Patient data were extracted from Danish national registers, which are linked through individuals’ unique registration number (CPR-number). The Danish health service has a long tradition of recording health service use and each contact with primary health care (e.g. general practitioner, public and private specialist, dentist, physiotherapist) and secondary health care (e.g. hospital admission, outpatient visits) is recorded with related data on age, sex, type of contact, speciality, fee/charge, diagnoses (secondary health care only) and procedure code. The present study was approved by the Danish Data Protection Agency (J. No. 2010-41-4305).
The analysis was conducted from a hospital sector perspective, as the relevant cancer types are almost exclusively diagnosed and treated at hospitals. New cancer patients were identified via specific ICD-10 diagnosis codes in the Danish National Cancer Register. Their annual hospital resource use was estimated based on hospital contacts recorded in the National Patient Register, which defines resource use by the DRG (Diagnosis Related Groups) system for admissions and by the Danish outpatient charges (DAGS charges) for outpatient visits (including emergency unit contacts)
]. The 2008 DRG and DAGS charges were used as cost estimates.
The cohort of cancer patients was defined as patients registered in the Danish National Cancer Register during 2004–2007 with anal, penile, vaginal or vulvar cancer as the primary localization. The patients were identified using the ICD-10 codes: C21 (anal cancer), C60 (penile cancer), C52 (vaginal cancer) and C51 (vulvar cancer).
For the cohort of cancer patients diagnosed during 2004–2007, health care use in 2006–2008 was compared with an age- and sex-matched cohort without cancer (controls). Five controls were identified for each cancer patient. The hospital costs associated with the controls were subtracted from the costs associated with the cancer patients (to identify the extra costs related to cancer), but this was done in regression analyses in which costs attributable to anal, penile, vaginal and vulvar cancer were estimated with cancer (yes/no) as an explanatory dummy variable. As a substantial number of the control patients incurred no health care costs (i.e. cost
0), a two-part model was applied
]. In this analysis, the probability that the patient had zero or non-zero costs was first predicted via logistic regression analysis. Secondly, the level of cost conditional on having positive costs was predicted using a generalized linear regression model (GLM), applying a log link function and assuming an inverse Gaussian distribution. Finally, the estimated health care costs were derived by multiplying the predictions from the two components (part 1 and 2) together.
When the data for this study was obtained, data on cancer incidence (the Danish National Cancer Register) was only available up until 2007, while data on DRG and
DAGS charges in the National Patient Register were only available from 2006 (given that the same, and relatively novel, version of the DRG- and DAGS-charges should be applied for all years). Therefore, we used a combined cross-sectional and longitudinal approach for the data analysis, where patients diagnosed during 2004–2007 had an associated resource use for 2006–2008. This allowed us to estimate the costs 0–12
months before the date of diagnosis (e.g. 2006 resource use data for a patient diagnosed in 2007) and the costs 0–12
months (e.g. 2008 resource use data for a patient diagnosed in 2006) and 25–36
months after the date of diagnosis (e.g. 2008 resource use data for a patient diagnosed in 2005). Results are thus presented as yearly cost estimates for the year before, the 1st
year, the 2nd
year and the 3rd
year after the date of diagnosis for patients alive at the respective times. We included costs for the year before diagnosis so that we could estimate the costs of initial examinations and diagnostics. Given the perspective of this analysis, the hospital cost estimates include medication related to hospital contacts, radiotherapy, chemotherapy and specialized rehabilitation.
Costs are presented in Euros. Future costs (i.e. cost estimates for the 2nd and 3rd years after diagnosis) were discounted using a 3% annual discount rate to represent their present value. When estimating the total average health care cost per patient, we adjusted for deaths during the observation period.
The number of diagnosed precancerous lesions was estimated based on data from the National Pathology Register. It was not possible to estimate treatment costs for these lesions (some of which progress to anogenital cancer) separately, however, as many of these cases are diagnosed and treated by privately practising specialists in the primary care sector. As diagnoses are not systematically registered for the primary care sector, it was impossible to distinguish contacts related to precancerous lesions from other contacts.
Data were analysed using SAS software version 9.2 (SAS Institute Inc., Cary, NC, USA).