Positive predictive value of primary inpatient discharge diagnoses of infection among cancer patients in the Danish National Registry of Patients
A partir des données du registre national danois des patients, cette étude développe et analyse un algorithme permettant d'identifier les infections nécessitant une hospitalisation chez des patients atteints de cancer
Purpose : Pharmacovigilance studies of cancer treatment frequently monitor infections. Predictive values of algorithms identifying disease depend on prevalence of the disease in the population under study. We therefore estimated the positive predictive value (PPV) of primary inpatient diagnosis of infection among cancer patients in the Danish National Registry of Patients (DNRP).
Methods : The algorithm to identify infections in the DNPR was based on International Classification of Diseases, 10th revision (ICD-10) codes. A physician blinded to the type of sampled infection, reviewed the medical charts, and assessed presence and type of infection. Using the physician global assessment (PGA) as gold standard, we computed PPVs with and without requiring agreement on infection type.
Results : We retrieved 266/272 (98%) medical charts. Presence of infection was confirmed in 261 patients, resulting in an overall PPV of 98% (95% confidence interval [CI]: 96-99%). When requiring agreement on infection type overall PPV was 77%. For skin infections, pneumonia and sepsis PPVs were 79%, 93% and 84, respectively. For these infections, we additionally calculated PPVs using evidence-based criteria as reference. PPV was similar for pneumonia, but lower for skin infections and sepsis.
Conclusions : The DNRP is suitable for monitoring infections requiring hospitalization among cancer patients
Annals of Epidemiology , résumé, 2013