Public hospitals in Paris edge closer to going entirely filmless

A common phrase in Parisian healthcare circles right now is “la région sans film.” Thanks to an ambitious enterprise-wide PACS project spanning the city’s extensive public hospital network, going filmless looks an achievable objective. Managers at the AP-HP (Assistance Publique- Hôpitaux de Paris) have sought to harmonize the practice and centralize the electronic organization over all its hospitals, allowing easy cross-reading between hospitals and protecting patient privacy, according to Dr. Daniel Reizine, a radiologist at the Lariboisière Hospital and coordinator of the project.

The goal, set in 2005, was to implement PACS across the organization within three years. The plan was to store all images online for at least five years. To achieve this, the PACS implementation was divided into three stages: storage, image viewing for radiologists, and image communication to clinicians. For each step, a transverse deployment over the AP-HP was used, harmonizing the practice between departments, and a centralized yet distributed architecture was developed, Reizine said.

The AP-HP is the main public hospital system of the city of Paris and its suburbs. Established in 1849, it is the largest public hospital system in Europe, with a total of 22,000 hospital beds (14,000 acute care, 8000 convalescent and long-term care). It provides healthcare, research, prevention, education, and emergency medical services and employs more than 90,000 people, including 15,000 medical doctors.

The AP-HP’s 47 hospitals handle 1.1 million hospitalizations, more than five million consultations, and a million emergencies a year. The system undertakes about two million radiological examinations per year on equipment that includes 36 CT scanners, 31 MR units, seven PETCT devices, and 37 SPECT systems. The modalities are linked to 37 radiological information systems (RIS) and are mainly located in 31 radiology departments and 16 nuclear imaging sections.

From the outset, the 47 hospitals were divided into hospitals with a PACS (two), which were excluded from the first stage of the project; hospitals without CT and MRI (25); and hospitals with at least one CT or one MR unit (20). The last group comprises 10 smaller hospitals, with 300 to 500 beds that carry out 50,000 to 100,000 examinations a year, and 10 larger hospitals, with more than 500 beds and 100,000 examinations a year.

The first step was to install 20 independent storage and viewing systems with a capacity of five years’ production storage in the DICOM lossless standard, Reizine said. After a pilot validation of the storage system and its link with both image capture systems and the backbone RIS, one hospital and then two hospitals per month were equipped. All other sites were connected to one of the 20 nodes. The second step was to build a single central database of all studies’ metadata, allowing a synchronization of the data across all 20 nodes.

The PACS software provider, Carestream, was selected after a tender process in 2004. The first node was installed in April 2005 after validation of the storage configuration and its links with imaging modalities and the RIS. Installation was complete in 18 hospitals by February 2006. Installation of the last two nodes was delayed until October 2006 due to local networking difficulties. Accessing of images by clinicians through the electronic medical record (EMR) began in mid-2008. The data center was completed in June 2008.

A new network was built for the imaging departments. A virtual network with a bandwidth of at least 100 Mb/sec was created between the radiology and IT departments. All existing modalities and workstations were connected within the new network. The same PACS software and hardware were used. One main server runs all PACS applications: archive manager, workflow manager, and web server. One storage bay (in RAID 5) is attached to the server, with a backup server and tape library for disaster recovery. All imaging modalities and the RIS are connected to the storage system.

Using floating and concurrent licenses and the web distribution offered by the PACS, it is simple to put all the existing applications-except the modalities’ workstations-on a single three-monitor environment, Reizine said.

For image viewing, clinicians must have a unique access code through the EMR. Access must be fast through a low-bandwidth network. An interface was built between the EMR, RIS, and PACS, and clinicians use the same web server and image data as radiologists (and the floating/concurrent licenses). The streaming technology of the PACS also helps.

Data are collected automatically and analyzed centrally. By October 2008, the system contained 3.6 million examinations from 20 sites stored in DICOM lossless standard. Every month, 120,000 new examinations are stored on the 20 distributed nodes. Between July 2007 and July 2008, 250,000 CT exams, 100,000 MR exams, and more than a million other exams were stored. CT accounted for 18% of the exams but uses 79% of disk space.

“Building an AP-HP imaging network would be useless if patient confidentiality were not strictly respected,” Reizine told delegates at the last RSNA meeting. “According to French law, access to patient files at a public hospital is restricted to the caregiving team.”

Each user receives a unique authentication login. Automatic logout occurs after 10 minutes of inactivity. Radiologists can access the global worklist of all exams stored in the local PACS; clinicians have access only to those patients they are in charge of. Access control is managed by the EMR. Remote reading by a radiologist can occur only if a locally authorized radiologist grants permission.

No image data are stored in the central system during the synchronization phase or during the examination viewing process. Network or hardware failures at one site or in the central database do not affect business continuity at other sites. Proper rollout was achieved at all sites within the initial budget, Reizine said. The system stores 150,000 examinations a month, using 5 TB.