotes
Conflict of interest statement
Dr. Fuller has received direct industry grant/in-kind support, honoraria and travel funding from Elekta AB. Prof. Langendijk was member of the Global Advisory Board of IBA (honorarium paid to the UMCG Research BV and the RayCare International Advisory Board of RaySearch). The department of Radiation Oncology of UMCG has research agreements with IBA, RaySearch, Siemens, Elekta, Mirada, and Leoni. All other authors declare no conflict of interest.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.org/10.1016/j.ejca.2022.10.011.
References
- [1].Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2021;71:209–49. [DOI] [PubMed] [Google Scholar]
- [2].Wyss A, Hashibe M, Chuang S-C, Lee Y-CA, Zhang Z-F, Yu G-P, et al. Cigarette, cigar, and pipe smoking and the risk of head and neck cancers: pooled analysis in the International Head and Neck Cancer Epidemiology Consortium. Am J Epidemiol 2013; 178:679–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [3].Jamal A, King BA, Neff LJ, Whitmill J, Babb SD, Graffunder CM. Current cigarette smoking among adults — United States, 2005–2015. MMWR Morb Mortal Wkly Rep 2016;65. [DOI] [PubMed] [Google Scholar]
- [4].Current cigarette smoking among adults—United States, 2011. JAMA 2013;309:539. [PubMed] [Google Scholar]
- [5].Marur S, Forastiere AA. Head and neck cancer: changing epidemiology, diagnosis, and treatment. Mayo Clin Proc 2008;83: 489–501. [DOI] [PubMed] [Google Scholar]
- [6].Benson E, Li R, Eisele D, Fakhry C. The clinical impact of HPV tumor status upon head and neck squamous cell carcinomas. Oral Oncol 2014;50:565–74. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [7].Marur S, D’Souza G, Westra WH, Forastiere AA. HPV-associated head and neck cancer: a virus-related cancer epidemic. Lancet Oncol 2010;11:781–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [8].Bourhis J, Overgaard J, Audry H, Ang KK, Saunders M, Bernier J, et al. Hyperfractionated or accelerated radiotherapy in head and neck cancer : a meta-analysis. Lancet 2006;368:843–54. [DOI] [PubMed] [Google Scholar]
- [9].Horiot JC, Le Fur R, N’Guyen T, Chenal C, Schraub S, Alfonsi S, et al. Hyperfractionation versus conventional fractionation in oropharyngeal carcinoma: final analysis of a randomized trial of the EORTC cooperative group of radiotherapy. Radiother Oncol 1992;25:231–41. [DOI] [PubMed] [Google Scholar]
- [10].Madani I, Duprez F, Boterberg T, van de Wiele C, Bonte K, Deron P, et al. Maximum tolerated dose in a phase I trial on adaptive dose painting by numbers for head and neck cancer q. Radiother Oncol 2011;101:351–5. [DOI] [PubMed] [Google Scholar]
- [11].Guerrero Urbano T, Clark CH, Hansen VN, Adams EJ, A’Hern R, Miles EA, et al. A phase I study of dose-escalated chemoradiation with accelerated intensity modulated radiotherapy in locally advanced head and neck cancer. Radiother Oncol 2007;85:36–41. [DOI] [PubMed] [Google Scholar]
- [12].Bahig H, Yuan Y, Mohamed ASR, Brock KK, Ping S, Wang J, et al. Clinical and translational radiation oncology magnetic resonance-based response assessment and dose adaptation in human papilloma virus positive tumors of the oropharynx treated with radiotherapy (MR-ADAPTOR): an R-IDEAL stage 2a-2b/bayesian phase II. Clin Transl Radiat Oncol 2018;13:19–23. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [13].Gillison ML, Trotti AM, Harris J, Eisbruch A, Harari PM, Adelstein DJ, et al. Radiotherapy plus cetuximab or cisplatin in human papillomavirus-positive oropharyngeal cancer (NRG Oncology RTOG 1016): a randomised, multicentre, non-inferiority trial. Lancet 2019;393:40–50. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [14].Mehanna H, Robinson M, Hartley A, Kong A, Foran B, Fulton-Lieuw T, et al. Radiotherapy plus cisplatin or cetuximab in low-risk human papillomavirus-positive oropharyngeal cancer (De-ESCALaTE HPV): an open-label randomised controlled phase 3 trial. Lancet 2019;393:51–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [15].Yom SS, Torres-Saavedra P, Caudell JJ, Waldron JN, Gillison ML, Xia P, et al. Reduced-dose radiation therapy for HPV-associated oropharyngeal carcinoma (NRG oncology HN002). J Clin Oncol 2021;39:956–65. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [16].Heukelom J, Hamming O, Bartelink H, Hoebers F, Giralt J, Herlestam T, et al. Adaptive and innovative Radiation Treatment FOR improving Cancer treatment outcomE (ARTFORCE); a randomized controlled phase II trial for individualized treatment of head and neck cancer. BMC Cancer 2013;13:1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [17].Thorwarth D, Soukup M, Alber M. Dose painting with IMPT, helical tomotherapy and IMXT: a dosimetric comparison. Radiother Oncol 2008;86:30–4. [DOI] [PubMed] [Google Scholar]
- [18].Fakhry C, Zhang Q, Nguyen-Tân PF, Rosenthal DI, Weber RS, Lambert L, et al. Development and validation of nomograms predictive of overall and progression-free survival in patients with oropharyngeal cancer. J Clin Oncol 2017;35:4057–65. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [19].Prince V, Bellile EL, Sun Y, Wolf GT, Hoban CW, Shuman AG, et al. Individualized risk prediction of outcomes for oral cavity cancer patients. Oral Oncol 2016;63:66–73. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [20].Emerick KS, Leavitt ER, Michaelson JS, Diephuis B, Clark JR, Deschler DG. Initial clinical findings of a mathematical model to predict survival of head and neck cancer. Otolaryngol Head Neck Surg 2013;149:572–8. [DOI] [PubMed] [Google Scholar]
- [21].Jan R, Jong B de, Hermans J, Molenaar J, r JJ, Cessie S. Prediction of survival in patients with head and neck cancer 2001. p. 718–24. [DOI] [PubMed] [Google Scholar]
- [22].Aerts HJWL, Velazquez ER, Leijenaar RTH, Parmar C, Grossmann P, Cavalho S, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun 2014;5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [23].Zhai T-T, van Dijk LV, Huang B-T, Lin Z-X, Ribeiro CO, Brouwer CL, et al. Improving the prediction of overall survival for head and neck cancer patients using image biomarkers in combination with clinical parameters. Radiother Oncol 2017; 124:256–62. [DOI] [PubMed] [Google Scholar]
- [24].Valliéres M, Kay-Rivest E, Perrin LJ, Liem X, Furstoss C, Aerts HJWL, et al. Radiomics strategies for risk assessment of tumour failure in head-and-neck cancer. Sci Rep 2017;7:1–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [25].Kwan J, Su J, Huang S, Ghoraie L, Xu W, Chan B, et al. Data from radiomic biomarkers to refine risk models for distant metastasis in oropharyngeal carcinoma. The Cancer Imaging Archive 2019. DOI: 10.79. [DOI] [PubMed] [Google Scholar]
- [26].Mohamed ASR, Cardenas CE, Garden AS, Awan MJ, Rock CD, Westergaard Sa, et al. Patterns-of-failure guided biological target volume definition for head and neck cancer patients: FDG-PET and dosimetric analysis of dose escalation candidate subregions. Radiother Oncol 2017;124:248–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [27].O’Sullivan B, Huang SH, Perez-Ordonez B, Massey C, Siu LL, Weinreb I, et al. Outcomes of HPV-related oropharyngeal cancer patients treated by radiotherapy alone using altered fractionation. Radiother Oncol 2012;103:49–56. [DOI] [PubMed] [Google Scholar]
- [28].Amin MB, Greene FL, Edge SB, Compton CC, Gershenwald JE, Brookland RK, et al. The Eighth Edition AJCC Cancer Staging Manual: continuing to build a bridge from a population-based to a more “personalized” approach to cancer staging. CA A Cancer J Clin 2017;67:93–9. [DOI] [PubMed] [Google Scholar]
- [29].Rubin DB, editor. Multiple imputation for nonresponse in surveys Hoboken, NJ, USA: John Wiley & Sons, Inc.; 1987. [Google Scholar]
- [30].van Dijk LV, Brouwer CL, van der Schaaf A, Burgerhof JGM, Beukinga RJ, Langendijk JA, et al. CT image biomarkers to improve patient-specific prediction of radiation-induced xerostomia and sticky saliva. Radiother Oncol 2017;122:185–91. [DOI] [PubMed] [Google Scholar]
- [31].van Dijk Lv, Thor M, Steenbakkers RJHM, Apte A, Zhai TT, Borra R, et al. Parotid gland fat related Magnetic Resonance image biomarkers improve prediction of late radiation-induced xerostomia. Radiother Oncol 2018;128:459–66. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [32].Zwanenburg A, Leger S, Valliéres M, Löck S. Image biomarker standardisation initiative - feature definitions 2016. ArXiv: 161207003. [Google Scholar]
- [33].Kowalski LP, Carvalho AL. Natural history of untreated head and neck cancer. Eur J Cancer 2000;36:1032–7. [DOI] [PubMed] [Google Scholar]
- [34].Coates A, Porzsolt F, Osoba D. Quality of life in oncology practice: prognostic value of EORTC QLQ-C30 scores in patients with advanced malignancy. Eur J Cancer 1997;33:1025–30. [DOI] [PubMed] [Google Scholar]
- [35].Datema FR, Ferrier MB, Vergouwe Y, Moya A, Molenaar J, Piccirillo JF, et al. Update and external validation of a head and neck cancer prognostic model. Head Neck 2013;35:1232–7. [DOI] [PubMed] [Google Scholar]
- [36].Beesley LJ, Hawkins PG, Amlani LM, Bellile EL, Casper KA, Chinn SB, et al. Individualized survival prediction for patients with oropharyngeal cancer in the human papillomavirus era. Cancer 2019;125:68–78. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [37].Grégoire V, Coche E, Cosnard G, Hamoir M, Reychler H. Selection and delineation of lymph node target volumes in head and neck conformal radiotherapy. Proposal for standardizing terminology and procedure based on the surgical experience. Radiother Oncol 2000;56:135–50. [DOI] [PubMed] [Google Scholar]
- [38].Ger RB, Zhou S, Elgohari B, Elhalawani H, Mackin DM, Meier JG, et al. Radiomics features of the primary tumor fail to improve prediction of overall survival in large cohorts of CT- and PET-imaged head and neck cancer patients. PLoS One 2019;14:1–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [39].Ger RB, Zhou S, Chi PCM, Lee HJ, Layman RR, Jones AK, et al. Comprehensive investigation on controlling for CT imaging variabilities in radiomics studies. Sci Rep 2018;8:1–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [40].Ho AS, Kraus DH, Ganly I, Lee NY, Shah JP, Morris LGT. Decision making in the management of recurrent head and neck cancer. Head Neck 2014;36:144–51. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.