Page 175 - Functional impairment and cues for rehabilitation of head and neck cancer patients -

A reactive tube placement protocol was used for all patients, with placement reactive to excessive weight loss (>5% over three months or >10% over six months), dehydration or proven aspiration based on videofluoroscopy. Tubes are removed in case oral intake is adequate and/ or aspiration is resolved. All patients are seen by the SLP and dietitian for clinical check-up and counseling before CRT and all were enrolled in the preventive swallowing exercise program according to the Institution’s protocol (29). We expect high compliance to the program because of intensive monitoring by an SLP at least until 90 days post CRT.
Endpoint definition
The endpoint of the prediction model was prolonged feeding tube dependency, defined as placement of a tube (NGT or PRG) before the end of CRT, which stayed in situ for more than ninety days, because by that time the acute local treatment-related toxicities have subsided and ongoing functional impairment like xerostomia and dysgeusia have become more stable. Also, in all patients with a feeding tube the need for the tube is reassessed every two weeks by the dietitian.
Statistical analysis
Analyses were performed using IBM® SPSS® Statistics 23.0 and R 3.3.2 (30, 31). P values < .05 were considered statistically significant. Univariable logistic regression analysis was used to assess the association of baseline variables with prolonged feeding tube dependency in this sample. Subsequently, a multivariable logistic regression model was developed, for which we considered known predictors based on theoretical considerations and pre-existing evidence. These candidate predictors were T-stage, BMI, dysphagia, weight loss and FOIS. The FOIS was dichotomized (7 = normal diet and < 7 = abnormal diet) due to the low number of patients with scores < 7. Variables were not subject to selection based on statistical significance (32), but variables with a contradicting sign of the regression coefficient (i.e., contradicting current clinical knowledge and/or biological plausibility) were excluded from the model. Odds ratios (OR) with corresponding 95% confidence intervals (95% CI) and p values of the final model are presented. Linear predictor scores were calculated for use in calibration and discrimination analysis.
Discrimination and calibration were assessed to evaluate the performance of the model. The area under the receiver-operation characteristic (ROC) curve was estimated to assess discriminative ability (0.5 = no discriminative ability and 1.0 = perfect discrimination). From a value of 0.7, the discriminative ability of the model is fair. For calibration (agreement between predicted and observed probabilities), the goodness-of-fit test (Hosmer-Lemeshow test) was used, with p values >.05 indicating good calibration. Bootstrapping analysis with 200 samples was used to internally validate the model and estimate shrinkage factors per predictor for future use. A nomogram (with regression coefficients after shrinkage) is presented to easily estimate the probability of prolonged tube dependency per patient.
7
Prediction model for tube feeding during CRT
173