When a man is newly diagnosed with prostate cancer, one of the most urgent questions his doctors face is whether the disease has already spread to his bones. Bone is the most common site of distant metastasis in prostate cancer, and the answer reshapes everything that follows: staging, treatment intensity, prognosis, and the choice of imaging studies. Yet deciding which patients truly need expensive and sometimes invasive bone-directed imaging has long been an imperfect exercise, relying on thresholds of prostate-specific antigen and clinical judgment. A new study from researchers at the First Affiliated Hospital of Anhui Medical University in Hefei, China, published in BMC Cancer, offers a data-driven answer. The team built and externally validated a diagnostic prediction model that combines routine clinicopathological information with a simple measure of systemic inflammation drawn from a standard complete blood count, aiming to estimate each patient’s individual probability of bone metastasis at the moment of diagnosis.

The research, led by Weibo Wang, Chengyang Tian, and Supeng Tai, with Jun Zhou and Guangyue Luo as corresponding authors, took advantage of a prospectively maintained database spanning two separate campuses of the same hospital system. The development cohort comprised 355 patients evaluated at the Jixi Road Campus, while 203 patients from the High-Tech Campus served as an entirely independent external validation cohort. This two-campus design matters because prediction models are notoriously prone to overfitting, performing brilliantly on the data used to build them and then collapsing when confronted with new patients. By holding out a geographically and administratively distinct patient population, the investigators subjected their model to one of the most demanding tests available in clinical prediction research: does it still work on people it has never seen?

Among the 355 development-cohort patients, 67, or 18.9 percent, were found to have bone metastasis. In the external validation cohort the figure was 48 of 203, or 23.6 percent. These substantial prevalence rates underscore why the question is clinically urgent: roughly one in five newly diagnosed men in these cohorts harbored skeletal disease, and identifying them accurately at the outset is essential for correct staging and treatment planning. The researchers began by examining five inflammatory-immune indices, all calculable from routine blood work: the neutrophil-to-lymphocyte ratio, the platelet-to-lymphocyte ratio, the lymphocyte-to-monocyte ratio, the systemic immune-inflammation index, and the systemic inflammation response index. These indices have attracted growing attention in oncology because they capture, in crude but reproducible form, the interplay between a patient’s immune system and the systemic inflammatory state that tumors both induce and exploit.

The statistical machinery behind the model was deliberately rigorous. Associations between each inflammatory-immune index and bone metastasis were tested in the development cohort using multivariable logistic regression, with restricted cubic splines employed to allow for nonlinear relationships rather than forcing the data into straight-line assumptions. Predictor selection then proceeded through a two-stage strategy: repeated least absolute shrinkage and selection operator regression, a penalized technique that shrinks unstable coefficients toward zero and is prized for its robustness against overfitting, followed by stepwise logistic regression to finalize the variable set. The result was a lean, four-variable model. The predictors that survived repeated selection were log-transformed prostate-specific antigen, clinical T stage, biopsy Grade Group as defined by the International Society of Urological Pathology, and the neutrophil-to-lymphocyte ratio.

Two of the five inflammatory indices earned their place in the final analysis on statistical grounds. The neutrophil-to-lymphocyte ratio, or NLR, and the systemic immune-inflammation index both remained associated with bone metastasis after full adjustment for other factors. Expressed per one-standard-deviation increase in the transformed index, the odds ratio for the NLR was 1.58, with a 95 percent confidence interval of 1.12 to 2.22, while the systemic immune-inflammation index carried an odds ratio of 1.44, with a confidence interval of 1.03 to 2.03. In plain terms, patients whose blood showed a higher relative abundance of neutrophils compared with lymphocytes were meaningfully more likely to have bone metastases, even after accounting for how advanced their tumors appeared by conventional measures. This fits a broader biological picture in which neutrophil-dominated inflammation can foster tumor progression and metastatic seeding, while lymphopenia may signal a blunted anti-tumor immune response.

How well did the model actually perform? In the development cohort it achieved an area under the receiver operating characteristic curve, or AUC, of 0.905, with a 95 percent confidence interval of 0.869 to 0.941. Because models built and evaluated on the same data tend to look better than they are, the team applied bootstrap resampling to obtain an optimism-corrected AUC of 0.898, with a confidence interval of 0.864 to 0.934, a negligible loss of performance. The decisive test came in the external validation cohort, where the AUC was 0.884, with a confidence interval of 0.828 to 0.941. An AUC near 0.9 represents strong discriminative ability, meaning the model separates patients with bone metastasis from those without it with high fidelity. Calibration plots, which assess whether predicted probabilities match observed frequencies, and the Brier score, which penalizes both discrimination and calibration errors, remained consistent across internal and external validation, as did clinical net benefit measured by decision curve analysis.

One of the study’s most clinically relevant findings concerns the incremental value of the blood-based marker. Clinicopathological factors alone, without the NLR, produced an AUC of 0.887. Adding the neutrophil-to-lymphocyte ratio raised this to 0.905, a change in AUC of 0.018 with a P value of 0.034, indicating a statistically significant but modest improvement. The authors are appropriately measured on this point: NLR provided complementary information rather than a revolution. In practical terms, a cheap, universally available blood count parameter sharpened a model already built on PSA, tumor stage, and biopsy grade. That kind of incremental gain, verified on independent patients, is exactly what guideline committees look for when deciding whether to endorse a new tool, and it suggests the inflammatory marker carries genuine biological signal about metastatic behavior rather than statistical noise.

The clinical use case the researchers envision is decision support for imaging. Current practice often relies on PSA thresholds and risk categories to determine which newly diagnosed men should undergo bone scintigraphy or, increasingly, PSMA-targeted PET/CT, a modality that is more sensitive but far more expensive and less widely available. A validated model that outputs an individualized probability of bone metastasis could help clinicians and patients weigh the trade-off between missing occult skeletal disease and subjecting low-risk men to unnecessary imaging, radiation exposure, cost, and incidental findings. The model’s strong negative predictive potential in low-probability patients, reflected in its calibration and net-benefit performance, is particularly relevant for safely deferring imaging in men whose disease is very unlikely to have spread.

The study also stands out for its methodological transparency. The authors followed the TRIPOD+AI reporting checklist for prediction model research, obtained ethics approval from the Ethics Committee of the First Affiliated Hospital of Anhui Medical University with informed consent from all participants or their legal representatives, and conducted the work in accordance with the Declaration of Helsinki. The research was supported by funding from the Anhui Provincial Department of Education, the Health Committee of Anhui Province, and an provincial innovation team program for male genitourinary diseases. The team declares no competing interests, and the article is open access under a Creative Commons license.

Caveats remain, as they do for any retrospective, single-institution modeling study. The cohorts come from one Chinese hospital system, and generalization to populations with different ancestry, PSA screening patterns, and diagnostic pathways will require further external validation elsewhere. The model estimates probability rather than replacing imaging outright; a low predicted risk still warrants clinical judgment, and a high predicted risk still demands confirmatory scans. Nevertheless, the study demonstrates a template for how oncology prediction models should be built: prespecified inflammatory biomarkers, penalized and repeated variable selection, bootstrap internal validation, and, crucially, a genuinely independent external cohort. If future multi-center studies replicate these results, a four-variable calculation performed at the bedside, drawing on numbers already sitting in the medical record, could quietly change how newly diagnosed prostate cancer is staged, sparing low-risk men unnecessary scans while ensuring that men with silent skeletal disease are found and treated sooner.

Subject of Research: A diagnostic prediction model using inflammatory-immune indices and clinicopathological factors to identify bone metastasis in newly diagnosed prostate cancer

Article Title: Bone metastasis in newly diagnosed prostate cancer: development and external validation of a diagnostic prediction model based on inflammatory-immune indices and clinicopathological factors to support imaging decisions

Article References: Wang, W., Tian, C., Tai, S., Tao, J., Yan, J., Wang, S., Yang, L., Zhou, J., & Luo, G. (2026). Bone metastasis in newly diagnosed prostate cancer: development and external validation of a diagnostic prediction model based on inflammatory-immune indices and clinicopathological factors to support imaging decisions. BMC Cancer. https://doi.org/10.1186/s12885-026-17147-z

Image Credits: AI Generated

DOI: 10.1186/s12885-026-17147-z

Keywords: prostate cancer, bone metastasis, neutrophil-to-lymphocyte ratio, prediction model, inflammatory-immune indices, PSA, external validation, diagnostic imaging, logistic regression, clinical decision support, BMC Cancer, cancer staging