P384 High incidence of hyperglycaemia in steroid treated hospitalised inflammatory bowel disease (IBD) patients and its risk factors identified by machine learning methods. (25th January 2019)

• Record Type: Journal Article
• Authors: McDonnell, M; Harris, R; Mills, T; Downey, L; Dharmasiri, S; Felwick, R; Borca, F; Phan, H; Cummings, F; Gwiggner, M
• Is Part Of: Journal of Crohn's and colitis. Volume 13(2019)Supplement 1
• Page Start: S299
• Page End: S300
• Publication Date: 2019-01-25
• Subjects: Inflammatory bowel diseases -- Periodicals; 616.344005
• DOI: 10.1093/ecco-jcc/jjy222.508
• Languages: English
• ISSNs: 1873-9946

Abstract

Abstract: Background: Glucocorticoids (GC) have been first-line treatment for hospitalised IBD patients for over 60 years, despite the introduction of biologic therapy. IBD patients often have systemic inflammation complicated by malnutrition leading to metabolic stress. The frequency of and specific risk factors for hyperglycaemia in hospitalised IBD patients receiving GC are unknown. Methods: In total, 93 consecutive IBD inpatients receiving intravenous hydrocortisone (IVH) for an acute flare had capillary blood glucose (CBG) monitoring automatically triggered by the electronic prescription. CBG, biomarkers, IBD severity scores (Harvey–Bradshaw, partial Mayo) and weight loss were prospectively recorded. Undiagnosed Diabetes Mellitus (DM) was defined as HbA1c >48 mmol/mol. Machine-learning (random forest regressor, RFR) was applied to the data to evaluate risk factors of hyperglycaemia. Results: Fifty-five per cent of hospitalised IVH-treated IBD patients met the WHO criteria of DM (CBG >11 mmol/l), while 22% and 8% had a CBG >14 mmol/l and >20 mmol/l, respectively. Only 8 patients had pre-existing DM, which was confirmed by admission HbA1c. RFR indicated disease severity score, duration of IVH, HbA1c and electrolyte imbalances (which affected 64%) were best predictors of hyperglycaemia. Relative importance of input features of RFR model for prediction of CBG_max (left). Predictive value from RFR model vs. true value for training data set (blue) and test data set (red) (right). Sixty-four per cent reported previous weight loss, which did not predict hyperglycaemia, although those with >5% weight loss had significantly lower admission serum potassium ( p = 0.0067). Admission serum potassium and preceding weight loss. Conclusions: Our data demonstrate that hyperglycaemia is common in IVH-treated inpatients, therefore CBG monitoring should be routine practice. Predictive modelling (RFR) identifies more severe disease activity, duration of IVH treatment and HbA1c as risk factors for hyperglycaemia. Preceding weight loss and electrolyte imbalance in the cohort demonstrate a tendency towards malnutrition-associated metabolic instability. The importance of IVH duration suggests hyperglycaemia risk may be physician modifiable. Alternative treatment strategies such as earlier introduction of biologics, rapid steroid taper and nutritional support could be used to minimise medication-associated metabolic instability in high-risk patients.