Remember me
In recent years, UK health care policy and guidance [-] and international research have highlighted the potential for artificial intelligence (AI)—advanced technology that can perform complex tasks associated with human intelligence [-]—to support and transform health care in areas such as radiology. Research has indicated potential benefits regarding a range of outcomes (eg, detection accuracy, error reduction or prevention, efficiency, decision-making, and reducing workforce burden) [,,]. However, there is mixed evidence for some outcomes (eg, diagnostic accuracy) [].
UK health policy increasingly recognizes the potential value of AI in supporting the efficiency and effectiveness of services provided by the National Health Service (NHS, the publicly funded health care system). In June 2023, NHS England (NHSE, a public body that runs and oversees the running of the NHS in England) announced the Artificial Intelligence Diagnostic Fund (AIDF), which has invested £21 million (US $27.93 million) to accelerate the deployment and implementation of AI diagnostic tools []. The fund focuses on chest x-ray (CXR) and chest computed tomography (CT) scans to improve the diagnosis of lung cancer and other conditions [] and potentially help to address the current unmet need for faster CXR reporting []. In the longer term, the NHS proposes that using AI to assist with the early detection of lung cancer may improve patient care and outcomes []. However, guidance from the National Institute for Health and Care Excellence (a public body that assesses cost-effectiveness of health care interventions and provides guidance and recommendations on use of these interventions) has identified numerous evidence gaps that must be addressed in relation to AI for chest diagnostic imaging, including gaps on time saving and resource use, adverse effects, performance in different patient groups, ease of use, and impact [].
In October 2023, the National Institute for Health and Care Research (NIHR) Rapid Service Evaluation Team (RSET) was commissioned to conduct a rapid evaluation of the implementation and outcomes of AIDF. Phase 1 aimed to establish existing evidence for implementing AI in radiology diagnostics, analyze early implementation of AIDF, and identify approaches for future evaluations of changes of this kind []. It comprised a systematic scoping review and an empirical study of network-led procurement of AI tools for chest diagnostic imaging, early deployment of AI tools within NHS hospital organizations; empirical work drew on qualitative, quantitative, and health economic perspectives []. The systematic scoping review identified potential benefits of AI in chest diagnostic imaging and radiology more broadly [-]. It also identified important gaps in knowledge, especially around real-world implementation, including how AI tools are procured, processes to support preparation for deployment, how staff, patients, and caregivers experience and perceive AI for diagnostics, and impact on service effectiveness and resources; this was partly because there are very few studies of real-world implementation of AI tools for radiology diagnostics [-]. The empirical study focused on procurement and early deployment of AI tools for chest diagnostics through AIDF [,,]. It established that these processes took longer than anticipated by national leadership. Findings suggested that procurement and preparation for deployment were complex sociotechnical processes, placing significant demands on already busy staff. Key obstacles included variations in local imaging technology, extensive governance processes, and variable local data infrastructure, while important facilitators included advice and support from program leadership, networks sharing expertise and capacity, commitment from hospital staff and AI suppliers, and dedicated project management [,,]. Phase 1 also delivered insights on implications for future evaluations, identifying several data sources that may support measurement of effectiveness and cost-effectiveness within the English NHS, but also noted that limitations to data availability need to be addressed [,].
Importantly, phase 1 was unable to evaluate the impact of AI on clinical practice, patient care and outcomes, and cost-effectiveness, and the perspectives of patients, caregivers, and those in the wider care pathway [,,].
The implementation of AI for chest diagnostic imaging is currently operating across 12 imaging networks of NHS trusts (organizations that provide a range of health care to local populations). Deploying AI tools in this way has the potential to change organization and delivery of care both within and across these organizations, with the aim of improving service delivery and outcomes at regional levels. Therefore, these programs may be conceptualized as examples of “major system change,” defined as “a coordinated, systemwide change affecting multiple organizations and care providers, with the goal of significant improvements in the efficiency of health care delivery, the quality of patient care, and population-level patient outcomes” ([], p422). The framework of Major System Change of Fulop et al [] outlines that it is necessary to evaluate all stages of implementation, including the decision to change, the decision on which model to implement, the implementation approach used, the implementation outcomes, and the intervention outcomes. This framework has been used to evaluate several national system changes within the English health care system (eg, reconfiguration of stroke services [,], specialist cancer services [], COVID-19 remote home monitoring [,], and prenatal exome sequencing []).
While research has shown that AI diagnostic tools have the potential to support and improve the detection of lung cancer, little is known about how effective and cost-effective these tools are, or how staff, patients, and caregivers experience them. For recommendations to be made regarding the implementation of AI diagnostic tools, these knowledge gaps need to be addressed [].
Aims and ObjectivesThis mixed methods evaluation of AI tools for chest diagnostic imaging aims to address previous research gaps by exploring the implementation of AI tools for chest diagnostic imaging, the impact and costs of implementing these models, and the experiences of patients, caregivers, and staff.
Research Questionsprovides details of our research questions (RQs) and which workstreams will address them.
Table 1. Workstreams and research questions.WSa and research questionSubquestions1aWS: workstream.
bRQ: research question.
cAI: artificial intelligence.
This is a multisite rapid study that combines qualitative, quantitative, and health economic methods. This evaluation (phase 2) was informed by the findings from the phase 1 evaluation of AI tools for chest diagnostic imaging) [-], scoping conversations (eg, with academics, clinicians, policy representatives, professional bodies, and third-sector organizations and regulators) and previous research, and was developed in collaboration with the study Patient and Public Involvement and Engagement (PPIE) group. The evaluation will take place over 12 months (January 2025 to December 2025).
The implementation of AI tools for chest diagnostic imaging may be seen as an example of Major System Change. This study will thus be informed by the Major System Change Framework [], which was designed to understand the processes, outcomes, and sustainability of such changes, in addition to the relationships between different stages of major system change () [].
We will employ a 2-level case study design, including 3 in-depth case study trusts and up to 9 light-touch case study trusts, with the aim to ensure our evaluation can contribute both depth and breadth in its lessons across the workstreams ().
Figure 1. How workstreams will contribute to addressing components of the major system change framework []. AIDF: Artificial Intelligence Diagnostic Fund; WS: workstream. Table 2. Summary of activity in in-depth and light-touch case study services.aRIS: radiology information system.
bNot applicable.
The study has been developed and will be performed in compliance with NIHR RSET equality, diversity, and inclusion requirements, with interpretation of findings and final project write reviewed in consultation with the study PPIE group, advisory group, and wider team ().
WorkstreamsWorkstream 1: Implementation, Staff, Patient, and Caregiver Experience of AI Tools for Chest Diagnostic Imaging (RQs 1-3)This workstream will focus on the implementation of AI tools for chest diagnostic imaging in NHS services, staff experience with using AI tools for chest diagnostic imaging, patient and caregiver experiences of receiving care supported by AI tools for chest diagnostic imaging, and factors influencing implementation and experiences. The study will be of qualitative design, comprising semistructured interviews, meeting observations, and documentary analysis ().
Table 3. Summary of primary data collection methods within workstreams 1 and 4.ActivityStudy participantsApprox. timeWorkstream 1aAI: artificial intelligence.
bPACS: picture archiving and communication systems.
cRIS: radiology information system.
dMDT: multidisciplinary team.
eGPs: general practices.
fED: emergency department.
gCXR: chest x-ray.
hCT: computed tomography.
iNA: not applicable.
jICB: integrated care board.
kAIDF: Artificial Intelligence Diagnostic Fund.
SampleSite SelectionThe study sample will comprise 3 of the 66 trusts as in-depth case studies, with up to 9 trusts as light-touch case studies. Site selection will be informed by our learning from the phase 1 evaluation [-] and an expression of interest process, whereby all trusts implementing AI for chest diagnostic imaging (through the national meetings) will be contacted for expressions of interest for study participation. Eligibility criteria for site selection are presented in and for wider data collection in and . Sites meeting implementation and data quality criteria have been identified through local network engagement and implementation progress tracking, with study inclusion guided in collaboration with imaging network and service leads.
Textbox 1. Eligibility criteria for site selection.All sites will have implemented artificial intelligence (AI) in their chest diagnostic imaging serviceIn-depth trusts will have sufficient prospects of good data quality (eg, local picture archiving and communication system [PACS] or radiology information system [RIS] data reports) to facilitate evaluation across workstreams 2 and 3Light-touch trusts will not be located in the same networks as the in-depth case studies.The study will seek to ensure representation across a range of service characteristics and contexts, includingThe purpose of the AI tool (prioritization, identification of lung cancer vs identification of other chest conditions)Scan type (chest x-ray vs chest computed tomography)AI tool supplierGeographical location (eg, urban, rural, and coastal)Other relevant characteristics that will support decision-making (eg, referral pathways, leadership approach [imaging network vs trust], and local PACS and RIS set-up [local arrangement vs regional platform])Textbox 2. Eligibility criteria for data collection.Staff interview participants
The National Health Service staff who work in or with the participating trusts, and who are involved in organization or delivery of care to patients receiving chest diagnostic imaging which have been supported by the artificial intelligence (AI) tools for chest diagnostic imaging; also, Artificial Intelligence Diagnostic Fund (AIDF) program staffOver the age of 18 yearsEnglish-speaking or able to participate in an interview with an interpreterAble to provide informed consentPatients and caregivers
Patients and their caregivers (including family members) who have had a chest x-ray or computed tomography (CT) scan that has been supported using AI for chest diagnostic imaging, at one of the 3 trusts included in this studyOver the age of 18 yearsEnglish-speaking or able to participate in an interview with an interpreterAble to provide informed consentWorkshop participants
National stakeholders with relevant job roles (eg, policymakers, commissioners, system leaders, and third-sector organizations) relating to the implementation of AI, or local staff involved in implementing AI from the eleven networks and 60 trusts implementing AI for chest diagnostic imaging as part of the AIDFOver the age of 18 yearsEnglish-speaking or able to participate in an interview with an interpreterAble to provide informed consentDocumentary analysis
Any documents pertaining to the implementation of AI for chest diagnostic imaging at the participating 3 trustsMeeting observations
Any meetings relevant to the implementation of AI chest diagnostic imaging at the participating 3 trustsTextbox 3. Exclusion criteria.Anyone under the age of 18 yearsAnyone who cannot provide informed consentPatients and caregivers for which the artificial intelligence tool was not involved in supporting their carePatients and caregivers at sites not included in this studyRecruitment and Consent: Initial IdentificationCase Study TrustsTo recruit participating trusts, we will present the study plans at existing AIDF network meetings and invite trusts to express interest in taking part. Sampling will be informed by findings from our phase 1 evaluation []. In addition, and to ensure a diverse sample, sites will be asked to provide some basic information to enable sites to be purposively sampled (eg, data availability, the purpose of the AI tool, type of scan, supplier, geographical location, referral pathway, leadership approach, and local picture archiving and communication system [PACS] or radiology information system [RIS] setup).
Staff InterviewsThe researchers will work with leads at each site to identify potential staff groups at their trust that may be appropriate for an interview. Researchers will contact potential participants via email to invite them to participate. Staff may also cascade details of the study (and an invite for anyone to contact the researchers if interested) to their staff networks to support recruitment.
Patient and Caregiver InterviewsThe researchers will work with staff leads or R&D contacts at each trust. The staff leads (or research nurses, if available) will contact potential patients and caregivers who meet the eligibility criteria (by telephone, email, or post) to share a study advert and see if they would be interested in participating in the study. Potential participants will be asked to contact the research team directly if they are interested in participating; alternatively, potential interviewees may ask the staff lead or R&D contact to securely pass on their details to the researcher (using the secure UCL Data Safe Haven) if preferred. The researcher will then contact the patient or caregiver to provide further information.
In the first phase of our evaluation, the team learned that services are taking varied approaches to informing patients about the use of AI in the diagnostic process, with some sites choosing not to inform people explicitly. Therefore, the invitation to be interviewed may be the first time patients are made aware that AI supported their diagnostic process: this may cause patients concern or a desire for more information. To accommodate this eventuality, patients will be made aware of the purpose of this study and signposted to national and local sources of information at each stage of the identification and recruitment process, for example, in invitation and recruitment documentation.
The team recognizes that hospital services are extremely busy. Therefore, when recruiting in-depth sites, we will ensure that the proposed approach to identification, invitation, and recruitment is feasible in these sites; further, we will work with local research nurses in sites where they are available to support our work.
Note: For the purposes of patient and caregiver interviews, sites will be classified as Patient Identification Centres.
Meeting ObservationsThe researchers will liaise with staff leads at each trust to identify appropriate meetings to observe. For each meeting type, we will liaise with the lead of the event (eg, meeting chair or lead trainer) regarding whether observation will be possible and appropriate.
WorkshopsTo recruit workshop participants, we will circulate study adverts via existing AIDF channels and networks, professional groups, social media, local third-sector organizations, and direct invitations.
AIDF StaffThe researchers will work with the AIDF program leads to identify staff that may be appropriate for an interview. Researchers will contact potential participants via email to invite them to participate.
Details of informed consent processes are presented under the Ethical Considerations section.
Data CollectionInterviewsInterviews will last 30-60 minutes, and will be semistructured and audio-recorded (subject to consent). Interview topic guides have been developed iteratively, informed by phase 1 findings [-], scoping conversations, the Major System Change Framework [], and previous research [-].
For in-depth case studies, the aim is to interview up to 32 staff members covering key clinical and organizational roles () and sampling patients and caregivers across a range of characteristics, including health outcome following review of scan (and therefore care pathway) and sociodemographic characteristics (eg, gender, age, ethnicity, and disability). We anticipate that patient characteristics may influence how interviewees experience diagnostic imaging services supported by AI (eg, many sociodemographic characteristics may influence AI performance, with underserved groups being disadvantaged). Sampling patients in this way, therefore, ensures a range of perspectives are captured.
The semistructured interviews will be guided by topic guides tailored to stakeholder groups (). Interviews will be transcribed verbatim by a professional transcription service and stored securely for analysis by the research team. In light-touch trusts, we will aim to interview up to 18 staff, with members of the NHSE AIDF program team also interviewed for their independent perspective ().
For all interviewees, sociodemographic information will be sought (on a voluntary basis), including job role and length of time in post (staff) or health outcome, comorbidities, age, gender, ethnicity, disability, sexuality, and employment status (patients and caregivers). All participants will also be informed that they are free to withdraw up to 2 weeks after the date of their interview.
Staff InterviewsStaff interviews will explore the interviewees’ role and professional background, their views on the reasons and drivers for implementing AI tools for chest diagnostic imaging, the aims, purpose, and function of AI tools, how AI tools are intended to be used and being used in trusts, how care is supported by AI in their trust, their experience of using AI for chest diagnostic imaging (including training, support, etc), perceived impacts and examples of perceived impacts, governance, data monitoring and evaluation, resource use, impacts on (in)equality, unintended consequences of using AI, barriers and facilitators to implementation and delivery, key learnings, and future use. Interviews will be scheduled to take place during regular working hours, as staff are not being compensated for their study participation.
Patient and Caregiver InterviewsPatient and caregiver interviews will explore the care they experienced, investigations and outcomes received to date (eg, scan received and process of receiving their report), information provision (eg, whether and in which ways they were informed about use of AI), the experience of the care they have received (things they liked and things they disliked), timeliness of care, and barriers and facilitators to their care experience. If previously unaware of the use of AI in their care, they will be provided with a short vignette that explains the use of AI in their local trust followed by questions on their knowledge and views on AI, how it can be used in health care, whether and how the use of AI had been communicated to them, whether they would like to find out more from their care providers (and if so, how), views on the benefits and challenges of AI, perceptions of the impact of AI, possible unintended consequences of its use, and how AI should be used in the future. At the end of the interview, all patients and caregivers will be provided with a secure link to add any further information they would like to share in written format. Participant information sheets and consent forms to be translated, with translation services available for the interview itself, if needed.
Meeting ObservationsActivities to be observed will include up to 30 meetings (up to 10 per in-depth trust) relevant to the use and governance of AI for chest diagnostic imaging, and health care affected by the implementation of AI in chest diagnostic imaging, AI implementation project meetings, AI training sessions, multidisciplinary team meetings, safety and quality committees (directorate- and trust-level), and regional oversight meetings.
Documentary AnalysisLocal documents pertaining to the implementation of AI for chest diagnostic imaging (eg, project plans, risk documents, meeting minutes, examples of anonymized AI reports, training materials, standard operating procedures, patient pathways, AI specifications, local audits, and evaluation plans) will be analyzed across all participating trusts.
Data AnalysisA medium Q thematic analysis approach [], combining inductive thematic analysis and the use of a coding framework [], will be used to analyze findings. Real-time notes will be used for postinterview completion of Rapid Assessment Procedure (RAP) sheets [], guided by RQs and the Major System Change Framework []. The categories used in the RAP sheet will be based on the interview topic guides. There will be flexibility to add categories during the research process.
Initial themes and subthemes will be developed using inductive thematic analysis [], the interim findings of which will be shared with key stakeholders throughout the study. The themes developed during the rapid analysis will be applied to interview transcripts and observation field notes to develop an in-depth coding framework for final themes and subthemes with cross-case comparisons made across the case study sites and staff characteristics (eg, to explore barriers or inequities relating to implementation, delivery, and patient experience), where possible.
Workstream 2: The Impact of AI Tools for Chest Diagnostic Imaging (RQ 4)This workstream focuses on evaluating the impact of AI tools for chest diagnostic imaging, performing mathematical modeling analysis on data from sample sites and relevant patient record datasets.
DesignQuantitative analysis of data derived from Hospital Episode Statistics (HES), Diagnostic Imaging Database (DID), and NHSE Benefit Register (where available) from all trusts that have implemented AI for chest diagnostic imaging through AIDF (n=63), including all networks involved in the AIDF program. More detailed analysis of imaging data from the 3 in-depth trusts. Given data availability limitations and the immaturity of postimplementation data, mathematical modeling of the chest diagnostic pathway will be informed by the sample data, supplemented with relevant evidence from published studies or the gray literature (these studies need not have investigated the use of AI).
Relevant data collected in workstream 1 will be analyzed to identify factors and implementation approaches likely to influence the impact of AI deployment.
SampleLocal data sampled from in-depth sites as described in workstream 1 unless where necessary (eg, due to data extraction or poor data quality issues). All 63 AIDF sites will be included for analyses of data sourced from HES, DID, and NHSE Benefit Metrics where it is available. Trusts that have not implemented AI for chest diagnostic imaging will be included as comparators.
MeasuresThe measures we will investigate will focus on (1) caseloads and workflow (eg, referral volumes, results of tests by category), (2) false positive and false negative results, (3) patient flow and waiting times (eg, time from CXR to CT scan, time to confirmed cancer diagnosis), (4) image processing and reporting (eg, turnaround times), (5) AI use and performance (eg, agreement with clinician, AI failure rates), (6) impact by patient characteristic, (7) influence on early detection (ie, the stage at which cancer is diagnosed), and (8) noncancer pathways outcomes.
False positives will include follow-up CT examinations with negative results (for CXR applications only) and subsequent cancer diagnoses that are negative. False negatives will include cancer cases that are missed by the diagnostic imaging supported by the AI. In all cases, the accuracy of the human reader with the AI tool in a decision support role is being measured.
Data CollectionWe will seek empirical data from the in-depth sites (aggregated or summary, RIS or PACS system data), together with data from currently accessible resources, for sample and comparator sites: DID, HES, and NHSE Benefits Registers.
A short feasibility study supported by expert advice will also be undertaken to assess the value of HES in its ability to support the analysis of how AI deployment may impact inequalities.
Further data on the performance of AI may be sought from the AI suppliers or trusts, depending on local arrangements, and the availability of linkage to cancer diagnostic data for sites will be explored. The proposed use of each source of data is summarized in .
The extent to which we can investigate these metrics will depend on what we are able to glean from these data sources, and their quality and completeness. Since many sites are in the early stages of deployment, this is not yet clear, and the influence of data quality and completeness is included within the workstream.
For example, clinical outcome measures can only be obtained from sites where data are linked between radiology systems and cancer registries, allowing us to chart patients’ diagnostic journeys to a definitive clinical outcome.
Understanding how outcomes differ for different types of patients, that is, understanding implications of deployment on inequalities, will require access to record-level data available from HES. The value of HES in supporting this analysis will need to be explored first to understand its capabilities and limitations, so, to this end, we will undertake a short feasibility study supported by expert advice.
Where we plan to use DID or HES for comparators, we will use longitudinal data both before and after deployments and apply statistical methods that will account for deployment at different times. When using patient-level data, trust factors can be included as random effects, and we could simultaneously explore any influence of patient characteristics. We cannot, however, account for simultaneous interventions that may be happening in comparator sites that aim to improve backlogs.
We will work alongside workstream 3 to develop a model of the lung cancer patient pathway supported by site data alongside available evidence from published sources on diagnostic accuracy, including resource constraints and efficiency. These will map the progress of patients from initial tests through to any confirmed cancer diagnosis with progressions dependent on the underlying cancer stage. The purpose of these models will be to link different levels of AI performance to outcomes such as volumes of follow-up tests, missed cancer diagnoses, and stage of cancer at first diagnosis. This would lead to more generalizable findings.
Once we scope data in HES and DID, we will know the degree to which we can investigate process outcomes (imaging processing, patient flow, etc). After speaking with individual sites regarding their data linkage, we will know whether we can look at clinical outcomes and match radiology data with the cancer registry. We will use existing evidence as well, but acquiring local data is important in centering this analysis around AIDF sites.
The nature of any potential bias that might affect our quantitative results have been, or will be, discussed with advisors, other stakeholders, and sites. We have attempted to request data from local sites and national datasets that include factors related to potential bias, where available. These will then be adjusted for in our analyses. Mindful that each trust organizes its local data differently, we are being thorough in our understanding of each field of data (in the form of continued correspondence with local data managers) to ensure that bias resulting from incorrect interpretation does not occur.
We will ensure that we define metrics that are less likely to be affected by missing data, that is, where a sample of cases will provide sufficient information. If there is missing data from local sites and it is important to our analysis, we will liaise with those sites to see if the issue can be rectified. In some cases, we can use proxy measures, for example, numbers of suspected cancers after x-ray as a proxy for numbers of follow-up CT scans.
Together with the work performed in workstream 3, these data and analyses will be used to develop a model of the lung cancer patient pathway from initial referral through to any confirmed cancer diagnosis. Evaluation of AI deployment will thus be enabled via generalizable study findings from the linkage of the impacts of AI deployment to diagnostic imaging and care pathway outcomes (eg, follow-up test volumes, missed cancer diagnoses, and stage of cancer at first diagnosis).
Table 4. Proposed data sources for workstreams 2 and 3.Data sourceProposed useAdvantagesDisadvantagesNHSEa Benefits RegistersMeasures relating to case volumes, the change in processing times, times to follow-up tests, resourcing, and cancer diagnosesaNHSE: National Health Service England.
bAIDF: Artificial Intelligence Diagnostic Fund.
cDID: Diagnostic Imaging Database.
dHES: Hospital Episode Statistics.
eGP: general practice.
fCT: computed tomography.
gAI: artificial intelligence.
hBy stage of diagnosis or associated with false negative or false positive diagnosis.
iRSET: Rapid Service Evaluation Team.
Workstream 3: The Cost and Cost-Effectiveness of AI Tools for Chest Diagnostic Imaging (RQ 5)This workstream focuses on evaluating aggregate costs and cost-effectiveness of AI tool deployment in the diagnostic chest imaging stage of the lung cancer care pathway, across a small sample of participating sites for chest diagnostic imaging.
DesignGiven the limitations of the resource capacity and timeframe of this project, a pragmatically designed economic model of the lung cancer care pathway will be developed, mapped to a simplified version of the National Optimal Lung Cancer Pathway () [].
Figure 2. Overview of the lung cancer diagnostic pathway. Adapted from National Optimal Lung Cancer Pathway, NHS England, 2020 []. A&E: accident and emergency; CT: computed tomography; CXR: chest x-ray; GP: general practice; LC: lung cancer; NICE: National Institute for Health and Care Excellence. *Includes follow-up CT or CXR for patients with indeterminate results. **Rapid diagnosis pathway, where detailed staging and fitness investigations are not needed to guide management (eg, patients with advanced disease not suitable for curative intent treatment). ***False negatives are presumed to re-present at A&E. The model will adopt a decision tree approach with an NHS and Personal Social Services perspective, and a lifetime time horizon. Patient flow and short-term costs and outcomes from the chest diagnostic imaging stage of the care pathway will be informed by data collected in this study. Long-term costs and outcomes for lung cancer patients (by stage at diagnosis) will be derived from published estimates from relevant studies in a UK setting and from appropriate health service datasets.
The comparator for AI deployment will be the usual care pathways for chest diagnostic imaging predeployment (without AI assistance). Data for these pathways will be drawn from the same sites pre-AIDF implementation, plus data from sites where AI tools for chest diagnostic imaging have not been implemented.
SampleThe model will be populated by relevant data collated and synthesized from workstream 2 (local datasets from the in-depth participant trusts), responses to a participant questionnaire (completed by both in-depth and light-touch participant trusts), relevant AIDF material (eg, procurement documentation), and any relevant information raised during participant interviews in workstream 1. Data required for populating the model, which is otherwise unavailable from these sources (eg, beyond the time or resource scope of this project), will be obtained from relevant health care datasets or published studies.
Where comparative (AI vs non-AI) estimates are used to inform model input parameters, these will be evaluated relative to baseline data from the respective sites, or from sites which do not use AI in the care pathway (where available). Pre- and postdeployment data obtained during the collection period of the evaluation will be converted to annualized estimates for costs and outcomes, weighted by the respective chest diagnostic imaging activity of the participating trusts.
MeasuresRelevant measures for informing the economic model are outlined in . Those to be obtained (where possible) from this study will include the measures listed in .
Measures for the economic modeling obtained from published studies or appropriate published datasets will include the items listed in .
Table 5. Key inputs for the economic modela.DescriptionCategorySourceb
Comments (0)