Doubly Robust Crowdsourcing

Large-scale labeled dataset is the indispensable fuel that ignites the AI revolution as we see today. Most such datasets are constructed using crowdsourcing services such as Amazon Mechanical Turk which provides noisy labels from non-experts at a fair price. The sheer size of such datasets mandates that it is only feasible to collect a few labels per data point. We formulate the problem of test-time label aggregation as a statistical estimation problem of inferring the expected voting score. By imitating workers with supervised learners and using them in a doubly robust estimation framework, we prove that the variance of estimation can be substantially reduced, even if the learner is a poor approximation. Synthetic and real-world experiments show that by combining the doubly robust approach with adaptive worker/item selection rules, we often need much lower label cost to achieve nearly the same accuracy as in the ideal world where all workers label all data points.

Influence of the trajectory of the urine output for 24 h on the occurrence of AKI in patients with sepsis in intensive care unit

Background Sepsis-associated acute kidney injury (S-AKI) is a common and life-threatening complication in hospitalized and critically ill patients. This condition is an independent cause of death. This study was performed to investigate the correlation between the trajectory of urine output within 24 h and S-AKI. Methods Patients with sepsis were studied retrospectively based on the Medical Information Mart for Intensive Care IV. Latent growth mixture modeling was used to classify the trajectory of urine output changes within 24 h of sepsis diagnosis. The outcome of this study is AKI that occurs 24 h after sepsis. Cox proportional hazard model, Fine–Gray subdistribution proportional hazard model, and doubly robust estimation method were used to explore the risk of AKI in patients with different trajectory classes. Results A total of 9869 sepsis patients were included in this study, and their 24-h urine output trajectories were divided into five classes. The Cox proportional hazard model showed that compared with class 1, the HR (95% CI) values for classes 3, 4, and 5 were 1.460 (1.137–1.875), 1.532 (1.197–1.961), and 2.232 (1.795–2.774), respectively. Competing risk model and doubly robust estimation methods reached similar results. Conclusions The trajectory of urine output within 24 h of sepsis patients has a certain impact on the occurrence of AKI. Therefore, in the early treatment of sepsis, close attention should be paid to changes in the patient's urine output to prevent the occurrence of S-AKI.

Sorafenib Combined with Chemoembolization for Locally Advanced Hepatocellular Carcinoma with Macroscopic Vascular Invasion: A Propensity Score Analysis

The purpose of this study was to compare the efficacy and safety of transarterial chemoembolization (TACE) plus sorafenib with those of TACE alone in patients with locally advanced hepatocellular carcinoma (HCC). Treatment-naïve patients with preserved hepatic reserve (Child–Pugh score ≤ 7) who received TACE plus sorafenib (n = 91) or TACE alone (n = 109) for locally advanced HCC with macrovascular invasion were retrospectively evaluated. Propensity score matching (PSM) was used to correct selection bias, and 63 pairs were created. In the entire study population, the median progression-free survival (PFS) and overall survival (OS) with TACE plus sorafenib were better than those with TACE alone. After PSM, the median PFS (7.0 vs. 4.3 months; p = 0.017) and OS (17.5 vs. 12.8 months; p = 0.049) were again significantly longer with TACE plus sorafenib than with TACE alone. Stratified Cox regression analysis and doubly robust estimation revealed that treatment type was significantly associated with both PFS and OS. In the subgroup analysis, TACE plus sorafenib did not show a significant survival benefit for patients with main portal vein or inferior vena cava invasion. Major complications were similar in both groups (p = 0.330). In conclusion, TACE plus sorafenib showed better survival outcomes than TACE alone in patients with locally advanced HCC.

Survival benefits and optimal timing for surgical intervention for infective endocarditis

Background Surgery is often indicated in patients with infective endocarditis (IE), but the survival benefits of surgical intervention have not been validated in large-scale studies. Although previous studies appeared to support early surgical intervention, the optimal timing of intervention remains uncertain. Purpose We aim to evaluate the benefits of surgery and identify the optimal timing of surgical intervention for patients with IE. Methods From a well-validated territory-wide database in Hong Kong, all patients aged 20 or above diagnosed with incident IE from 2000–2019 were included. Patients were divided into those who received surgical intervention within 1 year of IE (surgical cohort) and those who did not (control cohort). The two cohorts were then compared using inverse probability weighting of the covariate balancing propensity score, which included demographics, comorbidities, and causative organism as covariates. Outcomes of interest include, at 1 year, all-cause death, and the development of complications. A Cox proportional hazards model was used to evaluate the association between surgical intervention and death, with “doubly-robust estimation” used to minimise the effect of confounders. For complications, a Fine-Gray model was used to account for competing risk. The surgical cohort was subdivided into early (≤7 days of hospitalisation) or late surgical intervention; a similar propensity score analytic approach was used to evaluate the effects of early vs. late intervention, with those who died within the 7 days excluded to ensure a fair comparison. Results A total of 5,657 patients (age 59.9±18.3 years, 37.2% females) were included, of which 930 (16.4%) received surgical intervention in 1 year. Overall, the surgical cohort had a 45% risk reduction in all-cause death (hazard ratio [HR] 0.55, 95% CI [0.46 to 0.65], P<0.001) (Figure). This association remained consistent in subgroup analysis stratified by age, sex, and causative organisms (Table 1). The surgical cohort also had a lower risk of complications, including acute kidney injury (HR 0.61, 95% CI 0.43 to 0.87, P=0.006), systemic embolism (HR 0.35 [0.23 to 0.55], P<0.001), ischaemic stroke (HR 0.37 [0.24 to 0.55], P<0.001), cardiac dysrhythmia (HR 0.79 [0.66 to 0.95], P=0.011), and pneumonia (HR 0.36 [0.26 to 0.49], P<0.001). In the surgical cohort, compared to those who had early surgery (N=181), those with delayed surgery had a lower risk of all-cause death (HR 0.58 [0.34 to 0.99], P=0.045) (Figure) and complications (Table 2) at 1 year. In those who had early surgery, patients who received ultra-early surgery (≤3 days of hospitalisation, N=104) did not have a significantly different risk of death (HR 1.19 [0.47 to 3.34], P=0.654). Conclusions Surgical intervention significantly reduced the risk of death and complications in patients with infective endocarditis. Delayed surgical intervention appeared to be more protective. FUNDunding Acknowledgement Type of funding sources: Public Institution(s). Main funding source(s): The Shenzhen Key Medical DisciplineThe Sanming Project of HKU-SZH Cardiology

Nonelderly Adult Cancer Survivors in High Deductible Health Plan: Healthcare Expenditure, Utilization and Access

Background: To compare healthcare expenditure, utilization and access between nonelderly adult cancer survivors enrolled in a high deductible health plan with a health savings account (“HDHP+HSA”), HDHP without HSA (“HDHP alone”) and low deductible health plan (“LDHP”). Methods: 1735 cancer survivors, aged 18–64 years, with continuous private coverage identified from the 2012–2017 Medical Expenditure Panel Survey: HDHP alone (n = 353), HDHP+HSA (n = 242) and LDHP (n = 1140). Healthcare expenditures, utilization and inability/delay obtaining medical care were analyzed using generalized linear regressions with inverse propensity score weighting and doubly robust estimation. Results: HDHP alone group (23,255 USD) had significantly higher total healthcare expenditure compared to HDHP+HSA (15,580 USD, p = 0.012) and LDHP (16,261 USD, p = 0.016). HDHP alone (6089 USD; p = 0.002) and HDHP+HSA (5743 USD; p = 0.012) groups had significantly higher out-of-pocket (OOP) expenditure compared to LDHP (4853 USD). HDHP alone (17,128 USD, p = 0.010) and LDHP (12,645 USD, p = 0.045) had significantly higher private insurer payments compared to HDHP+HSA (9216 USD). No differences were found in utilization or inability/delay obtaining medical care across groups. Conclusions: Non-elderly adult cancer survivors with continuous coverage and comparable sociodemographic characteristics enrolled in HDHP with HSA displayed the lowest healthcare costs compared to HDHP without HSA and LDHP. HDHP+HSA had a significantly higher OOP expenditure than LDHP. No significant differences were observed in utilization or access among groups.

Refusal of surgery and survival outcomes in endometrial cancer

ObjectivesThe goal of this study was to determine the impact refusal of surgery has on overall survival in patients with endometrial cancer.MethodsFrom January 2004 to December 2015, the National Cancer Database was queried for patients with pathologically proven endometrial cancer who were recommended surgery and refused. Inverse probability of treatment weighting was used to account for differences in baseline characteristics between patients who underwent surgery and those who refused. Kaplan–Meier analyses and doubly robust estimation with multivariate Cox proportional hazards modeling were used to analyze overall survival.ResultsOf the 300 675 patients identified, 534 patients (0.2%) were recommended surgical treatment but refused: 18% (95/534) were age ≤40 years. The 5-year overall survival for all patients who refused surgery was significantly decreased compared with patients who underwent surgery (29.2% vs 71.9%, P<0.01). This was demonstrated at ages 41–64 years (65.5% vs 91.0%, P<0.01) and ≥65 years (23.4% vs 75.3%, P<0.01). The 5-year overall survival did not meet statistical significance at age ≤40 years (90.1% vs 87.8% P<0.19). However, there were few patients in this cohort. On multivariate analysis, factors associated with refusal of surgery included: Medicaid insurance, Black race, Hispanic Race, Charlson Comorbidity Index scores of 2 or greater, stage II or III, and if patient received external beam radiation therapy alone. Factors associated with undergoing surgery included: age greater than 41, stage IB, and if the patient received brachytherapy.ConclusionsRefusal of surgery for endometrial cancer is uncommon and leads to decreased overall survival.

The Implications of Treatment Delays In Adjuvant Therapy For Resected Cholangiocarcinoma Patients

Purpose: To understand factors associated with timing of adjuvant therapy for cholangiocarcinoma and the impact of delays on overall survival (OS).Methods: Data from the National Cancer Database (NCDB) for patients with non-metastatic bile duct cancer from 2004 to 2015 were analyzed. Patients were included only if they underwent surgery and adjuvant chemotherapy and/or radiotherapy (RT). Patients who underwent neoadjuvant or palliative treatments were excluded. Pearson’s chi-squared test and multivariate logistic regression analyses were used to assess the distribution of demographic, clinical, and treatment factors. After propensity-score matching with inverse probability of treatment weighting, OS was compared between patients initiating therapy past various time points using Kaplan Meier analyses and doubly-robust estimation with multivariate Cox proportional hazards modeling.Results: In total, 7,733 of 17,363 (45%) patients underwent adjuvant treatment. The median time to adjuvant therapy initiation was 59 days (interquartile range 45-78 days). Age over 65, black and Hispanic race, and treatment with RT alone were associated with later initiation of adjuvant treatment. Patients with larger tumors and high grade disease were more likely to initiate treatment early. After propensity score weighting, there was an OS decrement to initiation of treatment beyond the median of 59 days after surgery. Conclusions: We identified characteristics that are related to the timing of adjuvant therapy in patients with biliary cancers. There was an OS decrement associated with delays beyond the median time point of 59 days. This finding may be especially relevant given the treatment delays seen as a result of COVID-19.

Farmland lease, high-rent threat and contract instability: evidence from China

PurposeThe purpose is to understand why contract instability occurs when small landowners lease their land to large landholders.Design/methodology/approachThe authors develop a contract theoretical model to understand the stability problem in the farmland lease contract in China, where most landowners are small landholders.FindingsResults from the doubly robust estimation method used on randomly selected interview data from 552 households in nine provinces of China indicate that contract instability can arise endogenously when large landholders sign a contract. The authors conclude that a suitable rent control regime or contract enforcement may be necessary to promote a large-scale farmland transfer in China.Originality/valueThe authors develop a contract theoretical model and apply it to the land rental market in China. Data used are original and collected from farmers located in nine provinces of China.