Dr Gwo-Tzer Ho, University Department of Gastroenterology, Western General Hospital, Edinburgh EH4 2XU, UK. E-mail: email@example.com
Background : The failure rate of medical therapy in severe ulcerative colitis is high. A risk index, to aid the identification of patients of not responding at an early stage to intravenous corticosteroid therapy, would be useful to facilitate second-line treatment or surgery.
Methods : We recruited 167 consecutive patients with severe ulcerative colitis between January 1995 and March 2002; and employed multiple logistic regression to analyse parameters within the first 3 days of medical therapy. We applied statistical modelling to formulate a risk score according to the likelihood of medical failure.
Results : Sixty-seven (40%) patients failed to respond to medical therapy. Multiple logistic regression analysis identified mean stool frequency and colonic dilatation within the first 3 days and hypoalbuminaemia as independent predictors of outcome (P < 0.001, 0.001 and 0.002 respectively). A numerical risk score was formulated based on these variables. Patients with scores of 0–1, 2–3 and ≥4 had a medical therapy failure rate of 11%, 43% and 85% respectively. Receiver–operator characteristic analysis of this score yielded area under curve of 0.88, with a sensitivity of 85% and specificity of 75% using score ≥4 in predicting non-response.
Conclusion : This risk score allows the early identification of patients with severe ulcerative colitis who would be suitable for second-line medical therapy or surgery.
The natural history of ulcerative colitis (UC) is variable. Whereas, many patients with UC suffer symptoms that can be managed with out-patient medical therapy, a proportion experience symptoms that prove refractory to treatment, and ultimately require colectomy. Historically, 15% of patients with UC will develop a severe attack requiring intensive in-patient medical therapy.1 Of these, consistent data have shown that 30–40% of patients will fail to respond to medical therapy and require urgent colectomy.2–7 The mortality associated with such attacks has fallen from levels of 31–61% to the present level of 1–2% following the introduction of high-dose corticosteroid therapy and the acceptance of a policy of early surgery in patients not responding to medical therapy.8
Although, corticosteroids remain the mainstay of medical therapy for severe UC, new therapeutic avenues opened following a randomized-controlled trial in which 80% of patients unresponsive to corticosteroids alone avoided emergency colectomy after the addition of ciclosporin to their treatment regime.9 Ciclosporin has been adopted in specialist centres in the UK, both as first-line10 and second-line11, 12 therapy in severe UC. Nevertheless, the recent survey performed by the British Society of Gastroenterology – inflammatory bowel disease (IBD) network on patients across the United Kingdom in 2001, showed that the only 13% of these patients receiving ciclosporin therapy.13 This may reflect on-going concerns regarding toxicity of this particular treatment and its lack of efficacy in maintaining remission.14 In fact, recent published trials provide further evidence of efficacy and safety of intravenous ciclosporin.15, 16 In addition, follow-up data has shown that in patients who achieved clinical response with ciclosporin and maintained on azathioprine, two-thirds of such patients continue to be in remission without surgery at 5 years after the initiation of ciclosporin, with all failures of therapy occurring within the first 18 months.17
For patients who fail to respond to corticosteroids, the use of further medical therapy rather than surgery needs careful thought, bearing in mind that much of the morbidity (and mortality) of UC is associated with delayed surgery.5, 18 Therefore, the selection of patients that may benefit from second-line therapy and importantly, the timing of the administration of therapy, are central to the management of severe UC. A risk index that can stratify patients according to their respective likelihood of not responding to intravenous corticosteroid therapy early in the course of treatment would be clinically pertinent to the issues discussed above. This risk index would also be of potential use in selecting patients for future novel therapies. The last few years have seen an expansion of new therapies with the steroid-sensitizing agent, basiliximab and anti-CD3 monoclonal antibody therapy being examples of such therapies.19, 20
In the present study, we have analysed early clinical parameters within a large cohort of patients presenting with acute severe UC to identify predictive factors of non-response to medical therapy. We have developed a simple objective risk score to aid the clinician in decision-making, and specifically help to select patients for either early second-line medical therapy or early surgery depending on the likelihood of non-response to standard medical therapy.
Although, two predictive models have been described in literature, these serve primarily to identify patients for early colectomy rather than to select patients for second-line medical therapy. In a prospective study involving 49 patients with severe UC, Travis et al. proposed that 85% of patients with a stool frequency of >8/day or 3–8/day with C-reactive protein (CRP) of 45 mg/dL after 3 days of intensive medical therapy will fail to respond and require colectomy.4 Lindgren et al. developed a regressional formulae to predict the likelihood of medical failure – number of bowel movements + 0.14 × CRP (mg/L) > 8.0, as the optimum theoretical cut-off level to predict failure of medical therapy.5
The proposed risk score developed from our cohort aimed to identify patients who are at low, intermediate and high likelihood of not responding to intensive medical therapy. Therefore, with the improvement in medical therapy in severe UC, treatment can be targeted at patients who will benefit most from early second-line medical therapy.
Patients were recruited from the gastroenterology units of two university teaching hospitals (Western General Hospital and Royal Infirmary of Edinburgh) and a large district general hospital (St John's Hospital, Livingston) in the Edinburgh and Lothian region, Scotland, covering a combined catchment area of 600 000 people.
All patients admitted for in-patient management of acute UC between January 1995 and March 2002 were identified using the regional database of medical/surgical admissions (Lothian Surgical Audit) and respective local hospital discharge databases. Ethical approval was obtained from the Lothian Research and Ethics Committee and informed consent was obtained from each patient involved.
The diagnosis of UC was confirmed on clinical, radiological and histological criteria.21 A severe episode of UC was defined according to the modified Truelove and Witts criteria as, ≥6 episodes of bloody diarrhoea/24 h with one or more of the following features: anaemia (haemoglobin < 10.5 g/dL), fever (>37.8 °C), tachycardia (pulse > 90/min) or elevated erythrocyte sedimentation rate (ESR, >30 mm/h). Colonic dilatation was defined by the diameter of the transverse colon on plain abdominal X-ray, ≥5.5 cm.
For each patient, all clinical data recorded during hospital admission were retrieved from case note review. Fifty-six variables within the first 3 days of medical therapy were recorded. These included demographic data, clinical observations, laboratory parameters, X-ray and endoscopic assessments of the severity of UC.
The primary outcome was categorized as response (no colectomy) or non-response to medical therapy (colectomy) within the period of hospitalization. Statistical modelling was also performed on secondary outcomes such as colectomy at 60 days following presentation (to account for patients who have undergone a semielective colectomy or those with an early severe relapse); and in the setting where intravenous ciclosporin was considered as failure of first-line medical therapy (to account for the confounding effect of second-line medical therapy).
Univariate analyses were initially performed to identify potential determinants of outcome of all 56 variables recorded. Of these, 24 potential variables were appropriate for further analysis following the removal of those variables with no prognostic significance (Appendix 1). Forward stepwise multiple logistic regression was then employed to identify predictive factors (with α > 0.1 taken as a level of exclusion) associated with our outcomes. To generate a clinical predictive model, variables of prognostic significance were categorized and re-entered into a logistic regression model. An integer score was attributed to each category of each variable according to its relative contribution in the regression model (as determined by its regression coefficient in the logistic equation). The scores were then grouped to provide a practical triage into low-, intermediate- and high-risk categories. All analyses were performed using the Minitab statistical software.
A total of 1211 admissions with inflammatory bowel disease were identified between January 1995 and March 2002. Comprehensive review of these cases led to the identification of 245 patients admitted to hospital with acute flare of ulcerative colitis, of whom, 167 (68%) patients fulfilled the criteria of Truelove and Witts. Sixty-eight patients (40%) failed to respond to medical therapy and required colectomy during that admission. Two of these patients died during the period of hospitalization (both postcolectomy) with pneumonia and arterial thrombosis of lower limb respectively as the causes of death. The median duration of admission for non-responders and responders was 26- and 11-days respectively.
The median age at presentation was 38 years (IQR: 27–54) (Table 1). Males accounted for 61% of the patient population. The median duration of relapse prior to admission was 4 weeks (IQR: 2–9.5 weeks). Seventy-six (45%) patients were experiencing the first attack of UC, 91 patients (55%) were known to have UC; and 42 (25%) patients had previous admissions with severe colitis. Only eight (5%) patients were current smokers. Extensive colitis (beyond the splenic flexure) was present in 131 (81%) patients. Of these, 62 (47%) patients with extensive colitis were non-responders. In comparison, only 10% of patients with disease limited to recto-sigmoid area failed to respond to medical therapy (P < 0.001).
Table 1. Univariate analysis of parameters in 167 patients presenting with acute severe ulcerative colitis (UC)
Overall (n = 167)
Non-responders (n = 68)
Responders (n = 99)
Odds ratio (95% CI)
1 The duration of relapse was the time of the onset of symptoms to hospital admission.
2 The disease extent was based on both colonoscopic and histological features. Subtotal disease extends beyond the splenic flexure.
3 Patients admitted with a first acute severe UC episode. Mean values of the first 3 days of therapy. * Colonic dilatation – transverse colon diameter >5.5 cm. † Median values of C-reactive protein (CRP) shown with interquartile range. ‡ Laboratory parameters were based on values from day 1 of medical therapy. For continuous variables in laboratory parameters, odds ratio calculated per unit of change. § Odds ratio for platelets calculated per change of 10 units.
All patients were treated with standard medical therapy of intravenous corticosteroids (methylprednisolone 60 mg/day or hydrocortisone 400 mg/day) (Table 1). In addition, 83% of patients received oral 5-aminosalicylate (5-ASA) preparations, 45% received concurrent topical therapy (rectal 5-ASA or corticosteroids), 71% received subcutaneous heparin for thromboembolic prophylaxis, and 13% received intravenous ciclosporin and total parenteral nutrition (TPN).
Twenty-one patients received intravenous ciclosporin at dosage of 4 mg/kg. All patients had no objective response to intravenous corticosteroid therapy. No patients in this group had evidence of colonic dilatation. The median time for the initiation of ciclosporin therapy was 10 days after the commencement of intravenous corticosteroids. The time to ciclosporin treatment, however, was shorter in the non-responders (7 days vs. 12 days). Overall, the stool frequency at presentation was 7.2/day (compared with 7.4 in no ciclosporin group, P = 0.81), and CRP was 11.0 mg/dL (compared with 9.7 in no ciclosporin group, P = 0.88). Serum albumin, however, was significantly lower in patients who had received ciclosporin (ciclosporin group, 29.1 g/L vs. no ciclosporin 32.0 g/L; P = 0.01). There were no significant early clinical or laboratory pre-treatment parameters, which were associated with failure of ciclosporin therapy. Nine patients responded, 12 required colectomy (P = 0.09). The median duration of ciclosporin treatment was 3.5 days (IQR: 2.0–7.5) and 7 days (IQR: 4.5–8.5) for non-responders and responders respectively. Of those that responded, three patients subsequently required colectomy during follow-up (median time to colectomy –2.4 years, range: 1.4–3.2 years).
Indications for surgery
Eighty-eight percentage of patients (60 of 68 patients) underwent surgery as a consequence of failed medical therapy. Of these, 10 patients also had radiological evidence of colonic dilatation (colonic dilatation within the first 3 days of treatment was only used for analysis). We have considered these as joint indications for surgery as despite colonic dilatation; these patients were commenced on intravenous corticosteroid therapy. Surgery was performed as a result of persistent diarrhoea and failure of colonic dilatation to resolve. A further five patients with early evidence of colonic dilatation were operated urgently (although none with evidence of perforation). The median time to surgery in all patients with colonic dilatation was 7 days (IQR: 5–9). In our institution, colonic dilatation is not an absolute indication for surgery and as evident from the time to surgery in these patients, the finding itself is not a ‘decision factor’ leading directly to surgery. Only one patient developed colonic perforation requiring urgent surgery. There were no cases of massive haemorrhage. The median time to colectomy was 9 days following admission (interquartile range, IQR: 7–15 days).
Univariate analysis of clinical and laboratory parameters
A significantly higher percentage of non-responders presented with stool frequency >8/day on admission (58.8% vs. 29.3%, P < 0.001, OR: 0.29, 95% CI: 0.15–0.56) (Table 1). On each of days 1–3, there was a consistently maintained significant difference in stool frequency between responders and non-responders, and we compared the discrimination of the mean stool frequency (MSF) over the first 3 days with data of each individual day (Table 1). This derived variable showed a more significant difference between responders and non-responders than the stool frequencies on any individual days. The MSF over the first 3 days of intravenous corticosteroid was higher in the non-responders compared with patients who avoided colectomy (8.05 vs. 5.2 respectively, P< 0.001, OR: 0.71, 95% CI: 0.62–0.81).
Evidence of hypoalbuminaemia (<30 g/L) was present in 47% and 28% of non-responders and responders respectively on admission (P = 0.005). Serum albumin was also noted to be significantly lower on day 1 of therapy in non-responders (mean ± s.d., g/L) 30.6 ± 5 g/L compared with responders, 34.1 ± 6 g/L (P = 0.001, OR: 1.10, 95% CI: 1.04–1.17). Similarly, CRP levels on day 1 of medical therapy (median ± IQR, mg/L) were also significantly higher in non-responders [6.9 mg/L, IQR: 2.8–19.2 compared with responders; 3.9 mg/L, IQR: 1.5–9.3 (P = 0.02)]. Colonic dilatation was present in 16 patients (9.6%) of these, 15 patients required colectomy (P = 0.001, OR: 0.04, 95% CI: 0.00–0.29). Weaker trends were observed in platelet count and ESR.
Other comparisons, in particular, the proportion of patients experiencing their first attack, the duration of relapse prior to hospitalization and clinical parameters such as neutrophil count, temperature, pulse rates and stool consistency were not associated with outcome.
Multiple logistic regression analysis and modelling
Mean stool frequency, the presence of colonic dilatation within the first 3 days and serum albumin on day 1 of medical therapy were identified as significant independent predictors of outcome in our cohort (P < 0.001, <0.001 and 0.002 respectively) (Table 2). Notably, CRP, platelet and ESR measurements implicated in the univariate analysis did not achieve statistical significance in multiple logistic regression analysis.
Table 2. Multiple logistic regression analysis of clinical parameters within first 3 days of medical therapy identified mean stool frequency (MSF), the presence of colonic dilatation and day 1 serum albumin as significant predictors of failure of medical therapy
95% confidence interval (CI)
MSF and serum albumin were grouped into the categories above using receiver–operator characteristic (ROC) curve analysis and adapted according to clinical relevance. These categories were re-entered into the logistic regression model. MSF < 4/day was used as reference index in MSF category. Formulation of integer risk score for each category was based on the strength of contribution to logistic equation based on the coefficient (for example, the coefficient of MSF > 9 is −4.27, therefore an integer score of 4 was given, coefficient of colonic dilatation is −3.8, therefore an integer score of 4 is given). MSF, mean stool frequency.
Day 1 serum albumin
MSF 4 ≤6/day
Stool 6 ≤9/day
Stool > 9/day
Serum albumin < 30 g/L
As significant differences were also observed on univariate analysis of the additional medical therapies (Table 2), multiple logistic regression analysis was carried out separately to assess the effect of these therapies on the outcomes of our patients. When considered together with the 24 parameters originally included in the earlier multiple logistic regression analysis, only TPN demonstrated significant association with non-response to medical therapy (Table 3). It is noteworthy that 5-ASA and subcutaneous heparin, which were implicated on univariate analyses did not achieve statistical significance. TPN, however, was not considered in statistical modelling as the median time to the commencement of this treatment was 6 days (IQR: 4–7 days) following the initiation of intravenous corticosteroid therapy. In addition, patients who were treated with TPN had a higher MSF (7.05 vs. the overall MSF of 6.34, serum albumin of 29.1 g/L vs. overall level of 32.7 g/L and CRP of 9.92 mg/dL vs. overall level of 4.4 mg/dL). This suggests that it is likely that the significant finding of TPN is related to underlying severity of disease.
Table 3. Multiple logistic regression of clinical parameters within first 3 days of medical therapy [5-aminosalicylate (5-ASA), topical therapy, subcutaneous heparin, total parenteral nutrition (TPN), ciclosporin]
95% confidence interval (CI)
Only TPN was shown to be significant in predicting outcome with mean stool frequency (MSF), colonic dilatation and day 1 albumin.
Day 1 serum albumin
In order to generate a scoring system, MSF and serum albumin values were categorized using integer cut-points guided by the receiver–operator characteristic (ROC) curve and observed relationship with outcomes.
This resulted in the categorization of MSF as: ≤4 stools/24 h; > 4 stools but ≤6 stools/24 h; >6 stools but ≤9 stools/24 h; and > 9 stools/24 h.
And serum albumin as: ≤30 g/L; and > 30 g/L.
Colonic dilatation was regarded as present or not present within the first 3 days of therapy.
To formulate a numerical risk score, we used the coefficients generated by the logistic regression equation, to derive an integer number, approximating the values of the coefficients for each of the categories above (Table 2).
We also explored different settings for modelling. These were:
A Analyses of outcomes at 60 days to account for early severe relapse or semielective surgery in partial responders.
B Analyses where ciclosporin treatment was regarded as failures of medical treatment.
C Analyses where colonic dilatation was excluded as some centres may regard this sign as an absolute indication for surgery.
In model A, a further nine patients required colectomy within 60 days (three partial and six complete responders). In model B, nine patients who had received ciclosporin (but no colectomy during admission) were regarded as primary treatment failure. For model C, 16 patients with colonic dilatation were excluded. Multiple logistic regression analysis failed to identify prognostic factors additional to stool frequency, serum albumin and colonic dilatation.
Using the coefficients of the regression analyses, the MSF of ≤4 stools/24 h, > 4 and ≤6 stools/24 h, > 6 and ≤ 9 stools and > 9 stools/24 h were given scores of 0, 1, 2 and 4 respectively. The presence of colonic dilatation was attributed a score of 4 (or score 0, in the absence of this sign) and hypoalbuminaemia (≤30 g/L), score of 1 (or score 0, albumin > 30 g/L) (Table 4). Using the scoring system, each patient in our study was scored and compared with his/her eventual outcomes.
Table 4. Integer risk score attributable to each category derived from the coefficients of the logistic regression equation
Overall risk core = [score attributable to mean stool frequency (0, 1, 2 or 4)] + [presence of colonic dilatation (0 or 4)] + [presence of hypoalbuminaemia (0 or 1)]. Minimum score = 0, maximum score = 9.
Mean stool frequency
4 ≤ 6
6 ≤ 9
< 30 g/L
For an individual patient, the total score was derived from the sum of the score attributed to MSF (0–4), colonic dilatation (0 or 4) and serum albumin (0 or 1). The minimum score possible (0) would be attributed to a patient with MSF < 4, no evidence of colonic dilatation or hypoalbuminaemia and a maximum score of 9, to a patient with MSF > 9 (4-points) together with the presence of both colonic dilatation (4-points) and hypoalbuminaemia (1-point).
The rates of non-response according to prognostic scores are demonstrated in Figure 1. All patients with scores of ≥ 6 failed to respond to medical therapy (three patients scoring 6, six patients scoring 7 and one patient scoring the maximum value of 9). Proportions of patients falling into each score categories were 19%, 23%, 9%, 25%, 13%, 5% and 5% respectively.
It is possible to triage patients to the groups of low- (score 0–1), intermediate- (score 2–3) and high-risk (score ≥ 4) groups with medical failure rates of 11%, 45% and 85% respectively. ROC analysis for this risk score yielded an area under the curve (AUC) of 0.876 (AUC of 1.0 indicating perfect test). The sensitivity and specificity for predicting non-response to medical therapy with scores of ≥ 4 was 85% and 75% respectively. The AUC of this current model for colectomy at 60 days following presentation, ciclosporin treatment was regarded as primary treatment failure and where colonic dilatation were excluded were 0.833, 0.810 and 0.807 respectively.
The present analysis involves the largest cohort of patients presenting with severe UC reported in detail. In a 7-year period, we reported on 167 consecutive patients with acute severe UC satisfying the modified Truelove and Witts criteria. Consistent with previous studies,2–4, 6, 7, 22 the failure rate of medical therapy is high (40%). On multiple logistic regression analysis, MSF and colonic dilatation within the first 3 days of therapy and hypoalbuminaemia on day 1 of treatment, were significantly predictive of the need for surgery within the period of hospitalization (P < 0.001, < 0.001 and 0.002 respectively). Using statistical modelling, we have formulated a numerical risk score that assesses the individual's likelihood of non-response. This risk score enables the stratification of our patients to those with low- (11%), intermediate- (45%) and high-risks (85%) of not responding to therapy within 3 days of medical therapy.
We have considered different settings for statistical modelling which are clinically relevant – colectomy at 60 days to account for early severe relapse or semielective surgery for partial responders with chronic continuous symptoms, where second-line therapy such as ciclosporin was considered as primary treatment failure and where colonic dilatation was excluded from analysis. Nevertheless, in our cohort, no further factors were identified from multiple logistic regression analysis. We have accounted for colonic dilatation in our statistical modelling, on the basis that some may consider this finding as an indication for surgery by itself. Colonic dilatation was included in our primary analysis as we believe that this is not an uncommon presentation of acute severe UC. Indeed, approximately 10% of patients in our cohort had radiological evidence of this within the first 3 days of hospitalization.
The C-reactive protein concentration was a notable exclusion from our multiple logistic regression analysis. The explanation for this bears evaluation but it is conceivable that in a preselected group with unequivocal severe disease, the absolute level of acute phase reactants may not be as useful as the change in response to intensive medical therapy. Differences observed in 5-ASA, TPN and heparin therapy in univariate analyses were thought to be due to confounding effect (e.g. physician-led reaction to illness and malnutrition).
Is this risk score relevant to clinical practice? First, the sensitivity (85%) and specificity (75%) of our risk score in predicting non-response to medical therapy at the cut-off point of ≥4 compared favourably with two other models described in the literature.4, 5 The ability to categorize patients on day 3 of standard intensive medical therapy involving intravenous corticosteroids allows the use of potentially toxic second-line therapy or early surgery targeted to the individual patient according to risk.
Of greatest clinical relevance is the identification of low- and high-risk groups in which surgery was either very unlikely or likely. We propose that in the intermediate group (34% of patients), close monitoring is essential with the option of second-line medical therapy at an early stage. It is often considered that in patients with poor prognostic features or fulminant disease that a delay in surgery may lead to an unacceptable increase in the risk of complications, and the case for early surgery (rather than further medical intervention) may be more compelling in patients with high-risk score. Nevertheless, there is considerable data to suggest that even within this group, a proportion of patients will respond to ciclosporin therapy.9–11,15, 16, 17, 23 The decision for ciclosporin therapy may be implemented in the high-risk group (which comprised of 25% of patients in our cohort) providing that close physician–surgeon liaison is in place.24
Although, the use of second-line medical therapy is complicated by other issues such as patient choice, familiarity and experience of the treating centre; data regarding efficacy and safety of ciclosporin therapy are emerging. Most recently, in a randomized-controlled trial, Van Assche et al. showed that response rate of 85% in acute severe UC can be achieved by using the 2 mg/kg dosage schedule. Short-term colectomy rate was remarkably low at 8.6%.16 In the UK, Rayner et al. in an earlier study also demonstrated that low-dose ciclosporin was efficacious and safe.10 Along with the recent report of favourable long-term response in initial responders to ciclosporin subsequently maintained on Azathioprine therapy,17 quality of life following successful ciclosporin therapy is comparable if not better than patients who have underwent surgery.25 Therefore, this should be considered in most patients as more than merely a bridge to colectomy. Our risk score hopefully provides the guidance in identifying and counselling patients, for clinicians on a wider scale. It is important to emphasize that like all predictive models, this is a statement of probability rather than one of determinism, to be used as an adjunct to clinical decision-making rather than to replace it. Clearly, prospective evaluation is necessary in independent populations to demonstrate the reproducibility of this risk score.
The primary development of this risk score therefore, has important implications in allowing targeted therapy aimed at the group of patients that will benefit most from early second-line medical therapy, selection of patients for early surgery, patient counselling and the stratification of patients according to risk, in future therapeutic trials. These, we believe, will improve the clinical management of severe UC.
Dr Gwo-Tzer Ho is supported by the Chief Scientist Office, Scottish Executive, United Kingdom.
Authors are indebted to Dr Kelvin Palmer, Dr Ian Penman, Dr Alan Shand, Dr Ian Arnott, Mr David Bartolo, Prof. Kenneth Fearon, Prof. Malcolm Dunlop, Mr David Hamer-Hodges, Mr Graeme Wilson, Ms Collete McColl (Western General Hospital, Edinburgh), Dr Andrew Williams, Mr David Anderson, Mr Donald McDonald, Mrs Anne Connolly (St John's Hospital, Livingston) and Mrs Molly Stewart (The Royal Infirmary of Edinburgh) for their help and support.