Stakeholder perspectives towards diagnostic artificial intelligence: a co-produced qualitative evidence synthesis.

Background: Diagnosis is a cornerstone of medical practice. Worldwide, there is increased demand for diagnostic services, exacerbating workforce shortages. Artificial intelligence (AI) technologies may improve diagnostic efficiency, accuracy, and access. Understanding stakeholder perspectives is key to informing implementation of complex interventions. We systematically reviewed the literature on stakeholder perspectives on diagnostic AI, including all English-language peer-reviewed primary qualitative or mixed-methods research. Methods: We searched PubMed, Ovid MEDLINE/Embase, Scopus, CINAHL and Web of Science (22/2/2023 and updated 8/2/2024). The Critical Appraisal Skills Programme Checklist informed critical appraisal. We used a 'best-fit' framework approach for analysis, using the Non-adoption, Abandonment, Scale-up, Spread, Sustainability (NASSS) framework. This study was pre-registered (PROSPERO CRD42022313782). Findings: We screened 16,577 articles and included 44. 689 participants were interviewed, and 402 participated in focus groups. Four stakeholder groups were described: patients, clinicians, researchers and healthcare leaders. We found an under-representation of patients, researchers and leaders across articles. We summarise the differences and relationships between each group in a conceptual model, hinging on the establishment of trust, engagement and collaboration. We present a modification of the NASSS framework, tailored to diagnostic AI. Interpretation: We provide guidance for future research and implementation of diagnostic AI, highlighting the importance of representing all stakeholder groups. We suggest that implementation strategies consider how any proposed software fits within the extended NASSS-AI framework, and how stakeholder priorities and concerns have been addressed. Funding: RK is supported by an NIHR Doctoral Research Fellowship grant (NIHR302562), which funded patient and public involvement activities, and access to Covidence.

Mortality in patients with Dupuytren's disease in the first 5 years after diagnosis: a population-based survival analysis.

Previous studies suggest that Dupuytren's disease is associated with increased mortality, but most studies failed to account for important confounders. In this population-based cohort study, general practitioners' (GP) data were linked to Statistics Netherlands to register all-cause and disease-specific mortality. Patients with Dupuytren's disease were identified using the corresponding diagnosis code and assessing free-text fields from GP consultations. Multiple imputations were performed to estimate missing values of covariates, followed by 1:7 propensity score matching to balance cases with controls on confounding factors. A frailty proportional hazard model was used to compare mortality between both groups. Out of 209,966 individuals, 2561 patients with Dupuytren's disease were identified and matched to at least four controls. After a median follow-up of 5 years, mortality was found to be actually reduced in patients with Dupuytren's disease. There was no difference in mortality secondary to cancer or cardiovascular disease. Future studies with longer average follow-up using longitudinal data should clarify these associations in the longer term.Level of evidence: III.

Dupuytren Disease: Prevalence, Incidence, and Lifetime Risk of Surgical Intervention. A Population-Based Cohort Analysis.

BACKGROUND: Health care burden attributable to Dupuytren disease (DD) is largely unknown. The authors determined (1) the prevalence and incidence of DD, (2) the incidence of first surgical intervention, and (3) the lifetime risk of surgical intervention in the United Kingdom National Healthcare Service. METHODS: In this population-based dynamic cohort analysis, data of the Clinical Practice Research Datalink was linked to Hospital Episode Statistics, to characterize the diagnosis and surgical treatment of DD. Secular trends of incidence of DD diagnosis and first surgical treatment were calculated for 2000 to 2013. A multistate Markov model was designed to estimate the lifetime risk of first surgical intervention. RESULTS: A total of 10,553,454 subjects were included in the analyses, 5,502,879 (52%) of whom were women. Of these, 38,707 DD patients were identified. Point prevalence in 2013 was 0.67% (99% CI, 0.66 to 0.68). The incidence of DD almost doubled from 0.30 (99% CI, 0.28 to 0.33) per 1000 person-years in 2000, to 0.59 (99% CI, 0.56 to 0.62) per 1000 person-years in 2013. The incidence of first surgical intervention similarly increased from 0.29 (99% CI, 0.23 to 0.37) to 0.88 (99% CI, 0.77 to 1.00) in the same period. A man or woman newly diagnosed with DD at age 65 has a lifetime risk of surgical intervention of 23% and 13%, respectively, showing only a very subtle decrease when diagnosed later in life. CONCLUSIONS: DD is an important health condition in the older population, because prevalence and incidence rates have almost doubled in the past decade. Estimated lifetime risk of surgical treatment is relatively low, but almost twice in men compared with women. CLINICAL QUESTION/LEVEL OF EVIDENCE: Risk, III.

A Computable Algorithm for Medication Optimization in Heart Failure With Reduced Ejection Fraction.

Background: Guideline-directed medical therapy (GDMT) optimization can improve outcomes in heart failure with reduced ejection fraction. Objectives: The objective of this study was to determine if a novel computable algorithm appropriately recommended GDMT. Methods: Clinical trial data from the GUIDE-IT (Guiding Evidence-Based Therapy Using Biomarker Intensified Treatment in Heart Failure) and HF-ACTION (Heart Failure: A Controlled Trial Investigating Outcomes of Exercise Training) trials were evaluated with a computable medication optimization algorithm that outputs GDMT recommendations and a medication optimization score (MOS). Algorithm-based recommendations were compared to medication changes. A Cox proportional-hazards model was used to estimate the associations between MOS and the composite primary end point for both trials. Results: The algorithm recommended initiation of angiotensin-converting enzyme inhibitor/angiotensin receptor blocker, beta-blockers, and mineralocorticoid receptor antagonists in 52.8%, 34.9%, and 68.1% of GUIDE-IT visits, respectively, when not prescribed the drug. Initiation only occurred in 20.8%, 56.9%, and 15.8% of subsequent visits. The algorithm also identified dose titration in 48.8% of visits for angiotensin-converting enzyme inhibitor/angiotensin receptor blockers and 39.4% of visits for beta-blockers. Those increases only occurred in 24.3% and 36.8% of subsequent visits. A higher baseline MOS was associated with a lower risk of cardiovascular death or heart failure hospitalization (HR: 0.41; 95% CI: 0.21-0.80; P = 0.009) in GUIDE-IT and all-cause death and hospitalization (HR: 0.61; 95% CI: 0.44-0.84; P = 0.003) in HF-ACTION. Conclusions: The algorithm accurately identified patients for GDMT optimization. Even in a clinical trial with robust protocols, GDMT could have been further optimized in a meaningful number of visits. The algorithm-generated MOS was associated with a lower risk of clinical outcomes. Implementation into clinical care may identify and address suboptimal GDMT in patients with heart failure with reduced ejection fraction.

Artificial Intelligence in Fracture Detection: A Systematic Review and Meta-Analysis.

Background Patients with fractures are a common emergency presentation and may be misdiagnosed at radiologic imaging. An increasing number of studies apply artificial intelligence (AI) techniques to fracture detection as an adjunct to clinician diagnosis. Purpose To perform a systematic review and meta-analysis comparing the diagnostic performance in fracture detection between AI and clinicians in peer-reviewed publications and the gray literature (ie, articles published on preprint repositories). Materials and Methods A search of multiple electronic databases between January 2018 and July 2020 (updated June 2021) was performed that included any primary research studies that developed and/or validated AI for the purposes of fracture detection at any imaging modality and excluded studies that evaluated image segmentation algorithms. Meta-analysis with a hierarchical model to calculate pooled sensitivity and specificity was used. Risk of bias was assessed by using a modified Prediction Model Study Risk of Bias Assessment Tool, or PROBAST, checklist. Results Included for analysis were 42 studies, with 115 contingency tables extracted from 32 studies (55 061 images). Thirty-seven studies identified fractures on radiographs and five studies identified fractures on CT images. For internal validation test sets, the pooled sensitivity was 92% (95% CI: 88, 93) for AI and 91% (95% CI: 85, 95) for clinicians, and the pooled specificity was 91% (95% CI: 88, 93) for AI and 92% (95% CI: 89, 92) for clinicians. For external validation test sets, the pooled sensitivity was 91% (95% CI: 84, 95) for AI and 94% (95% CI: 90, 96) for clinicians, and the pooled specificity was 91% (95% CI: 81, 95) for AI and 94% (95% CI: 91, 95) for clinicians. There were no statistically significant differences between clinician and AI performance. There were 22 of 42 (52%) studies that were judged to have high risk of bias. Meta-regression identified multiple sources of heterogeneity in the data, including risk of bias and fracture type. Conclusion Artificial intelligence (AI) and clinicians had comparable reported diagnostic performance in fracture detection, suggesting that AI technology holds promise as a diagnostic adjunct in future clinical practice. Clinical trial registration no. CRD42020186641 © RSNA, 2022 Online supplemental material is available for this article. See also the editorial by Cohen and McInnes in this issue.

Perspectives: A surgeon's guide to machine learning.

The exponential increase in the volume and complexity of healthcare data presents new challenges to researchers and clinicians in analysis and interpretation. The requirement for new strategies to extract meaningful information from large, noisy datasets has led to the development of the field of big data analytics. Artificial intelligence (AI) is a general-purpose technology in which machines carry out tasks traditionally thought to be only achievable by humans. Machine learning (ML) is an approach to AI in which machines can "learn" to perform tasks in an automated process, rather than being explicitly programmed by a human. Research aiming to apply ML techniques to classification, prediction and decision-making problems in healthcare has increased 61-fold from 2005 to 2019, mirroring this sense of early promise. The field of healthcare ML is relatively young, and many critical steps are needed before adoption into clinical practice, including transparent, unbiased development and reporting of algorithms. Articles claiming that machines can outperform, or replace, doctors in high-level tasks, such as diagnosis or prognostication, must be carefully appraised. It is critical that surgeons have an understanding of the principles and terminology of AI and ML to evaluate these claims and to take an active role in directing research. This article is an up-to-date review and primer for surgeons covering the core tenets of ML applied to surgical problems, including algorithm types and selection, model training and validation, interpretation of common outcome metrics, current and future reporting guidelines and discussion of the challenges and limitations in this field.

Low rate of subsequent surgery and serious complications following intra-articular steroid injection for base of thumb osteoarthritis: national cohort analysis.

OBJECTIVES: Intra-articular steroid injection is commonly used to treat base of thumb osteoarthritis (BTOA), despite a lack of large-scale data on safety and effectiveness. We estimate the incidence of serious complications and further procedures following BTOA injection, including the risk of post-operative serious surgical site infection for subsequent operative intervention. METHODS: Hospital Episode Statistics data linked to mortality records from 1 April 1998 to 31 March 2017 were used to identify all BTOA injections undertaken in adults in the National Health Service secondary care in England. Patients were followed up longitudinally until death or 31 March 2017. A multivariable regression with a Fine and Gray model adjusting for the competing risk of mortality in addition to age, sex and socioeconomic deprivation was used to identify factors associated with progression to further procedure. Secondary outcomes included serious complications after injection and subsequent surgical site infection. RESULTS: A total of 19 120 primary injections were performed during the 19-year period in 18 356 patients. Of these 76.5% were female; mean age 62 years (s.d. 10.6); 50.48% underwent further procedure; 22.40% underwent surgery. Median time to further intervention was 412 days (IQR 110-1945). Female sex was associated with increased risk of proceeding to surgery. Serious complication rate following injection was 0.04% (0.01-0.08) within 90 days. Of those proceeding to surgery 0.16% (0.06-0.34) presented with a wound infection within 30 days and 90 days, compared with an overall post-operative wound infection rate of 0.03% (0.02-0.05). CONCLUSIONS: Very low rates of serious complications were identified following BTOA injections performed in secondary care; only one in five patients proceeded to subsequent surgery. CLINICAL TRIAL REGISTRATION: clinicaltrials.gov, https://www.clinicaltrials.gov, NCT03573765.

Nongenetic Factors Associated with Dupuytren's Disease: A Systematic Review.

BACKGROUND: Dupuytren's disease is a common complex disease caused by genetic and nongenetic factors. The role of many nongenetic risk factors is still unclear and debatable. This study aimed to systematically review the association between Dupuytren's disease and nongenetic risk factors. METHODS: A search strategy was developed based on the Population, Exposure, Comparison, Outcomes and Study framework. A Preferred Reporting Items for Systematic Reviews and Meta-Analyses-compliant literature search was conducted in MEDLINE, Embase, Scopus, Web of Science, Cochrane Database of Systematic Reviews, and Cochrane Central Register of Controlled Trials from inception to November of 2018. Title and abstract and then full-text screening against eligibility criteria was performed independently by two reviewers, and consensus was achieved by a third reviewer. The Effective Public Health Practice Project and the Oxford Centre for Evidence Based Medicine tools were used to assess study quality and to evaluate the level of evidence of included studies, respectively. RESULTS: Reviewers identified 4434 studies, of which 54 were included in the analysis. There was strong evidence for the association between Dupuytren's disease and advanced age, male sex, family history of Dupuytren's disease, and diabetes mellitus. Furthermore, heavy alcohol drinking, cigarette smoking, and manual work exposure showed a significant dose-response relationship. The quality of the included studies was mainly low or moderate, and most studies were level 3 or 4 on the Oxford Centre for Evidence Based Medicine scale. CONCLUSIONS: The study results show a strong association between Dupuytren's disease and advanced age, male sex, family history of Dupuytren's disease, diabetes mellitus, heavy alcohol drinking, cigarette smoking, and manual work exposure. Further studies are required to explain the causal relationship of these associations.

Dupuytren's Disease Predicts Increased All-Cause and Cancer-Specific Mortality: Analysis of a Large Cohort from the U.K. Clinical Practice Research Datalink.

BACKGROUND: Dupuytren's disease is a common, chronic, fibroproliferative disease of the palmar fascia. The cause is unclear but includes genetic predisposition alongside environmental factors. Several studies have suggested an association between Dupuytren's disease and excess mortality. The authors aimed to evaluate this association in adult patients in the United Kingdom and identify the causes of mortality. METHODS: The authors used a large primary care database (Clinical Practice Research Datalink) linked to the Office of National Statistics to identify patients with Dupuytren's disease between January 1, 1995, and December 31, 2013. Each patient was matched by age, sex, and general practitioner to five control patients without the disease. Cox regression models were used to study the association between Dupuytren's disease and all-cause and cause-specific mortality, adjusting for confounders. RESULTS: A total of 41,965 Dupuytren's disease patients and 209,825 control patients were identified. The all-cause mortality rate was increased in both unadjusted (hazard ratio, 1.48; 99% CI, 1.29 to 1.70; p < 0.0001) and multivariable adjusted (hazard ratio, 1.43; 99% CI, 1.25 to 1.65; p < 0.0001) models in patients with Dupuytren's disease, 12 years after diagnosis. Excess mortality was secondary to a wide range of causes, including cancer (hazard ratio, 1.66; 99% CI, 1.27 to 2.17; p < 0.0001), an effect that persisted after adjustment for confounders. CONCLUSIONS: There is excess mortality associated with Dupuytren's disease that can be partially explained through environmental factors. From time of diagnosis in primary care, there is a 12-year window of opportunity for intervention to reduce the impact of these factors. The authors observed an increased risk of cancer mortality independent of confounders, and hypothesize a shared genetic risk between Dupuytren's disease and cancer. CLINICAL QUESTION/LEVEL OF EVIDENCE: Risk, II.

Measuring Limb Volume: Accuracy and Reliability of Tape Measurement Versus Perometer Measurement.

BACKGROUND: Accurate limb volume measurement is key in the assessment of outcomes in lymphedema microsurgery. There are two commonly used methods as follows: manual circumferential measurement (tape) or Perometer measurement. There are no data on the intra- and interclass correlation of either method, making it difficult to establish a gold standard of limb volume measurement. We aim to assess the intra- and interclass correlation of each method to establish the most appropriate method for clinical practice and future research studies, aiming to compare the accuracy and reliability of tape measurement as assessed against Perometer measurement. METHODS AND RESULTS: Student volunteers and experts (lymphedema practitioners) were each asked to perform repeat tape and Perometer measurements on the upper or lower limb of one healthy volunteer. Perometer measurements were globally more accurate than tape (average SE [Perometer]: 23.23 vs. 77.21 [tape]). For intraobserver reliability, experts outperformed students in all domains tested, with little difference in intraobserver reliability using tape or Perometer (average Cronbach's alpha 0.9597 [expert)] vs. 0.6033 [student]). CONCLUSIONS: We recommend that, for increased interobserver reliability, the Perometer provides a more reliable standard of limb volume measurement.

Sentinel node biopsy status is strongly predictive of survival in cutaneous melanoma: Extended follow-up of Oxford patients from 1998 to 2014.

INTRODUCTION: Sentinel lymph node biopsy (SLNB) is widely used as a key investigatory tool for cutaneous melanoma, with results incorporated into the latest AJCC staging guidelines. We present the results of our extended follow-up of sentinel lymph node biopsy for melanoma over a sixteen-year period. METHODS: Data were collected prospectively from June 1998 to December 2014 from a single tertiary skin cancer referral centre. Chi-squared analysis was used to analyse patient demographics and primary tumour pathology. Survival analysis was conducted using Cox regression models and Kaplan-Meier survival curves. RESULTS: Over a sixteen-year period 1527 patients underwent SLNB in 1609 basins, with 2876 nodes harvested. 347 patients (23%) had a positive biopsy. The most common primary tumour sites for males was the back (32%); women had a significantly higher number of melanomas occurring on the lower and upper limbs (45% and 26% respectively) [all p < 0.0001, Chi-squared]. Mean follow-up time was 4.9 years. Patients with a positive SLNB at diagnosis were significantly more likely to die from melanoma (subhazard ratio 5.59, p = 0.000, 95% CI 3.59-8.69). Breslow thickness and ulceration were also significant predictors of melanoma-specific mortality. For patients with a primary Breslow >4.0 mm ten-year disease free survival was 52% for SLNB negative and 26% for SLNB positive patients. For Breslow thicknesses of 2.01-4 mm these values were 66% and 32% respectively. CONCLUSIONS: Sentinel lymph node biopsy status is strongly predictive of survival across all thicknesses of primary cutaneous melanoma.

A true branchial fistula in the context of branchiootic syndrome: challenges of diagnosis and management.

BACKGROUND: The presence of a branchial fistula with communication both internally and externally: a 'true' branchial fistula is rare, and may arise in the context of autosomal dominant conditions such as branchiootic syndrome and branchiootorenal syndrome. STUDY: We discuss the case of a true branchial fistula, which recurred after initial surgical excision, in a patient with branchiootic syndrome. The residual tract was dissected in a second operation through stepladder neck incisions and removed in toto via an intraoral approach. No renal abnormalities were detected on investigation with ultrasound. DISCUSSION: Incomplete excision of a branchial sinus is likely to cause recurrence however intraoperative visualisation of the tract can can sometimes prove challenging. An combined intraoral and external approach aids delineation and tract definition when there is a true branchial fistula and can therefore facilitate a complete excision. Suspicion of an hereditary aetiology should be raised in patients with bilateral or preauricular features, or a positive family history, which may then prompt additional renal and genetic investigation.