Numerous patients diagnosed with head and neck cancer have difficulties with oral intake attributable to (1) the tumor itself and/or (2) radiotherapy, with or without radiosensitizer. To prevent weight loss and dehydration, a significant portion of patients require tube feeding. Tube feeding can be administered through either a nasogastric tube (NGT), a percutaneous gastrostomy, which can be inserted either radiologically (PRG) or endoscopically (PEG), or a surgical gastrostomy (Witzel technique). Compared to a gastrostomy, an NGT is indicated for short-term nutritional support. An NGT may also offer advantages for swallowing rehabilitation and help to reduce the risk of long-term functional complications.

A gastrostomy is indicated when tube feeding is anticipated to be required for a duration of four weeks or longer. It offers better physical mobility, a more discreet appearance since it remains hidden under clothing, and overall enhancement in quality of life. On the other hand, gastrostomy procedures carry tube-placement-related risks and infectious complications. In the Netherlands it is therefore recommended to place a gastrostomy only when indicated, and not systematically in all patients treated with radiotherapy with radiosensitizer (Willemsen, 2022). Recent survey research indicates that tube feeding protocols in patients with head and neck cancer undergoing chemoradiotherapy vary significantly across cancer centers in the Netherlands. Approximately half of the head and neck centers have their own protocol outlining indications for prophylactic gastrostomy placement. In addition, some cancer centers recommend prophylactic gastrostomy tubes for every patient, while others only place reactive NGTs (Kok, 2022). Ideally, patients at risk of requiring tube feeding for ≥ 4 weeks should be identified before cancer treatment initiation to facilitate timely gastrostomy placement and mitigate potential wound healing complications. Given the challenge in predicting which patients would benefit from prophylactic gastrostomy placement, prediction models can be used to estimate the individual probability of requiring tube feeding for ≥ 4 weeks. In recent years, some prediction models have been developed and described in the literature. It is our aim to evaluate the effectiveness of the use of such prediction models compared to not using a model.

Description of studies

Not applicable.

Results

Not applicable.

Willemsen (2022) developed and validated a prediction model to estimate tube feeding dependency ≥ 4 weeks in patients with locally advanced head and neck squamous cell carcinoma (LAHNSCC) treated with chemoradiotherapy or bioradiotherapy (CRT/BRT). The model was developed using data from LAHNSCC patients treated at the University Medical Centre Utrecht (UMCU) between 2013 and 2016 and patients treated at the Netherlands Cancer Institute (NKI-AVL) between 2014 and 2017. External validation of the model was conducted on patients treated at Maastricht University Medical Centre+ (MUMC+) and Radboud University Medical Centre (RUMC) between 2013 and 2016. The development cohort comprised 409 patients, while the validation cohort included 334 patients. The mean age in the development cohort was 60.2 years and 58.5 years in the validation cohort. The percentage of male patients was 67% in the development cohort and 66.5% in the validation cohort. The percentage of female patients was 33.0% in the development cohort and 33.5% in the validation cohort. Multivariable analysis showed predictive value for pre-cancer treatment weight change (%), texture modified diet at baseline, poor Eastern Cooperative Oncology Group (ECOG) performance status, tumor site (oral cavity, oropharynx, and hypopharynx), nodal involvement, mean radiation dose to the contralateral parotid gland and the oral cavity. The prediction model's performance was assessed using the receiver operating characteristic curve (ROC), showing an area under the curve (AUC) of 72.8% for internal validation. The model calibration was considered good based on the Hosmer-Lemeshow test statistics (p=0.46). External validation in the combined MUMC+ and RUMC sample yielded an AUC of only 62.4%. In separate external validations, the MUMC+ sample had a higher AUC of 70.8%, while the RUMC sample had a lower AUC of 55.3%. The positive and negative predictive values for high risk of tube feeding dependency at 4 weeks were 81.8% and 42.3%, respectively.

Karsten (2019) developed and validated a prediction model to estimate the risk of prolonged (> 90 days) feeding tube dependency in patients with head and neck squamous cell carcinoma (HNSCC) treated with primary, cisplatin-based chemoradiotherapy (CRT). The model was developed using data from HNSCC patients treated in the NKI-AVL between January 2008 and October 2016. The cohort consisted of 336 patients. The mean age in this cohort was 60.0 years. The percentage of male and female patients was 77.3% and 22.7% respectively. Multivariable analysis showed predictive value for BMI, weight loss, Functional Orale Intake Scale, and tumor stage. The prediction model's performance was assessed using the ROC, showing an AUC of 69% for internal validation. The model calibration was considered satisfactory, based on the Hosmer-Lemeshow test statistics (p=0.254). External validation of the model was not performed.

Gaito (2021) developed and validated a prediction model to estimate the risk of requiring reactive enteral feeding through an NGT for more than four weeks during radiotherapy in patients with advanced head and neck squamous cell carcinoma (HNSCC). The model was developed using data from HNSCC patients identified in the institutional electronic database for retrospective analysis between January 2016 and January 2018. The cohort consisted of 225 patients. In this cohort 47.1% was <60 years of age, 14.7% between 60-70 and 38.2% >70. The percentage of male and female patients was 71% and 29% respectively. Multivariable analysis showed predictive value for tumor site (oral cavity and superior pharyngeal constrictor), advanced tumor stage, chemotherapy drug (none versus any drug), and mean radiation dose to the contralateral parotid gland. The prediction model's performance was assessed using the ROC, showing an AUC of 74.5% for internal validation. The model calibration was considered satisfactory, based on the Hosmer-Lemeshow test statistics (p=0.567). External validation of the model was not performed.

Giraud (2025) developed a predictive model using Artificial Intelligence (AI) for the need for enteral nutrition in patients who receive curative radiotherapy for oropharyngeal, oral cavity, hypopharyngeal, and laryngeal cancer. Two models were trained on a prospective monocentric cohort of 230 patients. The first model predicted an outcome combining severe or early fast weight loss, or severe oral intake impairment (grade 3 anorexia or dysphagia or the prescription of enteral nutrition). The second outcome only combined oral intake impairment criteria. The authors trained a gradient boosted tree with a nested cross validation for Bayesian optimization on a prospective cohort and predictive performances were reported on the external multicentric real-world cohort of 410 patients from 3 centers in France. For the first and second outcome, the model yielded a ROC curve AUC of 81 % and 80%, an accuracy of 77 % and 77 %, a positive predictive value of 77 % and 72 %, a specificity of 78 % and 79 % and a sensitivity of 75 % and 73 %. The negative predictive value was 80 % and 80 %. Overall, the most contributing predictor was concomitant chemotherapy. The model is not yet clinically implementable because it is not available for use.

McClelland (2018) performed a systematic review to investigate the optimal timing for placing a PEG tube (proactive versus reactive) in patients with advanced head and neck cancer undergoing chemoradiotherapy. The authors concluded that a proactive PEG approach can prevent malnutrition during treatment and improve quality of life at six months post-treatment, despite potentially leading to long-term PEG dependency. Clinical and dosimetric parameters such as age, percentage of weight loss at diagnosis, and radiation dose to specific muscles can help identify patients at risk of malnutrition. Additional research is needed to further evaluate the impact of PEG placement on swallowing outcomes and quality of life for these patients.

In conclusion, Willemsen (2022) and Giraud (2025) conducted an external validation of the model in addition to internal validation. In the study of Willemsen, the predictive values of the model (AUC) for the internal validation were acceptable, while those for the external validation were low. Whether applying the selection criteria described in this model may also affect patient-related outcomes, cannot be determined based on this study.

In the study of Giraud, the predictive value of the model in the external validation was high. It is unknown whether the model performs equally well with data from Dutch patients with head and neck cancer.

A comparative study is needed, in which the application of the model is compared to not applying the model (usual care/another model/no model). Such a study was not found.

Level of evidence of the literature

Not applicable.

A systematic review of the literature was performed to answer the following question:

What is the effectiveness of a prediction model in identifying which head and neck cancer patients would benefit from prophylactic gastrostomy placement prior to radiotherapy with or without radiosensitizer?

If the comparison described above is not found in the literature, the identified models and their performance will be described.

Relevant outcome measures

Critical outcome measures for decision making were: completion of cancer treatment as planned, the necessity of reactive feeding tube placement during treatment.

Important outcome measures for decision making were: weight loss, quality of life (QoL), swallowing function recovery, survival, and complications from tube feeding and/or tube placement.

A priori, the guideline development group did not predefine the outcome measures listed above, but instead used the definitions of the outcome measures employed in the included studies (Table 1).

The guideline development group defined minimal clinically (patient) important differences for patients with head and neck cancer per outcome measure as follows:

Table 1. Outcome measures and minimal clinically important differences

Search and select (Methods)

The databases Medline (via OVID) and Embase (via Embase.com) were searched with relevant search terms until 15^th of March 2024. The detailed search strategy is available on request (info@richtlijnendatabase.nl). The systematic literature search resulted in 120 hits. Studies were selected based on the following criteria: systematic review or meta-analysis, randomized controlled trial or other comparative studies that met the PICO. Seventeen studies were initially selected based on title and abstract screening. After reading the full text, 17 studies were excluded (see the Table with reasons for exclusion under the tab Methods), and no studies were included.

Results

There were no studies found which met the exact PICO criteria. However, there were some additional publications that were considered relevant by the committee, which are further described in the ‘results’ section.