French hospital database PMSI-MCO
The French computerised hospital database for medicine, surgery and obstetrics (PMSI-MCO: Programme Médicalisé des Systèmes d’Informations – Médecine Chirurgie Obstétrique) covers all short-term hospitalisations in France (about. 20 million public and private admissions per annum). It comprises a standardised case record compiled at discharge after each short-term hospitalisation and encodes administrative and medical data with standardised classifications and nomenclatures.
The system is derived from the US DRG classification and employs a decision tree that leads each hospitalisation towards its appropriate diagnostic or surgical procedures group (Groupe Homogène de Malades: GHM). Accessible data for all patients include age, gender, admission and discharge addresses, primary and secondary diagnoses, treatment procedures, DRG codes, admission duration, intensive care resources deployed, hospital identification, public or private sector, etc.
Nomenclature adopted the Xth revision of the International Classification of Diseases (ICD-10) of the World Health Organization [10], defining conditions by one primary diagnosis and up to 20 secondary diagnoses per hospitalisation, plus the French classification of medical procedures “La Classification Commune des Actes Médicaux” (CCAM), available as a full listing [11]. Thus medico-administrative (PMSI) variables plus one surgical act were sufficient to classify a hospitalisation as a ‘surgical’ DRG. After 2004, global hospitalisation budgets were progressively replaced by new budgets linked to specific activities (T2A Tarification A l’Activité). The new procedure was fully in place by 2008. Official tariffs per DRG in French hospitals are revised and issued every 6 months by the French Ministry of Health [12]. These tariffs per DRG are charged to the Sickness Funds for each hospitalisation. These tariffs were not used for our costing.
Study patients
The study was limited to surgery and included patients of all ages. DRG classification was performed at hospital discharge in 2008, hence some patients hospitalised in 2007 were included and others not discharged until 2009 were necessarily excluded. All hospitalisations extended to at least 2 nights. Shorter hospitalisations were less documented and no death was recorded in these short stays. Patient inclusion was based on surgical DRGs identified in the French system by a third letter coding ‘C’ for surgery (Chirurgie). A surgical procedure was the primary criterion specifying such DRGs.
Procedures such as aortic aneurysm repair are not classified as surgery.
Bleeding-related complication and/or blood product transfusion status
Bleeding-related complications and/or blood product transfusions were identified according to diagnosis and procedure codes relating to bleeding complications. Patients were divided into two groups, i.e. with bleeding (WB) or without bleeding (WoB) provided one of the following codes was retrieved for the hospitalisation: (1) Procedures codes (i.e. 11 CCAM codes: Additional file 1) for secondary haemostasis following index surgery (not preventative transfusions); or (2) one of the four following ICD Diagnosis codes: T81.0 for haemorrhage or haematoma complicating a procedure; Z513 for blood products transfusion; Y63.0 for natural blood and blood products; and Y446 for transfusion reactions.
It should be noted that no information was available concerning which blood products were transfused, since their costs were not included in the database. As diagnoses and codes for transfusion procedures were confounded, actual transfusion rates may be underestimated by the lack of specific coding. Also, surgical patients identified by bleeding, transfusions, or other haemorrhagic outcomes may have been coded for a manifestation of their underlying condition, or for a complication of surgery, with no possibility of identifying the real cause of bleeding. Hence in this manuscript ‘bleeding’ necessarily confounds bleeding complications and transfusions. Data concerning pre-operative bleeding risks were not available.
Statistical analysis
Hospitalisations associated with bleeding complications were extracted for each diagnosis-related group and their frequencies were ranked across DRGs. Surgical DRGs presenting where ≥10% of patients presented with bleeding were selected for further analysis. Control patients were those without bleeding (WoB) drawn from similar surgical DRGs. The ≥10% criterion was chosen empirically to include sufficient WB and WoB group sizes for comparisons of costs and LOS outcomes. DRGs were merged when they involved patients within the same organ classification undergoing similar surgical procedures with bleeding differences (<3%) (Additional file 2) to increase the power of the statistical analysis.
Comparisons of WB and WoB groups within each merged DRG were based on patient characteristics (age and gender) and DRG aspects unrelated to the classification, e.g., number of procedures and secondary diagnoses. Patient groups were compared for age, gender, number of procedures, number of diagnoses, adjusted length of stay (LOS) and costs, using non-parametric statistics (Wilcoxon). Cost estimates were derived from the French National Costs Scale based on information from a sample of hospitals using similar accounting methods and linked to the PMSI. This enabled statistical and economic analyses to calculate average costs per DRG. Separate scales were applied to public and private sectors. Cost estimates per DRG comprised total hospital costs including overheads and public sector physicians’ salaries and private sector physicians’ fees not included in the tariffs. The foregoing standard costs were applied to DRGs according to the hospital sector. No difference was observed in the bleeding rates between sectors.
Rates of hospitalisations for WB and WoB surgery exceeding the PMSI average LOS were compared using multivariate logistic regression that included confounding variables. The analysis was kept simple and objective, i.e. bleeding versus non-bleeding cases. Confounding variables were selected that primarily permitted comparisons of DRGs (with and without bleeding) according to patient characteristics (age and gender), LOS, numbers of diagnoses and procedures. Convergence within the model and statistically significant covariates were checked. No centre effect was included in the analyses as hospitals belonging to regional networks could not be precisely identified in the database.
Prior to comparisons, costs of hospitalisation with and without bleeding were calculated per surgical group and complication status. Adjustments for age, gender, numbers of procedures and secondary diagnoses were performed for the WB complication groups based on the difference observed between the WB group and the WoB group taking the WoB group as the reference. Transfer of patients to other acute care hospitals was rare and can be ignored.
Statistical analyses were performed using SAS Statistical Software #9.2 (SAS Institute Inc., North Carolina, USA) with a 5% significance threshold for all tests.