External Validation Of The Strate'S Predictive Model Of Severity In Lower Intestinal Bleeding

Introduction: Acute lower intestinal bleeding (LIB) is a common medical emergency. Recently Strate et al. developed and validated a predictive model to assess the severity of acute LIB. Generalisability of a prediction rule in clinical practice requires an external validation in other clinical contexts. Objective: External validation of the Strate severity index for LIB. Patients and Methods: Design: Prospective, cohort study. Patients (validation cohort): Patients admitted to our hospital with the diagnosis of LIB from April 2005 to April 2007. Strate predictive model: Severity was defined as transfusion of ≥ 2 units of red blood cells, and/or decrease in hematocrit of ≥ 20% within the first 24 hours, and/or recurrent bleeding (rectal bleeding together with decrease in hematocrit of ≥ 20% and/or additional blood transfusions, and/or readmission due to the same condition within 1 week of discharge). Independent predictive factors were: heart rate > 100 bpm, blood pressure < 115 mm Hg, comorbid conditions, rectal bleeding in the first 4 hours after admission, nontender abdominal exam and aspirine use. Patients were clasified, according to the number of accumulated factors, into low risk (no risk factors), moderate (1 to 3 risk factors) and high risk (> 3 risk factors). Statistics: Observed risk of severe bleeding in each group was compared to the risk predicted from the Strate cohort. Results: A total number of 410 patients were admitted during the study period. 159 (38.7%) of them fulfilled the severity criteria. The observed and predicted risks of severe bleeding were respectively: 16% (7/42) vs. 9% (1/11) (p = 0.53) in the low risk group, 39% (130/333) vs. 43% (85/197) (p = 0.35) in the moderate risk group and 63% (22/35) vs 79% (37/44) (p = 0.05) in the high risk group. The area under the ROC curve was 0.65 for the validation cohort and 0.76 for the original Strate cohort. Conclusion: Strate model can stratify patients with LIB into different severity groups. However its discriminative ability worsens outside the context where the model was developed, therefore its generalized use remains limited.