Study population

Incident cervical cancer cases diagnosed from January 1, 2005, to December 31, 2009, were identified from the Ontario Cancer Registry by use of the International Classification of Diseases , Version 9 diagnosis code (180.x). Information on cancer stage also was extracted from the Ontario Cancer Registry. We excluded non-Ontario residents, those with a previous cancer diagnosis before the current cervix cancer, and those who were diagnosed only at the time of death.

We selected a noncancer comparison group from the Registered Persons Data Base files, which contains information on basic demographics for anyone who has ever had an Ontario health card number. From the Registered Persons Data Base, we created a pool for a potential comparison group containing all noncancer females born during the same period of cervical cancer patients and matched cases and noncancer comparison subjects (“controls”) on year of birth. To each potential control subject, the diagnosis date of her corresponding case was assigned as her index date. For cases and all potential controls, we determined their rurality and neighborhood income by using the Statistics Canada’s postal code-conversion method. We evaluated their baseline health status, denoted by the Charlson-Deyo comorbidity score, and major physical problems and mental health problems of Aggregated Diagnostic Group (ADG) conditions during the 2 years before the index date, using diagnostic information from inpatient care, emergency department visits, and physician visits. We also estimated their healthcare utilization (the resource utilization band, or RUB) during 2 years before the index date, using the John’s Hopkins University’s Adjusted Clinical Group technique.

We then used the propensity score (PS) matching method to select 5 controls for each case of cervical cancer case. The PS was estimated by the use of a logistic regression model in which being a cervical cancer case was regressed on rurality, income quintile, RUB, mental ADG, major physical ADG, and Charlson comorbidity score. Cases and controls were matched on exact year of birth and the logit of the PS with a caliper of 0.2 SDs of the logit of the PS.

Phase of care definitions

Cases were followed from index date to death or March 31, 2013, and controls from index date to death, the death date of her matched case, or March 31, 2013, whichever came earliest. We divided the entire follow-up period for each subject into 4 clinically relevant costing phases based on index date and date of death: (1) prediagnosis phase (1 year before index date); (2) initial phase (from index date to 1 year after); (3) continuing phase (1 year after index date to 1 year before death); and (4) terminal phase (the last year of life). If the case died during the follow-up but her matched controls did not, then the controls’ observation time parallel to the case’s terminal phase was assigned as their terminal phase. Each subject had a whole year of observation time during prediagnosis phase, but not all of them had the following 3 phases. For those with less than 36 months of follow-up, we applied the algorithm used in previous studies to assigned their observational time first into the terminal phase (if they died), then into initial phase, and last into the continuing phase.

Costs of healthcare

We identified and costed all healthcare-related resources used by cases and controls and paid for by the Ontario Ministry of Health and Long-Term Care. These resources included inpatient hospitalization, emergency department visit, same-day surgery, physician services, prescription drugs, rehabilitation, laboratory services, services from nonphysician provider, radiation therapy, chemotherapy, long-term care, home care, and complex continuing care. Data sources and methods for costing these resources are included in the Appendix .

All costs were converted into 2012 Canadian dollars ($1.00CDN = $1.00USD), using the consumer price index of healthcare. Ethics approval was obtained from the Research Ethics Board of Sunnybrook Health Sciences Centre.

Analysis

We compared baseline characteristics of cases and their matched controls by using a paired t -test for continuous variables and the McNemar test for categorical variables. The Kaplan-Meier survival curve was used to describe the survival probability for cases of different cancer stages, and controls, along with log-rank test to compare the survival distributions of different groups.

We described the percentage of subjects who have used each healthcare resource during the prediagnosis, initial, continuing, and terminal phases. The costs (total and by resource) for each subject in each costing phase were then calculated. Because the continuing phase can be longer than 1 year, we reported standardized annual costs in this phase. We identified cost drivers in each phase by calculating the percentage of total cost for each resource input.

Because cases and controls were closely matched, cost differences within a matched group represent the incremental costs for treating cervical cancer, adjusted for matched variables. We estimated the incremental costs and their 95% confidence intervals using the generalized estimating equations, with the cost for an individual as the independent variable, and case/control status as the predicting variables, accounting for the cluster effect of matched case-control group.

We used multivariate quantile regression models to identify the predictors of greater total cost in each costing phase among cervical cancer cases, with patient demographics, cancer stage, and comorbidities as predictor variables.