An intentional approach to managing bias in general purpose embedding models.

Advances in machine learning for health care have brought concerns about bias from the research community; specifically, the introduction, perpetuation, or exacerbation of care disparities. Reinforcing these concerns is the finding that medical images often reveal signals about sensitive attributes in ways that are hard to pinpoint by both algorithms and people. This finding raises a question about how to best design general purpose pretrained embeddings (GPPEs, defined as embeddings meant to support a broad array of use cases) for building downstream models that are free from particular types of bias. The downstream model should be carefully evaluated for bias, and audited and improved as appropriate. However, in our view, well intentioned attempts to prevent the upstream components-GPPEs-from learning sensitive attributes can have unintended consequences on the downstream models. Despite producing a veneer of technical neutrality, the resultant end-to-end system might still be biased or poorly performing. We present reasons, by building on previously published data, to support the reasoning that GPPEs should ideally contain as much information as the original data contain, and highlight the perils of trying to remove sensitive attributes from a GPPE. We also emphasise that downstream prediction models trained for specific tasks and settings, whether developed using GPPEs or not, should be carefully designed and evaluated to avoid bias that makes models vulnerable to issues such as distributional shift. These evaluations should be done by a diverse team, including social scientists, on a diverse cohort representing the full breadth of the patient population for which the final model is intended.

Detecting shortcut learning for fair medical AI using shortcut testing.

Machine learning (ML) holds great promise for improving healthcare, but it is critical to ensure that its use will not propagate or amplify health disparities. An important step is to characterize the (un)fairness of ML models-their tendency to perform differently across subgroups of the population-and to understand its underlying mechanisms. One potential driver of algorithmic unfairness, shortcut learning, arises when ML models base predictions on improper correlations in the training data. Diagnosing this phenomenon is difficult as sensitive attributes may be causally linked with disease. Using multitask learning, we propose a method to directly test for the presence of shortcut learning in clinical ML systems and demonstrate its application to clinical tasks in radiology and dermatology. Finally, our approach reveals instances when shortcutting is not responsible for unfairness, highlighting the need for a holistic approach to fairness mitigation in medical AI.

Editor's Choice - Assessment of Correlation Between Mean Size of Infrarenal Abdominal Aortic Aneurysm at Time of Intact Repair Against Repair and Rupture Rate in Nine Countries.

OBJECTIVE: This study aimed to analyse the mean abdominal aortic aneurysm (AAA) diameter for repair in nine countries, and to determine variation in mean AAA diameter for elective AAA repair and its relationship to rupture AAA repair rates and aneurysm related mortality in corresponding populations. METHODS: Data on intact (iAAA) and ruptured infrarenal AAA (rAAA) repair for the years 2010-2012 were collected from Denmark, England, Finland, Germany, Hungary, New Zealand, Norway, Sweden, and the USA. The rate of iAAA repair and rAAA per 100 000 inhabitants above 59 years old, mean AAA diameter for iAAA repair and rAAA repair, and the national rates of rAAA were assessed. National cause of death statistics were used to estimate aneurysm related mortality. Direct standardisation methods were applied to the national mortality data. Logistic regression and analysis of variance model adjustments were made for age groups, sex, and year. RESULTS: There was a variation in the mean diameter of iAAA repair (n = 34 566; range Germany = 57 mm, Denmark = 68 mm). The standardised iAAA repair rate per 100000 inhabitants varied from 10.4 (Hungary) to 66.5 (Norway), p<.01, and the standardised rAAA repair rate per 100 000 from 5.8 (USA) to 16.9 (England), p<.01. Overall, there was no significant correlation between mean diameter of iAAA repair and standardised iAAA rate (r2 = 0.04, p = .3). There was no significant correlation between rAAA repair rate (n = 12 628) with mean diameter of iAAA repair (r2 = 0.2, p = .1). CONCLUSION: Despite recommendations from learned society guidelines, data indicate variations in mean diameter for AAA repair. There was no significant correlation between mean diameter of AAA repair and rates of iAAA repair and rAAA repair. These analyses are subject to differences in disease prevalence, uncertainties in rupture rates, validations of vascular registries, causes of death and registrations.

Implementation of a Digitally Enabled Care Pathway (Part 1): Impact on Clinical Outcomes and Associated Health Care Costs.

BACKGROUND: The development of acute kidney injury (AKI) in hospitalized patients is associated with adverse outcomes and increased health care costs. Simple automated e-alerts indicating its presence do not appear to improve outcomes, perhaps because of a lack of explicitly defined integration with a clinical response. OBJECTIVE: We sought to test this hypothesis by evaluating the impact of a digitally enabled intervention on clinical outcomes and health care costs associated with AKI in hospitalized patients. METHODS: We developed a care pathway comprising automated AKI detection, mobile clinician notification, in-app triage, and a protocolized specialist clinical response. We evaluated its impact by comparing data from pre- and postimplementation phases (May 2016 to January 2017 and May to September 2017, respectively) at the intervention site and another site not receiving the intervention. Clinical outcomes were analyzed using segmented regression analysis. The primary outcome was recovery of renal function to ≤120% of baseline by hospital discharge. Secondary clinical outcomes were mortality within 30 days of alert, progression of AKI stage, transfer to renal/intensive care units, hospital re-admission within 30 days of discharge, dependence on renal replacement therapy 30 days after discharge, and hospital-wide cardiac arrest rate. Time taken for specialist review of AKI alerts was measured. Impact on health care costs as defined by Patient-Level Information and Costing System data was evaluated using difference-in-differences (DID) analysis. RESULTS: The median time to AKI alert review by a specialist was 14.0 min (interquartile range 1.0-60.0 min). There was no impact on the primary outcome (estimated odds ratio [OR] 1.00, 95% CI 0.58-1.71; P=.99). Although the hospital-wide cardiac arrest rate fell significantly at the intervention site (OR 0.55, 95% CI 0.38-0.76; P<.001), DID analysis with the comparator site was not significant (OR 1.13, 95% CI 0.63-1.99; P=.69). There was no impact on other secondary clinical outcomes. Mean health care costs per patient were reduced by £2123 (95% CI -£4024 to -£222; P=.03), not including costs of providing the technology. CONCLUSIONS: The digitally enabled clinical intervention to detect and treat AKI in hospitalized patients reduced health care costs and possibly reduced cardiac arrest rates. Its impact on other clinical outcomes and identification of the active components of the pathway requires clarification through evaluation across multiple sites.

Research activity and the association with mortality.

INTRODUCTION: The aims of this study were to describe the key features of acute NHS Trusts with different levels of research activity and to investigate associations between research activity and clinical outcomes. METHODS: National Institute for Health Research (NIHR) Comprehensive Clinical Research Network (CCRN) funding and number of patients recruited to NIHR Clinical Research Network (CRN) portfolio studies for each NHS Trusts were used as markers of research activity. Patient-level data for adult non-elective admissions were extracted from the English Hospital Episode Statistics (2005-10). Risk-adjusted mortality associations between Trust structures, research activity and, clinical outcomes were investigated. RESULTS: Low mortality Trusts received greater levels of funding and recruited more patients adjusted for size of Trust (n = 35, 2,349 £/bed [95% CI 1,855-2,843], 5.9 patients/bed [2.7-9.0]) than Trusts with expected (n = 63, 1,110 £/bed, [864-1,357] p<0.0001, 2.6 patients/bed [1.7-3.5] p<0.0169) or, high (n = 42, 930 £/bed [683-1,177] p = 0.0001, 1.8 patients/bed [1.4-2.1] p<0.0005) mortality rates. The most research active Trusts were those with more doctors, nurses, critical care beds, operating theatres and, made greater use of radiology. Multifactorial analysis demonstrated better survival in the top funding and patient recruitment tertiles (lowest vs. highest (odds ratio & 95% CI: funding 1.050 [1.033-1.068] p<0.0001, recruitment 1.069 [1.052-1.086] p<0.0001), middle vs. highest (funding 1.040 [1.024-1.055] p<0.0001, recruitment 1.085 [1.070-1.100] p<0.0001). CONCLUSIONS: Research active Trusts appear to have key differences in composition than less research active Trusts. Research active Trusts had lower risk-adjusted mortality for acute admissions, which persisted after adjustment for staffing and other structural factors.

Assessment of short-term clinical outcomes following salvage esophagectomy for the treatment of esophageal malignancy: systematic review and pooled analysis.

BACKGROUND: Combined chemoradiotherapy is increasingly being used as definitive treatment for locoregional esophageal malignancy. Patients with residual or recurrent localized cancer are often selectively considered for salvage esophagectomy (SALV). The aim of this pooled analysis was to compare short-term clinical outcomes from SALV following definitive chemoradiotherapy with those from planned esophagectomy following neoadjuvant chemoradiotherapy (NCRS). METHODS: MEDLINE, EMBASE, Cochrane, trial registries, conference proceedings and reference lists were searched for relevant comparative studies. Primary outcome measures were in-hospital mortality, anastomotic leak and pulmonary complications. Secondary outcomes were length of hospital stay, negative (R0) resection margin, and estimated blood loss. RESULTS: Eight studies comprising 954 patients; 242 (SALV) and 712 (NCRS) were included. SALV was associated with a significantly increased incidence of post-operative mortality (9.50 vs. 4.07 %; pooled odds ratio [POR] = 3.02; p < 0.001), anastomotic leak (23.97 vs. 14.47 %; POR = 1.99; p = 0.005), pulmonary complications (29.75 vs. 16.99 %; POR = 2.12; p < 0.001), and an increased length of hospital stay (weighted mean difference = 8.29 days; 95 % CI 7.08-9.5; p < 0.001). There were no significant differences between the groups in the incidence of negative resection margins or estimated blood loss. CONCLUSIONS: SALV has poorer short-term outcomes when compared with planned esophagectomy following neoadjuvant chemoradiotherapy. Patients and multidisciplinary tumor boards should be made aware of these differences in outcomes and SALV should be reserved for practice in high-volume institutions.

The clinical and economic costs of delirium after surgical resection for esophageal malignancy.

OBJECTIVE: The aim of this study was to identify preoperative risk factors and postoperative consequences that are associated with the occurrence of delirium after esophagectomy for malignancy. BACKGROUND: Delirium is an underdiagnosed, serious complication after major surgery, particularly in the elderly population. METHODS: All patients undergoing esophagectomy for cancer (1991-2011) were included. Patients with and without delirium were compared with respect to medical comorbidities, use of neoadjuvant therapy, operative outcomes, postoperative complications, overall cost, and survival. RESULTS: Of the 500 patients included in this analysis, 46 (9.2%) patients developed postoperative delirium. Patients with delirium had higher ASA and Charlson comorbidity index scores. Delirium was associated with a longer hospital (14 ± 7.5 vs 10.9 ± 5.7; P < 0.05) and intensive care unit stay (3.6 ± 3.8 vs 2.7 ± 16.9; P < 0.05) and an increased incidence of pulmonary complications and increased hospital costs. Delirium was preceded by another complication in 32.6% of cases but by a septic complication in only 19.6% of cases. Age was the only preoperative predictor of postoperative delirium in multivariate modeling (P < 0.05). No differences were noted in the use of neoadjuvant chemoradiotherapy or survival. CONCLUSIONS: This study demonstrates that postoperative delirium is associated with a more complicated and costly recovery after esophagectomy and that age is independently predictive of its development. Delirium has often been thought to be the sequela of other complications; however, this study demonstrates that it presents in isolation or precedes other complications in 67.4% of cases. Focused screening will likely allow targeted preventative strategies to be used in the perioperative period to reduce complications and costs associated with delirium.

A systematic review of protocols for the three-dimensional morphologic assessment of abdominal aortic aneurysms using computed tomographic angiography.

The morphology of infrarenal abdominal aortic aneurysms (AAAs) directly influences the perioperative outcome and long-term durability of endovascular aneurysm repair. A variety of methods have been proposed for the characterization of AAA morphology using reconstructed three-dimensional (3D) computed tomography (CT) images. At present, there is lack of consensus as to which of these methods is most applicable to clinical practice or research. The purpose of this review was to evaluate existing protocols that used 3D CT images in the assessment of various aspects of AAA morphology. An electronic search was performed, from January 1996 to the end of October 2010, using the Embase and Medline databases. The literature review conformed to PRISMA statement standards. The literature search identified 604 articles, of which 31 studies met inclusion criteria. Only 15 of 31 studies objectively assessed reproducibility. Existing published protocols were insufficient to define a single evidence-based methodology for preoperative assessment of AAA morphology. Further development and expert consensus are required to establish a standardized and validated protocol to determine precisely how morphology relates to outcomes after endovascular aneurysm repair.

Outcomes assessment of the surgical management of esophageal cancer in younger and older patients.

BACKGROUND: The aim of this study was to assess the influence of age on disease presentation, clinical and pathologic staging, postoperative outcomes, costs, and long-term survival after esophagectomy for esophageal malignancy. METHODS: All patients undergoing esophagectomy for cancer between 1991 and 2011 were prospectively enrolled in an Institutional Review Board approved database. RESULTS: A total of 493 patients underwent surgical resection during the study period; 58 (11.76%) of these patients were 50 years or less (44 ± 4.7) and 435 patients were greater than 50 years (67 ± 8.44). There was no difference in clinical stage; however, patients 50 years or less were more likely to have adenocarcinoma and reduced Charlson comorbidity index and younger patients tended to have a more delayed presentation as manifested by an increased period of dysphagia and a greater degree of weight loss. In the 50 or less age group there was a significantly greater use of neoadjuvant therapy in stage II patients and the use of neoadjuvant chemotherapy significantly decreased with increasing age. Surgery in the 50 or less age group was associated with significantly reduced intensive care unit stay, incidence of postoperative complications, and overall costs. Multivariate analysis also confirmed associations between increasing age and increased incidence of postoperative complications and cost. There were no significant differences in pathologic stage, positive resection margins, incidence of complete response to neoadjuvant therapy, or in overall survival. CONCLUSIONS: This study demonstrates younger patients have fewer complications and lower overall treatment costs after esophagectomy. In spite of having a more delayed presentation, younger patients presented with a similar stage and demonstrated similar overall survival.

The Glasgow Aneurysm Score does not predict mortality after open abdominal aortic aneurysm in the era of endovascular aneurysm repair.

OBJECTIVE: Endovascular aneurysm repair (EVAR) has reduced early adverse outcomes from abdominal aortic aneurysm (AAA) repair. Preferential use of EVAR may have altered the profile of patients who undergo open repair. The validity of scoring systems such as the Glasgow Aneurysm Score (GAS), devised when open surgery was the only treatment, required reappraisal. METHODS: Patients were identified from a database of patients undergoing elective infrarenal aneurysm repair at seven United Kingdom centers, and the GAS was calculated for each patient. Discrimination and calibration were calculated to determine the performance of the model in this setting using the C statistic, tertile analysis, and the χ(2) test. Univariate analysis was performed to determine if a new iteration of the GAS could be produced. RESULTS: We identified 330 patients who met the inclusion criteria. There were 18 deaths ≤30 days of surgery (5.4%). The average (standard deviation) GAS was 78.6 (8.8) for the survivors and 81.9 (10.4) for nonsurvivors (P = .122). The C statistic was 0.625 (95% confidence interval, 0.481-0.769; P = .75) suggesting a discriminatory ability not much better than chance alone. Despite this, calibration of the model was good. There was no significant difference in the comorbidities of either group, so no recalibration of the GAS could be performed. CONCLUSION: The GAS did not discriminate between survivors and nonsurvivors after open AAA repair in this cohort. In the era of EVAR, it is possible that the GAS does not predict the outcome of open AAA repair. An alternative explanation is that patients with risk factors for poor outcomes from EVAR, such as adverse AAA morphology, are being selected out for open repair.

Continuous wound infusion of local anaesthetic agents following colorectal surgery: systematic review and meta-analysis.

AIM: To provide a specific review and meta-analysis of the available evidence for continuous wound infusion of local anaesthetic agents following midline laparotomy for major colorectal surgery. METHODS: Medline, Embase, trial registries, conference proceedings and article reference lists were searched to identify randomised, controlled trials of continuous wound infusion of local anaesthetic agents following colorectal surgery. The primary outcomes were opioid consumption, pain visual analogue scores (VASs), return to bowel function and length of hospital stay. Weighted mean difference were calculated for continuous outcomes. RESULTS: Five trials containing 542 laparotomy wounds were eligible for inclusion. There was a significant decrease in post-operative pain VAS at rest on day 3 (weighted mean difference: -0.43; 95% CI: -0.81 to -0.04; P = 0.03) but not on post-operative day 1 and 2. Local anaesthetic infusion was associated with a significant reduction in pain VAS on movement on all three post-operative days (day 1 weighted mean difference: -1.14; 95% CI: -2.24 to -0.041; P = 0.04, day 2 weighted mean difference: -0.97, 95% CI: -1.91 to -0.029; P = 0.04, day 3 weighted mean difference: -0.61; 95% CI: 1.01 to -0.20; P = 0.0038). Local anaesthetic wound infusion was associated with a significant decrease in total opioid consumption (weighted mean difference: -40.13; 95% CI: -76.74 to -3.53; P = 0.03). There was no significant decrease in length of stay (weighted mean difference: -20.87; 95% CI: -46.96 to 5.21; P = 0.12) or return of bowel function (weighted mean difference: -9.40; 95% CI: -33.98 to 15.17; P = 0.45). CONCLUSION: The results of this systematic review and meta-analysis suggest that local anaesthetic wound infusion following laparotomy for major colorectal surgery is a promising technique but do not provide conclusive evidence of benefit. Further research is required including cost-effectiveness analysis.

Efficacy of closed suction drainage in lower limb arterial surgery: a meta-analysis of published clinical trials.

BACKGROUND: Suction drains are widely used in vascular surgery, despite the absence of specific evidence that they confer benefit to patients. There has been no systematic review of the available evidence, though drainage has been shown to confer no benefit, or indeed harm, across a variety of surgical disciplines. Accordingly, a systematic review and meta-analysis of the current evidence base for closed suction drainage following surgical lower limb revascularization was undertaken. METHODS: Medline, Embase, trial registries, conference proceedings, and article reference lists were searched to identify randomized controlled trials of the use of surgical drains. The primary outcomes were wound infection, seroma/lymphocele formation, and hematoma formation. Pooled odds ratios were calculated for categorical outcomes. RESULTS: Four trials containing 429 groin wounds were eligible for inclusion. There was no significant effect on wound infection, seroma/lymphocele formation, or hematoma formation. CONCLUSION: Our meta-analysis suggests that no benefit is conferred by wound drainage following lower limb revascularization. The practice incurs avoidable expense and should not be routinely used.