Using hospital in-patient data to assess comorbidity in cancer patients: Is it fit for purpose?
Session type: Poster / e-Poster / Silent Theatre session
Comorbidities impact on cancer treatment and prognosis, but this information is not routinely collected by most cancer registries. One approach is to use other routinely-collected data to identify comorbid conditions and classify patients according to an established index. The aim of the present work is to describe challenges experienced in assessing comorbidity using hospital in-patient data in a study of factors influencing prostate cancer (PC) treatment.
PC cases (ICD10 C61) diagnosed 2002-2008, registered with the National Cancer Registry Ireland, and with a hospital in-patient episode within one year of diagnosis were included (n=9659). Comorbidity (assessed by Charlson and Elixhauser indices) was determined from diagnoses on hospital episodes. Multinomial logistic regression was used to investigate associations between comorbidities and treatment receipt. A survey of urologists and radiotherapists was conducted.
Prevalence of any comorbidities among PC patients varied according to the index employed: 17% (Charlson), 27% (Elixhauser). Uncomplicated hypertension, chronic obstructive pulmonary diseases, diabetes and fluid and electrolytic disorders were the most common conditions. Men with comorbid conditions were less likely to have microscopic verification of their tumours at diagnosis and more likely to have more advanced disease. Concerning treatment receipt, some inconsistent and counter-intuitive results were observed. Using Charlson, men with a single comorbid condition were more likely to undergo radical prostatectomy (RRR=1.51; 95%CI 1.07-2.14) than radiotherapy than men with no comorbidities; the association was stronger using Elixhauser (RRR=3.31; 2.59-4.25). Conditions reported by clinicians as influencing treatment were mostly rare.
Several challenges were identified. Findings varied by index. Some conditions included within indices may be diagnosed because of cancer and treatment, rather than representing "true" comorbidities. Conditions included in indices may not be those considered clinically important in determining treatment and clinically important conditions may be rare. Using routine data to assess comorbidity is not straightforward.