Results for aim A: feasibility of setting up a cardiovascular magnetic resonance registry by linking sources of routinely collected data

Parts of this text have been reproduced from Brierley et al.27 © Article author(s) (or their employer(s) unless otherwise stated in the text of the article) 2018. All rights reserved. No commercial use is permitted unless otherwise expressly granted. This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.

Aim A: to determine whether or not it is feasible to set up a cardiovascular magnetic resonance registry in this patient population by linking sources of routinely collected data

• Objective A1: to implement consent and establish patient consent rate.
• Objective A2: to provide evidence of whether or not data linkage and extraction can be carried out across PPCI centres.
• Objective A3: to determine whether or not the registry can be compiled from HES and the PEDW rather than multiple HISs.

The number of data about ‘usual care’ provided by UK NHS hospitals that are being collected and stored electronically in HISs is rapidly increasing. Previous studies have shown that it is possible to use multiple HISs to retrieve and compile these patient-level data into a research database.61–63 This study tested the feasibility of setting up a large multicentre registry for patients who activate the PPCI pathway by linking data from multiple HISs within several hospitals. If the registry was feasible, we had the longer-term aim of using registry data to evaluate the clinical effectiveness and cost-effectiveness of CMR in this population.

Screening and consent

Standard consent was successfully implemented at all participating hospitals. Across the four participating hospitals, a total of 2462 patients were screened between 13 May 2013 and 4 September 2014. Of these, 1670 (68%) consented to participate in the study. Consent rates ranged from 59% to 74% (an average of 68% across the four hospitals). We investigated various options for obtaining consent during the study: obtaining consent at discharge (rather than straight after the procedure), and an opt-out model on the procedural consent form (adding a paragraph to the procedural consent form explaining that the clinical data collected would be used for research unless patients ‘opted out’ by ticking a box on the form). These were not implemented as they were deemed not feasible.

Detailed screening data were available for only one hospital (hospital A, Figure 5). A total of 1210 patients were screened in hospital A; 856 were identified through the cardiology database ‘query’ that was set up to identify eligible participants for the study and 354 were identified independently by the research nurse team. The other hospitals did not provide detailed screening data. We anticipated that local record linkage between databases held by a hospital would allow patients who gave consent to be distinguished from those who did not. Thus, it was intended that standard aspects of data collection for prospective studies, for example a screening log with reasons for ineligibility and a record of consent, would be implemented in our study as routine reports from the database. In hospital A, we set out to identify eligible patients through a database ‘query’ set up in the local cardiology database. It was hoped that this screening model would be exported to the other participating hospitals. However, when we started recruiting, it became clear that this screening model did not work, as many patients were identified independently of the database query. The other hospitals were, therefore, allowed to implement their own methods of screening, involving a combination of electronic processes (e.g. e-mail alerts or database searches for new potentially eligible patients) and manual processes (e.g. checking admission books) by the research nurses and/or the local principal investigator.

FIGURE 5 Flow chart of patient recruitment in hospital A. (Figure not reproduced here; see the original article.)

Recruitment

A total of 1670 participants were recruited to the study between May 2013 and September 2014. Cumulative recruitment by hospital is shown in Figure 6. Recruitment over time increased steadily across all hospitals. Hospital A recruited the largest number of patients.

FIGURE 6 Recruitment over time, by hospital. (Figure not reproduced here; see the original article.)

Because we did not have screening data for three of the hospitals, we estimated the approximate number of eligible patients over the recruiting period from the data between January and December 2014 provided directly by the BCIS. This registry provides comparative data on the provision of PCI in the UK and describes the quality and patterns of care, the process of care and outcomes for patients from all PCI centres. We extrapolated the number of PPCI procedures carried out during the recruitment period at each of the hospitals from the annual data provided for each hospital. The estimated rate of recruitment at each hospital is shown in Table 10. The average recruitment rate across the four hospitals was 52%, but there was large variation between hospitals (22–86%). These recruitment figures do not take into account (1) patients who did not meet the study eligibility criteria or (2) patients who activated the PPCI pathway but had emergency angiography only (i.e. patients with unobstructed arteries).

Data linkage and extraction across hospitals

Figure 7 shows the flow chart of data availability. Anonymised data were available for 705 out of the 792 (89%) patients who were identified as eligible but who did not consent or were not approached. Clinical and biochemistry data were most readily available (for > 80% of participants), whereas data on medications were least readily available (for only 44% of participants).

FIGURE 7 Flow chart of data availability. a, The number of patients with their index procedure identified in hospital episode data (exact match on day or 1 day out) or inpatient or outpatient or A&E data available in the year following the index procedure. Note that 1612 out of 1670 (97%) patients had their index procedure identified in hospital episode data (exact match on day or 1 day out). Reproduced from Brierley et al.27 © Article author(s) (or their employer(s) unless otherwise stated in the text of the article) 2018. All rights reserved. No commercial use is permitted unless otherwise expressly granted. This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

Figure 8 shows the flow chart of data submitted from each hospital. One of the hospitals (hospital B) provided the data in a linked format; linkage was undertaken centrally for the other three hospitals with data linked on a local hospital identifier. Anonymised data were available for > 80% of patients from three hospitals, but for only 50% of patients from one hospital. Clinical data were available for 93% of patients at all hospitals (range 82–98%). The availability of ECHO data was variable; one hospital was unable to submit any ECHO data, one hospital submitted ECHO data for only 34% of patients and two submitted data for > 75% of patients. ECHO data were submitted as free-text reports, rather than coded data. Biochemistry data were available for ≥ 98% of patients from three hospitals; one hospital was unable to submit biochemistry data for any of its patients. Data about medications on discharge were available for 97% of patients from one hospital; the other three hospitals were unable to submit these data.27 Although we requested biochemistry test results relevant to the cardiovascular disease population [creatinine, full blood count (serial), electrolytes (serial), troponin (admission; 8 hour), cholesterol (on admission), blood glucose (serial) and C-reactive protein (serial)], the three hospitals that provided biochemistry data sent all tests conducted during the specified time frame. However, it was feasible to extract specific test results from the biochemistry data sets and identify specific medications from the medications data sets.

FIGURE 8 Flow chart of data availability, by hospital. Shading indicates that no data were available. Reproduced from Brierley et al.27 © Article author(s) (or their employer(s) unless otherwise stated in the text of the article) 2018. All rights reserved. No commercial use is permitted unless otherwise expressly granted. This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

There were a number of issues with data that were provided by hospitals, such as the inclusion of ‘test’ patients at one hospital, discrepancies between hospitals on the format and naming of the fields (e.g. BCIS fields were provided as: ‘yes/no’, ‘1/2’ or ‘1. yes/2. no’) and the delay in receiving extracts. Once identified, ‘test’ patients were removed at source and discrepancies between formatting and naming of fields were addressed by hardcoding.

Cardiovascular magnetic resonance exposure

Cardiovascular magnetic resonance reports were submitted for all consented patients who received CMR in hospitals A and B. Like the ECHO data, CMR data were in free-text format; the hospital that provided linked data (hospital B) extracted the CMR variables of interest from the free-text reports. Information about whether or not a CMR was ordered for clinical need or research purposes (i.e. patient participating in another research study) was not recorded routinely. From information provided by hospital A, the proportion ordered for clinical need was 67%.

Data linkage with Hospital Episode Statistics and Patient Episode Database Wales

Hospital Episode Statistics, ONS and PEDW data were provided from the first recruitment date across all hospitals (May 2013) up to 30 September 2014 (Swansea), 31 March 2015 (Bristol and Leeds) and 30 April 2015 (Cardiff). Data linkage with HES, PEDW and vital status was achieved for 1655 out of 1670 (99%) patients across all hospitals. The linkage for each HES or PEDW data set is shown in Table 11. No data were supplied for patients for whom the NHS number was not recorded.

A further review of the linked data demonstrated that it was possible to identify an admission that matched the index admission for 1554 (93%) patients; this increased to 1612 (97%) patients when the dates of admissions recorded in hospital episode data and the date of the index admission recorded in HISs were allowed to differ by 1 day. For some patients (n = 22, hospital D), it was not possible to identify the index admission in HES because the event date was not recorded in the HIS. This also meant that it was not possible to identify admissions, outpatient appointments or A&E visits in the year after the index admission for these patients. Data availability is shown in Table 12.

Identifying the eligible study cohort and cardiovascular magnetic resonance exposure in hospital episode data

We identified 98% (range 92–100% across hospitals) of patients who underwent PPCI and 85% (range 46–99% across hospitals) of patients who had an emergency angiogram but no PCI in hospital episode data.

We could not identify all CMR exposure in HES. Admissions including the date of the CMR were identified for 109 of the 189 patients with CMR recorded in HISs (57%) and a record of CMR was found in the hospital episode data for 49 of these 109 (45%) patients. Patients who had a CMR dated after discharge, based on the HIS, were assumed to have had the CMR as an outpatient (n = 81); only 7 (9%) of these patients had a matching outpatient date in the hospital episode data. None of these seven outpatient episodes included any code for CMR (procedure codes are not well recorded in the HES outpatients’ data set). Data availability is shown in Table 13 and the details of CMR identification in the hospital episode data are shown in Figure 9.

FIGURE 9 Details of CMR recording in HES/PEDW data. (Figure not reproduced here; see the original article.)

Summary of main findings

Identifying the eligible population

Patients who activated the PPCI pathway and received PCI were easy to identify from catheter laboratory databases, since consistent terminology was used to define these patients. A proportion of these patients recruited in the study were identified independently of these databases. Patients who activated the PPCI pathway but did not receive PPCI were not consistently defined in catheter laboratory databases; these patients required further manual checks by catheter laboratory staff who assisted the research nurses in identifying eligible patients. Some of these patients were screened out, possibly because of (1) misunderstanding by catheter laboratory staff or research nurses over eligibility (especially at the beginning of the study) and (2) the lack of consistent terminology used to describe these patients in catheter laboratory databases, leading to confusion about whether or not to include them.

Implementing consent

We asked all participating hospitals to implement a conventional system of written consent so that we could establish a consent rate and ensure that we had consent to obtain follow-up data from HES and PEDW. We met the feasibility objective of implementing conventional prospective consent at all hospitals. Consent rates ranged from 59% to 74%27 with respect to numbers considered eligible to participate, and from 22% to 86% with respect to approximate BCIS volume in the period of recruitment; these percentages were lower for some hospitals than we had anticipated for a study in which participants have no active involvement. The main reason for the lower than expected consent rate, and the variation between hospitals, was that many patients identified as eligible were missed; PPCI patients are asleep immediately following the procedure, rarely confined to bed after the first 4–6 hours and many are discharged within 24 hours or transferred to referring hospitals. Some of the missed patients activated the PPCI pathway at the weekend, when research nurses were not always available. In some centres, patients did not get admitted to the critical care ward or general ward after the PPCI procedure. We implemented postal consent to invite identified eligible patients who were not approached in hospital, but return rates were low. Some patients were unable to cope with the emergency nature of their condition and the large amount of new information they were given as part of their care; for some, the request to participate in a research study seemed to be yet another thing to think about and it was easier for them to decline to take part.

Data linkage and extraction across hospitals

We did not meet the feasibility criterion for data linkage and extraction from HISs for all of the requested data sets. We prespecified that, for each participating hospital, the data items requested should be available (i.e. extracted from sources and linked) for > 90% of patients. Only one hospital provided all of the requested data sets (anonymised, clinical, ECHO, biochemistry and medication on discharge) for > 80% of eligible (for the anonymised data set) and consented patients. Some hospitals had entire data sets missing: medications on discharge were not available from hospitals B, C and D; information about ECHO tests was not available from hospital B; and biochemistry data were not available from hospital D. There were several reasons for missing data, for example:

• Some hospitals were unable to extract data from local stand-alone databases that were designed for a specific purpose, for example logging biochemistry requests or storing and presenting test results.27 Such local databases were typically old and did not have an interface for data extraction; data would have had to be retrieved manually for each patient in turn, rather than submitting a ‘batch’ query. The study was not resourced for this activity. Moreover, obtaining data in this way was considered infeasible for a national registry (as well as time consuming) and hospitals were not requested to do so.
• Some data (e.g. medications on discharge, bedside ECHO) were not recorded electronically and were available only in paper form (sometimes scanned but not coded/searchable).
• Non-bedside imaging tests (ECHO and CMR) were available electronically but as free-text reports rather than coded data; variables of interest were impossible to extract from these reports by conventional text-searching methods because of the variability with which the information was recorded.27

Cardiovascular magnetic resonance exposure

We were able to obtain CMR reports for all consented patients, although imaging reports were largely in free-text format, so individual CMR parameters would have had to be extracted manually. We concluded that this would not be feasible for an extended registry.

Follow-up/outcome data

It would be feasible to obtain 12 months’ follow-up data after the index admission through HES in an extended registry. Admissions (inpatient data set), outpatient appointments and A&E data could be identified for all patients for whom a NHS number was available and who had a date recorded for their index admission. Because we obtained hospital episode data only for patients who we believed to be eligible, we cannot comment on the risk that eligible patients might not have been identified.

Compiling the registry from Hospital Episode Statistics

It is not currently feasible to compile the registry from HES data rather than multiple HISs. Although we could identify the eligible population and obtain follow-up data, we could not identify the CMR exposure within 10 weeks of the index admission. This was mainly because the CMR is often performed at an outpatient clinic and the HES outpatient data quality, particularly for clinical information, is poor (i.e. lack of data on diagnoses or procedures).