Population
Inclusion criteria: adult patient, without opposition to the extraction of data from her/his file and the follow-up on day 30, scheduled for one surgical model interventions or a control intervention. The following interventions were studied as ERP models since specific guidelines exist for ERP in these models, they are frequent (≥ 600 annual stays at the APHP) and carried out in around ten different services (which will allow an analysis of the determinants at the service level. We chose total knee arthroplasty (approximately 1,500 interventions per year on 12 sites, 10 of which are active (i.e > 30 interventions / year), left colectomy for cancer or non-cancer pathology (600 annual interventions on 17 sites, 9 of which have an activity > 30 interventions / year) and hysterectomy (approximately 570 procedures per year on 16 sites, 12 of which have an activity > 30 procedures / year). This audit covered both the target intervention (ERP interventions) and control interventions of the same severity not specifically targeted by the ERP approach (i.e. having not benefited from specific clinical paths). Three other surgical models were analysed as control to observe spontaneous evolution of medical process: total hip arthroplasty for orthopaedic surgery; gastrectomy for visceral surgery and ovariectomy for gynaecological surgery
Each selected centre had more than 30 interventions per year for the selected surgical model and 30 randomly selected files were analysed per centre (20 files for ERP interventions and 10 for control interventions)
Non-inclusion criteria: none
Intervention
The comparison scheme was “before-and-after” type for the two 2015–2017 periods comparison. We consider the year 2015 as a “reference” year and the year 2017 as a year of full implementation, 2016 being for launching.
Between the two evaluations period, an institutional awareness-raising phase on ERP in 2016 was composed of 5 steps: 1. One day of sensibilization on ERP in the institution in April 2016 with a meeting of 200 health care providers and institutional stake holders; 2. Development of a Massive Open Online Course on ERP in July 2017; 3. Diffusion of national update on ERP by Haute Autorité de Santé in June 2016 to all surgical and anaesthesia departments of APHP (4); 4. Institutional access to GRACE group in 2016 for all surgical and anaesthesiology departments offering access to guidelines, scientific literature and possibility to organize survey and benchmarking (5); 5. Preparation of participation to a regional program organized by Regional Health Agency (Agence Régionale de Santé) offering inter institutional collaboration on ERP program. This was the initiative of ten surgical departments out of 29 involved in the survey participate in 2017 to the Regional Health Agency training program offering inter institutional collaboration on ERP program. These centres were involved as learning centres or teaching expert based on the existence of ERP protocols and previous evaluation with GRACE group. Four participate in orthopaedic surgery (2 expert, 2 learning centres), 3 in visceral surgery (1 expert, 2 learning) and 3 in gynaecological surgery (1 expert, 2 learning).
Outcomes
The main demographic (age, sex), clinical (main pathology, comorbidities, ASA score, autonomy) and sociologic (living alone) characteristics were collected simultaneously. The Charleson score was calculated for all patients.
Patients’ files were used to obtain duration of stay and incidence of complications after surgery. A list of eight complications was used for identification in the patient’s file (i.e. transfert to intensive care unit, new surgery, bleeding, infection requiring antibiotic, pulmonary embolism or venous thrombosis, allergy, aggravation of existing medical condition, other). It was defined in the patient’s chart whether this complication has prolonged the hospital stay or if a new hospitalization in the same institution occurred within 30 days after surgery.
For each ERP model intervention, a specific grid was designed based on existing ERP guidelines in 2016 for each surgery to collect data describing the patient management process pre, per and postoperatively (6). The cut off value of 70% for ERP item application (i.e. 70% of a particular item is applied on the patient population) was considered as reflecting sufficient appropriation (7).
It was performed when research assistant visit the hospital for data collection with a representative of surgical and anaesthesiology department to describe level of development of ERP, resources issue and difficulties. Concerning services, the characteristics collected concerned: structure (equipment) and resource (personnel, qualification) data; commitment to quality improvement approach (Morbi mortality review; quality professional improvement), existing clinical pathways. The same independent dedicated staff trained to data collection, collected all data and performed interview in each centre
Statistical analysis
Primary outcome was the length of stay of the different surgical models selected, over a 2 periods of one year, before (year 2015) then after (year 2017) the institutional awareness-raising phase on ERP. The secondary objectives were: effects on the length of hospital stays in each centre following an awareness phase on the practice of ERP; evaluation of the conformity of practices with regard to ERP before and after awareness-raising phase; evaluation of the incidence of 30 days post-intervention complications, in each intervention, before and after the implementation of the program
We postulated that the implementation of the ERP would modify the compliance rate from 60% of non-compliant files “before” to 40% of non-compliant files “after”, then requiring, for a two-sided alpha risk of 5% and 80% power, 2 groups of 110 files, ie 220 files in total (110 before and 110 after) per model intervention.
However, there is probably a cluster effect at hospital level, all the interventions of a model being supported by a single anaesthesia service and a single surgical team, and therefore the observations of the same hospital tend to resemble each other and provide less information than the same number of independent observations. Considering the literature on cluster effect for process-related variables, we retained a 0.02 ICC (8). Then, starting from 25 files on average per hospital and model, the design effect is around 1.5, and 380 files must be studied per model intervention (190 before and 190 after) to obtain the same information amount.
The number of “control” interventions is halved.
All comparisons were made with Chi2 test or Fischer test for discrete variable and t-test or Wilcoxon rank-sum test for continuous variable between 2015 and 2017 period for the whole population and each centre for control and ERP surgical models.
A linear mixed model was used to identified variables related to LOS among patient's characteristics, by taking into account the existing correlation between files from the same surgical department.
Missing data were analysed depending on the type of item and best available interpretation. Most of items that were not available in the patient’s chart were considered as absent and non-performed.