chapter 14 enhancing access to uk renal registry data through innovative online …€¦ · ·...

Chapter 14Enhancing Access to UK Renal RegistryData through Innovative Online DataVisualisations

Afzal Chaudhrya,b, Terry Feestb

aCambridge University NHS Hospitals Foundation Trust, Cambridge, UK; bUK Renal Registry, Bristol, UK

Summary

. It is now possible to interactively interrogate,analyse and produce displays of UK Renal Registrydata online at http://www.renalreg.com/n_portal/pages/main/registryportal.php. This enables anyindividual to interrogate the validated anonymisedUKRR datasets.

. This bespoke web-based interactive data portalprovides a focussed point of access to a variety ofgraphical display formats and analyses of UKRRdata.

. Centre-specific reports can be produced – includinga colour-coded dashboard summary as well as bothfunnel plots and longitudinal statistical processcontrol charts for a range of clinical parameters.

. Interactive flash-based longitudinal Statistical Pro-cess Control charts are available on a per-centreand per-parameter basis allowing for a moredetailed review of performance over time.

. There are Rosling/Gapminder-style motion chartson a per-parameter basis simultaneously detailingperformance and activity data from multiple centresinteractively over time (more details below).

. There is an interactive graphical pivot chart solutionusing OLAP (online analytical processing) tech-nology allowing users to design and export theirown charts/analyses in real-time using UKRR data.

. The portal will empower the UK renal communityin the comparative analysis of delivered renal careultimately hopefully leading to enhanced qualityimprovement over time.

257

Background

The UK Renal Registry (UKRR) has grown fromstrength-to-strength since its inception in 1997 andnow receives data from every renal centre in England,Wales and Northern Ireland with a more limited datasetbeing contributed annually by the Scottish Renal Regis-try. Each year the report provides detailed comparisonsof the activity and performance of each of the submittingcentres across a range of clinical parameters. Whilst thisdata and its associated analyses are valued by a widerange of stakeholders (the renal centres themselves,the NHS Renal healthcare Commissioners and alliedpatient-related groups e.g. the National Kidney Federa-tion), it is recognised that as the volume of data andanalyses grows with time, so does the need to presentthis increasingly complex information in an accessibleand clinically informative manner which is responsiveto, and reflects the nature of, the enquiries made bythose seeking to access the data. In particular, the needto reflect changes in activity and performance overtime (longitudinal data) as opposed to the pre-dominantly ‘snapshot’ cross-sectional data contained inthe annual report is essential if the UKRR is to achieve

its goal of monitoring renal care in the UK thus leadingto improvements in the quality and efficiency of this verysame care.

Online Interactive Geographical Maps

It was in this context that in 2009 the UKRR launchedthe world’s first interactive maps (figure 14.1 and table14.1) detailing the achievement of quality measures inthe care of dialysis patients spanning a five-year period.Initially these maps were confined to a variety ofhaematological and biochemical parameters (table14.2) based around healthcare commissioning geo-graphies. They were soon followed by the addition ofnational public health datasets (cardiovascular mortality,indices of social deprivation and ethnic distributionscourtesy of the East Midlands Public Health Observa-tory, table 14.2) based on these same geographies allow-ing for a limited exploration of correlations betweenthese parameters and UKRR data. Renal centre-basedmaps displaying UKRR data alone are now also available(table 14.2).

Fig. 14.1. Interactive Geographical Maps – http://www.renalreg.com/Maps/map_root/maps.html

258

The UK Renal Registry The Thirteenth Annual Report

The Statistical Comparison of Data

Historically many of the graphs in the various UKRRreports describing, for example, the proportion ofpatents in a given centre achieving a given biochemicalstandard, have been presented in the form of a

‘caterpillar’ plot with centres listed on the x-axis inorder of the percentage achievement. At the same timeit was recognised that these types of plots perhapsencouraged inappropriate statistical comparisonsbetween centres. In the context of the online maps, theguidance states:

Table 14.1. UK Renal Registry online interactive map types

Geography basis Type Description

Area-basedCentre-based

Single View containing a thematic map, table and chart indicating spatial/geographical patterns and temporal trends for a specific indicator of interest.


Double View containing two thematic maps allowing for the comparison of patternsand relationships between two indicators for the same geography type.


Area Profile View allowing users to select a geographical ‘area’ and to chart a cross-sectionof selected key indicators. This provides an ‘at-a-glance’ assessment of theselected ‘area’ in comparison with other ‘areas’, including regional and/ornational geography regions.

Centre-based Funnel Plot View of cross-sectional data containing a thematic map, table and chartindicating spatial/geographical patterns as well as a funnel plot with upper andlower 95% and 99.9% confidence intervals.

Table 14.2. UK Renal Registry online interactive map datasets

Country Geography UKRR data Public Health Data

England Primary CareTrust

RRT incidence and prevalence ratesBone mineral metabolism (Ca, PO4, PTH)Control of anaemia (Hb, ferritin)Haemodialysis adequacy

Circulatory disease mortality 2005–2007a

Index of Multiple Deprivation 2007b

Ethnicity – Office of National Statistics 2006c

NorthernIreland

Local District RRT incidence and prevalence ratesBone mineral metabolism (Ca, PO4, PTH)Control of anaemia (Hb, ferritin)Haemodialysis adequacy

Circulatory disease mortality 2005–2007d

Northern Ireland Multiple Deprivation Index2005d

Ethnicity – 2001 Censusc

Scotland Health Board RRT incidence and prevalence rates Circulatory disease mortality 2006–2008e

Scottish Index of Multiple Deprivation 2009e


Wales Local HealthBoard

RRT incidence and prevalence ratesBone mineral metabolism (Ca, PO4, PTH)Control of anaemia (Hb, ferritin)Haemodialysis adequacy

Circulatory disease mortality 2002–2004f

Welsh Index of Multiple Deprivation 2008g


All Renal Centre As maps above, plus (where data is available):Data ‘completeness’Acid-base control (Bicarbonate)Blood pressure control

No data aggregated by renal centre available

Data Sources:a Compendium of Clinical and Health IndicatorsbNational Statistics, Department of Communities and Local Governmentc Office for National Statisticsd The Northern Ireland Statistics and Research Agencye ScotPHOf National Public Health Service for WalesgWelsh Assembly Government

259

Chapter 14 Access to UKRR data

‘Caution should be used when comparing databetween many areas using maps such as these. Unlessyou have a priori comparisons in mind and makeonly those specific comparisons then there will beproblems.

. . . The funnel plots allow for the identification ofrenal centres which fall outside of the upper or lower95% and 99.9% confidence intervals. These compare allcentres with the average rather than being centre tocentre comparisons like you are tempted to make fromthe ‘caterpillar’ style plots. You are still making multiplecomparisons (comparing each of about 70 centres withthe average) but using the 99.9% intervals instead ofthe 95% ones makes some adjustment for this.’

More recently therefore the UKRR has made increaseduse of funnel plots for presenting such data in the annualreport.

Contemporaneously with the introduction of thesemaps, Dr Alex Hodsman, Renal Registry ResearchFellow, was advancing the application of additionalrobust and well-validated statistical analytical techniquesto UKRR data. One area of focus in her research wasthe use of statistical process control charts, originallydeveloped by Dr Walter Shewart for the Bell TelephoneCompany in 1924 (see: http://en.wikipedia.org/wiki/Walter_A._Shewhart).

Dr Shewart realised that variation in data can occurfor a number of reasons. Assuming that data collectionand processing is robust then there are principally twotypes of variation: ‘common cause’ and ‘special cause’variation. All processes demonstrate some degree ofrandom variation, which is known as ‘common cause’variation and a particular process is said to be ‘incontrol’/’stable’ if it demonstrates only ‘common cause’variation. However, unexpected events or situationsarising in a process can result in ‘special cause’ variation.In such circumstances, a process is said to be ‘out ofcontrol’/‘unstable’. Variation of this type needs to beidentified and, if genuine, investigated further.

Statistical process control (SPC) charts can be used todifferentiate between these two types of variation (usinga variety of ‘rules’) and can be plotted as either cross-sectional or longitudinal charts using either static (asnapshot in time) or dynamic (a series of data pointsover time) data respectively. Funnel plots are a meansof displaying cross-sectional/static data whilst longi-tudinal run charts can be used to display longitudinal/dynamic data (see: http://www.indicators.scot.nhs.uk/SPC/Main.html).

The UK Renal Registry Online Data Portal

Following on from:

. considerable positive feedback regarding the inter-active maps,

. a desire to extend the statistical rigor applied toUKRR data (as outlined above), and

. feedback from many centres about their ownaccessibility to UKRR data with increasingly fre-quent requests for timely access

the UKRR now wishes to report the extension of thisonline interactive strategy to the deployment of abespoke interactive data portal (Summer 2011). Theaim of the portal is to provide a focussed point ofaccess to a variety of graphical display formats andanalyses of UKRR data, with figures 14.2A, B, C and Dshowing the different data presentation options:

A. Centre-specific reports – a distillation of annualUKRR data including a colour-coded dashboardsummary as well as both funnel plots and longitu-dinal statistical process control charts for a rangeof clinical parameters. The dashboard describesfor each clinical parameter:. the centre’s performance (percentage achieve-ment of standard) for that year

. a numerical and colour-coded comparison withthe previous year

. whether or not the centre is an outlier on afunnel plot in that year

. whether or not the centre’s performance exhi-bits ‘special cause’ variation over time on alongitudinal SPC

. the mean percentage achievement of all centresin the same region for that year

. a colour coded median rank (with 95% confi-dence intervals) amongst all centres for thatparameter based upon a standard statisticalsimulation model.

B. Interactive flash-based longitudinal SPC charts ona per-centre and per-parameter basis allowing fora more detailed review of performance over time.These charts are the interactive correlates of thoseavailable in the centre-specific reports.

C. Rosling/Gapminder-style motion charts on aper-parameter basis simultaneously detailingperformance and activity data from multiple cen-tres interactively over time (more details below).

260


D. An interactive graphical pivot chart solution usingOLAP technology allowing users to design andexport their own charts/analyses in real-timeusing UKRR data.

Rosling Motion Charts

Hans Rosling, Professor of International Health atthe Karolinska Institute, co-founded the GapminderFoundation (see: http://www.gapminder.org/) whichdeveloped the motion chart software system. Thiswas most notably popularised in a much admiredtalk given by Professor Rosling at the Technology,

Entertainment and Design Conference in 2006. (See:http://www.ted.com/talks/hans_rosling_shows_the_best_stats_you_ve_ever_seen.html)

A motion chart graphical system is now freelyavailable for use as part of the Google visualisation pro-gramming interface and it is this system which has beenutilised in the UKRR data portal. Motion charts are apotentially very powerful graphing system able to displayup to five parameters at any one time (representedby: ubble graphic size, colour/appearance, x-position,y-position and time, respectively). This enables complexdata interplays to be explored and investigated in anintuitive manner far in excess of what can usually beachieved with a two dimensional static scatterplot typeof chart.

Fig. 14.2. Interactive Online Data Portal – http://www.renalreg.com/n_portal

261

Chapter 14 Access to UKRR data

OLAP Pivot Charts

An OLAP (online analytical processing) cube is adata structure that allows for the rapid analysis of datacategorised by a number of dimensions. Each of theelements of a dimension can be described using ahierarchy which is a series of parent–child relationshipsin a generational type structure. So, each parent isderived by the aggregation of its child members and inturn, parent members at any one level may be furtheraggregated as the children of another parent at thenext higher level. As an example, September 2003 canbe aggregated into Quarter 3 2003, which in turn canbe aggregated into Year 2003 – all of these membersbelong to the ‘Time’ dimension.

The numerical data in the cube goes to form themeasures of the cube. These are usually summary calcu-lations (e.g. minimum, maximum, mean, total etc.) forthe various data points aggregated across the dimensionsof the cube at the level of the hierarchy selected (e.g. themean percentage attainment of a given standard for allcentres in Quarter 3 2003).

One of the strengths of an OLAP cube is the ability todrill-up and to drill-down through the hierarchical levelsof one or more dimensions (e.g. starting by looking atdata aggregated by year, then drilling down to dataaggregated by quarter and so on). Another is the abilityto rapidly re-orientate the cube so as to look at thesame data but from an entirely different perspectivee.g. instead of looking at data through the ‘Time’ dimen-sion, the data could be viewed from the perspective of the‘Location’ dimension which might start with the fourHome countries of the UK, drilling down to a regionallevel and then down to the level of the individual renalcentres themselves. In reality, data is usually exploredand aggregated simultaneously across multiple dimen-sions (e.g. geographical region, time, RRT modalityetc.) – but the power of the OLAP cube means thatchanges in the selected dimensions/hierarchies arereflected in the displayed graphic within seconds ratherthan in the minutes or hours that might be needed ifaggregating this data via a traditional relational databasequery.

In the implementation of an OLAP cube used inthis data portal, the OLAP processing is done in theFlash/Adobe Flex front end client running on the localcomputer following transfer of all the raw data from aMySQL database on the UKRR web server. A more

efficient strategy would be for the OLAP processing totake place on the UKRR’s server following submissionof a query and for only the results of the query to bepassed to the requesting client. This is to be consideredin a future development.

At present the power of the OLAP cube is tied to aninteractive pivot chart style of front end where the usermay ‘drag and drop’ parameters and in addition, mayselect individual members of different hierarchies inorder to generate their own personalised charts fromvalidated UKRR datasets. These charts may then beexported in either a jpg of pdf format for use locally.

Future Plans and Summary

All of the online interactive visualisations use the samevalidated UKRR datasets as those used in each respectiveyear’s annual report. Currently (Summer 2011), theonline maps house data from the 2003–2008 UKRRdatasets (except for the funnel plots which house the2002–2007 datasets). The centre-specific reports in theonline portal are derived from the 2007 dataset whilstthe SPC charts, the motion charts, and the OLAP pivotchart all use data from the 2002–2007 datasets.

Over the next few months more of the UKRR’s annualdatasets will be uploaded into portal. More complexcomparisons will be developed to take advantage of thepower of the motion charts and as the volume of datagrows so the technical structure of the OLAP chartingsystem will be revised to improve the responsiveness ofthe system as outlined above. In addition, the aim is tomore fully integrate the geographical maps into thedata portal itself. Further refinements will also be madein response to the feedback received from users of theportal.

We believe that this work – ‘Enhancing Access ToRegistry Data Through Innovative Online Data Visuali-sations’ – builds strongly on the wealth of informationarising from the high-quality validated UKRR datasets,and that both the maps and the portal will empowerand engage the UK renal community in the comparativeanalysis of delivered renal care ultimately leading toenhanced quality improvement over time.

Conflicts of interest: none

262


chapter 14 enhancing access to uk renal registry data through innovative online …€¦ · ·...

Documents