A novel prediction tool for mortality in patients with acute lower gastrointestinal bleeding requiring emergency hospitalization: a large multicenter study

Abstract

The study aimed to identify prognostic factors for patients with acute lower gastrointestinal bleeding and to develop a high-accuracy prediction tool. The analysis included 8254 cases of acute hematochezia patients who were admitted urgently based on the judgment of emergency physicians or gastroenterology consultants (from the CODE BLUE J-study). Patients were randomly assigned to a derivation cohort and a validation cohort in a 2:1 ratio using a random number table. Assuming that factors present at the time of admission are involved in mortality within 30 days of admission, and adding management factors during hospitalization to the factors at the time of admission for mortality within 1 year, prognostic factors were established. Multivariate analysis was conducted, and scores were assigned to each factor using regression coefficients, summing these to measure the score. The newly created score (CACHEXIA score) became a tool capable of measuring both mortality within 30 days (ROC-AUC 0.93) and within 1 year (C-index, 0.88). The 1-year mortality rates for patients classified as low, medium, and high risk by the CACHEXIA score were 1.0%, 13.4%, and 54.3% respectively (all P < 0.001). After discharge, patients identified as high risk using our unique predictive score require ongoing observation.

Figure 1

Flow chart.

Figure 2

Predictive ability of the CACHEXIA score for 30-day mortality. (A) Comparison of scoring methods in the derivation cohort (n = 5459). (B) 30-day death rates by risk category for the derivation cohort (n = 5459). (C) Comparison of scoring systems in the validation cohort (n = 2795). (D) 30-day mortality rates by risk category in the validation cohort (n = 2795). P values were determined using Fisher’s exact test with Bonferroni correction (B,D). *P value was 0.003, **P value was < 0.001. ROC-AUC receiver operator characteristic curves of the area under the curve, CCI Charlson comorbidity index.

Figure 3

Predictive ability of the CACHEXIA score for 1-year mortality. (A) Cumulative probability of mortality according to risk category in the derivation cohort (n = 4030). (B) One-year mortality rates by risk category in the derivation cohort (n = 4030). (C) The cumulative death probability according to risk category in the validation cohort (n = 2054). (D) One-year mortality rates by risk category in the validation cohort (n = 2054). P values were calculated using the Cox proportional hazards model (A,C) and Fisher’s exact test with Bonferroni correction (B,D). Cases with missing data were excluded from the full case analysis (B,D). ROC-AUC receiver operator characteristic curves of the area under the curve, CCI Charlson comorbidity index.

Figure 4

Cumulative probability of mortality according to risk factors using the Kaplan–Meier method. (A) BMI (< 17.0, 17.0–18.4, and ≥ 18.5). (B) Albumin level (< 2.5, 2.5–2.9, and ≥ 3.0 g/dL). (C) Performance status (1, 2–3, and 4). (D) BUN (< 25.0, 25.0–29.9, and > 30.0 mg/dL). (E) CRP (< 1.0, 1.0–2.9, and ≥ 3.0 mg/dL). (F) Amount of blood transfusion (none, 1–7 units, and ≥ 8 units). P values were calculated using the Cox proportional hazards model. BMI body mass index, BUN blood urea nitrogen, CRP C-reactive protein.