Prediction of 90-day mortality among cancer patients with unplanned hospitalisation: a retrospective validation study of three prognostic scores

Background

Accurate prediction of 90-day mortality in hospitalised cancer patients is critical for guiding personalised treatment decisions and optimising oncologic care. However, existing prognostic models often lack sufficient precision, particularly in distinguishing between high- and low-risk patients. In this retrospective study, we independently evaluated the prognostic performance of three scoring systems—the Prognostic Score for Hospitalised Cancer Patients (PROMISE), the Gustave Roussy Immune (GRIm) score, and the C-reactive protein–Triglyceride–Glucose Index (CTI)—in patients admitted for unplanned hospitalisations.

Methods

This retrospective observational study was conducted at the Medical Oncology Clinic of Ankara Etlik City Hospital, Turkey, and included patients aged 18 years or older with a diagnosis of cancer who were hospitalised unexpectedly between February 2023 and February 2024. Laboratory data were retrieved from the institutional hospital information system. The PROMISE score was calculated using its original specification via the online tool (https://promise.vhio.net/). The GRIm score was calculated based on neutrophil-to-lymphocyte ratio (NLR), albumin, and lactate dehydrogenase (LDH). The CTI score was computed as: CTI = [0.412 × ln (C-reactive protein [CRP])] + ln [Triglyceride × Glucose/2], with a cut-off value of 4.78. A PROMISE–CTI Combined score was derived using regression-based weighting. Risk stratification was performed for all three scores using validated thresholds. Statistical analyses included Kaplan–Meier survival analysis, log-rank tests, univariable and multivariable logistic regression to assess predictors of 90-day mortality, and receiver operating characteristic (ROC) curve analysis to evaluate discriminatory performance.

Findings

Among 1657 hospitalised cancer patients screened during the study period, 1109 met the inclusion criteria and were included in the analysis. PROMISE and GRIm scores were calculated for all 1109 patients, while CTI score was assessed in 333 patients with complete laboratory data. The 90-day mortality rate was 63.7% (n = 707). High PROMISE score (OR: 3.32, 95% CI: 1.40–7.86; p = 0.006) and high CTI score (OR: 2.85, 95% CI: 1.32–6.18; p = 0.008) were associated with increased 90-day mortality. Low PROMISE score (OR: 0.22, 95% CI: 0.10–0.49; p = 0.001) and low CTI score (OR: 0.35, 95% CI: 0.17–0.73; p = 0.003) were associated with reduced 90-day mortality. High GRIm score (OR: 1.83, 95% CI: 0.83–2.91; p = 0.07) and low GRIm score (OR: 0.73, 95% CI: 0.47–1.20; p = 0.08) were not significantly associated with 90-day mortality. The area under the curve (AUC) of the PROMISE–CTI Combined score was 0.884 (95% CI: 0.849–0.919; p < 0.0001). Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy of the PROMISE–CTI Combined score were 92.4%, 81.1%, 85.3%, 89.6%, and 86.7%, respectively.

Interpretation

The PROMISE score demonstrated strong discriminatory ability in predicting 90-day mortality among cancer patients admitted for unplanned hospitalisations. Integration of the CTI score further improved risk stratification by incorporating nutritional and inflammatory markers. The PROMISE–CTI Combined score may serve as a practical clinical tool for short-term prognostic assessment in this setting. Prospective, multicentre, randomised studies are needed to confirm the clinical utility and generalisability of the PROMISE–CTI Combined score.

Funding

This study received no funding.