Using electronic patient records to assess the effect of a complex antenatal intervention in a cluster randomised controlled trial—data management experience from the DESiGN Trial team

Abstract

Background The use of electronic patient records for assessing outcomes in clinical trials is a methodological strategy intended to drive faster and more cost-efficient acquisition of results. The aim of this manuscript was to outline the data collection and management considerations of a maternity and perinatal clinical trial using data from electronic patient records, exemplifying the DESiGN Trial as a case study. Methods The DESiGN Trial is a cluster randomised control trial assessing the effect of a complex intervention versus standard care for identifying small for gestational age foetuses. Data on maternal/perinatal characteristics and outcomes including infants admitted to neonatal care, parameters from foetal ultrasound and details of hospital activity for health-economic evaluation were collected at two time points from four types of electronic patient records held in 22 different electronic record systems at the 13 research clusters. Data were pseudonymised on site using a bespoke Microsoft Excel macro and securely transferred to the central data store. Data quality checks were undertaken. Rules for data harmonisation of the raw data were developed and a data dictionary produced, along with rules and assumptions for data linkage of the datasets. The dictionary included descriptions of the rationale and assumptions for data harmonisation and quality checks. Results Data were collected on 182,052 babies from 178,350 pregnancies in 165,397 unique women. Data availability and completeness varied across research sites; each of eight variables which were key to calculation of the primary outcome were completely missing in median 3 (range 1–4) clusters at the time of the first data download. This improved by the second data download following clarification of instructions to the research sites (each of the eight key variables were completely missing in median 1 (range 0–1) cluster at the second time point). Common data management challenges were harmonising a single variable from multiple sources and categorising free-text data, solutions were developed for this trial. Conclusions Conduct of clinical trials which use electronic patient records for the assessment of outcomes can be time and cost-effective but still requires appropriate time and resources to maximise data quality. A difficulty for pregnancy and perinatal research in the UK is the wide variety of different systems used to collect patient data across maternity units. In this manuscript, we describe how we managed this and provide a detailed data dictionary covering the harmonisation of variable names and values that will be helpful for other researchers working with these data. Trial registration Primary registry and trial identifying number: ISRCTN 67698474. Registered on 02/11/16.

Evaluation of the reported data linkage process and associated quality issues for linked routinely collected healthcare data in Multimorbidity research: a systematic review

“…The two main themes emerging from the reported issues regarding the dataset were misclassi cation and missingness. This nding is consistent with previous research using routinely collected healthcare data for research (42). A poorly or improperly recorded variable could lead to huge discrepancies between a person's actual disease status and the status they are assigned in the study.…”

Evaluation of the Growth Assessment Protocol (GAP) for antenatal detection of small for gestational age: The DESiGN cluster randomised trial

“…While the learning process of care providers may delay full programme effectiveness, an alternative “pioneering effect” may be working in the opposite direction [ 33 ]. Other limitations include issues related to the availability, or format, of data that are inherent in the use of routinely collected data, though we followed clear protocols in harmonisation and linkage of data from multiple electronic systems to minimise any variations in data quality between the randomised arms [ 15 ]. Missingness for characteristics (including customisation factors) was dealt with by multiple imputation, which is dependent on the assumption that results after inclusion of variables in the imputation model will be consistent between those with and without missing data.…”

“…Data were obtained from 4 types of routinely collected electronic patient record system at each cluster: maternity, ultrasound, neonatal, and administrative [ 15 ]. Additional data were collected to assess compliance with the intervention in allocated clusters from review of a subset of women’s paper maternity records ( n = 120 per cluster).…”

“…Data management was performed to harmonise and amalgamate datasets from all clusters. This process has previously been described in detail and published [ 15 ]. The approach for multiple imputation of missing data is summarised in the S2 Appendix (page 5).…”

Improving antenatal detection of small‐for‐gestational‐age fetus: economic evaluation of Growth Assessment Protocol

“…All data required for analysis of clinical outcomes and costs were obtained from routine electronic patient records (EPRs). Our data collection and management methods, including a detailed description of data quality checks, have been published previously 11 .…”