Blood Sugar Spike Plus Frailty Flags Fivefold Death Risk in Heart Patients

When a patient with coronary heart disease is rushed into the intensive care unit, clinicians face an urgent and often agonizing question: how likely is this person to die within the next month? Traditional scoring systems weigh age, vital signs, and diagnoses, but they can miss two crucial dimensions of risk that operate on very different timescales. One is the sudden metabolic storm that critical illness unleashes on the body, driving blood glucose to dangerous heights even in people who have never had diabetes. The other is the slow, quiet erosion of physiological reserve known as frailty, which determines how much punishment a body can absorb before it fails. A new study published in BMC Endocrine Disorders suggests that reading these two signals together, rather than separately, may dramatically sharpen the prediction of early death in critically ill cardiac patients.

The research, conducted by Hui-Jian Wang, Yi-Jun Huang, and Huabin He at the Second Affiliated Hospital of Nanchang University and Jiujiang NO.1 People’s Hospital in China, mined the MIMIC-IV database, a vast, de-identified repository of intensive care records from Beth Israel Deaconess Medical Center. The team identified 5,580 adult ICU patients with coronary heart disease and followed their outcomes for thirty days. Among this cohort, 496 patients, or 8.9 percent, died within a month of admission. The central question was whether a simple mathematical pairing of two admission-time measurements could outperform conventional clinical models in spotting who would fall into that group.

The first measurement is the stress hyperglycemia ratio, or SHR. Unlike a raw glucose reading, which can be inflated by many transient factors, the SHR is a ratio: the glucose measured on admission divided by an estimate of the patient’s chronic average glucose, derived from glycated hemoglobin, or HbA1c. Because HbA1c reflects blood sugar levels over the preceding two to three months, the ratio isolates the glucose surge attributable to the acute stress of illness itself. A high SHR signals that the body is mounting an exaggerated metabolic response to injury, a state linked in prior research to inflammation, oxidative damage, and worse outcomes after heart attacks and infections.

The second measurement is a laboratory-based frailty index, abbreviated FI-Lab. Frailty is classically assessed through questionnaires and physical performance tests, which are impractical in a busy ICU. The FI-Lab sidesteps this problem by constructing a frailty score entirely from routine laboratory parameters and vital signs, such as markers of inflammation, organ function, and physiological instability, that are already collected at admission. The result is a quantitative snapshot of the patient’s baseline physiological reserve, computable from data that exists in every hospital record. Patients were sorted into tertiles, or thirds, of both SHR and FI-Lab, creating nine possible combined groups.

The results of the joint stratification were striking. Using multivariable Cox regression, which adjusts for confounding factors, the researchers found that patients in the highest tertile of both SHR and FI-Lab faced by far the gravest danger. Compared with the reference group, their adjusted hazard ratio for thirty-day all-cause mortality was 5.86, with a 95 percent confidence interval of 3.41 to 10.08. In plain terms, a critically ill cardiac patient whose body was both mounting an intense stress-glucose response and carrying a depleted physiological reserve was nearly six times more likely to die within thirty days than an otherwise comparable patient at the low end of both measures.

Beyond the raw risk association, the team tested whether adding SHR and FI-Lab to a baseline clinical model genuinely improved its predictive power. Three complementary statistical tools were deployed. The area under the receiver operating characteristic curve, or AUC, which measures how well a model separates those who die from those who survive, rose to 0.8166 when the two new variables were included. Integrated discrimination improvement, or IDI, quantified a gain of 0.025 in the model’s ability to assign higher predicted probabilities to patients who actually died. Net reclassification improvement, or NRI, showed that 22.9 percent more patients were correctly moved into appropriate risk categories. Together, these metrics indicate that the combined metabolic and frailty assessment adds real, incremental information beyond what conventional clinical variables capture.

To push the analysis further, the researchers built machine learning models as a supplementary predictive exercise. Among the algorithms tested, Random Forest achieved the best performance on the test set, reaching an AUC of 0.846. Crucially, the team did not treat the algorithm as a black box. They applied SHAP analysis, or SHapley Additive exPlanations, a technique borrowed from cooperative game theory that assigns each input variable a quantified contribution to every individual prediction. The SHAP results confirmed that both the frailty index and the stress hyperglycemia ratio ranked among the most important features driving the model’s outputs, validating the central hypothesis of the study through an entirely independent analytical lens.

The conceptual elegance of the approach lies in how the two markers complement each other. SHR captures the acute, dynamic dimension of critical illness, the intensity of the metabolic assault unfolding in real time, while FI-Lab captures the chronic, static dimension, the accumulated vulnerability of a body whose reserves have been worn down by age, disease, and inflammation. Neither alone tells the full story. A patient with a modest glucose spike but profound frailty may be as endangered as one with a dramatic metabolic surge but robust reserves. By stratifying patients on both axes simultaneously, clinicians can identify the small subgroup, roughly one in nine combinations of the tertile scheme, where risk concentrates most heavily, potentially guiding decisions about monitoring intensity, treatment aggressiveness, and conversations with families.

Several caveats deserve attention. The study is retrospective, drawing on data from a single academic medical center captured in the MIMIC-IV database, and the findings will need prospective validation in independent cohorts before they can change bedside practice. The SHR calculation depends on HbA1c availability at admission, which may introduce selection considerations, and the FI-Lab, while practical, is a proxy rather than a gold-standard frailty assessment. The authors also note that the work received no external funding and was conducted under the Declaration of Helsinki using fully anonymized data, with the original collection approved by the MIT and BIDMC Institutional Review Boards. Still, the message is compelling: two numbers already sitting in every ICU chart, when read together, can reveal a nearly sixfold gradient in mortality risk. As predictive medicine moves toward richer, multiparameter portraits of the patient, this study offers a vivid example of how combining an acute stress signal with a measure of long-term resilience can turn routine laboratory data into a life-saving early warning system.

Subject of Research: Combined use of the stress hyperglycemia ratio and a laboratory-based frailty index to predict thirty-day mortality in critically ill patients with coronary heart disease using the MIMIC-IV database.

Article Title: Combined assessment of stress hyperglycemia ratio and frailty for mortality risk prediction in critically ill patients with coronary heart disease: insights from the MIMIC-IV database

Article References: Wang, H.-J., Huang, Y.-J., & He, H. (2026). Combined assessment of stress hyperglycemia ratio and frailty for mortality risk prediction in critically ill patients with coronary heart disease: insights from the MIMIC-IV database. BMC Endocrine Disorders. https://doi.org/10.1186/s12902-026-02516-3

Image Credits: AI Generated

DOI: 10.1186/s12902-026-02516-3

Keywords: stress hyperglycemia ratio, frailty index, coronary heart disease, intensive care unit, mortality prediction, MIMIC-IV, machine learning, Random Forest, SHAP, HbA1c, risk stratification, critical illness