6-8PB of PACS Images Stored in Local Service Provider Data Centers – BridgeHead

This article was written by Shaun Smale, solutions consultant at BridgeHead Software Ltd.

There are an estimated 6-8PB of Picture Archive and Communication System (PACS) images stored in Local Service Provider (LSP) data centres as a result of the National Programme for IT.

This data is stored on SAN technology replicated across two locations plus a local cache at each hospital. Collectively, this equates to more than 30PB of raw SAN storage set aside for radiology data alone.
 
The number of radiology examinations has doubled in the last six years and yet the UK is still behind many other countries on the number of exams performed per head of population. If you then add the amount of clinical data the other ‘ologies generate ‘ into the equation, the overall storage requirement is immense and growing.
 
The challenge IT managers face right now is to balance storage cost versus clinical performance.
 
A fully populated 45TB SAN uses about 5Kw of power and generates 15 thousand British Thermal Unit [BTU] of heat per hour. If you do the maths, that equates to an electricity bill of approximately £611 a year for every useable terabyte of SAN storage just to keep the disk spinning and the system cooled. If, for arguments sake, 25TB of your archive is data that has not been accessed in the last ten years it has cost the Trust £153,000 in electricity alone to store this data. These numbers can be doubled for a resilient data centre solution; and do not include service cost, start-up and the normal five year refresh cost for the SAN hardware.
 
However, the same data could be stored on tapes that cost £12.50/TB and have the added benefit of consuming no electricity and generating no heat whilst they are inactive. Two tape copies for resilience, with one off-site, pushes the ten year storage costs for 25TB up to £625 for the tapes, plus relatively small additional costs such as transport, secure physical storage cost, and of course the tape drive and jukebox to read the tape if required.
 
While there has been a lot of rounding and estimation to get to these figures, plus SAN storage is admittedly becoming more affordable, one cannot escape the fact that storing inactive data with a very low probability of ever being viewed again on expensive SAN services is not cost effective for Trusts or the taxpayer.
 
How much stored clinical data is redundant?
Although it is difficult to ratify such figures, it is suggested that, with sufficient local cache to accommodate the last 18 months of images (on-line), the probability of retrieving an off-line image held in a long-term archive is between 2 and 6%.
 
NHS guideline’s for records management (NHS Code of Practice Part 2, annex D1: Health Records Retention Schedule) set out requirements for data retention based on the age and status of the patient. They are complex and dynamic, thereby making it extremely difficult to construct a suitable algorithm to identify consistent criteria for ‘deletion or culling’ of images. In the days of film, part of departmental workflow was to review and cull films. In the current digital age, there is a reluctance to delete any data; instead there is a tendency to store all data for the default 30 years and usually this data is stored on expensive SAN technology.
 
It is estimated that 80%  of a persons imaging is conducted in the last year of their life. Retention guidelines could be constructed to store data for eight years after death, unless other factors such as litigation or research are involved. Even if a large percentage of data needs to be stored for regulatory purposes, thereby extending the time it should be retained, it is unlikely to be viewed for clinical purposes again. In such cases, this data does not need to be stored on fast access media. In fact, it would not be unreasonable to argue that this data could even be stored on offline media.
 
BridgeHead Software believes that strategically managing all healthcare data, across all storage media types, is the most efficient way to control and drive down storage costs.

Clinical data, in active use, should be kept on high performance SAN, whether it is a database, raw data for number crunching (such as genome sequencing), or unreported DICOM images from one or more departments.

However, once the clinical episode is closed, the probability of access falls and the data could be moved to cheaper, lower performance, storage where it can be quickly retrieved as a prior study. After some time of inactivity, the copy could be held on tape alone, still accessible, but would take a few minutes (rather thans) to access. Finally, if the data is held only for regulatory purposes, the copy could be held on off-site tape. It may take days to recall, but costs very little to store. Other storage media, locations, vendors and technologies can be included in the mix to get the right balance of performance, resilience and reduced recovery time in case of an accidental deletion, loss, corruption, outage or disaster – an approach often referred to as a tiered storage architecture.
 
BridgeHead also believe that a Trust’s retention policy needs to be granular and make use of all sorts of data. For example, where an active oncology episode may need to review prior images annually, a suspected fracture following a skiing accident for the same patient may never be looked at again. For a typical LSP PACS, both studies are archived in expensive replicated SAN arrays housed in data centres and may be kept for 30 years.
 
What can be done?
By using BridgeHead’s Healthcare Data Management (HDM) solution, policies can be configured so that each file can be individually managed. This means that a copy can be stored on the most appropriate media based on the probability of it being accessed by a clinician.
 
Pre-fetch techniques are used by applications such as PACS to move files from their long-term stores or archives to the local cache. Where this is triggered by a request for a new exam, information contained in the request can be used to intelligently select prior images that are relevant to the current clinical episode.

In the absence of a scheduled examination, a planned visit, such as annual follow-up at the clinic could be used as a trigger. Here, there is less information available and so the application may simply retrieve all data for that patient irrespective of whether is it likely to be viewed.

This has the down side of filling the finite space of the local cache with irrelevant data. In the case of an unscheduled visit, such as an emergency trauma, there would be insufficient time to schedule a pre-fetch, an ad-hoc retrieval from the archive would be triggered. BridgeHead’s HDM Solution uses all these events to ensure that a copy of the file is transferred to the fastest storage media, in case the clinician needs it, without filling valuable space in the applications cache.
 
In conclusion, as storage costs fall, and capacity increases, there is a temptation to keep everything. However, this will soon be overtaken by the colossal amount of data being generated. Hard choices and intelligent decisions will be required to maximise efficiency and reduce storage costs of all healthcare data to improve clinical performance and, ultimately, patient care.