Researcher Experience: Dr Feifei Bu

In this first Research Experience post of 2021 we hear from Dr Feifei Bu, Senior Research Fellow in the Department of Behavioural Science and Health at the University College London (UCL). Feifei first started working with administrative data in 2014 when she worked with the National Pupil Database linked to Understanding Society survey data (UK Household Longitudinal Study). In 2015, she joined the University of Stirling and started working on projects that were using administrative extensively. In particular, she worked with Scottish Morbidity Record (SMR) data linked with the Social Care Survey (now Source) and Healthy Ageing in Scotland (HAGIS). From there, her interest in carrying out research using administrative data continued into her current position at UCL where she has worked with Hospital Episode Statistics (HES) linked with English Longitudinal Study of Ageing (ELSA). She has also worked with de-identified Whole Systems Integrated Care (WSIC) data. All in all, Feifei has been carrying out research using administrative datasets for around seven years.

Overview of my research

My work using administrative data has been mainly around health service utilisation. Collaborating with colleagues from Stirling and Dundee, we had looked at the cost of hospital admissions for people with cognitive spectrum disorders using SMR data. In 2019, I worked on a project on the relationships between social factors and health outcomes amongst older adults using ELSA linked with HES. We looked at how loneliness and social isolation were associated with the risk of hospitalisation related to fall, cardiovascular disease and respiratory disease respectively. More recently, I led a project looking at how patient activation (a measure of people’s knowledge, skills and confidence to manage their own health and wellbeing) was related to the usage of different health care services, including GP and non-GP primary care, elective and emergency inpatient admissions, outpatient and A&E attendances. At the moment, I am involved in an ESRC funded project looking at how indoor temperature is related to secondary care health service utilisation using ELSA linked with HES.


Summary of any challenges faced

Unlike survey data that are usually thoroughly cleaned and well documented, administrative data often require some extra work. Based on my own experience, for example, the episode order variable comes with the SMR or HES data cannot be taken for granted. In some cases, it could be important to further sort them into the correct order. Also, it may take some detective work to find out what a specific variable measures or how data were collected in practice and by who—this could be critical for data interpretation.

A unique strength of administrative data is that they offer objective and detailed measures that are usually unavailable in surveys. However, as these data were not collected for research purposes, there is often a lack of other critical information that we would like to take into account in our research. If data linkage is not possible, this is an even tougher challenge than the one above.

Due to data protection purposes, administrative data often need to be analysed in a safe setting, like a data safe haven. This can usually be accessed via a remote desktop connection, but in some cases, you might need to go to a secure access point that is not necessarily local. This will slow down your progress significantly. Some administrative data are stored in data warehouses, in which case researchers need to extract data that are relevant to them using programming language, like SQL. In other instances, researchers may not have access to the data warehouse directly and data extraction need to be done by a data analyst. This would require a lot of planning ahead as well as communication back and forth. Finally, data access is time-limited in most cases. It may ‘expire’ before getting everything published. This is something that needs to be taken into account when applying for data access.

Working with administrative data is like learning to tame a dragon—albeit challenging, it is also exciting and rewarding!


Thoughts for fellow and future eCRUSADers

As previous Researcher Experience posts have mentioned already, the access application can take a long time to go through. It is important to plan ahead especially if you are on a tight schedule—either for your PhD or other funded projects.

It is important to acknowledge the limitations of administrative data, in particular, the lack of critical information that need to be ‘controlled for’ in analyses. We should not rule out the possibility that survey data may serve our research purposes better. Here is a note to myself, and to be shared with eCRUSADers: our passion for data should not outweigh a solid research design.

Public Benefit Privacy Panel Timelines

Project: Social Care Survey linked to Scottish Morbidity Record

Preparation of PBPP application: – December 2015- April 2016 (approximately 4 months)

Submission to initial PBPP approval: April 2016 – August 2016 (approximately 4 months)

PBPP approval to data access: August 2016 – April 2018 (approximately 2 years)

Publications using administrative data

Bu, F., Abell, J., Zaninotto, P., & Fancourt, D. (2020). A longitudinal analysis of loneliness, social isolation and falls amongst older people in EnglandSci Rep, 10 (1), 20064. doi:10.1038/s41598-020-77104-z

Bu, F., Zaninotto, P., & Fancourt, D. (2020). Longitudinal associations between loneliness, social isolation and cardiovascular eventsHeart. doi:10.1136/heartjnl-2020-316614

Bu, F., Philip, K., & Fancourt, D. (2020). Social isolation and loneliness as risk factors for hospital admissions for respiratory disease among older adultsThorax. doi:10.1136/thoraxjnl-2019-214445

Hapca, S., Guthrie, B., Cvoro, V., Bu, F., Rutherford, A. C., Reynish, E., & Donnan, P. T. (2018). Mortality in people with dementia, delirium, and unspecified cognitive impairment in the general hospital: prospective cohort study of 6,724 patients with 2 years follow-upClin Epidemiol, 10, 1743-1753. doi:10.2147/CLEP.S174807

Leave a Reply

Your email address will not be published. Required fields are marked *