Preoperative malnutrition and risk of re-revision and mortality following revision arthroplasty – Correspondence

We appreciate the work of Yao et al, who address a service-relevant question of whether a simple preoperative prognostic nutritional index (PNI) can inform re-revision risk management after aseptic revision arthroplasty. Their large single-center cohort proposes a pragmatic cut-point (~46.6) with moderate discrimination (AUC around 0.72), describes a markedly asymmetric predictive profile with high negative predictive value (NPV ≈ 0.91) but very low positive predictive value (PPV ≈ 0.02), and shows that the association is predominantly driven by periprosthetic joint infection (PJI) rather than noninfectious failures. They further report a U-shaped PNI–mortality pattern with a nadir around 47.5, suggesting that both marked malnutrition and possible “over-nutrition” may be harmful[1]. Building on this clinically proximate work, we offer several deployment-oriented suggestions to maximize safe translation, aligned with contemporary AI and prediction-model reporting standards such as the TITAN guideline[2]. First, bedside decisions depend more on calibrated absolute risks than on relative or rank-based metrics. A model with an apparently “good” AUC can still be poorly calibrated and mislead clinicians when deciding whether to escalate care[3]. We therefore encourage Yao et al (and future investigators) to report calibration-in-the-large and calibration slope, present graphical calibration plots across clinically meaningful risk groups (e.g., deciles or PNI-defined strata), and apply bootstrap optimism-correction to attenuate overfitting[3,4]. Where the calibration slope is <1, uniform shrinkage or penalization followed by post-shrinkage recalibration should be considered[3]. These steps are increasingly viewed as minimum good practice in multivariable prediction modelling, rather than optional extras[3,4]. Second, because PNI in this setting behaves primarily as a rule-out tool, its clinical value hinges on how risk thresholds are chosen and communicated. A very low PPV is unsurprising when event rates are low, but clinicians still need to know at which predicted risk (e.g., 5–20% probability of PJI-related re-revision) the benefits of intensified optimization or surveillance outweigh the burdens[5]. Decision-curve analysis can demonstrate whether PNI-informed strategies yield higher net benefit than “treat none” or “treat all” across relevant threshold ranges[5,6]. We would also welcome benchmarking against simple baselines such as albumin-only or lymphocyte-only indices, to show that the composite PNI truly adds value over its components[5]. Explicit reporting of clinically endorsed risk thresholds would greatly enhance the interpretability of NPV and PPV in shared decision making. Third, to ease adoption, we suggest providing a simple translation toolkit. A one-page lookup table could map common PNI values (e.g., 40, 45, 46.6, 50) to corresponding (albumin, lymphocyte) pairs and to expected NPV and PPV under the study event rate, alongside a worked example of intercept-and-slope recalibration for centers with different baseline PJI prevalence[3,7]. Because the prognostic signal appears infection-specific, we believe PJI-oriented endpoints should be foregrounded in any electronic alerts, order sets or perioperative “checklists” built around the model. In older or frailer patients, parallel sensitivity analyses using the Geriatric Nutritional Risk Index might help triangulate risk and support consistency with geriatric-nutrition literature[8], ideally aligned with the 2018 evidence-based PJI definition[9]. Finally, in patient communication, we would avoid labelling PNI as “highly accurate” in isolation. Instead, PNI could be presented as a low-cost screening tool that, once crossing a decision-curve-supported threshold, triggers a standard optimization bundle (nutrition support, anemia and iron management, glycemic control, smoking cessation), with pre-theater re-measurement to capture delta-PNI and potential risk attenuation over time[4–6]. This framing emphasizes trajectories and modifiable risk, which may be more meaningful to patients than a single static probability. In sum, Yao et al deliver a credible, actionable signal linking preoperative malnutrition with infection-driven re-revision and complex mortality patterns. By adding robust calibration, decision-curve evidence, transparent thresholds with recalibration guidance, and validation aligned with modern TRIPOD- and TITAN-style expectations for prediction and AI-adjacent tools[2–7], their work can convert statistical accuracy into deployable, infection-specific decision support across revision arthroplasty services[1–9].