Heat map showing percentage of adult population at high risk of diabetes using a statistical smoothing technique. 

Context in source publication

Context 1

... distribution of a variable but allows additional mapping of factors that might influence this variable in each locality. Had the map focused on prevalence of diabetes rather than a risk score we could have included known risk factors. We avoided this, however, to avoid any overlap or collinearity between the variables of the QDScore. This exploratory study was made possible by a number of key partnerships. The work was led by DN, a public health registrar who had previously worked at Tower Hamlets Primary Care Trust and was on an academic attachment with the Centre for Primary Care and Public Health within the medical school, with input from the Department of Geography. An initiative to improve and maintain data quality of general practice records across the Primary Care Trust had been in place for several years, led by the Clinical Effectiveness Group within the Centre for Primary Care and Public Health. Key relationships and infrastructure including data sharing and governance arrangements were thus already in place to enable Clinical Effectiveness Group staff to securely download and audit data from the electronic medical records of 35 of 36 practices in the district which used the same electronic record system, the Egton Medical Information System (EMIS), and had recently moved to a web-based version of this system enabling remote access. 17 The study was classed as service ‘audit’ and deemed outwith its remit by the local National Health Service Research Ethics Committee in January 2011. The local information governance group representing the general practices at the Primary Care Trust agreed to the study and advice on data handling, and mapping was also sought from the National Information Governance Board. The tasks of identifying, extracting, manipulating, sharing, summarising and presenting our data, especially those derived from the electronic medical records of a large cohort of general practice patients, presented complex practical, technical and information governance challenges. To capture these, we prospectively collected a data set comprising documents (protocols, service-level agreements, agendas and minutes of meetings) and corre- spondence (letters, emails, notes of telephone calls). Those represented in this data set included the NHS Research and Ethics Board, University Departments, Tower Hamlets Primary Care Trust, local general practitioners and public health specialists, and the National Information Governance Board. We analysed this data set by applying a theoretical framework developed previously to study the complex organisational, social and political issues involved in introducing a nationally shared electronic medical record. 42 Specifically, we considered: (1) information governance challenges; (2) practical challenges, such as the ease with which procedures could actually be carried out and (3) technical challenges including issues of data security, downloading and interoperability. Completeness of general practice records in our selected cohort aged 25 e 79 years without diabetes (data set 1, n 1⁄4 163 275) was as follows: age (100%), gender (100%), ethnicity (92.1%), Townsend deprivation score (99.7%), body mass index (76.4%), smoking status (96.3%) and family history of diabetes (21.5%). Of the data that were used in the mapping (n 1⁄4 157 045) 9.48% of people (n 1⁄4 14 885) were at high risk of developing type 2 diabetes within 10 years. This is in addition to the 7% of the adult population of Tower Hamlets already known to have type 2 diabetes. 43 Records could not be generated or were removed if (1) the general practice was not able to share the data for technical reasons (n 1⁄4 3922) or patient permission was withheld (n 1⁄4 187), (2) the individual record contained no postcode (n 1⁄4 29) or lower super output area was not calculable from the available postcode (n 1⁄4 275), (3) the geographic location was outside Tower Hamlets (n 1⁄4 1813) or (4) there was a mismatch between records in set 1 and set 2 (n 1⁄4 4). This left 157 045 records for analysis (96.2%) representing 33 of 36 general practices. Reducing the list of restaurants to those with a major business purpose of takeaway food resulted in a total sample of 371 outlets. The basic map (figure 1A) illustrates the variation in prevalence of high diabetes risk across lower super output areas in Tower Hamlets, with a maximum of 17.3% of the non-diabetic population being at high risk (not including the 7% already diabetic). General practices and hospitals are also shown in figure 1A. The areas of highest prevalence for diabetes risk were distributed on either side of the main east-west highway which transects the district and corresponds with well-known deprived housing estates and high-rise blocks of flats on either side of this road. A basic map of Index of Multiple Deprivation scores by lower super output area (figure 1B) showed a near- identical geographical distribution with high diabetes risk. The heat map (figure 2) shows the same information as figure 1A but displayed as a globally smoothed surface over the entire geographic area. The prevalence of high diabetes risk in this smoothed version of the data varied from 5.1% to 13.8%. This way of visualising the data depicts d somewhat more dramatically d a high-risk ‘hot’ band running west to east through the deprived housing estates and much lower risk ‘cool’ areas in the more affluent riverside in the south and parkside in the north of the district. The heat map is free from the visual lower super output area administrative boundaries that are commonly used in maps of the ‘basic’ type. The resulting map is likely more intuitive for users to interpret due to the colour scheme, and there are no boundaries to disrupt the visualisation of diabetes risk. The ring map (figure 3) shows prevalence of high diabetes risk by middle super output area. In this depiction of the data, prevalence of diabetes risk ranges from 3.8% to 13.7%. Each middle super output area is shown linked to a band of three social and environmental indicators, which are often suggested to influence poorer health. 44 These are (from the inside out) fast food outlets per head of population, percentage of non-green space and population density per square kilometre. Overall, the ring map provided a striking visual display of type 2 diabetes risk in the areas that corresponded to known deprivation, and the ring provided a relatively new way of displaying social and environmental determinants of health at a small area level. The ring provides a dashboard of indicators of wider determinants of health that appeared most useful when locally applied to specific population groups of 5000 e 7200 persons. It demonstrates the sort of putative environmental determinants that public health specialists may want to map as part of routine health needs assessment to inform interventions at small area level. As we had anticipated, the information governance challenges were substantial and were as time consuming as the technical ones. In order to access the data from general practice records, permission had to be obtained from both the local information governance committee of the Primary Care Trust and the National Information Governance Board. In addition, because we considered that this project had a research element, we were also required to seek advice from the local National Health Service Research Ethics Committee and from the university’s Research and Development Office (who both deemed the project ‘audit’). Potentially identifiable data from patient records had to be handled securely under a protocol advised by the National Information Governance Board. This kept postcode information separate from clinical variables with pseudonymised conversion to lower super output area. Information governance issues were thus time consuming and required specialised knowledge and formal permissions, but they were not ...