scholarly journals An Objective Search for Unrecognized Bias in Validated COVID-19 Prediction Models

2021 ◽  
Author(s):  
Hossein Estiri ◽  
Zachary Strasser ◽  
Sina Rashidian ◽  
Jeffrey Klann ◽  
Kavishwar Wagholikar ◽  
...  
Keyword(s):  
Predictive Models ◽  
Medical Records ◽  
Prediction Models ◽  
Evaluation Studies ◽  
Protected Group ◽  
Increased Risk ◽  
Using Data ◽  
Bias Evaluation ◽  
Algorithmic Bias

The growing recognition of algorithmic bias has spurred discussions about fairness in artificial intelligence (AI) / machine learning (ML) algorithms. The increasing translation of predictive models into clinical practice brings an increased risk of direct harm from algorithmic bias; however, bias remains incompletely measured in many medical AI applications. Using data from over 56 thousand Mass General Brigham (MGB) patients with confirmed severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), we evaluate unrecognized bias in four AI models developed during the early months of the pandemic in Boston, Massachusetts that predict risks of hospital admission, ICU admission, mechanical ventilation, and death after a SARS-CoV-2 infection purely based on their pre-infection longitudinal medical records. We discuss that while a model can be biased against certain protected groups (i.e., perform worse) in certain tasks, it can be at the same time biased towards another protected group (i.e., perform better). As such, current bias evaluation studies may lack a full depiction of the variable effects of a model on its subpopulations. If the goal is to make a change in a positive way, the underlying roots of bias need to be fully explored in medical AI. Only a holistic evaluation, a diligent search for unrecognized bias, can provide enough information for an unbiased judgment of AI bias that can invigorate follow-up investigations on identifying the underlying roots of bias and ultimately make a change.

scholarly journals Hemodialysis services: are public policies turned to guaranteeing the access?

2015 ◽  
Vol 31 (7) ◽  
pp. 1505-1516 ◽  
Author(s):  
Ana Rita Barbieri ◽  
Crhistinne Cavalheiro Maymone Gonçalves ◽  
Maria de Fátima Meinberg Cheade ◽  
Cristina Souza ◽  
Daniel Henrique Tsuha ◽  
...  
Keyword(s):  
Medical Records ◽  
Health Informatics ◽  
Ecological Study ◽  
Public Policies ◽  
National Average ◽  
Mato Grosso ◽  
Number Of Patients ◽  
Using Data ◽  
Mato Grosso Do Sul

The increasing incidence of chronic renal failure in Brazil and the consequential expansion of hemodialysis as a choice for treatment in final stage have to be taken into account to guarantee access to those in need. The ecological study conducted in Mato Grosso do Sul State, Brazil, in 2012, using data from the Brazilian Health Informatics Department (DATASUS) and from the analysis of medical records in 12 clinics, identified and mapped patients on hemodialysis, the distance they travelled and the estimated number of patients. The prevalence of hemodialysis patients in Mato Grosso do Sul State, about 55 per 100,000 inhabitants, is similar to the national average. The analyses indicated concentration of patients in counties with clinics and also geographical gaps that generate displacement of over 100km for more than 16% of patients. The results point to the necessity of strengthening public policies that consider, for decision-making, the decentralization of service, the expansion of home care and the follow-up education for professionals.

scholarly journals Covid19Risk.ai: An open source repository and online calculator of prediction models for early diagnosis and prognosis of Covid-19

2021 ◽  
Author(s):  
Iva Halilaj ◽  
Avishek Chatterjee ◽  
Yvonka van Wijk ◽  
Guangyao Wu ◽  
Brice van Eeckhout ◽  
...  
Keyword(s):  
Open Source ◽  
Predictive Models ◽  
Prediction Models ◽  
Patient Specific ◽  
Diagnosis And Prognosis ◽  
Online Calculator ◽  
User Friendly ◽  
Type 3 ◽  
Proof Of Principle

AbstractObjectiveThe current pandemic has led to a proliferation of predictive models being developed to address various aspects of COVID-19 patient care. We aimed to develop an online platform that would serve as an open source repository for a curated subset of such models, and provide a simple interface for included models to allow for online calculation. This platform would support doctors during decision-making regarding diagnoses, prognoses, and follow-up of COVID-19 patients, expediting the models’ transition from research to clinical practice.MethodsIn this proof-of-principle study, we performed a literature search in PubMed and WHO database to find suitable models for implementation on our platform. All selected models were publicly available (peer reviewed publications or open source repository) and had been validated (TRIPOD type 3 or 2b). We created a method for obtaining the regression coefficients if only the nomogram was available in the original publication. All predictive models were transcribed on a practical graphical user interface using PHP 8.0.0, and published online together with supporting documentation and links to the associated articles.ResultsThe open source website https://covid19risk.ai/ currently incorporates nine models from six different research groups, evaluated on datasets from different countries. The website will continue to be populated with other models related to COVID-19 prediction as these become available. This dynamic platform allows COVID-19 researchers to contact us to have their model curated and included on our website, thereby increasing the reach and real-world impact of their work.ConclusionWe have successfully demonstrated in this proof-of-principle study that our website provides an inclusive platform for predictive models related to COVID-19. It enables doctors to supplement their judgment with patient-specific predictions from externally-validated models in a user-friendly format. Additionally, this platform supports researchers in showcasing their work, which will increase the visibility and use of their models.

Tear Rates of the Ipsilateral ACL Graft and the Contralateral Native ACL Are Similar following ACL Reconstruction

2020 ◽  
Author(s):  
Nadia Nastassia Ifran ◽  
Ying Ren Mok ◽  
Lingaraj Krishna
Keyword(s):  
Acl Reconstruction ◽  
Medical Records ◽  
Cruciate Ligament ◽  
Acl Graft ◽  
Increased Risk ◽  
Significant Difference ◽  
The Mean ◽  
Anterior Cruciate ◽  
Mean Time

AbstractThe aim of the study is to compare the tear rates of ipsilateral anterior cruciate ligament (ACL) grafts and the contralateral native ACL as well as to investigate the correlation of gender, age at time of surgery, and body mass index (BMI) with the occurrence of these injuries. The medical records of 751 patients who underwent ACL reconstruction surgery with follow-up periods of 2 to 7 years were retrospectively analyzed. Survival analyses of ipsilateral ACL grafts and contralateral native ACL were performed. Univariate and multivariate logistic regression analyses were performed to identify risk factors that were associated with these injuries. The tear rates of the ipsilateral ACL graft and contralateral ACL were 5.86 and 6.66%, respectively with no significant difference between groups (p = 0.998). The mean time of tears of the ipsilateral ACL and contralateral ACL was also similar (p = 0.977) at 2.64 and 2.78 years, respectively after surgery. Both the odds of sustaining an ipsilateral ACL graft and contralateral ACL tear were also significantly decreased by 0.10 (p = 0.003) and 0.14 (p = 0.000), respectively, for every 1-year increase in age at which the reconstruction was performed. However, graft type, gender, and BMI were not associated with an increased risk of these injuries. There was no difference between tear rates of ipsilateral ACL graft and contralateral ACL following ACL reconstruction. Patients who undergo ACL reconstruction at a young age are at an increased risk of both ipsilateral graft and contralateral ACL rupture after an ACL reconstruction. Patients who are young and more likely to return to competitive sports should be counselled of the risks and advised to not neglect the rehabilitation of the contralateral knee during the immediate and back to sports period of recovery. This is a Level III, retrospective cohort study.

Cardiac Outcomes in a Prospective Cohort of Adult Non-Hodgkin Lymphoma Survivors

Blood ◽  
2011 ◽  
Vol 118 (21) ◽  
pp. 2656-2656
Author(s):  
Carrie A. Thompson ◽  
Hongxiu Luo ◽  
Matthew J Maurer ◽  
Cristine Allmer ◽  
Thomas M. Habermann ◽  
...  
Keyword(s):  
Heart Disease ◽  
Hodgkin Lymphoma ◽  
Valvular Heart Disease ◽  
Mayo Clinic ◽  
Cumulative Incidence ◽  
Medical Records ◽  
P Value ◽  
Non Hodgkin Lymphoma ◽  
Increased Risk

Abstract Abstract 2656 Background Treatment of non-Hodgkin lymphoma (NHL) can lead to development of cardiovascular disease (CVD). We sought to describe the cumulative incidence of CVD in adult NHL survivors diagnosed in the recent treatment era (since 2002) and identify clinical and treatment predictors for its development. Methods All patients were from the Mayo component of the Molecular Epidemiology Resource (MER) of the University of Iowa/Mayo Clinic Lymphoma Specialized Program of Research Excellence (SPORE). The MER offers enrollment to all consecutive patients with newly diagnosed NHL who are US residents and age >18 years. Clinical data from the time of diagnosis and treatment data are abstracted from medical records using a standard protocol. Patients are prospectively contacted via telephone or in person per protocol every 6 months for the first 3 years from diagnosis and yearly afterwards to assess disease status and development of comorbid conditions. CVD events, including heart failure (HF), myocardial infarction (MI), arrhythmia, pericarditis, and valvular heart disease, occurring after diagnosis were identified during follow-up and validated against medical records. HF was validated with the Cardiovascular Health Study Criteria and/or the Framingham Criteria. MI was validated using case definition standards of coronary heart disease, while arrhythmia, pericarditis, and valvular heart disease were validated using clinical definitions. The prevalence of CVD and associations between CVD and clinical characteristics (sex, age) and treatment (radiation, anthracyclines) were performed using Cox models with a competing risk approach. Results 1164 patients with NHL were enrolled into the MER at Mayo Clinic between 9/1/2002–2/28/2008. 646 were male (56%) and median age at diagnosis was 62 years (range 20–93). Median follow-up of all cases was 59 months (range 1–105). 131 patients reported CVD prior to the diagnosis of NHL and were excluded from analyses. An additional 76 patients did not have follow-up and were excluded. Of the 957 remaining patients, 75 (7.8%) self-reported a new diagnosis of CVD. Of these, 71 cases had available medical records. 57 of the 71 reviewed cases (80%) were validated (18 HF, 9 MI, 21 arrhythmia, 2 pericarditis, and 10 valvular heart disease). Cumulative incidence of CVD at 1, 3, 5, and 7 years was 1.3%, 3.7%, 5.2%, and 7.4%, respectively. Median time from NHL diagnosis to CVD was 26.5 months (range 1–84). Older age was associated with increased risk of overall CVD (p-value<0.001). Gender (p=0.59), radiation therapy (p=0.61), and anthracycline treatment (p=0.25) were not associated with the incidence of overall CVD. Among types of CVD, anthracycline use was associated with development of HF (HR=5.30; p-value=0.008) and arrhythmia (HR=2.68; p-value=0.04). Radiation was associated with development of arrhythmia (HR=2.73; p-value=0.03), while older age was associated with development of HF (HR=1.36 per 5 year increment; p-value=0.003) and arrhythmia (HR=1.25 per 5 year increment; p-value=0.02). Conclusions The risk of CVD in patients with NHL is approximately 1% per year after the initial diagnosis of lymphoma. The most commonly occurring CVDs in this cohort of NHL survivors were arrhythmia and HF. Treatment with anthracyclines and radiation are associated with increased risk of developing some types of CVD. 80% of self-reported CVD events in NHL survivors were validated using epidemiologic criteria. Future studies will include building models incorporating comorbid health conditions and lifestyle factors to determine risk of CVD as well as the impact of CVD on quality of life. Disclosures: No relevant conflicts of interest to declare.

scholarly journals Smoking as an independent predictor of reoperation after lumbar laminectomy: a study of 500 cases

2015 ◽  
Vol 22 (3) ◽  
pp. 288-293 ◽  
Author(s):  
Mohamad Bydon ◽  
Mohamed Macki ◽  
Rafael De la Garza-Ramos ◽  
Daniel M. Sciubba ◽  
Jean-Paul Wolinsky ◽  
...  
Keyword(s):  
Medical Records ◽  
Independent Predictor ◽  
Multiple Logistic Regression Analysis ◽  
Wound Dehiscence ◽  
Chronic Obstructive ◽  
Obstructive Pulmonary Disease ◽  
Single Level ◽  
Increased Risk ◽  
Spinal Degeneration

OBJECT This study aimed to identify the factors predicting an increased risk for reoperation in patients who had undergone a lumbar laminectomy. METHODS The authors retrospectively reviewed the electronic medical records of all patients who had undergone firsttime, bilateral laminectomy at 1, 2, or 3 levels for lumbar spondylosis at the authors' institution. Patients who underwent fusion, laminotomy, discectomy, or complete facetectomy were excluded. The patients' preoperative symptoms and comorbidities were also obtained from their medical records. RESULTS Over an average follow-up period of 46.8 months, of 500 patients who had undergone laminectomy at 1, 2, or 3 levels, 81 patients (16.2%) developed subsequent spinal disorders that required a reoperation. A multiple logistic regression analysis identified smoking as an independent predictor of reoperation (OR 2.15, p = 0.01). Smoking was also an independent predictor of reoperation after a single-level laminectomy (OR 11.3, p = 0.02) and after a multilevel (that is, involving 2 or 3 levels) laminectomy (OR 1.98, p = 0.05). For 72 patients undergoing reoperation only for spinal degeneration, smoking remained an independent, statistically significant predictor of reoperation (OR 2.06, p = 0.04). Nine patients underwent reoperation for nondegenerative conditions (hematoma, wound infection, or wound dehiscence), and in these patients, chronic obstructive pulmonary disease was the only statistically significant predictor of reoperation (OR 8.92, p = 0.03). CONCLUSIONS Smoking was the strongest predictor of reoperation in patients who had undergone single-level laminectomy, multilevel laminectomy, or reoperation for progression of spinal degeneration. These findings suggest that smokers have worse outcomes of lumbar decompression than nonsmokers.

scholarly journals Risk factors for systemic thromboembolism after left chamber cardiac thrombi --- A retrospective multi-center study

2019 ◽  
Author(s):  
wei zhou ◽  
Shun-yi Shi ◽  
Yuan Ji ◽  
Xin Chen ◽  
Jun Huang ◽  
...  
Keyword(s):  
Medical Records ◽  
Heart Function ◽  
Multivariate Logistic Regression ◽  
Increased Risk ◽  
All Cause Mortality ◽  
History Of ◽  
Multi Center Study ◽  
Relationship Of ◽  
The Relationship

Abstract Background : We aimed to characterize the independent predictors of systemic thromboembolism (ST) after left chamber thrombi. Methods: A retrospective analysis on the medical records of 175 patients diagnosed with left chamber thrombi by transthoracic echocardiography (TTE) at three centers were carried out. Multivariate logistic regression was performed to determine the relationship of each characteristic with ST. Multivariate Cox proportional survival analysis was conducted, with covariate adjustments, to identify predictors of all-cause mortality. Results: During a median 42 months of follow-up (25th–75th percentile: 20–62 months), 24 (13.7%) patients had ST, and 62 (35.4%) died. History of diabetes and thrombus mobility were independent predictors of ST (P = 0.003, P = 0.02, respectively). There was a significant association between abnormal ejection fraction (EF) and all-cause mortality (P = 0.003). Conclusions: The morbidity associated with ST and the increased risk for mortality associated with left chamber cardiac thrombi relates to medical history, thrombus state, and diminished heart function.

scholarly journals SARS-CoV-2 infections in people with PCD: neither frequent, nor particularly severe

2020 ◽  
Author(s):  
Eva SL Pedersen ◽  
Myrofora Goutaki ◽  
Amanda L Harris ◽  
Lucy Dixon ◽  
Michele Manion ◽  
...  
Keyword(s):  
Incidence Rate ◽  
Severe Symptom ◽  
Demographic Data ◽  
Severe Disease ◽  
Median Number ◽  
Disease Course ◽  
Increased Risk ◽  
Using Data ◽  
Ciliary Dyskinesia

AbstractPeople with pre-existing chronic health conditions are reportedly at high risk of getting the coronavirus disease (COVID-19) and of having a severe disease course but little data exist on rare diseases such as Primary Ciliary Dyskinesia (PCD). We studied risk and severity of SARS-CoV-2 infections among people with PCD using data from the COVID-PCD, a participatory study that collects data in real-time directly from people with PCD. Data was collected using online questionnaires. A baseline questionnaire collected information on demographic data, information about the PCD diagnosis and severity. A short weekly questionnaire collected information about current symptoms and incident SARS-CoV-2 infections. 578 people participated in the COVID-PCD by December 7, 2020, with a median number of follow-up weeks of 9 (interquartile range: 4-19 weeks). 256 (45%) of the participants had been tested for SARS-CoV-2 and 12 tested positive prior to study entry or during study follow up (2.1% of the total included population, 95% confidence interval (CI) 1.1-3.6%). 4 people tested positive during the study follow-up, corresponding to an incidence rate of 2.5 per 100 person-years (95% CI: 0.9-6.5). Overall, reported severity was mild with two reporting no symptoms, eight reporting mild symptoms, one reporting severe symptom without hospitalisation, and one reporting hospitalisation for 9 days. The study suggests that with careful personal protection, people with PCD do not seem to have an increased risk of infection with SARS-COV-2, nor an especially severe disease course.Take home messageIn this longitudinal study of people with PCD followed weekly via online questionnaires, the incidence rate of COVID-19 and the proportion of participants infected were low, and the observed severity mostly mild.

scholarly journals 751Investigation of the obesity paradox in kidney cancer: mystifying association or myth?

2021 ◽  
Vol 50 (Supplement_1) ◽  
Author(s):  
Alicia Heath ◽  
Joanna Clasen ◽  
Elio Riboli ◽  
Ghislaine Scelo ◽  
David Muller
Keyword(s):  
Kidney Cancer ◽  
Cox Regression ◽  
Obesity Paradox ◽  
Hazard Ratios ◽  
Increased Risk ◽  
Incidence And Mortality ◽  
All Cause Mortality ◽  
Using Data ◽  
Restricted Cubic Splines

Abstract Background An “obesity paradox” has been reported in kidney cancer, whereby obesity is a risk factor, yet appears to be associated with better survival. To evaluate this paradox, we investigated the association between pre-diagnostic adiposity and renal cell carcinoma (RCC) incidence and mortality. Methods Using data from 363,521 men and women in the European Prospective Investigation into Cancer and Nutrition (EPIC), Cox regression models yielded confounder-adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for RCC incidence and mortality in relation to BMI modelled continuously and using restricted cubic splines. RCC-specific and all-cause mortality were evaluated among cases. Results During a mean follow-up of 14.9 years, 936 incident RCC cases were identified, 383 of whom died (278 due to RCC). Each 5 kg/m2 increment in BMI was associated with 27% and 46% higher RCC incidence and mortality (HRs=1.27, 95% CI 1.18-1.37 and 1.46, 95% CI 1.28-1.66, respectively). Comparing a BMI of 35 with 22 kg/m2, HRs for RCC incidence and mortality were 1.88 (95% CI 1.54-2.30) and 2.37 (95% CI 1.68-3.35), respectively. Among RCC cases, HRs per 5 kg/m2 increment in BMI were 1.22 (95% CI 1.07-1.41) for RCC-specific mortality and 1.18 (95% CI 1.04-1.34) for all-cause mortality. Similar, positive linear associations were evident for waist circumference and waist-to-hip ratio. Conclusions Obesity was associated with increased RCC incidence and mortality, and worse prognosis among cases. Key messages The kidney cancer-obesity paradox does not appear to be real. Higher adiposity is associated with an increased risk of incident and fatal RCC.

P3771Validation of risk scoring system predicting for progression of atrial fibrillation: the Fushimi AF Registry

2019 ◽  
Vol 40 (Supplement_1) ◽  
Author(s):  
H Ogawa ◽  
Y An ◽  
K Ishigami ◽  
Y Aono ◽  
S Ikeda ◽  
...  
Keyword(s):  
Risk Factors ◽  
Atrial Fibrillation ◽  
External Validation ◽  
Scoring Systems ◽  
Japanese Patients ◽  
Urban District ◽  
Increased Risk ◽  
Risk Scoring ◽  
Using Data

Abstract Background Atrial fibrillation (AF) increases the risks of thromboembolism and death. Progression from paroxysmal to sustained types (persistent or permanent) of AF is sometimes seen in clinical practice. We recently reported that progression of AF was associated with increased risk of clinical adverse events in Japanese AF patients. However, risk stratification schemes of predicting the progression of AF has not been fully established. Methods The Fushimi AF Registry, a community-based prospective survey, was designed to enroll all of the AF patients in Fushimi-ku, Kyoto, which is a typical urban district of Japan with a population of 283,000. Follow-up data were available for 4,454 patients. We investigated the risk factors of AF progression and validated the performance of various risk scoring systems predicting for progression of AF, such as APPLE, BASE-AF2, HATCH, and MB-LATER score, using data from 995 paroxysmal AF patients (mean age; 72.6±11.4 years, female; 42.2%, mean CHA2DS2-VASc score; 3.26±1.67) whose echocardiogram data were obtained at baseline. Results Of 995 AF patients, during the median follow-up of 1,477 days, progression from paroxysmal to sustained AF occurred in 160 patients (16.1%; 4.0 per 100 person-years). On a multivariate model, we indicated that history of AF ≥2 years (odds ratio [OR] 1.83; 95% confidence interval [CI] 1.28–2.61), left atrial diameter ≥40 mm (OR 1.45; 95% CI 1.02–2.08), daily drinker (OR 1.56; 95% CI 1.24–2.81), and cardiomyopathy (OR 2.58; 95% CI 1.17–5.69) were significantly associated with higher incidence of AF progression. Our model had better predictive potential for AF progression (area under curve [AUC] 0.612; 95% CI 0.566–0.658) than the APPLE (AUC 0.553; 95% CI 0.508–0.598; p=0.06), BASE-AF2 (AUC 0.571; 95% CI 0.526–0.617; p=0.04), CHADS2 (AUC 0.508; 95% CI 0.462–0.554; p<0.01), CHA2DS2-VASc (AUC 0.501; 95% CI 0.453–0.548; p<0.01), HATCH (AUC 0.502; 95% CI 0.456–0.548; p<0.01), and MB-LATER (AUC 0.528; 95% CI 0.483–0.572; p<0.01) score. Conclusion We identified 4 risk factors which may be useful to predict for progression of AF in Japanese patients. External validation of our model in other cohorts is needed. Acknowledgement/Funding Boehringer, Bayer, Pfizer, Bristol-Myers, Astellas, AstraZeneca, Daiichi Sankyo, Novartis, MSD, Sanofi and Takeda. Japan Agency for Medical Research

scholarly journals Prevalence of Coronavirus Disease 2019 Infection and Outcomes Among Symptomatic Healthcare Workers in Seattle, Washington

2020 ◽  
Vol 71 (10) ◽  
pp. 2702-2707 ◽  
Author(s):  
Nandita S Mani ◽  
Jehan Z Budak ◽  
Kristine F Lan ◽  
Chloe Bryson-Cahn ◽  
Allison Zelikoff ◽  
...  
Keyword(s):  
United States ◽  
Healthcare Workers ◽  
Clinical Characteristics ◽  
The United States ◽  
Limited Data ◽  
Healthcare Workforce ◽  
Increased Risk ◽  
Using Data ◽  
The University

Abstract Background Healthcare workers (HCWs) who serve on the front lines of the coronavirus disease 2019 (COVID-19) pandemic have been at increased risk for infection due to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) in some settings. Healthcare-acquired infection has been reported in similar epidemics, but there are limited data on the prevalence of COVID-19 among HCWs and their associated clinical outcomes in the United States. Methods We established 2 high-throughput employee testing centers in Seattle, Washington, with drive-through and walk-through options for symptomatic employees in the University of Washington Medicine system and its affiliated organizations. Using data from these testing centers, we report the prevalence of SARS-CoV-2 infection among symptomatic employees and describe the clinical characteristics and outcomes among employees with COVID-19. Results Between 12 March 2020 and 23 April 2020, 3477 symptomatic employees were tested for COVID-19 at 2 employee testing centers; 185 (5.3%) employees tested positive for COVID-19. The prevalence of SARS-CoV-2 was similar when comparing frontline HCWs (5.2%) with nonfrontline staff (5.5%). Among 174 positive employees reached for follow-up at least 14 days after diagnosis, 6 reported COVID-related hospitalization; all recovered. Conclusions During the study period, we observed that the prevalence of positive SARS-CoV-2 tests among symptomatic HCWs was comparable to that of symptomatic nonfrontline staff. Reliable and rapid access to testing for employees is essential to preserve the health, safety, and availability of the healthcare workforce during this pandemic and to facilitate the rapid return of SARS-CoV-2–negative employees to work.

Sign in / Sign up

Export Citation Format

Share Document