Abstract
Introduction
Cystectomy with urinary diversion (i.e. bladder removal surgery) is commonly studied using large health administrative databases. These databases often use diagnoses or procedure codes with unknown accuracy to identify cystectomy patients, thereby resulting in significant misclassification bias. The primary objective of this study is to develop a predictive model that will return an accurate probability that patients recorded in the discharge abstract database have undergone cystectomy with urinary diversion, stratified by type of urinary diversion (continent vs incontinent). Secondary objectives of this study include: 1) to internally validate our predictive model to determine its accuracy using a cohort of all adults admitted to The Ottawa Hospital (TOH) within the study period; and 2) compare the accuracy of this model to that of code-based algorithms used previously in published studies to identify cystectomy.
Methods
A gold standard reference cohort (GSC) of all patients who underwent cystectomy and urinary diversion at TOH between 2009 and 2019 was created by using the SIMS registry within the TOH data warehouse which captures all primary surgical procedures performed. The GSC was then confirmed by manual chart review to ensure accuracy. Through ICES, the GSC was linked to the provincial Discharge Abstract Database (DAD), physician billing records (OHIP), and Ontario Cancer Registry (OCR) and a new combined dataset containing all admissions at TOH during the study period was created. Clinical information, billing, and intervention codes within these databases were reviewed and the co-variables thought to be predictive of cystectomy were selected a priori. A multinomial logistic regression model (i.e. The Ottawa Cystectomy Identification Model or OCIM) was created using these co-variables to determine the probability of a patient undergoing cystectomy, stratified by continent vs incontinent diversion, during an admission in the DAD. Using the OCIM and bootstrap imputation methods, co-variable values and 95% confidence intervals were calculated. The values of these same co- variables were then measured using a code algorithm (the presence of either a procedure code or billing code for cystectomy with incontinent or continent diversion). Misclassification bias was then measured by comparing the values of co-variables using the OCIM or code algorithm to the true values obtained from the gold standard reference cohort.
Results
Five hundred patients were included in the GSC [median age 68.0 (IQR 13.0); 75.6% male; 55.6% incontinent diversion]. The prevalence of cystectomy within the DAD over the study period was 0.12% (500/428697 total admissions). Sensitivity and positive predictive values for cystectomy codes were 97.1% and 58.6% for incontinent diversions and 100.0% and 48.4% for continent diversions, respectively. The OCIM accurately predicted cystectomy with incontinent diversion (c-statistic [C] 0.999, Integrated Calibration Index [ICI] 0.000) and cystectomy with continent diversion (C:1.000, ICI 0.000) probabilities. Misclassification bias was lower when identifying cystectomy patients using the OCIM with bootstrap imputation compared to the use of the code algorithm alone.
Conclusions
A model using administrative data accurately returned the probability that cystectomy by diversion type occurred during a hospitalization. Using this model to impute cystectomy status minimized misclassification bias.
| Identifer | oai:union.ndltd.org:uottawa.ca/oai:ruor.uottawa.ca:10393/45908 |
| Date | 01 February 2024 |
| Creators | Ross, James |
| Contributors | Van Walraven, Carl G. |
| Publisher | Université d'Ottawa / University of Ottawa |
| Source Sets | Université d’Ottawa |
| Language | English |
| Detected Language | English |
| Type | Thesis |
| Format | application/pdf |
Page generated in 0.0018 seconds