New Blood Ratio Flags Short-Term Death Risk in Heart Failure, but Falls Short of Existing Indices

A simple ratio built from three routine blood tests has been linked to short-term survival in patients hospitalized with heart failure, but the researchers behind the finding are urging caution about how far it should be trusted. In a retrospective observational study published in Frontiers in Nutrition, a team led by Gengxu He of Hebei North University examined the neutrophil-to-prognostic nutritional index ratio, or NPNR, a laboratory composite that divides a patient’s neutrophil count by the prognostic nutritional index, itself calculated from serum albumin plus five times the lymphocyte count. Higher values of this ratio were associated with greater 28-day and 90-day all-cause mortality in a Chinese discovery cohort, yet the marker failed to outperform established inflammatory indices and showed weaker, incomplete replication in an independent American intensive care cohort.

The discovery data came from the Zigong Heart Failure Database, an electronic health record resource built at the Fourth People’s Hospital of Zigong in Sichuan Province, China. Of 2,008 consecutive patients hospitalized with heart failure between December 2016 and June 2019, 1,885, or 93.9 percent, had the neutrophil count, lymphocyte count, and serum albumin measurements needed to compute NPNR. Laboratory values were drawn from the first day of hospitalization. Because the ratio’s distribution was strongly right-skewed, the investigators analyzed it on a log2 scale, so each hazard ratio describes the change in mortality risk associated with a doubling of the index. Outcomes were tracked at fixed horizons of 28 days, 90 days, and, in sensitivity analyses, 180 days.

The statistical picture was constrained by a striking scarcity of deaths. Only 32 patients died within 28 days and 36 within 90 days with known event times, meaning the fully adjusted model, which estimated seven regression coefficients, operated at roughly 4.6 to 5.1 events per parameter, well below the levels conventionally considered robust against overfitting. Nevertheless, after adjustment for age, sex, the Charlson Comorbidity Index, New York Heart Association functional class, B-type natriuretic peptide, estimated glomerular filtration rate, and serum sodium, each doubling of NPNR was associated with a 73 percent higher hazard of death at 28 days (hazard ratio 1.728, 95 percent confidence interval 1.130 to 2.642) and a 68 percent higher hazard at 90 days (hazard ratio 1.684, 95 percent confidence interval 1.124 to 2.524). Simpler four-parameter core models and Firth-penalized regressions designed for sparse events produced similar estimates, and multiple imputation, ridge penalization, and analyses excluding patients with liver disease or malignancy all preserved the direction of the association.

When patients were divided into cohort-specific tertiles, a clear risk gradient emerged. Twenty-eight-day mortality climbed from 0.48 percent in the lowest tertile to 1.11 percent in the middle and 3.50 percent in the highest; the 90-day figures were 0.48, 1.59, and 3.82 percent. In the fully adjusted models, the top tertile carried roughly a fourfold higher hazard than the bottom at both horizons, with significant tests for trend. But the lowest tertile contained only three deaths at each time point, so the confidence intervals around the tertile-specific estimates stretched extremely wide, from about 1.3 to 15.4 for the top group. Kaplan-Meier curves separated cleanly across tertiles, with log-rank p values below 0.001, yet the authors emphasized that the continuous exposure remained the more reliable measure.

Restricted cubic spline analyses added an intriguing wrinkle to the shape of the relationship. In the Zigong cohort, the exposure-response curves suggested nonlinearity at both time horizons, with overall and nonlinearity p values of 0.012 and 0.009 at 28 days and 0.016 and 0.013 at 90 days. Risk appeared to rise steeply across lower-to-moderate NPNR values before approaching a plateau, though confidence intervals ballooned at the upper end of the distribution where observations and deaths were scarce. The authors explicitly cautioned that this pattern does not support any fixed clinical cutoff for the index.

Discrimination testing told a more sobering story. The area under the receiver operating characteristic curve for NPNR was 0.727 at 28 days and 0.717 at 90 days, indicating moderate stand-alone performance. In paired DeLong comparisons, however, NPNR showed no statistically significant superiority over the prognostic nutritional index, the neutrophil-to-lymphocyte ratio, or the systemic immune-inflammation index at either horizon. Adding NPNR to a base clinical model raised the optimism-corrected C-index from 0.717 to 0.738 at 28 days and from 0.708 to 0.735 at 90 days, but the bootstrap-corrected confidence intervals for these increments crossed zero, and the corrected changes in area under the curve did so as well. Likelihood ratio tests reached p = 0.013 at both horizons, yet the authors interpreted the incremental prognostic information as exploratory and not demonstrably clinically meaningful.

To probe whether the association traveled beyond a single health system, the team turned to MIMIC-IV, a de-identified database of hospital and intensive care unit records from Beth Israel Deaconess Medical Center in Boston covering 2008 to 2022. After retaining the earliest eligible ICU stay per patient, the replication cohort comprised 1,150 unique adults with heart failure, a far sicker population than the Zigong group: 90-day mortality reached 37.7 percent overall, compared with roughly 2 percent in China, and the median NPNR was more than twice as high at 0.274 versus about 0.115. Mortality again climbed across tertiles, from 21.9 to 35.2 percent at 28 days and 30.2 to 45.7 percent at 90 days, with significant tests for trend and Kaplan-Meier separation.

The continuous association, however, weakened considerably in the American cohort. In the fully adjusted model, each doubling of NPNR yielded a hazard ratio of 1.112 for 28-day mortality, borderline significant at p = 0.049 with a confidence interval touching 1.000, and 1.081 for 90-day mortality, which did not reach conventional significance at p = 0.096. Substituting continuous standardized age for the dichotomized version nudged both p values above 0.05. Spline analyses in MIMIC-IV found significant overall associations but no clear nonlinearity, producing curves closer to monotonic than the plateau-shaped Zigong relationship. The authors characterized this as partial external replication of the direction of the association rather than formal validation of a prediction model, noting that differences in baseline risk, ICU acuity, sampling timing, and albumin distribution could all contribute to the attenuated estimates.

Biologically, the interpretation demands care. Neutrophils are central effectors of innate inflammation and can drive endothelial dysfunction, microvascular injury, and tissue damage through protease release, oxidative stress, and neutrophil extracellular trap formation, processes relevant to heart failure progression. But albumin is a negative acute-phase reactant whose concentration falls with inflammation, hepatic dysfunction, renal impairment, and congestion, not merely with poor nutrition. Because NPNR is mathematically constructed from neutrophils, lymphocytes, and albumin, it overlaps by definition with the neutrophil-to-lymphocyte ratio, the prognostic nutritional index, and the systemic immune-inflammation index, so differences in these components across tertiles cannot establish a distinct biological pathway. The authors concluded that NPNR is best regarded as an exploratory laboratory composite associated with systemic inflammation and overall illness severity rather than a genuine nutritional marker, particularly since validated nutritional assessments were unavailable.

The study’s limitations are substantial and were stated plainly. Beyond the sparse death counts, left ventricular ejection fraction was missing for 67.96 percent of the Zigong patients, preventing classification into reduced, mildly reduced, or preserved ejection fraction phenotypes and limiting generalizability. Residual confounding and reverse causation cannot be excluded in a retrospective design, serial inflammatory measurements were absent, and the markedly different acuity of the two cohorts precludes direct comparison of effect sizes. An exploratory subgroup interaction with diabetes at 90 days survived false discovery rate correction but was deemed hypothesis-generating only. The authors’ bottom line is measured: NPNR can be computed cheaply from routine blood tests and may serve as an adjunctive signal of inflammatory and illness burden, but it demonstrated no superior discrimination over existing indices and should not be used independently for treatment decisions or routine clinical thresholds. Prospective studies with fuller heart failure phenotyping and larger event counts will be needed to determine whether the ratio adds anything that established markers do not already provide.

Subject of Research: Association of the neutrophil-to-prognostic nutritional index ratio with short-term mortality in heart failure

Article Title: Neutrophil-to-prognostic nutritional index ratio and short-term mortality in heart failure: an exploratory association without demonstrated superiority over PNI, NLR, or SII

Article References: Yuan, Z., Li, M., Zhao, L., Geng, H., Ji, Q., Zuo, K., Luo, Y., Liang, G., & He, G. (2026). Neutrophil-to-prognostic nutritional index ratio and short-term mortality in heart failure: an exploratory association without demonstrated superiority over PNI, NLR, or SII. Frontiers in Nutrition, 13, Article 1995940. https://doi.org/10.3389/fnut.2026.1995940

Image Credits: AI Generated

DOI: 10.3389/fnut.2026.1995940

Keywords: heart failure, NPNR, prognostic nutritional index, systemic inflammation, mortality, MIMIC-IV, Cox regression, biomarkers, retrospective study, neutrophils, albumin, Neutrophil-to-prognostic