External validation of a prediction model that identifies adverse outcomes in hospitalised patients with diabetes

Krishnarajah Nirantharakumar, Nicola Adderley, Tom Marshall, Gerry Rayman, Susan Mallett, Sandip Ghosh, Funke Akiboye, Srikanth Bellary, Konstantinos Toulis, Jamie Coleman

Research output: Contribution to journalAbstractpeer-review

Abstract

Aim: To externally validate a prediction model that identifies adverse outcomes in hospitalised patients with diabetes.

Methods: Data source: Inpatient data for 2014 from Heart of England Foundation Trust (HEFT) and Ipswich General Hospital.

Participants: Patients with a diagnostic code of diabetes.

Candidate variables: Age; gender; ethnicity; admission type; ITU admission; insulin therapy; albumin; sodium; potassium; haemoglobin; C-Reactive Protein; Estimated Glomerular Filtration Rate; and neutrophil count.

Outcome definition: Adverse outcome is a composite outcome (excessive length of stay or death).

Performance assessment: To assess discrimination, a receiver operating characteristic curve was constructed and Area Under the Curve (AUC) calculated. Calibration was assessed by plotting predicted probabilities of outcome against observed probabilities of outcome.

Results: There were 16,568 admissions with diabetes to HEFT in 2014. The model discriminated well between those with and without an adverse outcome in HEFT (AUC:0.758, 95% CI: 0.747–0.768; sensitivity: 72.6%; specificity: 66.0%; and Positive Predictive Value (PPV): 45.1%). These are in comparison to the internally validated model derived at University Hospitals Birmingham NHS Foundation Trust (AUC: 0.802; sensitivity: 76%; specificity: 70%; and PPV: 49%). Calibration plot suggested predicted probabilities were similar to observed probabilities. Ipswich General Hospital produced comparable results to that of HEFT.

Conclusion: The externally validated prediction model is a novel tool that can be used to improve care pathways for inpatients with diabetes. Further research utilising the model to develop care pathways and assess clinical utility is needed.
Original languageEnglish
Article numberA4 (P376)
Pages (from-to)6
Number of pages1
JournalDiabetic Medicine
Volume34
Issue numberSuppl. 1
DOIs
Publication statusPublished - 8 Mar 2017

Keywords

  • diabetes
  • Inpatient
  • Prediction model
  • validation

Fingerprint

Dive into the research topics of 'External validation of a prediction model that identifies adverse outcomes in hospitalised patients with diabetes'. Together they form a unique fingerprint.

Cite this