PATIENTS AND METHODS

The study included 153 children referred to the paediatric gastroenterology department outpatient clinic at Birmingham Children’s Hospital. This department serves as a secondary and tertiary referral centre for children with gastrointestinal disorders. They presented with various gastrointestinal and nutritional symptoms compatible with IBD. Investigations routinely performed included the following: haemoglobin, platelet count, ESR, serum CRP, and serum albumin. ESR was estimated by the Westergen method. CRP was measured using the dry chemistry immunological method and a Vitros 250 analyser (Ortho Clinical Diagnostics, Johnson & Johnson, USA). These five widely used screening tests for IBD were subjected to a systematic evaluation.

Further investigations such as small bowel barium contrast radiology, upper gastrointestinal endoscopy, and colonoscopy were undertaken at the discretion of a paediatric gastroenterologist, being governed by the specific clinical presentation in each case. Diagnoses of Crohn’s disease and ulcerative colitis were based on conventional clinical, radiological, endoscopic, and histological criteria.

In total 103 children were found to have IBD (age range 1–17.6 years, median 11.6). Of these, 60 had Crohn’s disease, 37 had ulcerative colitis, and six indeterminate colitis. Amongst those with Crohn’s disease presenting clinical features included: anorexia and weight loss (n = 53), abdominal pain (n = 50), diarrhoea without blood (n = 24), bloody diarrhoea (n = 13), passage of blood without diarrhoea (n = 1), perianal Crohn’s disease (n = 27), finger clubbing (n = 15), short stature (n = 11), arthralgia (n = 8), and erythema nodosum (n = 3). Those with ulcerative colitis presented with bloody diarrhoea (n = 28), diarrhoea without blood (n = 6), passage of blood without diarrhoea (n = 3), abdominal pain (n = 23), anorexia and weight loss (n = 21), short stature (n = 2), and arthralgia (n = 1).

The 50 children who did not have IBD (age range 2–17 years, median 9) acted as the control group. They included 30 with recurrent abdominal pain (29 functional abdominal pain, one abdominal migraine), 11 with diarrhoea (five coeliac disease, one gastroenteritis, five no identified disorder), three with vomiting (one gastro-oesophageal reflux, two no identified disorder), and one with unexplained intermittent rectal bleeding.

Data analysis

Receiver operating characteristic (ROC) curves were obtained for platelet count, ESR, CRP, and albumin concentrations to explore the sensitivity and specificity of these tests using varying threshold values. Based on these analyses threshold values could be determined for both platelet count and ESR that were associated with optimal sensitivities and specificities. The optimal platelet count threshold level was >350×10⁹/l, giving a sensitivity of 85% and a specificity of 85%. For ESR the optimal threshold level was >10 mm/h with a sensitivity of 82% and a specificity of 78%. ROC curves for CRP and albumin showed poor sensitivities with values of 60% or worse for plausible thresholds. For further statistical analysis, cut-off values were taken from the published literature (albumin <35 g/l; CRP >5 mg/l).¹ For haemoglobin, ROC curve analysis was not employed because the normal reference ranges are age specific. The selected cut-off values for haemoglobin corresponded to the lower limits of established age related reference ranges (1–3 y, <110 g/l; >3 to 6 y, <117 g/l; >6 to 18 y, <120 g/l).

Binary logistic regression analysis was used to investigate how the occurrence of IBD depended on the five screening tests in combination, and a stepwise method was used to find the combination of tests that best discriminated between the presence and absence of IBD. These analyses were done using both original test measurements as predictors and also using binary indicator variables showing when a test crossed the selected threshold from the ROC analysis above.

RESULTS

Logistic regression analysis

Binary logistic regression analysis using the selected cut-off values as binary indicators for each test proved to be more effective in discriminating between IBD patients and controls than using the actual test values, so only the former results are presented. Table 1 presents the results of this analysis. Tests with good discriminatory power show a combination of a high odds ratio together with a low p value (that is, a significant difference between the groups). Platelets and haemoglobin were the tests that best differentiated between the control and IBD groups. Albumin and CRP did not show significant discriminatory power either individually or in conjunction with the other tests, and were thus not included in further data analysis. This approach did not result in any patient with IBD being missed. For ESR, the conclusion derived from the ROC curve and binary logistic regression analysis was that this test was an inessential predictor of IBD when used in conjunction with platelets and haemoglobin. However, if ESR analysis were to be dropped, three patients with IBD might have gone without further investigations. Subsequent analysis therefore focused on the reliability of haemoglobin, platelet count, and ESR in screening for IBD.

Comparison of accuracy indices

Finally, diagnostic accuracy indices were calculated using the above cut-off values for the combination of haemoglobin, platelets, and ESR. The logistic regression function as a predictor of IBD had a sensitivity of 89.5% and a specificity of 77.6%. As this is a very similar predictive performance to simpler, easily implemented rules using the same tests, we detail only the latter. By using a rule of “any two or three positive tests out of three predict IBD”, 85.7% sensitivity (95% CI 77.2 to 92.0%), and 89.8% specificity (95% CI 77.8 to 96.6%) was obtained. For platelets and haemoglobin only and using a rule of “any one or two positive tests out of two predict IBD”, 90.8% sensitivity (95% CI 83.3 to 95.7%) and 80.0% specificity (95% CI 65.7 to 89.8%) was achieved. This confirms the result of logistic regression that ESR is an inessential predictor of IBD when used in conjunction with platelets and haemoglobin. Diagnostic accuracy indices for Crohn’s disease and ulcerative colitis were calculated separately (table 2). Using “at least two of the triple test combination abnormal”, and “at least one of the dual test combination abnormal” gave comparably good reliability indices both for Crohn’s disease and ulcerative colitis.

DISCUSSION

There are a number of published descriptive studies on the value of haemoglobin, platelet count, ESR, CRP, and albumin as screening tests for IBD.^1,2 In this study, however, a rigorous analysis was performed to determine the reliability of these blood tests both individually and in various combinations. Parameters of test reliability including sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. Both sensitivity and specificity are affected by sample size, and so 95% confidence intervals were determined for these estimates.

This study found that in children attending a paediatric gastroenterology department, combined measurement of haemoglobin and platelet count is an appropriate screening strategy for ulcerative colitis and Crohn’s disease. In patients presenting with gastrointestinal symptoms or nutritional concerns, the presence of anaemia or thrombocytosis was associated with a diagnostic sensitivity of 90.8% (95% CI 83.3 to 95.7%), a specificity of 80% (95% CI 65.7 to 89.8%), a PPV of 90%, and NPV of 81% for IBD. Similarly values were obtained for both Crohn’s disease and ulcerative colitis.

Beattie et al reported haemoglobin, platelet count, ESR, CRP, and albumin results in 39 children with IBD (26 with Crohn’s disease and 13 with ulcerative colitis) and 37 controls.¹ All children with Crohn’s disease had at least one test abnormality, but in 8% with ulcerative colitis all of the tests were normal. In that study all with Crohn’s disease and 60% with ulcerative colitis had an increased CRP. Hypoalbuminaemia was found in 35% with Crohn’s disease and 15% with ulcerative colitis. However, our study found that although increased CRP and hypoalbuminaemia were common in IBD, these did not contribute significantly to the reliability of the blood test panel.

The association between thrombocytosis and both Crohn’s disease and ulcerative colitis has been recognised for many years.^2–6 Beattie et al also reported that the platelet count was increased in 88% with Crohn’s disease and 70% with ulcerative colitis.¹ This is in keeping with our study, and our analysis confirmed that platelet count was an important constituent in the test panel.

Another study reported the use of serum CRP, α₁ acid glycoprotein, α₁ antitrypsin, complement factor 9, factor B, and ESR in 18 children with IBD and 12 controls.⁷ Although the results were significantly different between the groups, there was an overlap and no single test was reliable for diagnosis.

Other laboratory tests that have been used in this context have included interleukin-1 receptor antagonist,⁸ eotaxin,⁹ anti-neutrophil cytoplasmic and anti-saccharomyces cerevisiae antibodies,¹⁰ tumour necrosis factor α,¹¹ and thrombopoietin.¹² Increased levels of these have been found in IBD but the reliability of these markers has not been studied adequately.

In conclusion, the simple combination of haemoglobin and platelet count provides a useful test panel for children with suspected IBD. However, it should not be relied on completely as both false positive and false negative results may occur. If clinical suspicion is high, further investigations are required.