Penile cancer treatment costs in England

The study was split into two phases. Firstly, to calculate annual patient numbers, along with mean annual total and per patient costs, a retrospective analysis was performed using data extracted from Hospital Episode Statistics (HES). Each hospital Finished Consultant Episode (FCE) results in the production of a summary of care discharge report, containing patient demographics, method of admission and discharge, the nature of treatment carried out during the stay including any comorbidities and complications, and the main diagnosis that preceded the hospital admission. Diagnoses are coded using the International Classification of Diseases, 10^th revision (ICD-10) either as primary, related, or significantly associated diagnoses.

Inpatient and outpatient records associated with any of the ICD-10 codes for penile cancer were extracted for the years 2006/07 to 2010/11 (2010/11 was representative of nine months provisional data). Since a hospital admission may include other ICD-10 code diagnoses where penile cancer is the secondary or tertiary diagnosis, a clinical expert verified that such instances should be included within the analyses to capture the broader burden of disease. ICD-10 codes for outpatient attendances were extracted for primary and secondary diagnoses only, as attendances are generally more disease specific in post-treatment care. The following ICD-10 codes were used to identify eligible records: C600 - Malignant neoplasm of prepuce, C601 - Malignant neoplasm of glans penis, C602 - Malignant neoplasm of body of penis, C608 - Malignant neoplasm, overlapping lesion of penis, C609 - Malignant neoplasm of penis, unspecified. Penile dysplasia was not included within this analysis as no specific ICD-10 code was available.

Approval for use of the HES data was provided by the information asset owner from the Health and Social Care Information Centre. The individual HES records extracted contained no sensitive data and were pseudonymised preventing the true identification of patients; analyses pertaining to HES records adhered to published regulations [10]. Ethical approval was not required as secondary analysis of HES data can be used to identify public health issues and for general medical research under existing protocol; the Health Research Authority decision tool corroborated this fact stating no ethical approval was required for this research [11].

Payment by Results is the payment framework for which healthcare providers are reimbursed by commissioners in England [12]. Nationally determined currencies and tariffs comprise the fundamental features of Payment by Results. The currency for inpatient care are Healthcare Resource Groups (HRG) which are setting-independent, clinically specified groups of diagnoses and interventions that consume similar levels of National Health Service resources. In contrast, the currency for outpatient care is based on treatment function codes.

Under Payment by Results, inpatient reimbursement is provided for a spell of care, which can potentially include multiple FCEs captured by HES. A spell is a more robust activity measure than an FCE because FCEs can easily be influenced (e.g. by transferring patients between consultants) in ways that spells cannot. To group episodes of care into spells, the 2010/11 local payment grouper was used [13]. A single core HRG was derived reflective of the nature of treatment given during a hospital stay.

Costs were considered from the healthcare payer perspective although ambulatory care, primary care, and costs pertaining to high cost drugs were not considered in the main analysis due to exclusion from HES. Hospital costs were calculated using the National Tariff 2010/11 [14] and Reference Costs 2006-10 [15] for HRGs excluded under the National Tariff. Costs were inflated using the consumer price index. Costs are presented (£, 2011 prices) as mean annual costs per patient and total costs per year (based on all inpatients and outpatients from 2006/07 to 2010/11). To compensate for the nine months provisional data in the final year, all costs, patient numbers and annual spells were multiplied by a correction factor (assuming no treatment seasonality) prior to mean and standard deviation calculations.

Under HRG4, some significant elements of cost and activity have been ‘unbundled’ from the core HRGs. The ‘unbundled component’ becomes a HRG in its own right as an addition to a core HRG. A spell of care may involve several procedures and care phases (for example diagnostic imaging, surgery and rehabilitation) each with an individual cost. Unbundling of the tariff separates the payment for that particular treatment, allowing different providers to receive the costs associated with different aspects of a procedure. Despite being constituents of unbundled HRGs, the procurement and delivery of chemotherapy and radiotherapy were disaggregated prior to analysis. Statistical analysis of the HES data was conducted using Statistical Analysis System Enterprise guide 4.3, and a costing algorithm was developed using Microsoft Excel 2007 with Visual Basic for Applications.

Due to limitations of the HES database which include the inability to distinguish cancer stage and type (initial versus recurrent; incident versus prevalent), the exclusion of treatment costs and the short data period in comparison to survival rates, a mathematical model of the treatment pathway was developed in order to estimate total per patient treatment costs. The treatment pathway was based on the European Association of Urology Guidelines on Penile Cancer [16], with probabilities for different treatment approaches taken from published local sources and expert opinion where necessary.

Firstly, a series of decision trees were constructed to estimate the costs of referral, staging and treatment for the primary tumour and local lymph nodes (Fig. 1). Each branch of the decision tree is associated with a probability based on publically available information and expert opinion (please see Additional file 1). A Markov extension to the model was then used to estimate the costs of follow-up, taking into account mortality and the probability of local, regional or distant relapse.

Figure 2 shows the structure of the Markov model. Given that the risk of relapse tends to decrease over time, local and regional relapse probabilities were assumed to follow a Weibull distribution, with the probability of the latter falling to zero after five years [4]. For distant relapse, the risk was assumed to be constant up to two years after which it was assumed to be zero. The risk of death from any of the states in the model were also assumed to be constant.

Values for all parameters detailed above, including the time dependent risks for local and regional relapse, were estimated by fitting the model to published data from a large study of outcomes in penile cancer patients across two specialist treatment centres, one in Sweden and one in the Netherlands [4]. This was achieved using a recursive algorithm, with goodness-of-fit assessed using ordinary least squares. Figure 3 shows the fit of the calibrated model to data on the various types of relapse.

Consistent with the approach used in the HES analysis, costs applied in the model were taken from the 2010/11 National Tariff where the HRG associated with a treatment or intervention was known and included within the tariff provisions. The Reference Costs for the same year were used for all off-tariff payments. Four core scenarios were examined: no price inflation and no correction for regional market forces, no price inflation and correction for regional market forces, price inflation (assumed to remain constant at 2011/12 levels [17]) and no correction for regional market forces, price inflation and correction for regional market forces. In addition, a one way sensitivity analysis was carried out on the following inputs: mode of admission, costs of treatment, intensity of follow-up, stage at presentation, approach to primary tumour management and approach to lymph node management. Full details of the model inputs are available in Additional file 1.