External validation of the Oakland Score to assess safe hospital discharge among adult patients with acute lower gastrointestinal bleeding in a single New Zealand Centre

Acute lower gastrointestinal bleeding (LGIB) is a common reason for hospital admission. However, the majority resolve spontaneously and only a minority require inpatient intervention. We aimed to describe the epidemiology and aetiology of acute LGIB admissions in our institution. We also aimed to validate the Oakland Score, which can identify patients at low risk of adverse outcome from LGIB, in our population and determine the proportion who could have safely avoided admission.


Introduction
Acute lower gastrointestinal bleeding (LGIB) typically manifests as haematochezia or per rectal bleeding and is a common Emergency Department (ED) presentation.In Western countries, the incidence is estimated to be 20 to 30 cases per 100 000 and is most commonly caused by diverticulosis. 1,26][7] This includes personnel, equipment and time, use of inpatient beds, blood transfusion, interventional radiology, and upper and lower GI endoscopy.Despite this, almost a quarter of these patients are discharged without a formal diagnosis. 3Given the incidence of LGIB is likely to increase with an ageing population on antiplatelet agents and anticoagulation, 8 a risk stratification tool could allow some patients to be managed safely in the community, thereby avoiding hospital admission and associated costs.
The Oakland score is a risk assessment tool derived from a nationwide UK study which aimed to identify stable patients with LGIB who were at low risk of experiencing adverse outcomes and could safely avoid hospital admission.This tool comprises seven routinely collected clinical variables and the risk score is calculated by summing the individual components. 9rom the original paper, a patient scoring ≤8 points at presentation has a 95% probability of safe discharge from the EDdefined as the absence of subsequent rebleeding, blood transfusion, therapeutic intervention to control bleeding, in-hospital death, and readmission with further LGIB within 28 daysand may be appropriately investigated and followed up as an outpatient.This score has been recommended for use in routine clinical practice in the UK by the British Society of Gastroenterology. 10 In an external validation study in the US population involving 46 128 patients, a threshold of 8 points identified 8.7% of patients presenting with LGIB as safe for discharge from the ED, with a sensitivity of 98.4%.Extension of the threshold to 10 points identified 17.8% of patients as safe for discharge from the ED whilst maintaining a sensitivity of 96.0%. 11tudies on the demographics and resource use of patients admitted with acute LGIB in a New Zealand setting are very limited. 6urrently, there is no standardized management pathway or triage tool advocated for use at a national level in assessment of these patients.We aimed to describe the epidemiology and aetiology of acute LGIB admissions in our institution.We also aimed to validate the Oakland score in a New Zealand population and estimate how many admissions could be safely avoided if it were routinely used.

Methods
This retrospective cohort study used the prospective, validated DIVA database (version 2022.1,Otago Clinical Audit, University of Otago).This database has been in routine use in our institution for the last 30 years and captures admissions under General Surgery only.We searched for adult patients aged ≥18 years admitted acutely to a single tertiary centre, Dunedin Hospital, between 1 January 2013 and 31 December 2020 with a primary diagnosis of per rectal bleeding.We excluded patients admitted electively, those with an alternate primary diagnosis, patients with inflammatory bowel disease, and patients with a clinical diagnosis or confirmed upper GI bleed.
The DIVA data were used to obtain patient details and gender, dates of admission and discharge, previous LGIB admission or readmission (if this occurred at Dunedin Hospital within the defined study period), type of inpatient management (i.e., conservative, endoscopic, radiological, and/ or operative), operative details where relevant, and discharge diagnosis.All other information including patient ethnicity, Charlsson score, previous LGIB admission or readmission (if not already captured by DIVA data), transfer from rural hospital, regular antiplatelet or anticoagulation, initial heart rate and systolic blood pressure, initial haemoglobin, and details of hospital admission including inpatient investigations and treatment were obtained from the electronic patient record (Health Connect South, HCS, Orion Health).All radiological investigations, as well as details of endoscopic, interventional radiology, and operative treatments are electronically uploaded and accessible through the HCS record.
We retrospectively calculated the Oakland score for patients based on clinical information documented on their admission to hospital.The digital rectal examination (DRE) variable was included in the Oakland score calculation if available from the electronic HCS record, otherwise omitted if unavailable as performed previously. 11We also determined whether or not patients would have been safe to discharge instead of admitting into hospital, which we defined as the absence of blood transfusion, haemostatic intervention to control bleeding, in-hospital death, and readmission with LGIB within 28 days. 9Patients were considered to have experienced an adverse outcome if they met any of these criteria.The absence of rebleeding (additional transfusion requirements and/or a decrease in haematocrit concentration of ≥20% after 24 h of clinical stability 12 ) was also included in Oakland's external validation study.We were unable to accurately assess this from the HCS records, however, we did capture any additional transfusion requirements within the blood transfusion outcome.Local ethics approval (University of Otago Research Ethics Committee, HD21/064) and M aori consultation were obtained.
This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement guidelines. 13he ability of the Oakland score to predict the composite outcome of safe discharge was assessed using area under the receiveroperating characteristic (AUROC) curve. 11Apart from the DRE variable and rebleeding outcome (as described above), patients with other missing Oakland score component variables or incomplete outcome data were excluded from this analysis.The probability for safe discharge was calculated using a threshold of 8 points or less, 9 or less, and 10 or less.P < 0.05 was considered statistically significant.Statistical analysis was performed using Stata version 17.0 (Statacorp LLC, College Station, Texas).

A total of 761 patients admitted to Dunedin Hospital with acute
LGIB were identified.Of these, 260 were excluded (Fig. 1).The demographics of the remaining 501 patients are shown in Table 1.
Overall, two thirds of the patients (344, 69%) did not experience an adverse outcome and therefore could have safely avoided admission.One third of the patients (157, 31%) experienced an adverse outcome and would not have safely avoided admission (Table 2).
A total of 360 patients (72%) were managed conservatively with observation alone or diagnostic endoscopy.Blood products were required in 159 (32%) with a median pre-transfusion haemoglobin concentration of 78 g/L (IQR 71-92).Only 34 (7%) required haemostatic intervention: endoscopic, radiological, or surgical intervention to control bleeding.The median length of stay was 2 days.Two patients died in-hospital and 27 (6%) were readmitted to hospital with LGIB within 28 days of discharge.The cause of bleeding was not specified in 52% of the cohort (Table 2).Data were available to retrospectively calculate the Oakland score for 422 patients (84%).Of these, the DRE outcome was not documented for 65 patients.When divided into two groups of patients who would and would not have safely avoided admission, the scores were bimodally distributed (Fig. 2).An Oakland score of 10 points or less identified 127 patients (30%) who would have had a 95% probability of safely avoiding admission.The AUROC for the Oakland score was 0.85 (95% CI 0.81-0.89)as shown in Figure 3.
There were six patients with an Oakland score of 10 points or less who would not have safely avoided admission.Two patients required a  Missing vital signs at presentation (initial heart rate and systolic blood pressure).blood transfusion, two patients underwent an examination under anaesthesia, and two patients were readmitted with LGIB within 28 days.The latter two patients were both managed conservatively on their readmission and discharged.There were no deaths.

Discussion
We aimed to describe the epidemiology and aetiology of acute LGIB admissions in our tertiary New Zealand hospital.Two thirds of patients admitted were managed conservatively and did not require any specific inpatient treatment.This finding, along with the fact that only 7% required haemostatic intervention, supports data from international studies which highlights that most acute LGIB is self-limiting and will resolve spontaneously. 3,4he most common inpatient treatment was blood transfusion, given to 26% of patients.The median haemoglobin prior to transfusion was 78 g/L, suggesting a relatively liberal approach to transfusion.This is somewhat in contrast to recommended  guidelines for a restrictive threshold of 70 g/L 14 , however, given the advanced age and comorbidity of some patients in our cohort, a higher threshold may have been clinically appropriate.Furthermore, our blood transfusion rates are similar to reported rates from Australia and the UK, and much lower compared to those in the USA. 3,5,11iverticular disease was the most frequent cause of bleeding reported, in keeping with reported literature. 3,5,11On the other hand, in over half of cases, the cause was not specified on the hospital discharge summary.Although this "unspecified" group does not necessarily represent an unknown source and is likely to be overrepresented due to incomplete or inaccurate electronic documentation, it illustrates the fact that most patients admitted with LGIB are not investigated acutely as an inpatient, and therefore a presumptive, rather than definitive, diagnosis of the cause of bleeding is often made based on the history and clinical features.
We also aimed to validate the Oakland score in a New Zealand population.We found that it was a good predictor with an AUROC of 0.85 (95% CI 0.81-0.89).We found that if a threshold of 10 points or less had been applied, 30% of patients could have avoided admission.This is a significantly higher proportion of patients identified compared with 17.8% in the external validation study in the USA using the same cut-off. 11This is possibly due, at least in part, to our smaller cohort and the fact that the Oakland score was not able to be calculated for 16% of patients.However, in our cohort baseline haemoglobin levels were relatively high, and 69% met the criteria for safely avoiding admission compared to only 48% in that study, which means that overall, our patient population was more likely to be correctly classified as low risk.
Application of the Oakland score into routine clinical practice may reduce hospitalization for acute LGIB.A threshold of 10 points or less identified 127 patients over an eight-year period, or 15 patients per year.Given a median length of stay of 2 days, avoidance of hospitalization for these patients would potentially save 30 bed-days annually.Minimizing unnecessary acute admissions is of particular importance in a COVID era, in which elective surgery is significantly restricted and during lockdowns led to one in seven patients not undergoing their planned cancer operation. 15Furthermore, the growing backlog of elective surgical cases requires a combination of strategies to tackle, 16 and ultimately the availability and capacity of bed resource is an essential component.
Although the majority of patients with an Oakland score of 10 points or less would have been safely discharged (i.e., would have been correctly identified as low risk), a small proportion may have represented.The potential for misclassification and potential for adverse outcomes is inherent to many clinical risk scoring tools, 17 and highlights the need to strike the balance between a suitable threshold which will be of practical use in a clinical setting, versus the risk of a type 2 error.We believe a threshold of 10 points or less strikes the right balance here, 11 and based on our data, the risk of clinical harm in this group appears to be low with a number needed to harm of 60 patients.However, use of this tool should not replace or override clinical judgement for any individual patient.Furthermore, the introduction of any new stratification tool into routine clinical use should be done so within a quality improvement project where outcomes and unintended consequences can be monitored until confidence in the clinical safety has been gained.
A number of limitations in this study should be noted.First of all, it was retrospective and only included patients admitted to hospital with LGIB and captured within the DIVA database.Secondly, details regarding inpatient management were obtained primarily from the electronic record.Since all blood product, investigation, and intervention details are routinely uploaded electronically in our institution, it is unlikely that a more thorough search process incorporating written records would have significantly altered our results in this regard.However, it was assumed that the heart rate and systolic blood pressure documented on the electronic ED assessment note and/ or discharge summary represented the initial vital signs at hospital presentation.In addition, previous LGIB admissions and readmissions were only captured if identified within the DIVA data, documented in the electronic HCS discharge summary, or uploaded as admission events on HCS (the latter limited to admissions to Dunedin hospital or other hospitals in the South Island occurring within the last 15 years).Thirdly, the rebleeding outcome was unable to be assessed and therefore was not included as part of the safe discharge criteria.This may have led to an underestimation of the proportion of patients not meeting the criteria for safe discharge in our study.However, any additional transfusion requirements were captured within the blood transfusion outcome and given a median length of stay of only 2 days, it is unlikely that significant rebleeding would have been a common occurrence.

Conclusion
This study has shown that the majority of patients admitted with acute LGIB in our institution are managed conservatively.It has also validated the use of the Oakland score to predict low risk patients suitable for discharge from the ED in a New Zealand setting.A structured introduction of this simple tool into everyday practice when assessing these patients, using a threshold of 10 points or less, could prevent unnecessary hospital admissions.

Fig. 2 .
Fig. 2. Distribution of Oakland scores in cohort, = 422; Green = Safe avoidance of admission, Red = Not safe avoidance of admission.

Table 1
Characteristics of patients admitted with LGIB.Continuous variables are presented as median (interquartile range), categorical variables as n

Table 2
Summary of inpatient management, outcomes, and cause of bleeding.Continuous variables presented as median (interquartile range), categorical variables as n (percentage).There are no missing data †Safe avoidance of admission defined as absence of blood transfusion, haemostatic intervention to control bleeding, in-hospital death, and readmission with LGIB within 28 days. 9‡ Note for interpretation, some patients received more than one management or intervention.§ As documented on hospital discharge summary or confirmed by inpatient investigations.