2. Triage
2.1 Triage purpose
This is an exercise which linkers will do alongside the data linkage enhancement team at the start of an engagement with the quality assurance framework for data linkage. There are two objectives from this work:
- Firstly, it will start a conversation to identify potential linkage issues
- Secondly, this log forms an important part of the documentation alongside the framework
It is worth noting that the questions have been formulated for NHS England use cases, and as such may not be applicable for all organisations.
If you would like to keep track of your triage responses, download our triage log here.
2.2 Multiple choice triage questions
The first section of questions are multiple choice to establish some of the project’s core tenets, and you'd be expected to choose one of the options listed in the bullet points.
| Dimension | Question | Options | Why is this question important? |
|---|---|---|---|
| Engagement start | What stage of your linkage project have you started engagement with the QAFDL? | Start of the project | This affects the order with which elements of the QAFDL should be considered |
| Some project work has been completed, but work is ongoing | |||
| Linkage has been completed, engaging with QAFDL for audit purposes | |||
| Linkage project type | What project type does this engagement fall under? | Type 1 | Categorising projects in this way can help with transfer of lessons learnt, and signposting the most relevant documentation |
| Type 2 | |||
| Type 3 | |||
| Type 4 | |||
| Types of linkage algorithm | Which type of linkage algorithm are you using? If you are not sure, carry out the Preparation phase on the QAF log to discover the best option for your project. Then, come back to this question and select the most appropriate answer. Are you constrained in using a pre-determined linkage algorithm (e.g. MPS)? | Simple join | The data linker might need to use a specific, previously established methods for linking data, which affects how we check the quality of the linkage. |
| Established linkage algorithms like MPS | |||
| Established linkage algorithms + additional rules OR Ad-hoc models | |||
| Commissioned to third-parties | |||
| Purpose | For what purpose are the data sets being linked? | Direct care | The purpose influences the quality assurance needed, for example, linked data used for direct patient care requires the highest quality. Depending on whether the linkage is defining a cohort or defining attributes you will want to have differing levels of sensitivity and recall. |
| Non-direct care | |||
| Risk | What is the risk associated with data linkage inaccuracies? (missing a link making the wrong one) | High risk | Understanding the risks involved allows us adjust our data linkage approach to ensure accuracy and reliability for its intended use. |
| Moderate risk | |||
| Low-risk | |||
| Profile of data content | What is the importance profile of the data ? | High profile health data | The project's significance is determined by how errors could impact patient safety and may be influenced by other factors like business or politics. |
| Normal profile | |||
| Sensitivity of data | How sensitive is the data we are using? | Sensitive data | We adjust our quality checks based on how sensitive the data is to protect it properly during linkage. |
| Non-sensitive data | |||
| Frequency | How often do you plan on linking these datasets? | One-off | The linkage frequency impacts how we approach the efficiency and repeatability of the process. |
| Repeated (yearly, monthly) | |||
| Repeated (weekly, daily) | |||
| Live | |||
| Expertise | How experienced is the person linking the data? | Intermediate+ | The skill level of the person linking the data affects quality; our framework guides less experienced staff through the necessary steps. |
| Novice |
2.3 Extended triage questions
This second section asks more extended questions to ascertain a more in-depth view of the project’s risk profile.
| Additional questions | Why is this question important? | Related QAFDL aspect | |
|---|---|---|---|
| Entity being linked | What entity is being linked (person, event, organisation,etc)? | Knowing what we're linking—people, events, or organisations—guides our choice of linking variables and the reference master dataset used. | Configuration of linkage parameters/settings |
| Common characteristics used | What common linkage fields do you have available in the two assets for linkage? (NHS numbers, MPS_ID, pseudo (NHS number or MPS_ID), DOB, etc | To link data successfully, we need to identify fields that are sufficient to identify a match. | Profiling |
| Metadata | Is metadata and documentation available for all datasets? | Having common linking fields might not be enough for a successful linkage. Understanding them is important to define how they need to be adjusted or derived. | Profiling |
| Benchmarking | What level of overlapping are you expecting from the data sets being linked? | Linked datasets may represent different groups, so knowing how much overlap to expect helps us to predict acceptable linkage rates. | Verification and validity |
| Involvement of reference master data (a "spine") | Do you need to link data sets to a reference master data set (a "spine") or between them or both? | Linking to a main reference dataset, or "spine," is best practice. It helps to connect different datasets and track changes over time, like when a patient moves house. | Techniques and tools |
| Synchronicity of data collection | Do the data sets being linked refer to the same time frame? And are the PIDs being collected at the same time than the analytical measures of interest? | If datasets span different times, we must adjust for these differences and consider how they might affect the linking process, potentially evaluating how errors in linkage change over time. | Profiling, assessment |
| Source of data | What categories do the data sets being linked fall into? How was the data collected? Is it clinically validated, or patient inputted? (e.g. Cohort from external researchers, Core assets, DARS assets, Scraped data , Collected by NHSE, Sent by OGD) | This context can provide lots of clues as to how best to link, what kind of quality we are expecting with respect to outputs, and what kind of quality assurance is needed post linkage. The source of data used in the linkage process can support decisions around other dimensions. For example, core assets are usually linked to reference master data sets via pipelines set up and maintained by the data engineering team. | Data access |
| Data types | What data types are involved in the linkage (numerical, free text, binary, strings, etc)? | The nature of the data we link (numbers, text, yes/no answers) affects how we prepare it for linking and which tools we choose. Data types need to be consistent across linkage variables between datasets. | Profiling |
| Data quality | What is the expected data quality of the data sets being linked? (high, low) | Data quality drives our choice of linking methods—better quality data allows for simpler linking techniques. | Assessment |
| Number of data sets | How many data sets will be linked? | When linking several datasets, we must consider how they'll group together and how to resolve any mismatches, shaping our evaluation process. | Configuration of linkage parameters/settings |
| Environments where the linkage is executed | Are you constrained in which environment will you be executing the linkage (e.g. DAE, DPS, UDAL, NDSR, RDS, FDP, etc)? | The setting where we link data can dictate what data fields, tools, and methods we can use, possibly restricting our options of linkage algorithm. | Computational resources, Safety, Configuration of linkage parameters/settings, Techniques and tools |
| Environments where the assets reside | Where do the assets reside (DAE, UDAL, RDS, etc)? | Knowing the location of our data helps anticipate and plan for any hurdles, like secure transfers and privacy protection needs. | Safety |
| Customer request | Is the customer asking for the highest level of quality assurance regardless of the triage outcomes? | A project might be commissioned by a customer that requires the highest level of quality assurance regardless of the project characteristics. The opposite may also occur, however, it is always recommended that the minimum suggested level of quality assurance is delivered to guarantee a professional outcome. |
Knowledge management, Quality of linkage |
| Resources | Which resources (and what type) do you have available to carry out the linkage? | Our desired quality level might be limited by available resources, and we need to be clear when we can't meet our ideal standards. | Uncertainty management |
Last update:
August 30, 2024
Created: August 30, 2024
Created: August 30, 2024