Centralised datasets

Case history

Dr Trudi is a consultant anaesthetist. She has been working in the Trust (local healthcare service responsible for the delivery of a wide range of clinical services) for 20 years and has a regular colorectal surgical list, with a major caseload including colectomies and rectal resections. Trudi also has regular major gynaecology and urology surgical lists. Trudi suspects that the length of stay of her major colorectal surgical patients appears higher than she would expect based on discussions with colleagues. Unfortunately, she has found it difficult to find any data to confirm her suspicions and has no way to compare these outcomes nationally. Lacking the bigger picture of the patient perioperative pathway, she pushes her concerns to the back of her mind and concentrates on delivering good anaesthetic care to her patients.

Question to be answered

Would access to higher-quality perioperative care data empower Trudi to drive local Quality Improvement (QI) in a more targeted and bespoke manner?

Discussion

Arguably, the most fundamental principle in QI methodology is that any intervention be preceded by accurate baseline data. Baseline data are often available and provide an indication of current performance and important substance to any subsequent statistical or trend analysis.

More often, problems begin to arise when there is no sustainable mechanism to maintain data collection and, in this circumstance, many QI processes can grind to a halt. The uncomfortable reality is that, too often, baseline data collection is conducted manually by transient clinicians, which prevents meaningful continuity for ongoing data collection. Progress has been made in the recent shift in emphasis from audit to QI, reflecting the importance of ongoing data collection to ensure that improved performance is maintained. However, Trust-level mechanisms (such as information technology systems) are frequently inadequate to enable QI continuity.

This is not to say that data are not being collected; they are, but it is important to look deeper into these data, specifically:

1. What data are we collecting?

2. Are the data relevant to clinicians?

3. Are the data only collected locally or do the data populate central databases?

Although perhaps not immediately obvious, this last question is arguably the most important.

Centralised datasets come in distinct forms. In the UK, the most abundant dataset, with the widest uptake, is the Hospital Episode Statistics (HES) dataset and its sub-datasets. It is housed within National Health Service (NHS) Digital (which was incorporated into NHS England in 2023) and populated by individual Trusts via their clinical coding departments.

Centralised datasets

When we talk about centralised datasets in the context of the NHS, we are essentially talking about HES. Although HES is arguably the most relevant to clinicians, other datasets exist, including national clinical audits, as well as national data on costings, pharmaceuticals and patient safety.

HES is a database containing details of all admissions, Emergency Department attendances and outpatient appointments at NHS hospitals in England. HES data are provided by individual Trusts via their coding departments at the point of patient attendance or discharge. HES is an extract of data from the Secondary Uses Service (SUS). Trusts send data from their systems to SUS in NHS Digital. NHS Digital takes extracts from SUS, runs some cleaning routines and produces the HES datasets. That this resource is largely unknown to clinicians, let alone accessed by them, indicates how minimally this extensive resource is currently used to inform front-line QI initiatives. What would be useful is a mechanism that can provide Trusts with a considered, succinct and relevant summary of the key perioperative metrics that might be used to drive QI at a Trust level. This has existed in many guises over the years and will, no doubt, continue to do so in the future.

The current incarnation and market dominator is the Getting It Right First Time (GIRFT) programme (Fig. 1). GIRFT sits within NHS England and is split into specialty-specific work streams including Anaesthetics and Perioperative Medicine. Although this article is not specifically an advocate of, nor an apologist for, GIRFT, we use this initiative to illustrate the concept of how HES data can be used to create centrally compiled data packs for Trusts to use at a local level. The stages of the GIRFT process include:

1. Clinical lead, GIRFT project manager and analysts sit down to design the data pack to probe current specific and relevant questions on the specialty.

2. Pilot data packs are produced and trialled in discussions with pilot Trusts.

3. Composition of data pack is confirmed and reports compiled for every Trust in England.

4. Clinical leads visit every Trust in England (Deep Dives) to present their bespoke data pack, benchmarked to all other Trusts, to identify ‘outlying’ (positive or negative) practice or outcomes.

5. Bespoke action plans are created.

6. GIRFT local implementation teams engage with Trusts to help implement the action plans.

Figure 1

Figure 1

Principles of the Getting It Right First Time (GIRFT) programme. Interpretation of data, including a pragmatic assessment of its accuracy and relevance in a clinical context, is paramount to driving effective local-level QI.

Source: Copyright © 2013, 2014, 2015 re-used with the permission of The Health & Social Care Information Centre. All rights reserved.

How might Trudi use a resource such as GIRFT?

Trudi has heard about the GIRFT data pack and approaches her Clinical Director, Steven, to see what the Trust Anaesthetics and Perioperative Medicine data pack looks like. What she sees interests her greatly (Fig. 2). It suggests that emergency readmissions across all surgical specialties are within normal limits, but that the complication rates after colectomy and rectal resection appear to be a negative outlier. Steven confirms that this was noticed at the recent GIRFT meeting, and an investigation into it forms one of the actions of the Anaesthetics and Perioperative Medicine action plan. Trudi offers to contribute to QI initiatives that follow on from this action plan. Importantly, she feels reassured that her initial gut instinct was correct and she can now use her positive energy to help improve local practice and, most important of all, patient outcomes.

Figure 2

Figure 2

Example of a GIRFT box and whiskers plot. Each dot represents an English NHS Trust, with the black diamond relating to the Trust in question. Here we can see that Trudi’s Trust appears to be an outlier for complications after colectomy and rectal resection (right). Interestingly, this seems to correlate with the higher than UK average 30-day readmission rates (left).

Source: Copyright © 2013, 2014, 2015 re-used with the permission of the Health & Social Care Information Centre. All rights reserved.

Interestingly, there also appears to be an increased rate of readmission of surgical patients to critical care postoperatively (Fig. 3), although Steven is less sure this is genuine as he suspects the Trust coding of surgical intensive care unit admissions is suboptimal.

Figure 3

Figure 3

Example of a GIRFT funnel plot. Each dot represents an English NHS Trust. Red indicates >3SDs above the mean, orange >2SDs above the mean, dark green >3SDs below the mean, and light green >2SDs below the mean. The larger diamond relates to the individual Trust in question, which appears to have a high readmission rate to intensive care that may warrant investigation.

Source: Copyright © 2013, 2014, 2015 re-used with the permission of The Health & Social Care Information Centre. All rights reserved.

Is this just another stick for central regulators to hit us with?

Genuinely, this is not the intention of initiatives such as GIRFT. Primarily, their function is to provide benchmarked data to drive locally improved outcomes. Indeed, whereas the example here is a ‘negative’ outlier, just as many ‘positive’ outliers are identified in data packs. For example, a length of stay in diabetic patients that is far below that of most other Trusts will prompt a discussion into perioperative diabetic pathways. The idea would be to put a Trust with increased length of stay in diabetic patients in touch with the well-performing Trust to promote peer-to-peer sharing of good practice, in an attempt to break down the NHS silos of both past and present.

Why should we believe the data?

Any GIRFT, or HES data in general, should be viewed with a (healthy) degree of cynicism. When presented with data portraying an uncomfortable picture of a Trust’s performance, it is natural to want to critique the data and defend their position. This may be justified. Typically, criticisms of the data fall into three broad categories:

• 1.  The data are wrong.

The key point here is that no one is pretending (or at least they should not be) that the data are perfect; it is about raising the question.

GIRFT packs deliberately do not claim statistical significance; they highlight >2 or >3 standard deviation variance, quartiles or deciles relative to peers. This is then used as a platform to ask ‘Does this feel right to you?’ If it can be easily explained away by poor data or old data that have since been acted on, then so be it, and the focus of conversation can move on. If, however, the data strike a chord, it may be the start of an internal process to address the potential underlying issues.

• 2.  The data are inaccurate.

This is a very important issue. These Trust data come from HES, which is populated from SUS. As Trusts populate HES data, a logical, if not brutal, conclusion must be drawn: ‘if your Trust data do not represent what you see in the Trust, it could be because the data that you have provided are of poor quality’.

Let us be honest, how many of us know where our clinical coding department is, let alone interact with them? Coding is, first and foremost, about accurately capturing the activity performed in a hospital and for the patient cohort they are treating. We code for information; finance is merely a by-product.

Does the health system in which we operate enable the coders to do their job well? Do we recognise the importance of their job? This piece is too short to go into more detail about how we, as clinicians, may engage with clinical coding and the range of benefits it can convey. But, suffice to say, the way to improve the quality of centralised datasets is to improve the coding of the information at the Trust level, at the point of patient contact. If this can be fine-tuned, we will have the world’s best, continuously populated, healthcare dataset to drive ongoing local QI. Unfortunately, we, like the rest of the world, are some distance away from this utopian reality.

• 3.  The data might be right, but we are different.

(For example, our patients are more complex or come from an area of higher deprivation.)

The GIRFT data packs attempt to address this by putting in demographic benchmarks such as age, socio-economic status and deprivation scores for a Trust’s catchment population. This should prove or disprove that the Trust truly is different, and so the subsequent data can be interpreted within the relevant context (Figure 4).

Figure 4

Figure 4

GIRFT bar chart comparing the age and deprivation patient mix in a Trust (inner black bar), compared with the England average (outer box). Here, the Trust has a slightly older patient demographic and higher levels of deprivation.

Source: Copyright © 2013, 2014, 2015 re-used with permission of the Health & Social Care Information Centre. All rights reserved.

What can you do?

Using resources available to you, obtain baseline data:

• Internal Trust audits

• HES-based datasets (seek out your coding department and see the GIRFT data pack)

• National audit projects (abundant with a varying degree of data sharing arrangements): e.g. NELA (Royal College of Anaesthetists), NHFD (Royal College of Physicians), ICNARC (Intensive Care National Audit and Research Centre).

Follow your usual Trust-level QI process.

If you see something that does not look right, question it and engage with your coding department, you might be surprised by what you learn.

Disclaimer

This Narrative article is adapted with permission from Case Studies in Perioperative Medicine, a UCL Press open access educational resource, available from https://doi.org/10.14324/111.444.9781787356917.04

Declarations and conflicts of interest