KEY POINTS
- NRS-RT was developed through literature review, multidisciplinary interviews, and two Delphi rounds, then prospectively evaluated in 254 radiotherapy patients. The final instrument contains 19 items, takes an estimated 5–8 minutes, and incorporates RT-specific variables including treatment site, treatment duration, and concurrent chemoradiotherapy.
- The tool uses a two-stage design: four initial screening questions followed, when positive, by a 15-item assessment scored from 0–100. ROC-derived categories were 0–6 normal, 7–14 mild, 15–30 moderate, and ≥31 severe, with respective discrimination AUCs of 0.990, 0.985, and 0.970 at the three boundaries.
- Against NRS-2002, a 7-point cutoff produced 92.9% sensitivity (95% CI 87.1–96.2%), 97.2% specificity (95% CI 93.0–99.0%), 96.3% PPV, 94.5% NPV, and 95.3% overall accuracy; Cohen's κ was 0.904.
- Agreement with the four-category PG-SGA classification was also high, with weighted κ 0.823 (95% CI 0.783–0.878), an exact category match in 82.3% (209/254) and agreement within one category in 99.2%. Internal consistency was acceptable rather than excellent, with Cronbach's α 0.778.
- The manuscript reports 42 patients considered higher risk by NRS-RT but missed by NRS-2002 and another 24 underestimated by PG-SGA. During prospective weight follow-up of these 66 patients, 42.4% lost >5% of baseline weight; concurrent chemoradiotherapy (OR 2.64, 95% CI 1.32–5.28) and lower albumin (OR 0.89 per 1 g/L increase) predicted discordance.
- There is an important internal reporting issue: the stated 66 discordant patients are difficult to reconcile with the reported NRS-RT/NRS-2002 binary accuracy of 95.3% in 254 patients, which mathematically implies far fewer binary disagreements. In addition, CFA showed suboptimal structural fit, all cutoffs were calibrated in the same dataset, and no comparison against treatment completion, toxicity, survival, or another hard clinical endpoint was performed.
CLINICAL TAKEAWAY
NRS-RT is an interesting attempt to incorporate treatment-specific nutritional risk into routine radiotherapy screening, and its initial diagnostic metrics are strong. However, this remains a single-centre derivation/validation study with imperfect construct validation and an apparent inconsistency in the reported discordance analysis; external prospective validation is needed before replacing established tools.