Conceptualising fairness: three pillars for medical algorithms and health equity

ObjectivesFairness is a core concept meant to grapple with different forms of discrimination and bias that emerge with advances in Artificial Intelligence (eg, machine learning, ML). Yet, claims to fairness in ML discourses are often vague and contradictory. The response to these issues within the scientific community has been technocratic. Studies either measure (mathematically) competing definitions of fairness, and/or recommend a range of governance tools (eg, fairness checklists or guiding principles). To advance efforts to operationalise fairness in medicine, we synthesised a broad range of literature.MethodsWe conducted an environmental scan of English language literature on fairness from 1960-July 31, 2021. Electronic databases Medline, PubMed and Google Scholar were searched, supplemented by additional hand searches. Data from 213 selected publications were analysed using rapid framework analysis. Search and analysis were completed in two rounds: to explore previously identified issues (a priori), as well as those emerging from the analysis (de novo).ResultsOur synthesis identified ‘Three Pillars for Fairness’: transparency, impartiality and inclusion. We draw on these insights to propose a multidimensional conceptual framework to guide empirical research on the operationalisation of fairness in healthcare.DiscussionWe apply the conceptual framework generated by our synthesis to risk assessment in psychiatry as a case study. We argue that any claim to fairness must reflect critical assessment and ongoing social and political deliberation around these three pillars with a range of stakeholders, including patients.ConclusionWe conclude by outlining areas for further research that would bolster ongoing commitments to fairness and health equity in healthcare.

Codeveloping an effective EMPA to maturity in an acute NHS Trust: an implementer report

IntroductionUniversity Hospitals Leicester has codeveloped, with Nervecentre, an Electronic Prescribing and Medicines Administration System that meets specific clinical and interoperability demands of the National Health Service (NHS).MethodsThe system was developed through a frontline-led and agile approach with a project team consisting of clinicians, Information Technology (IT) specialists and the vendor’s representatives over an 18-month period.ResultsThe system was deployed successfully with more than a thousand transcriptions during roll-out. Despite the high caseload and novelty of the system, there was no increase in error rates within the first 3 months of roll-out. Healthcare professionals perceived the new system as efficient with improved clinical workflow, and safe through an integrated medication alert system.DiscussionThis case study demonstrates how NHS trusts can successfully co-develop, with vendors, new IT systems which meet interoperability standards such as Fast Healthcare Interoperability Resources, while improving front line clinical experience.ConclusionAlternative methods to the ‘big bang’ deployment of IT projects, such as ‘gradual implementation’, must be demonstrated and evaluated for their ability to deliver digital transformation projects in the NHS successfully.

Driving digital health transformation in hospitals: a formative qualitative evaluation of the English Global Digital Exemplar programme

BackgroundThere is currently a strong drive internationally towards creating digitally advanced healthcare systems through coordinated efforts at a national level. The English Global Digital Exemplar (GDE) programme is a large-scale national health information technology change programme aiming to promote digitally-enabled transformation in secondary healthcare provider organisations by supporting relatively digitally mature provider organisations to become international centres of excellence.AimTo qualitatively evaluate the impact of the GDE programme in promoting digital transformation in provider organisations that took part in the programme.MethodsWe conducted a series of in-depth case studies in 12 purposively selected provider organisations and a further 24 wider case studies of the remaining organisations participating in the GDE programme. Data collected included 628 interviews, non-participant observations of 190 meetings and workshops and analysis of 9 documents. We used thematic analysis aided by NVivo software and drew on sociotechnical theory to analyse the data.ResultsWe found the GDE programme accelerated digital transformation within participating provider organisations. This acceleration was triggered by: (1) dedicated funding and the associated requirement for matched internal funding, which in turn helped to prioritise digital transformation locally; (2) governance requirements put in place by the programme that helped strengthen existing local governance and project management structures and supported the emergence of a cadre of clinical health informatics leaders locally; and (3) reputational benefits associated with being recognised as a centre of digital excellence, which facilitated organisational buy-in for digital transformation and increased negotiating power with vendors.ConclusionThe GDE programme has been successful in accelerating digital transformation in participating provider organisations. Large-scale digital transformation programmes in healthcare can stimulate local progress through protected funding, putting in place governance structures and leveraging reputational benefits for participating provider organisations, around a coherent vision of transformation.

Development and validation pathways of artificial intelligence tools evaluated in randomised clinical trials

ObjectiveGiven the complexities of testing the translational capability of new artificial intelligence (AI) tools, we aimed to map the pathways of training/validation/testing in development process and external validation of AI tools evaluated in dedicated randomised controlled trials (AI-RCTs).MethodsWe searched for peer-reviewed protocols and completed AI-RCTs evaluating the clinical effectiveness of AI tools and identified development and validation studies of AI tools. We collected detailed information, and evaluated patterns of development and external validation of AI tools.ResultsWe found 23 AI-RCTs evaluating the clinical impact of 18 unique AI tools (2009–2021). Standard-of-care interventions were used in the control arms in all but one AI-RCT. Investigators did not provide access to the software code of the AI tool in any of the studies. Considering the primary outcome, the results were in favour of the AI intervention in 82% of the completed AI-RCTs (14 out of 17). We identified significant variation in the patterns of development, external validation and clinical evaluation approaches among different AI tools. A published development study was found only for 10 of the 18 AI tools. Median time from the publication of a development study to the respective AI-RCT was 1.4 years (IQR 0.2–2.2).ConclusionsWe found significant variation in the patterns of development and validation for AI tools before their evaluation in dedicated AI-RCTs. Published peer-reviewed protocols and completed AI-RCTs were also heterogeneous in design and reporting. Upcoming guidelines providing guidance for the development and clinical translation process aim to improve these aspects.

Are child health information services a viable source of accurate vaccination data for clinicians working in paediatric emergency departments in England?

Vaccination is a global success story, yet UK coverage remains undertarget for a number of diseases. The paediatric emergency department (PED) offers the potential for opportunistic vaccination interventions.ObjectivesTo map the Greater Manchester (GM) Child Health Information System network to see if it was a viable source of vaccination data for clinicians working in the PED as a case study.MethodsPostprimary care vaccination management systems for GM were visualised using a systems mapping approach, with data obtained from the Office for National Statistics and commissioners in the GM Health and Social Care Partnership.ResultsOnce vaccination data left primary care, it passed through 1 of 10 local child health information services (CHISs), using an assortment of different information technology systems, after which it shed individual identifiers and was aggregated within national systems. None of the existing GM CHISs were accessible to PED practitioners.ConclusionMore work needs to be done to explore possible alternative sources of accurate vaccination data during a PED consultation.

Association of persistent acute kidney injury and renal recovery with mortality in hospitalised patients

ObjectivesAcute kidney injury (AKI) affects up to one-quarter of hospitalised patients and 60% of patients in the intensive care unit (ICU). We aim to understand the baseline characteristics of patients who will develop distinct AKI trajectories, determine the impact of persistent AKI and renal non-recovery on clinical outcomes, resource use, and assess the relative importance of AKI severity, duration and recovery on survival.MethodsIn this retrospective, longitudinal cohort study, 156 699 patients admitted to a quaternary care hospital between January 2012 and August 2019 were staged and classified (no AKI, rapidly reversed AKI, persistent AKI with and without renal recovery). Clinical outcomes, resource use and short-term and long-term survival adjusting for AKI severity were compared among AKI trajectories in all cohort and subcohorts with and without ICU admission.ResultsFifty-eight per cent (31 500/54 212) had AKI that rapidly reversed within 48 hours; among patients with persistent AKI, two-thirds (14 122/22 712) did not have renal recovery by discharge. One-year mortality was significantly higher among patients with persistent AKI (35%, 7856/22 712) than patients with rapidly reversed AKI (15%, 4714/31 500) and no AKI (7%, 22 117/301 466). Persistent AKI without renal recovery was associated with approximately fivefold increased hazard rates compared with no AKI in all cohort and ICU and non-ICU subcohorts, independent of AKI severity.DiscussionAmong hospitalised, ICU and non-ICU patients, persistent AKI and the absence of renal recovery are associated with reduced long-term survival, independent of AKI severity.ConclusionsIt is essential to identify patients at risk of developing persistent AKI and no renal recovery to guide treatment-related decisions.

Exploring stakeholder attitudes towards AI in clinical practice

ObjectivesDifferent stakeholders may hold varying attitudes towards artificial intelligence (AI) applications in healthcare, which may constrain their acceptance if AI developers fail to take them into account. We set out to ascertain evidence of the attitudes of clinicians, consumers, managers, researchers, regulators and industry towards AI applications in healthcare.MethodsWe undertook an exploratory analysis of articles whose titles or abstracts contained the terms ‘artificial intelligence’ or ‘AI’ and ‘medical’ or ‘healthcare’ and ‘attitudes’, ‘perceptions’, ‘opinions’, ‘views’, ‘expectations’. Using a snowballing strategy, we searched PubMed and Google Scholar for articles published 1 January 2010 through 31 May 2021. We selected articles relating to non-robotic clinician-facing AI applications used to support healthcare-related tasks or decision-making.ResultsAcross 27 studies, attitudes towards AI applications in healthcare, in general, were positive, more so for those with direct experience of AI, but provided certain safeguards were met. AI applications which automated data interpretation and synthesis were regarded more favourably by clinicians and consumers than those that directly influenced clinical decisions or potentially impacted clinician–patient relationships. Privacy breaches and personal liability for AI-related error worried clinicians, while loss of clinician oversight and inability to fully share in decision-making worried consumers. Both clinicians and consumers wanted AI-generated advice to be trustworthy, while industry groups emphasised AI benefits and wanted more data, funding and regulatory certainty.DiscussionCertain expectations of AI applications were common to many stakeholder groups from which a set of dependencies can be defined.ConclusionStakeholders differ in some but not all of their attitudes towards AI. Those developing and implementing applications should consider policies and processes that bridge attitudinal disconnects between different stakeholders.

Association between the chronology of gestation and the morphometrical skin characteristics at childbirth: a development of predictive model

ObjectiveThe structural maturation of the skin is considered a potential marker of pregnancy dating. This study investigated the correlation between the morphometrical skin characteristics with the pregnancy chronology to propose models for predicting gestational age.MethodsA cross-sectional analysis selected 35 corpses of newborns. The biopsy was performed up to 48 hours after death in the periumbilical abdomen, palm and sole regions. Pregnancy chronology was based on the obstetric ultrasound before 14 weeks. The dimensions of the skin layers, area of glands and connective fibrous tissue were measured with imaging software support. Univariate and multivariate regression models on morphometric values were used to predict gestational age.ResultsGestational age at birth ranged from 20.3 to 41.2 weeks. Seventy-one skin specimens resulted in the analysis of 1183 digital histological images. The correlation between skin thickness and gestational age was positive and strong in both regions of the body. The highest univariate correlation between gestational age and skin thickness was using the epidermal layer dimensions, in palm (r=0.867, p<0.001). The multivariate modelling with the thickness of the abdominal epidermis, the dermis and the area of the sebaceous glands adjusted had the highest correlation with gestational age (r=0.99, p<0.001).ConclusionThe thickness of the protective epidermal barrier is, in itself, a potential marker of pregnancy dating. However, sets of values obtained from skin morphometry enhanced the estimation of the gestational age. Such findings may support non-invasive image approaches to estimate pregnancy dating with various clinical applications.

Information processing by community health nurses using mobile health (mHealth) tools for early identification of suicide and depression risks in Fiji Islands

ObjectivesHigh rates of depression and suicide and a lack of trained psychiatrists have emerged as significant concerns in the low-income and middle-income countries (LMICs) such as the Pacific Island Countries (PICs). Readily available smartphones were leveraged with community health nurses (CHNs) in task-sharing for early identification of suicide and depression risks in Fiji Islands, the largest of PICs. This investigation examines how CHNs can efficiently and effectively process patient information about depression and suicide risk for making diagnostic and management decisions without compromising safety. The research is driven by the theoretical framework of text comprehension (knowledge representation and interpretation) and decision-making.MethodsMobile health (mHealth) Application for Suicide Risk and Depression Assessment (ASRaDA) was designed to include culturally useful clinical guidelines for these disorders. A representative sample of 48 CHNs was recruited and presented with two clinical cases (depression and suicide) in a simulated setting under three conditions: No support, paper-based and mobile-based culturally valid guideline support. Data were collected as the nurses read through the scenarios, ‘thinking aloud’, before summarising, diagnoses and follow-up recommendations. Transcribed audiotapes were analysed using formal qualitative discourse analysis methods for diagnostic accuracy, comprehension of clinical problems and reasoning patterns.ResultsUsing guidelines on ASRaDA, the CHNs took less time to process patient information with more accurate diagnostic and therapeutic decisions for depression and suicide risk than with paper-based or no guideline conditions. A change in reasoning pattern for nurses’ information processing was observed with decision support.DiscussionAlthough these results are shown in a mental health setting in Fiji, there are reasons to believe they are generalisable beyond mental health and other lower-to-middle income countries.ConclusionsCulturally appropriate clinical guidelines on mHealth supports efficient information processing for quick and accurate decisions and a positive shift in reasoning behaviour by the nurses. However, translating complex qualitative patient information into quantitative scores could generate conceptual errors. These results are valid in simulated conditions.

Increasing enrolment in a national VA transitions of care programme: a pre–post evaluation of a data dashboard and nudge-based intervention

BackgroundThe rural transitions nurse programme (TNP) is a care coordination intervention for high-risk veterans. An interactive dashboard was used to provide real-time performance metrics to sites as an audit and feedback tool. One-year post implementation, enrolment goals were not met. Nudge emails were introduced to increase TNP veteran enrolment. This study evaluated whether veteran enrolment increased when feedback occurred through a dashboard plus weekly nudge email versus dashboard alone.Setting/populationThis observational study included veterans who were hospitalised and discharged from four Veterans Health Administration hospitals participating in TNP.MethodsVeteran enrolment counts between the dashboard phase and dashboard plus weekly nudge email phase were compared. Nudge emails included run charts of enrolment data. The difference of means for weekly enrolment between the two phases were calculated. After 3 months of nudge emails, a survey assessing TNP transitions nurse and physician champion perceptions of the nudge emails was distributed.ResultsThe average enrolment for the four TNP sites during the ~20-month dashboard only phase was 4.23 veterans/week. The average during the 3-month dashboard plus nudge email phase was 4.21 veterans/week. The difference in means was −0.03 (p=0.73). Adjusting for time trends had no further effect. Four nurses responded to the survey. Two nurses reported neutral and two reported positive perceptions of the nudge emails.ConclusionDrawing attention to metrics, through nudge emails, maintained, but did not increase TNP veteran discharges compared to dashboard feedback alone.