Introduction C.difficile-induced diarrhoea is often refractory to treatment and can be life-threatening. As yet, there is no prospectively derived tool with which to predict treatment failure (TF) in C.difficile infection (CDI). We performed a fully powered, pragmatic, prospective observational study to create such a tool.

Methods Patients with confirmed CDI were consented to enter the study within 24 hours of testing positive. Demographic data (including age, antibiotic and PPI use, smoking, ward type), clinical variables (pulse, PB, temperature) and blood test results (FBC, urea, creatinine, CRP, albumin) and faecal calprotectin on the day of diagnosis were collected. TF was defined as occurrence of any of: death while admitted, colectomy, ongoing diarrhoea at day 7, recurrent diarrhoea at <30 days after initial CDI diagnosis. Case level re-structuring was used to account for missing data and forward stepwise binary regression to derive a predictive model. The model was internally validated by bootstrapping and assessed by Receiver Operated Characteristic (ROC) analysis.

Results 122 patients were recruited and primarily treated by their routine clinical team with metronidazole (n = 89) or vancomycin (n = 29). 63 patients (52%) failed treatment: 28 died during their admission, 43 had continuing diarrhoea at day 7, 16 had recurrent diarrhoea within 30 days and 1 had a colectomy (some patients had TF on >1 criteria). TF rate was the same whether metronidazole or vancomycin was primary therapy. Of the variables measured, only age and serum albumin predicted TF (age, p = 0.029; albumin, p = 0.0001). An equation with which to predict individual patients’ risk of TF was then derived: for ease of clinical application, a simple read off table was derived allowing prediction of outcome using the patient’s age and serum albumin. The model correctly predicted TF in 79% of cases. By ROC analysis, the model initially had an Area Under the Curve (AUC) of 0.76; in the internal validation assessment the AUC was 0.75.

Conclusion A prospectively and internally validated tool with which to predict treatment failure in CDI has been derived. The tool consists of 2 variables (age and serum albumin) on the day of diagnosis of CDI. The predictive tool could be used to highlight those who might benefit from more intensive treatment, for example using fidaxomicin or faecal microbial transplant as primary CDI therapy.

Disclosure of Interest None Declared