A retrospective analysis was performed using data extracted from the French Medical Information System (Programme de Médicalisation des Systèmes d'Information - PMSI) for 2007 which covers all French public and private hospitals, except military and psychiatric hospitals. In fact, since 2004, all French hospitals has adopted a prospective payment system based on case-mix, called "Tarification à l'Activité". Each hospital stay resulted in the production of a standard discharge summary ("Résumé Standard de Sortie" RSS) following inpatient conventional stays, day-hospital stays or sessions. The RSS contains information on the nature of the treatment and work-up (the examinations) carried out during the stay, on the main diagnosis that led to the hospital admission, on co-morbidities or possible complications. Diagnoses are coded using the International Classification of Diseases, 10th revision (ICD-10) either as primary, related, or significant associated diagnosis. The RSS is then integrated into a Diagnosis Related Group (DRG) used for classification of hospital stays.
All the hospital stays and sessions performed during a specific year are summarized into standardized discharge reports and collected in a national database called PMSI. This database is the basis of hospitals funding, but also allows to estimate various indicators by disease, like the number of patients treated annually, the number of stays per patient or the burden of the disease, with an exhaustiveness close to 98% in 2007 .
The SAE database (''Statistique Annuelle des Etablissements de santé'')  relies on information collected through an exhaustive and compulsory survey covering all French hospitals. It was used to extract data on radiotherapy sessions for private sector, as they are not available in the PMSI database.
Since the introduction of a DRG type prospective per case payment in 2004, a list of so-called expansive drugs has been set. Conversely to other drugs whose cost are included in the DRG tariff, drugs on the list are reimbursed 100% to the hospital based on a national reference reimbursement tariff and the EMI, conditional to the adherence of prescription recommendations. The FICHCOMP database contains the expenses of drugs on the list, by drug and per stay (for public hospitals only).
Cases were extracted from the PMSI database using the ICD-10 codes referring to head and neck cancers. They were classified into five categories corresponding to five localizations: oral cavity (including lip, tongue, gum, floor of the mouth, palate, and mouth; coded as C00, C02-06), salivary glands (including parotidis glands, salivary glands; coded as C07-08), oropharynx (including base of tongue tonsil, oropharynx; coded as C01, C09-10), pharynx other than oropharynx (including rhinopharynx, pyriform sinus, hypopharynx, others; coded C11-14) and larynx (coded as C32).
Number of stays
The annual number of stays was assessed after extraction of all hospital stays with one of the above-mentioned head and neck cancers as a primary diagnosis in the hospital database. Since a single hospital stay may include other ICD codes with head and neck cancers coded as related or significant associated diagnosis (and not primary diagnosis), a medical interpretation by a PMSI specialist was required to assess whether the specific hospital stay had a direct link with the diagnoses of interest. All medical stays were classified according to the type of management: medical, surgical, exploration or palliative care. Distinction was made between full stays and one-day hospital setting (defined as hospitalizations of less than 48 hours). Diagnosis Related Groups (DRG) were evaluated, as well as the length of stay for full stays.
Number of chemotherapy and radiotherapy sessions
The number of sessions referred both to chemotherapy and radiotherapy sessions. Data on chemotherapy sessions were extracted from the PMSI database, by selecting all hospital stays with a primary diagnosis coded as Z51.1 (''Chemotherapy session of neoplasm'') and an associated or a related diagnosis coded as a head and neck cancer. In the same way, data on radiotherapy sessions were extracted using the Z51.0 code (''Radiotherapy session'') as primary diagnosis. It only concerned the radiotherapy sessions performed in the public sector (radiotherapy sessions performed in the private sector are usually not reported in the PMSI database). The SAE database was used to estimate the overall annual number of radiotherapy sessions performed in public and private hospitals in 2007 and to calculate the ratio between public and private radiotherapy sessions. The annual number of radiotherapy sessions performed in the private sector was then estimated by applying the ratio to the number of radiotherapy sessions for head and neck cancers retrieved in the PMSI database in the public sector.
Number of patients
Since a patient may have several hospital stays or sessions during a year, the number of patients hospitalized at least once in 2007 for head and neck cancers was obtained by linking all hospital stays and sessions, based on patient's identification number. This number is built using the patient's social security number, date of birth and gender. After anonymisation, data are sent to the regional agency of hospitalization ("Agence Régionale d'Hospitalisation - ARH"). This number allows linking all hospital stays that occurred in public and private sectors in 2007 by patient (except for radiotherapy session performed in the private sector). It is then possible to estimate the total number of head and neck cancer patients that were hospitalized at least once in 2007, by sex and age group. In 2007, respectively 98% and 87% of all hospital stays and sessions were successfully chained, meaning that the total number of patients with radiotherapy or chemotherapy sessions was slightly underestimated.
Costs were considered from the healthcare payer perspective. It included hospitalization costs and expenses of innovative drugs, related to public and private sectors. Ambulatory costs and indirect costs related to productivity loss were not considered in the main analysis.
Hospital costs were calculated using the official 2008 diagnosis related group (DRG) tariffs for public hospitals (that were the tariffs available when the analysis was performed). Tariffs included nursing care, treatments, drugs, accommodation and investment costs for hospitalized patients. For public hospitals, it also covers medical and technical acts. For private hospitals, costs were estimated using the official 2008 DRG tariffs for private hospitals to which physician's fees were added as they are not included in private DRG tariffs and are reimbursed on a fee-for-service basis (source: ENCC 2006). Since no data are available on the cost for radiotherapy session in the private setting, its cost was estimated using those calculated for the public sector. Costs are presented as mean annual cost per patient (based on patients for whom data were firmly linked) and total cost per year (for all patients hospitalized in 2007) and were expressed by gender, cancer localization, type of care and sector (public or private).
The economic impact of expensive drugs was estimated by linking the chemotherapy sessions or stays for palliative care extracted from the PMSI database with the FICHCOMP database. It allows evaluating the proportion of these stays including the administration of an expensive drug, as well as the cost of expensive drugs. Since the FICHCOMP database was accessible for the first time for the year 2008, this was performed by using H&N cancers related stays of 2008 (by using the same methodology than in 2007).
The economic burden of H&N cancers attributable to HPV infection was estimated by using the specific prevalence of HPV infection by localization (data extracted from the international published literature, no available data for salivary glands and pharynx cancers other than oropharynx).