Predictive Factors for 30-day Mortality and Predictive Score
The analysis included 8,254 patients (median [interquartile range] age, 74 [63–82] years; 4,973 [60.2%] men). Except for platelet counts of < 150,000/mm3 and the use of computed tomography for diagnosis, the baseline characteristics of the derivation (n = 5,459) and validation (n = 2,795) cohorts were similar (Table 1).
In the derivation cohort, 51 (0.9%) of 5,459 patients died within 30 days; 7 (0.1%) patients died of bleeding-related causes (Supplementary Table 4). The univariate analysis revealed 26 baseline parameters linked to 30-day mortality. Multivariate logistic regression analysis identified the following six factors as risk factors for 30-day mortality: PS ≥ 2, albumin level < 3.0 g/dL, blood urea nitrogen (BUN) ≥ 25 mg/dL, C-reactive protein (CRP) ≥ 1.0 mg/dL, comorbid metastatic cancer, and cirrhosis (Table 2). Based on the coefficients from the multivariate analysis, we created a novel weighted score for predicting 30-day mortality (maximum 11 points) utilizing these six factors (Table 2). For the derivation cohort, the ROC-AUC of the new score was 0.92 (95% CI, 0.88–0.96), which was significantly higher than existing clinical risk scores (ROC-AUC: Sengupta, 0.89; NOBLADS, 0.84; CCI, 0.78; Oakland, 0.71) (Fig. 2a). Based on the novel score, patients in the derivation cohort were categorized into low-score ≤ 1 (n = 3,927), medium-score 2–4 (n = 1,009), and high-score ≥ 5 (n = 132) groups. The 30-day mortality rates for the low-, medium-, and high-score groups were 0.1% (n = 5), 2.3% (n = 23), and 18.2% (n = 24), respectively (P < .001 for each comparison) (Fig. 2b).
In the validation cohort, the ROC-AUC for the new score was 0.90 (95% CI, 0.85–0.95) (Fig. 2c). Patients in low-, medium-, and high-score groups had 30-day mortality rates of 0.2% (n = 4), 1.2% (n = 6), and 20.6% (n = 13), respectively. The mortality rates were significantly higher in the medium-score group compared with the low-score group and in the high-score group compared with the low- and medium-score groups (low vs. medium score, P = .003; high vs. medium/low scores, P < .001) (Fig. 2d).
Predictive Factors for 1-year Mortality and Predictive Score
Mortality within 1 year was assessed in 6,084 cases after excluding the 2,170 cases (Fig. 1). The characteristics and in-hospital care of the derivation and validation cohorts were similar, except for five factors: platelet count < 150,000/mm3, diabetes mellitus, the use of CT for diagnosis, surgery for hemostasis, and bleeding due to cancer (Supplementary Table 2). In the derivation cohort, 163 (3.0%) of 4,030 patients died within 1 year. Univariate analysis showed that 34 factors were associated with 1-year mortality among the baseline characteristics and in-hospital management data. The Cox proportional hazards model showed 10 risk factors for 1-year mortality, including PS ≥ 2, albumin level < 3.0 g/dL, BUN ≥ 25 mg/dL, CRP ≥ 1.0 mg/dL, comorbid metastatic cancer, cirrhosis, body mass index (BMI) < 18.5, blood transfusion during hospitalization, blood malignancy and solid cancer, and bleeding from malignancy (hematochezia due to cancer) (Table 3). Based on the coefficients from the multivariate analysis, we created a novel weighted score for predicting 1-year mortality (maximum 17 points) (Table 3).
Table 3
Predictors of 1-year mortality by Cox proportional hazards model (derivation cohort, n = 4,030)
Characteristics | Crude hazard ratio (95% CI) | P value* | Adjusted hazard ratio (95% CI) | Coefficient (95% CI) | P value* | Score points |
Age ≥ 70 y | 1.98 (1.39–2.84) | < .001 | | | | |
Sex (male) | 1.17 (0.85–1.61) | .347 | | | | |
BMI ≤ 18.5 | 4.01 (2.89–5.55) | < .001 | 1.80 (1.21–2.68) | 0.59 (1.89–0.99) | .004 | 1 |
Current drinker | 0.75 (0.53–1.06) | .102 | | | | |
Current smoker | 0.69 (0.43–1.12) | .136 | | | | |
Performance status ≥ 2 | 4.29 (3.09–5.96) | < .001 | 2.22 (1.49–3.31) | 0.80 (0.40–1.20) | < .001 | 2 |
Hemodynamics | | | | | | |
Systolic blood pressure ≤ 100 mmHg | 2.92 (2.08–4.09) | < .001 | | | | |
Heart rate ≥ 100/min | 1.61 (1.14–2.27) | .007 | | | | |
Symptom | | | | | | |
Altered mental status | 1.18 (0.64–2.18) | .588 | | | | |
Abdominal pain | 0.81 (0.53–1.23) | .319 | | | | |
Diarrhea | 1.17 (0.70–1.95) | .562 | | | | |
Laboratory data | | | | | | |
Hemoglobin < 12 g/dL | 3.91 (2.55–6.00) | < .001 | | | | |
WBC > 10000/mm3 | 1.26 (0.88–1.80) | .210 | | | | |
Platelet count < 150000/mm3 | 2.66 (1.92–3.69) | < .001 | | | | |
Albumin level > 3.0 g/dL | 6.77 (4.95–9.26) | < .001 | 2.03 (1.33–3.09) | 0.71 (0.28–1.13) | .001 | 1 |
INR ≥ 1.5 | 1.99 (1.33–2.98) | < .001 | | | | |
Hematocrit ≤ 35% | 3.61 (2.44–5.33) | < .001 | | | | |
BUN ≥ 25 mg/dL | 3.24 (2.38–4.40) | < .001 | 1.90 (1.32–2.73) | 0.64 (0.28–1.01) | .001 | 1 |
Creatinine ≥ 1.5 mg/dL | 3.35 (2.42–4.64) | < .001 | | | | |
CRP ≥ 1.0 mg/dL | 3.08 (2.27–4.19) | < .001 | 1.70 (1.15–2.51) | 0.53 (0.14–0.92) | .008 | 1 |
Home medication | | | | | | |
Low-dose aspirin | 1.03 (0.71–1.49) | .884 | | | | |
Antiplatelet drug (non-aspirin) | 0.96 (0.62–1.49) | .842 | | | | |
Warfarin | 1.16 (0.68–1.97) | .584 | | | | |
Direct oral anticoagulant | 1.29 (0.73–2.27) | .381 | | | | |
NSAIDs | 2.03 (1.35–3.04) | < .001 | | | | |
Acetaminophen | 2.32 (1.19–4.55) | .014 | | | | |
Corticosteroid | 2.02 (1.25–3.25) | .004 | | | | |
Comorbidity | | | | | | |
Previous diverticular bleeding | 0.34 (0.18–0.67) | < .001 | | | | |
Hypertension | 0.72 (0.53–0.98) | .038 | | | | |
Dyslipidemia | 0.77 (0.53–1.10) | .148 | | | | |
Diabetes mellitus | 1.26 (0.89–1.81) | .197 | | | | |
Diabetes complication | 1.57 (0.80–3.08) | .187 | | | | |
Hemiplegia | 1.47 (0.60–3.57) | .399 | | | | |
Cerebrovascular disease | 1.16 (0.73–1.70) | .611 | | | | |
Chronic obstructive pulmonary disease | 1.73 (0.88–3.39) | .110 | | | | |
Dementia | 2.72 (1.65–4.49) | < .001 | | | | |
Collagen disease | 1.61 (0.91–2.83) | .102 | | | | |
Ischemic heart disease | 0.81 (0.52–1.27) | .354 | | | | |
Heart failure | 2.40 (1.62–3.57) | < .001 | | | | |
Previous peptic ulcer | 1.48 (0.91–2.41) | .119 | | | | |
Renal failure | 2.16 (1.53–3.04) | < .001 | | | | |
Peripheral arterial disease | 1.49 (0.79–2.83) | .219 | | | | |
Chronic hepatitis | 1.74 (0.85–3.54) | .128 | | | | |
Cirrhosis | 5.64 (3.41–9.32) | < .001 | 4.44 (2.46–8.01) | 1.49 (0.90–2.08) | < .001 | 3 |
Blood malignancy and nonmetastatic solid cancer† | 7.82 (5.72–10.68) | < .001 | 3.03 (1.97–4.66) | 1.11 (0.68–1.54) | < .001 | 2 |
Metastatic cancer | 14.96 (10.58–21.16) | < .001 | 4.48 (2.75–7.31) | 1.50 (1.01–1.99) | < .001 | 3 |
Diagnostic procedure | | | | | | |
CT | 0.80 (0.58–1.09) | .161 | | | | |
Therapeutic procedures | | | | | | |
Endoscopic treatment | 0.77 (0.53–1.12) | .171 | | | | |
Interventional radiology | 1.03 (0.33–3.21) | .965 | | | | |
Surgery | 2.34 (1.10–4.99) | .028 | | | | |
Blood transfusion during hospitalization | 3.51 (2.58–4.77) | < .001 | 1.92 (1.37–2.82) | 0.65 (0.27–1.04) | .001 | 1 |
The final diagnosis of hematochezia | | | | | | |
Colonic diverticular bleeding | 0.34 (0.25–0.48) | < .001 | | | | |
Rectal ulcer | 5.43 (3.19–9.24) | < .001 | | | | |
Angioectasia | 2.23 (0.99–5.05) | .054 | | | | |
Upper gastrointestinal bleeding | 3.59 (1.89–6.80) | < .001 | | | | |
Small intestinal bleeding | 2.24 (1.10–4.56) | .026 | | | | |
Malignancy | 6.43 (4.17–9.92) | < .001 | 2.61 (1.56–4.36) | 0.96 (0.45–1.47) | < .001 | 2 |
Others | 2.61 (1.71–4.00) | < .001 | | | | |
In-hospital outcomes | | | | | | |
Length of hospital stay ≥ 8 days | 2.43 (1.74–3.38) | < .001 | | | | |
Rebleeding during hospitalization | 1.01 (0.65–1.57) | .956 | | | | |
Thromboembolism | 4.54 (2.01–10.27) | < .001 | | | | |
Abbreviations: BMI, body mass index; BUN, blood urea nitrogen; CI, confidence interval; CRP, C-reactive protein; CT, computed tomography; INR, international normalized ratio; NSAIDs, nonsteroidal anti-inflammatory drugs; WBC, white blood cell count. |
* Bold values indicate P < .05. |
† Blood malignancy was included with the comorbidity of leukemia and lymphoma. |
In the derivation cohort, the c-statistic for the novel score was 0.87 (95% CI, 0.84–0.90). Patients were categorized into low-score ≤ 4 (n = 2,867), medium-score 5–9 (n = 594), and high-score ≥ 10 (n = 81) groups. A log-rank test revealed that patients in the high-score group had significantly higher probabilities of death compared with patients in the low-score group (HR, 84.20; 95% CI, 52.6–137.9; P < .001) and the medium-score group (HR, 14.25; 95% CI, 9.32–21.8; P < .001) (Fig. 3a). The 1-year mortality rates for the low-, medium-, and high-score groups were 1.0%, 13.4%, and 54.3%, respectively (all: P < .001) (Fig. 3b).
The c-statistic for novel scores in the validation cohort was 0.84 (95% CI, 0.80–0.89). Compared to the low-score group, the high-score group (HR, 33.10; 95% CI, 15.8–69.4; P < .001) and the intermediate-score group (HR, 9.25; 95% CI, 5.6–15.3; P < .001) had significantly higher probabilities of death (Fig. 3c). The 1-year mortality rates for the low-, medium-, and high-score groups were 1.6%, 14.4%, and 38.5%, respectively (low vs. intermediate and low vs. high: P < .001; intermediate vs. high: P = .012) (Fig. 3d).
Group analysis was carried out to investigate the association between 1-year mortality and various factors, which were split into three categories. In the Cox proportional hazards model, a decrease in BMI (< 17.0, 17.0–18.4, and ≥ 18.5) and albumin levels (< 2.5, 2.5–2.9, and ≥ 3.0 g/dL) significantly increased the probability of 1-year mortality (all: P < .001) (Fig. 4a, b). In contrast, increases in PS (1, 2–3, and 4), BUN (< 25.0, 25.0–29.9, > 30.0 mg/dL), CRP (< 1.0, 1.0–2.9, and > 3.0 mg/dL), and the amount of blood transfusion (none, 1–7 units, and 8 units) significantly increased the probability of 1-year mortality (all: P < .001) (Fig. 4c–f).